2005-04-16 22:20:36 +00:00
|
|
|
/************************************************************************
|
2005-09-07 04:36:56 +00:00
|
|
|
* s2io.c: A Linux PCI-X Ethernet driver for Neterion 10GbE Server NIC
|
2007-03-10 02:28:32 +00:00
|
|
|
* Copyright(c) 2002-2007 Neterion Inc.
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
* This software may be used and distributed according to the terms of
|
|
|
|
* the GNU General Public License (GPL), incorporated herein by reference.
|
|
|
|
* Drivers based on or derived from this code fall under the GPL and must
|
|
|
|
* retain the authorship, copyright and license notice. This file is not
|
|
|
|
* a complete program and may only be used when the entire operating
|
|
|
|
* system is licensed under the GPL.
|
|
|
|
* See the file COPYING in this distribution for more information.
|
|
|
|
*
|
|
|
|
* Credits:
|
2005-08-03 19:24:33 +00:00
|
|
|
* Jeff Garzik : For pointing out the improper error condition
|
|
|
|
* check in the s2io_xmit routine and also some
|
|
|
|
* issues in the Tx watch dog function. Also for
|
|
|
|
* patiently answering all those innumerable
|
2005-04-16 22:20:36 +00:00
|
|
|
* questions regaring the 2.6 porting issues.
|
|
|
|
* Stephen Hemminger : Providing proper 2.6 porting mechanism for some
|
|
|
|
* macros available only in 2.6 Kernel.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Francois Romieu : For pointing out all code part that were
|
2005-04-16 22:20:36 +00:00
|
|
|
* deprecated and also styling related comments.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Grant Grundler : For helping me get rid of some Architecture
|
2005-04-16 22:20:36 +00:00
|
|
|
* dependent code.
|
|
|
|
* Christopher Hellwig : Some more 2.6 specific issues in the driver.
|
2005-08-03 19:24:33 +00:00
|
|
|
*
|
2005-04-16 22:20:36 +00:00
|
|
|
* The module loadable parameters that are supported by the driver and a brief
|
|
|
|
* explaination of all the variables.
|
2006-04-21 23:05:41 +00:00
|
|
|
*
|
2005-08-03 19:24:33 +00:00
|
|
|
* rx_ring_num : This can be used to program the number of receive rings used
|
|
|
|
* in the driver.
|
2006-04-21 23:05:41 +00:00
|
|
|
* rx_ring_sz: This defines the number of receive blocks each ring can have.
|
|
|
|
* This is also an array of size 8.
|
2005-10-31 21:55:31 +00:00
|
|
|
* rx_ring_mode: This defines the operation mode of all 8 rings. The valid
|
2007-07-23 06:20:51 +00:00
|
|
|
* values are 1, 2.
|
2005-04-16 22:20:36 +00:00
|
|
|
* tx_fifo_num: This defines the number of Tx FIFOs thats used int the driver.
|
2005-08-03 19:24:33 +00:00
|
|
|
* tx_fifo_len: This too is an array of 8. Each element defines the number of
|
2005-04-16 22:20:36 +00:00
|
|
|
* Tx descriptors that can be associated with each corresponding FIFO.
|
2006-04-21 23:05:41 +00:00
|
|
|
* intr_type: This defines the type of interrupt. The values can be 0(INTA),
|
2007-09-15 20:11:34 +00:00
|
|
|
* 2(MSI_X). Default value is '2(MSI_X)'
|
2007-10-05 19:39:21 +00:00
|
|
|
* lro_enable: Specifies whether to enable Large Receive Offload (LRO) or not.
|
2006-04-21 23:05:41 +00:00
|
|
|
* Possible values '1' for enable '0' for disable. Default is '0'
|
|
|
|
* lro_max_pkts: This parameter defines maximum number of packets can be
|
|
|
|
* aggregated as a single large packet
|
2007-02-24 06:59:39 +00:00
|
|
|
* napi: This parameter used to enable/disable NAPI (polling Rx)
|
|
|
|
* Possible values '1' for enable and '0' for disable. Default is '1'
|
|
|
|
* ufo: This parameter used to enable/disable UDP Fragmentation Offload(UFO)
|
|
|
|
* Possible values '1' for enable and '0' for disable. Default is '0'
|
|
|
|
* vlan_tag_strip: This can be used to enable or disable vlan stripping.
|
|
|
|
* Possible values '1' for enable , '0' for disable.
|
|
|
|
* Default is '2' - which means disable in promisc mode
|
|
|
|
* and enable in non-promiscuous mode.
|
2008-02-20 21:44:07 +00:00
|
|
|
* multiq: This parameter used to enable/disable MULTIQUEUE support.
|
|
|
|
* Possible values '1' for enable and '0' for disable. Default is '0'
|
2005-04-16 22:20:36 +00:00
|
|
|
************************************************************************/
|
|
|
|
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/types.h>
|
|
|
|
#include <linux/errno.h>
|
|
|
|
#include <linux/ioport.h>
|
|
|
|
#include <linux/pci.h>
|
2005-06-26 22:22:14 +00:00
|
|
|
#include <linux/dma-mapping.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/netdevice.h>
|
|
|
|
#include <linux/etherdevice.h>
|
|
|
|
#include <linux/skbuff.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/stddef.h>
|
|
|
|
#include <linux/ioctl.h>
|
|
|
|
#include <linux/timex.h>
|
|
|
|
#include <linux/ethtool.h>
|
|
|
|
#include <linux/workqueue.h>
|
2005-08-03 19:35:55 +00:00
|
|
|
#include <linux/if_vlan.h>
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
#include <linux/ip.h>
|
|
|
|
#include <linux/tcp.h>
|
|
|
|
#include <net/tcp.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
#include <asm/system.h>
|
|
|
|
#include <asm/uaccess.h>
|
2005-08-03 19:24:33 +00:00
|
|
|
#include <asm/io.h>
|
2006-02-03 09:45:12 +00:00
|
|
|
#include <asm/div64.h>
|
2006-08-15 06:00:14 +00:00
|
|
|
#include <asm/irq.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* local include */
|
|
|
|
#include "s2io.h"
|
|
|
|
#include "s2io-regs.h"
|
|
|
|
|
2008-07-10 03:50:13 +00:00
|
|
|
#define DRV_VERSION "2.0.26.25"
|
2005-10-04 11:51:45 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* S2io Driver name & version. */
|
2005-08-03 19:24:33 +00:00
|
|
|
static char s2io_driver_name[] = "Neterion";
|
2005-10-04 11:51:45 +00:00
|
|
|
static char s2io_driver_version[] = DRV_VERSION;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-07-23 06:20:51 +00:00
|
|
|
static int rxd_size[2] = {32,48};
|
|
|
|
static int rxd_count[2] = {127,85};
|
2005-10-31 21:55:31 +00:00
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static inline int RXD_IS_UP2DT(struct RxD_t *rxdp)
|
2005-08-03 19:27:09 +00:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = ((!(rxdp->Control_1 & RXD_OWN_XENA)) &&
|
|
|
|
(GET_RXD_MARKER(rxdp->Control_2) != THE_RXD_MARK));
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Cards with following subsystem_id have a link state indication
|
|
|
|
* problem, 600B, 600C, 600D, 640B, 640C and 640D.
|
|
|
|
* macro below identifies these cards given the subsystem_id.
|
|
|
|
*/
|
2005-08-03 19:36:55 +00:00
|
|
|
#define CARDS_WITH_FAULTY_LINK_INDICATORS(dev_type, subid) \
|
|
|
|
(dev_type == XFRAME_I_DEVICE) ? \
|
|
|
|
((((subid >= 0x600B) && (subid <= 0x600D)) || \
|
|
|
|
((subid >= 0x640B) && (subid <= 0x640D))) ? 1 : 0) : 0
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
#define LINK_IS_UP(val64) (!(val64 & (ADAPTER_STATUS_RMAC_REMOTE_FAULT | \
|
|
|
|
ADAPTER_STATUS_RMAC_LOCAL_FAULT)))
|
|
|
|
|
2007-09-06 10:51:14 +00:00
|
|
|
static inline int is_s2io_card_up(const struct s2io_nic * sp)
|
|
|
|
{
|
|
|
|
return test_bit(__S2IO_STATE_CARD_UP, &sp->state);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Ethtool related variables and Macros. */
|
|
|
|
static char s2io_gstrings[][ETH_GSTRING_LEN] = {
|
|
|
|
"Register test\t(offline)",
|
|
|
|
"Eeprom test\t(offline)",
|
|
|
|
"Link test\t(online)",
|
|
|
|
"RLDRAM test\t(offline)",
|
|
|
|
"BIST Test\t(offline)"
|
|
|
|
};
|
|
|
|
|
2007-02-24 07:03:22 +00:00
|
|
|
static char ethtool_xena_stats_keys[][ETH_GSTRING_LEN] = {
|
2005-04-16 22:20:36 +00:00
|
|
|
{"tmac_frms"},
|
|
|
|
{"tmac_data_octets"},
|
|
|
|
{"tmac_drop_frms"},
|
|
|
|
{"tmac_mcst_frms"},
|
|
|
|
{"tmac_bcst_frms"},
|
|
|
|
{"tmac_pause_ctrl_frms"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"tmac_ttl_octets"},
|
|
|
|
{"tmac_ucst_frms"},
|
|
|
|
{"tmac_nucst_frms"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"tmac_any_err_frms"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"tmac_ttl_less_fb_octets"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"tmac_vld_ip_octets"},
|
|
|
|
{"tmac_vld_ip"},
|
|
|
|
{"tmac_drop_ip"},
|
|
|
|
{"tmac_icmp"},
|
|
|
|
{"tmac_rst_tcp"},
|
|
|
|
{"tmac_tcp"},
|
|
|
|
{"tmac_udp"},
|
|
|
|
{"rmac_vld_frms"},
|
|
|
|
{"rmac_data_octets"},
|
|
|
|
{"rmac_fcs_err_frms"},
|
|
|
|
{"rmac_drop_frms"},
|
|
|
|
{"rmac_vld_mcst_frms"},
|
|
|
|
{"rmac_vld_bcst_frms"},
|
|
|
|
{"rmac_in_rng_len_err_frms"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_out_rng_len_err_frms"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"rmac_long_frms"},
|
|
|
|
{"rmac_pause_ctrl_frms"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_unsup_ctrl_frms"},
|
|
|
|
{"rmac_ttl_octets"},
|
|
|
|
{"rmac_accepted_ucst_frms"},
|
|
|
|
{"rmac_accepted_nucst_frms"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"rmac_discarded_frms"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_drop_events"},
|
|
|
|
{"rmac_ttl_less_fb_octets"},
|
|
|
|
{"rmac_ttl_frms"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"rmac_usized_frms"},
|
|
|
|
{"rmac_osized_frms"},
|
|
|
|
{"rmac_frag_frms"},
|
|
|
|
{"rmac_jabber_frms"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_ttl_64_frms"},
|
|
|
|
{"rmac_ttl_65_127_frms"},
|
|
|
|
{"rmac_ttl_128_255_frms"},
|
|
|
|
{"rmac_ttl_256_511_frms"},
|
|
|
|
{"rmac_ttl_512_1023_frms"},
|
|
|
|
{"rmac_ttl_1024_1518_frms"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"rmac_ip"},
|
|
|
|
{"rmac_ip_octets"},
|
|
|
|
{"rmac_hdr_err_ip"},
|
|
|
|
{"rmac_drop_ip"},
|
|
|
|
{"rmac_icmp"},
|
|
|
|
{"rmac_tcp"},
|
|
|
|
{"rmac_udp"},
|
|
|
|
{"rmac_err_drp_udp"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_xgmii_err_sym"},
|
|
|
|
{"rmac_frms_q0"},
|
|
|
|
{"rmac_frms_q1"},
|
|
|
|
{"rmac_frms_q2"},
|
|
|
|
{"rmac_frms_q3"},
|
|
|
|
{"rmac_frms_q4"},
|
|
|
|
{"rmac_frms_q5"},
|
|
|
|
{"rmac_frms_q6"},
|
|
|
|
{"rmac_frms_q7"},
|
|
|
|
{"rmac_full_q0"},
|
|
|
|
{"rmac_full_q1"},
|
|
|
|
{"rmac_full_q2"},
|
|
|
|
{"rmac_full_q3"},
|
|
|
|
{"rmac_full_q4"},
|
|
|
|
{"rmac_full_q5"},
|
|
|
|
{"rmac_full_q6"},
|
|
|
|
{"rmac_full_q7"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"rmac_pause_cnt"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_xgmii_data_err_cnt"},
|
|
|
|
{"rmac_xgmii_ctrl_err_cnt"},
|
2005-04-16 22:20:36 +00:00
|
|
|
{"rmac_accepted_ip"},
|
|
|
|
{"rmac_err_tcp"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rd_req_cnt"},
|
|
|
|
{"new_rd_req_cnt"},
|
|
|
|
{"new_rd_req_rtry_cnt"},
|
|
|
|
{"rd_rtry_cnt"},
|
|
|
|
{"wr_rtry_rd_ack_cnt"},
|
|
|
|
{"wr_req_cnt"},
|
|
|
|
{"new_wr_req_cnt"},
|
|
|
|
{"new_wr_req_rtry_cnt"},
|
|
|
|
{"wr_rtry_cnt"},
|
|
|
|
{"wr_disc_cnt"},
|
|
|
|
{"rd_rtry_wr_ack_cnt"},
|
|
|
|
{"txp_wr_cnt"},
|
|
|
|
{"txd_rd_cnt"},
|
|
|
|
{"txd_wr_cnt"},
|
|
|
|
{"rxd_rd_cnt"},
|
|
|
|
{"rxd_wr_cnt"},
|
|
|
|
{"txf_rd_cnt"},
|
2007-02-24 07:03:22 +00:00
|
|
|
{"rxf_wr_cnt"}
|
|
|
|
};
|
|
|
|
|
|
|
|
static char ethtool_enhanced_stats_keys[][ETH_GSTRING_LEN] = {
|
2006-04-21 23:20:22 +00:00
|
|
|
{"rmac_ttl_1519_4095_frms"},
|
|
|
|
{"rmac_ttl_4096_8191_frms"},
|
|
|
|
{"rmac_ttl_8192_max_frms"},
|
|
|
|
{"rmac_ttl_gt_max_frms"},
|
|
|
|
{"rmac_osized_alt_frms"},
|
|
|
|
{"rmac_jabber_alt_frms"},
|
|
|
|
{"rmac_gt_max_alt_frms"},
|
|
|
|
{"rmac_vlan_frms"},
|
|
|
|
{"rmac_len_discard"},
|
|
|
|
{"rmac_fcs_discard"},
|
|
|
|
{"rmac_pf_discard"},
|
|
|
|
{"rmac_da_discard"},
|
|
|
|
{"rmac_red_discard"},
|
|
|
|
{"rmac_rts_discard"},
|
|
|
|
{"rmac_ingm_full_discard"},
|
2007-02-24 07:03:22 +00:00
|
|
|
{"link_fault_cnt"}
|
|
|
|
};
|
|
|
|
|
|
|
|
static char ethtool_driver_stats_keys[][ETH_GSTRING_LEN] = {
|
2005-08-03 19:29:20 +00:00
|
|
|
{"\n DRIVER STATISTICS"},
|
|
|
|
{"single_bit_ecc_errs"},
|
|
|
|
{"double_bit_ecc_errs"},
|
2006-04-21 23:20:22 +00:00
|
|
|
{"parity_err_cnt"},
|
|
|
|
{"serious_err_cnt"},
|
|
|
|
{"soft_reset_cnt"},
|
|
|
|
{"fifo_full_cnt"},
|
2007-09-17 20:05:35 +00:00
|
|
|
{"ring_0_full_cnt"},
|
|
|
|
{"ring_1_full_cnt"},
|
|
|
|
{"ring_2_full_cnt"},
|
|
|
|
{"ring_3_full_cnt"},
|
|
|
|
{"ring_4_full_cnt"},
|
|
|
|
{"ring_5_full_cnt"},
|
|
|
|
{"ring_6_full_cnt"},
|
|
|
|
{"ring_7_full_cnt"},
|
2007-10-05 19:39:21 +00:00
|
|
|
{"alarm_transceiver_temp_high"},
|
|
|
|
{"alarm_transceiver_temp_low"},
|
|
|
|
{"alarm_laser_bias_current_high"},
|
|
|
|
{"alarm_laser_bias_current_low"},
|
|
|
|
{"alarm_laser_output_power_high"},
|
|
|
|
{"alarm_laser_output_power_low"},
|
|
|
|
{"warn_transceiver_temp_high"},
|
|
|
|
{"warn_transceiver_temp_low"},
|
|
|
|
{"warn_laser_bias_current_high"},
|
|
|
|
{"warn_laser_bias_current_low"},
|
|
|
|
{"warn_laser_output_power_high"},
|
|
|
|
{"warn_laser_output_power_low"},
|
|
|
|
{"lro_aggregated_pkts"},
|
|
|
|
{"lro_flush_both_count"},
|
|
|
|
{"lro_out_of_sequence_pkts"},
|
|
|
|
{"lro_flush_due_to_max_pkts"},
|
|
|
|
{"lro_avg_aggr_pkts"},
|
|
|
|
{"mem_alloc_fail_cnt"},
|
|
|
|
{"pci_map_fail_cnt"},
|
|
|
|
{"watchdog_timer_cnt"},
|
|
|
|
{"mem_allocated"},
|
|
|
|
{"mem_freed"},
|
|
|
|
{"link_up_cnt"},
|
|
|
|
{"link_down_cnt"},
|
|
|
|
{"link_up_time"},
|
|
|
|
{"link_down_time"},
|
|
|
|
{"tx_tcode_buf_abort_cnt"},
|
|
|
|
{"tx_tcode_desc_abort_cnt"},
|
|
|
|
{"tx_tcode_parity_err_cnt"},
|
|
|
|
{"tx_tcode_link_loss_cnt"},
|
|
|
|
{"tx_tcode_list_proc_err_cnt"},
|
|
|
|
{"rx_tcode_parity_err_cnt"},
|
|
|
|
{"rx_tcode_abort_cnt"},
|
|
|
|
{"rx_tcode_parity_abort_cnt"},
|
|
|
|
{"rx_tcode_rda_fail_cnt"},
|
|
|
|
{"rx_tcode_unkn_prot_cnt"},
|
|
|
|
{"rx_tcode_fcs_err_cnt"},
|
|
|
|
{"rx_tcode_buf_size_err_cnt"},
|
|
|
|
{"rx_tcode_rxd_corrupt_cnt"},
|
|
|
|
{"rx_tcode_unkn_err_cnt"},
|
2007-09-17 20:05:35 +00:00
|
|
|
{"tda_err_cnt"},
|
|
|
|
{"pfc_err_cnt"},
|
|
|
|
{"pcc_err_cnt"},
|
|
|
|
{"tti_err_cnt"},
|
|
|
|
{"tpa_err_cnt"},
|
|
|
|
{"sm_err_cnt"},
|
|
|
|
{"lso_err_cnt"},
|
|
|
|
{"mac_tmac_err_cnt"},
|
|
|
|
{"mac_rmac_err_cnt"},
|
|
|
|
{"xgxs_txgxs_err_cnt"},
|
|
|
|
{"xgxs_rxgxs_err_cnt"},
|
|
|
|
{"rc_err_cnt"},
|
|
|
|
{"prc_pcix_err_cnt"},
|
|
|
|
{"rpa_err_cnt"},
|
|
|
|
{"rda_err_cnt"},
|
|
|
|
{"rti_err_cnt"},
|
|
|
|
{"mc_err_cnt"}
|
2005-04-16 22:20:36 +00:00
|
|
|
};
|
|
|
|
|
2007-10-18 08:00:15 +00:00
|
|
|
#define S2IO_XENA_STAT_LEN ARRAY_SIZE(ethtool_xena_stats_keys)
|
|
|
|
#define S2IO_ENHANCED_STAT_LEN ARRAY_SIZE(ethtool_enhanced_stats_keys)
|
|
|
|
#define S2IO_DRIVER_STAT_LEN ARRAY_SIZE(ethtool_driver_stats_keys)
|
2007-02-24 07:03:22 +00:00
|
|
|
|
|
|
|
#define XFRAME_I_STAT_LEN (S2IO_XENA_STAT_LEN + S2IO_DRIVER_STAT_LEN )
|
|
|
|
#define XFRAME_II_STAT_LEN (XFRAME_I_STAT_LEN + S2IO_ENHANCED_STAT_LEN )
|
|
|
|
|
|
|
|
#define XFRAME_I_STAT_STRINGS_LEN ( XFRAME_I_STAT_LEN * ETH_GSTRING_LEN )
|
|
|
|
#define XFRAME_II_STAT_STRINGS_LEN ( XFRAME_II_STAT_LEN * ETH_GSTRING_LEN )
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-10-18 08:00:15 +00:00
|
|
|
#define S2IO_TEST_LEN ARRAY_SIZE(s2io_gstrings)
|
2005-04-16 22:20:36 +00:00
|
|
|
#define S2IO_STRINGS_LEN S2IO_TEST_LEN * ETH_GSTRING_LEN
|
|
|
|
|
2005-08-03 19:34:11 +00:00
|
|
|
#define S2IO_TIMER_CONF(timer, handle, arg, exp) \
|
|
|
|
init_timer(&timer); \
|
|
|
|
timer.function = handle; \
|
|
|
|
timer.data = (unsigned long) arg; \
|
|
|
|
mod_timer(&timer, (jiffies + exp)) \
|
|
|
|
|
2007-09-14 11:39:19 +00:00
|
|
|
/* copy mac addr to def_mac_addr array */
|
|
|
|
static void do_s2io_copy_mac_addr(struct s2io_nic *sp, int offset, u64 mac_addr)
|
|
|
|
{
|
|
|
|
sp->def_mac_addr[offset].mac_addr[5] = (u8) (mac_addr);
|
|
|
|
sp->def_mac_addr[offset].mac_addr[4] = (u8) (mac_addr >> 8);
|
|
|
|
sp->def_mac_addr[offset].mac_addr[3] = (u8) (mac_addr >> 16);
|
|
|
|
sp->def_mac_addr[offset].mac_addr[2] = (u8) (mac_addr >> 24);
|
|
|
|
sp->def_mac_addr[offset].mac_addr[1] = (u8) (mac_addr >> 32);
|
|
|
|
sp->def_mac_addr[offset].mac_addr[0] = (u8) (mac_addr >> 40);
|
|
|
|
}
|
2008-11-22 01:28:55 +00:00
|
|
|
|
2005-08-03 19:35:55 +00:00
|
|
|
/* Add the vlan */
|
|
|
|
static void s2io_vlan_rx_register(struct net_device *dev,
|
2008-11-22 01:28:55 +00:00
|
|
|
struct vlan_group *grp)
|
2005-08-03 19:35:55 +00:00
|
|
|
{
|
2008-01-24 10:08:59 +00:00
|
|
|
int i;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *nic = netdev_priv(dev);
|
2008-01-24 10:08:59 +00:00
|
|
|
unsigned long flags[MAX_TX_FIFOS];
|
|
|
|
struct mac_info *mac_control = &nic->mac_control;
|
|
|
|
struct config_param *config = &nic->config;
|
|
|
|
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++)
|
|
|
|
spin_lock_irqsave(&mac_control->fifos[i].tx_lock, flags[i]);
|
2005-08-03 19:35:55 +00:00
|
|
|
|
|
|
|
nic->vlgrp = grp;
|
2008-01-24 10:08:59 +00:00
|
|
|
for (i = config->tx_fifo_num - 1; i >= 0; i--)
|
|
|
|
spin_unlock_irqrestore(&mac_control->fifos[i].tx_lock,
|
|
|
|
flags[i]);
|
2005-08-03 19:35:55 +00:00
|
|
|
}
|
|
|
|
|
2008-02-20 22:09:15 +00:00
|
|
|
/* Unregister the vlan */
|
2008-11-22 01:28:55 +00:00
|
|
|
static void s2io_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
|
2008-02-20 22:09:15 +00:00
|
|
|
{
|
|
|
|
int i;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *nic = netdev_priv(dev);
|
2008-02-20 22:09:15 +00:00
|
|
|
unsigned long flags[MAX_TX_FIFOS];
|
|
|
|
struct mac_info *mac_control = &nic->mac_control;
|
|
|
|
struct config_param *config = &nic->config;
|
|
|
|
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++)
|
|
|
|
spin_lock_irqsave(&mac_control->fifos[i].tx_lock, flags[i]);
|
|
|
|
|
|
|
|
if (nic->vlgrp)
|
|
|
|
vlan_group_set_device(nic->vlgrp, vid, NULL);
|
|
|
|
|
|
|
|
for (i = config->tx_fifo_num - 1; i >= 0; i--)
|
|
|
|
spin_unlock_irqrestore(&mac_control->fifos[i].tx_lock,
|
|
|
|
flags[i]);
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Constants to be programmed into the Xena's registers, to configure
|
|
|
|
* the XAUI.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define END_SIGN 0x0
|
2006-03-04 02:33:57 +00:00
|
|
|
static const u64 herc_act_dtx_cfg[] = {
|
2005-08-03 19:36:55 +00:00
|
|
|
/* Set address */
|
2005-08-12 17:15:59 +00:00
|
|
|
0x8000051536750000ULL, 0x80000515367500E0ULL,
|
2005-08-03 19:36:55 +00:00
|
|
|
/* Write data */
|
2005-08-12 17:15:59 +00:00
|
|
|
0x8000051536750004ULL, 0x80000515367500E4ULL,
|
2005-08-03 19:36:55 +00:00
|
|
|
/* Set address */
|
|
|
|
0x80010515003F0000ULL, 0x80010515003F00E0ULL,
|
|
|
|
/* Write data */
|
|
|
|
0x80010515003F0004ULL, 0x80010515003F00E4ULL,
|
|
|
|
/* Set address */
|
2005-08-12 17:15:59 +00:00
|
|
|
0x801205150D440000ULL, 0x801205150D4400E0ULL,
|
|
|
|
/* Write data */
|
|
|
|
0x801205150D440004ULL, 0x801205150D4400E4ULL,
|
|
|
|
/* Set address */
|
2005-08-03 19:36:55 +00:00
|
|
|
0x80020515F2100000ULL, 0x80020515F21000E0ULL,
|
|
|
|
/* Write data */
|
|
|
|
0x80020515F2100004ULL, 0x80020515F21000E4ULL,
|
|
|
|
/* Done */
|
|
|
|
END_SIGN
|
|
|
|
};
|
|
|
|
|
2006-03-04 02:33:57 +00:00
|
|
|
static const u64 xena_dtx_cfg[] = {
|
2006-04-21 23:18:03 +00:00
|
|
|
/* Set address */
|
2005-04-16 22:20:36 +00:00
|
|
|
0x8000051500000000ULL, 0x80000515000000E0ULL,
|
2006-04-21 23:18:03 +00:00
|
|
|
/* Write data */
|
|
|
|
0x80000515D9350004ULL, 0x80000515D93500E4ULL,
|
|
|
|
/* Set address */
|
|
|
|
0x8001051500000000ULL, 0x80010515000000E0ULL,
|
|
|
|
/* Write data */
|
|
|
|
0x80010515001E0004ULL, 0x80010515001E00E4ULL,
|
|
|
|
/* Set address */
|
2005-04-16 22:20:36 +00:00
|
|
|
0x8002051500000000ULL, 0x80020515000000E0ULL,
|
2006-04-21 23:18:03 +00:00
|
|
|
/* Write data */
|
|
|
|
0x80020515F2100004ULL, 0x80020515F21000E4ULL,
|
2005-04-16 22:20:36 +00:00
|
|
|
END_SIGN
|
|
|
|
};
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Constants for Fixing the MacAddress problem seen mostly on
|
|
|
|
* Alpha machines.
|
|
|
|
*/
|
2006-03-04 02:33:57 +00:00
|
|
|
static const u64 fix_mac[] = {
|
2005-04-16 22:20:36 +00:00
|
|
|
0x0060000000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0040600000000000ULL, 0x0000600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
0x0020600000000000ULL, 0x0000600000000000ULL,
|
|
|
|
0x0040600000000000ULL, 0x0060600000000000ULL,
|
|
|
|
END_SIGN
|
|
|
|
};
|
|
|
|
|
2006-07-24 23:52:49 +00:00
|
|
|
MODULE_LICENSE("GPL");
|
|
|
|
MODULE_VERSION(DRV_VERSION);
|
|
|
|
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Module Loadable parameters. */
|
2008-02-20 22:07:51 +00:00
|
|
|
S2IO_PARM_INT(tx_fifo_num, FIFO_DEFAULT_NUM);
|
2006-07-24 23:52:49 +00:00
|
|
|
S2IO_PARM_INT(rx_ring_num, 1);
|
2008-02-20 21:44:07 +00:00
|
|
|
S2IO_PARM_INT(multiq, 0);
|
2006-07-24 23:52:49 +00:00
|
|
|
S2IO_PARM_INT(rx_ring_mode, 1);
|
|
|
|
S2IO_PARM_INT(use_continuous_tx_intrs, 1);
|
|
|
|
S2IO_PARM_INT(rmac_pause_time, 0x100);
|
|
|
|
S2IO_PARM_INT(mc_pause_threshold_q0q3, 187);
|
|
|
|
S2IO_PARM_INT(mc_pause_threshold_q4q7, 187);
|
|
|
|
S2IO_PARM_INT(shared_splits, 0);
|
|
|
|
S2IO_PARM_INT(tmac_util_period, 5);
|
|
|
|
S2IO_PARM_INT(rmac_util_period, 5);
|
|
|
|
S2IO_PARM_INT(l3l4hdr_size, 128);
|
2008-02-20 22:07:51 +00:00
|
|
|
/* 0 is no steering, 1 is Priority steering, 2 is Default steering */
|
|
|
|
S2IO_PARM_INT(tx_steering_type, TX_DEFAULT_STEERING);
|
2005-08-03 19:41:38 +00:00
|
|
|
/* Frequency of Rx desc syncs expressed as power of 2 */
|
2006-07-24 23:52:49 +00:00
|
|
|
S2IO_PARM_INT(rxsync_frequency, 3);
|
2007-07-23 06:23:54 +00:00
|
|
|
/* Interrupt type. Values can be 0(INTA), 2(MSI_X) */
|
2007-09-15 20:11:34 +00:00
|
|
|
S2IO_PARM_INT(intr_type, 2);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/* Large receive offload feature */
|
2007-10-05 19:39:21 +00:00
|
|
|
static unsigned int lro_enable;
|
|
|
|
module_param_named(lro, lro_enable, uint, 0);
|
|
|
|
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/* Max pkts to be aggregated by LRO at one time. If not specified,
|
|
|
|
* aggregation happens until we hit max IP pkt size(64K)
|
|
|
|
*/
|
2006-07-24 23:52:49 +00:00
|
|
|
S2IO_PARM_INT(lro_max_pkts, 0xFFFF);
|
|
|
|
S2IO_PARM_INT(indicate_max_pkts, 0);
|
2007-01-31 18:28:08 +00:00
|
|
|
|
|
|
|
S2IO_PARM_INT(napi, 1);
|
|
|
|
S2IO_PARM_INT(ufo, 0);
|
2007-02-24 06:59:39 +00:00
|
|
|
S2IO_PARM_INT(vlan_tag_strip, NO_STRIP_IN_PROMISC);
|
2006-07-24 23:52:49 +00:00
|
|
|
|
|
|
|
static unsigned int tx_fifo_len[MAX_TX_FIFOS] =
|
|
|
|
{DEFAULT_FIFO_0_LEN, [1 ...(MAX_TX_FIFOS - 1)] = DEFAULT_FIFO_1_7_LEN};
|
|
|
|
static unsigned int rx_ring_sz[MAX_RX_RINGS] =
|
|
|
|
{[0 ...(MAX_RX_RINGS - 1)] = SMALL_BLK_CNT};
|
|
|
|
static unsigned int rts_frm_len[MAX_RX_RINGS] =
|
|
|
|
{[0 ...(MAX_RX_RINGS - 1)] = 0 };
|
|
|
|
|
|
|
|
module_param_array(tx_fifo_len, uint, NULL, 0);
|
|
|
|
module_param_array(rx_ring_sz, uint, NULL, 0);
|
|
|
|
module_param_array(rts_frm_len, uint, NULL, 0);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* S2IO device table.
|
2005-08-03 19:24:33 +00:00
|
|
|
* This table lists all the devices that this driver supports.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
static struct pci_device_id s2io_tbl[] __devinitdata = {
|
|
|
|
{PCI_VENDOR_ID_S2IO, PCI_DEVICE_ID_S2IO_WIN,
|
|
|
|
PCI_ANY_ID, PCI_ANY_ID},
|
|
|
|
{PCI_VENDOR_ID_S2IO, PCI_DEVICE_ID_S2IO_UNI,
|
|
|
|
PCI_ANY_ID, PCI_ANY_ID},
|
|
|
|
{PCI_VENDOR_ID_S2IO, PCI_DEVICE_ID_HERC_WIN,
|
2005-08-03 19:24:33 +00:00
|
|
|
PCI_ANY_ID, PCI_ANY_ID},
|
|
|
|
{PCI_VENDOR_ID_S2IO, PCI_DEVICE_ID_HERC_UNI,
|
|
|
|
PCI_ANY_ID, PCI_ANY_ID},
|
2005-04-16 22:20:36 +00:00
|
|
|
{0,}
|
|
|
|
};
|
|
|
|
|
|
|
|
MODULE_DEVICE_TABLE(pci, s2io_tbl);
|
|
|
|
|
2007-05-14 23:37:30 +00:00
|
|
|
static struct pci_error_handlers s2io_err_handler = {
|
|
|
|
.error_detected = s2io_io_error_detected,
|
|
|
|
.slot_reset = s2io_io_slot_reset,
|
|
|
|
.resume = s2io_io_resume,
|
|
|
|
};
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static struct pci_driver s2io_driver = {
|
|
|
|
.name = "S2IO",
|
|
|
|
.id_table = s2io_tbl,
|
|
|
|
.probe = s2io_init_nic,
|
|
|
|
.remove = __devexit_p(s2io_rem_nic),
|
2007-05-14 23:37:30 +00:00
|
|
|
.err_handler = &s2io_err_handler,
|
2005-04-16 22:20:36 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
/* A simplifier macro used both by init and free shared_mem Fns(). */
|
|
|
|
#define TXD_MEM_PAGE_CNT(len, per_each) ((len+per_each - 1) / per_each)
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
/* netqueue manipulation helper functions */
|
|
|
|
static inline void s2io_stop_all_tx_queue(struct s2io_nic *sp)
|
|
|
|
{
|
2008-07-17 08:56:23 +00:00
|
|
|
if (!sp->config.multiq) {
|
|
|
|
int i;
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
for (i = 0; i < sp->config.tx_fifo_num; i++)
|
|
|
|
sp->mac_control.fifos[i].queue_state = FIFO_QUEUE_STOP;
|
|
|
|
}
|
2008-07-17 08:56:23 +00:00
|
|
|
netif_tx_stop_all_queues(sp->dev);
|
2008-02-20 21:44:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void s2io_stop_tx_queue(struct s2io_nic *sp, int fifo_no)
|
|
|
|
{
|
2008-07-17 08:56:23 +00:00
|
|
|
if (!sp->config.multiq)
|
2008-02-20 21:44:07 +00:00
|
|
|
sp->mac_control.fifos[fifo_no].queue_state =
|
|
|
|
FIFO_QUEUE_STOP;
|
2008-07-17 08:56:23 +00:00
|
|
|
|
|
|
|
netif_tx_stop_all_queues(sp->dev);
|
2008-02-20 21:44:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void s2io_start_all_tx_queue(struct s2io_nic *sp)
|
|
|
|
{
|
2008-07-17 08:56:23 +00:00
|
|
|
if (!sp->config.multiq) {
|
|
|
|
int i;
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
for (i = 0; i < sp->config.tx_fifo_num; i++)
|
|
|
|
sp->mac_control.fifos[i].queue_state = FIFO_QUEUE_START;
|
|
|
|
}
|
2008-07-17 08:56:23 +00:00
|
|
|
netif_tx_start_all_queues(sp->dev);
|
2008-02-20 21:44:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void s2io_start_tx_queue(struct s2io_nic *sp, int fifo_no)
|
|
|
|
{
|
2008-07-17 08:56:23 +00:00
|
|
|
if (!sp->config.multiq)
|
2008-02-20 21:44:07 +00:00
|
|
|
sp->mac_control.fifos[fifo_no].queue_state =
|
|
|
|
FIFO_QUEUE_START;
|
2008-07-17 08:56:23 +00:00
|
|
|
|
|
|
|
netif_tx_start_all_queues(sp->dev);
|
2008-02-20 21:44:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void s2io_wake_all_tx_queue(struct s2io_nic *sp)
|
|
|
|
{
|
2008-07-17 08:56:23 +00:00
|
|
|
if (!sp->config.multiq) {
|
|
|
|
int i;
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
for (i = 0; i < sp->config.tx_fifo_num; i++)
|
|
|
|
sp->mac_control.fifos[i].queue_state = FIFO_QUEUE_START;
|
|
|
|
}
|
2008-07-17 08:56:23 +00:00
|
|
|
netif_tx_wake_all_queues(sp->dev);
|
2008-02-20 21:44:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void s2io_wake_tx_queue(
|
|
|
|
struct fifo_info *fifo, int cnt, u8 multiq)
|
|
|
|
{
|
|
|
|
|
|
|
|
if (multiq) {
|
|
|
|
if (cnt && __netif_subqueue_stopped(fifo->dev, fifo->fifo_no))
|
|
|
|
netif_wake_subqueue(fifo->dev, fifo->fifo_no);
|
2008-07-09 06:14:24 +00:00
|
|
|
} else if (cnt && (fifo->queue_state == FIFO_QUEUE_STOP)) {
|
2008-02-20 21:44:07 +00:00
|
|
|
if (netif_queue_stopped(fifo->dev)) {
|
|
|
|
fifo->queue_state = FIFO_QUEUE_START;
|
|
|
|
netif_wake_queue(fifo->dev);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* init_shared_mem - Allocation and Initialization of Memory
|
|
|
|
* @nic: Device private variable.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: The function allocates all the memory areas shared
|
|
|
|
* between the NIC and the driver. This includes Tx descriptors,
|
2005-04-16 22:20:36 +00:00
|
|
|
* Rx descriptors and the statistics block.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int init_shared_mem(struct s2io_nic *nic)
|
|
|
|
{
|
|
|
|
u32 size;
|
|
|
|
void *tmp_v_addr, *tmp_v_addr_next;
|
|
|
|
dma_addr_t tmp_p_addr, tmp_p_addr_next;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct RxD_block *pre_rxd_blk = NULL;
|
2007-01-31 18:32:57 +00:00
|
|
|
int i, j, blk_cnt;
|
2005-04-16 22:20:36 +00:00
|
|
|
int lst_size, lst_per_page;
|
|
|
|
struct net_device *dev = nic->dev;
|
2005-09-02 19:15:29 +00:00
|
|
|
unsigned long tmp;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct buffAdd *ba;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2007-05-10 08:22:25 +00:00
|
|
|
unsigned long long mem_allocated = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
|
|
|
|
|
|
|
/* Allocation and initialization of TXDLs in FIOFs */
|
|
|
|
size = 0;
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
size += config->tx_cfg[i].fifo_len;
|
|
|
|
}
|
|
|
|
if (size > MAX_AVAILABLE_TXDS) {
|
2006-07-24 23:52:49 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Requested TxDs too high, ");
|
2005-08-03 19:39:56 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "Requested: %d, max supported: 8192\n", size);
|
2006-07-24 23:52:49 +00:00
|
|
|
return -EINVAL;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
size = 0;
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
size = config->tx_cfg[i].fifo_len;
|
|
|
|
/*
|
|
|
|
* Legal values are from 2 to 8192
|
|
|
|
*/
|
|
|
|
if (size < 2) {
|
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Invalid fifo len (%d)", size);
|
|
|
|
DBG_PRINT(ERR_DBG, "for fifo %d\n", i);
|
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Legal values for fifo len"
|
|
|
|
"are 2 to 8192\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
lst_size = (sizeof(struct TxD) * config->max_txds);
|
2005-04-16 22:20:36 +00:00
|
|
|
lst_per_page = PAGE_SIZE / lst_size;
|
|
|
|
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
int fifo_len = config->tx_cfg[i].fifo_len;
|
2007-01-31 19:09:29 +00:00
|
|
|
int list_holder_size = fifo_len * sizeof(struct list_info_hold);
|
2007-09-14 11:28:50 +00:00
|
|
|
mac_control->fifos[i].list_info = kzalloc(list_holder_size,
|
2005-08-03 19:24:33 +00:00
|
|
|
GFP_KERNEL);
|
|
|
|
if (!mac_control->fifos[i].list_info) {
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG,
|
2005-04-16 22:20:36 +00:00
|
|
|
"Malloc failed for list_info\n");
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_allocated += list_holder_size;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
int page_num = TXD_MEM_PAGE_CNT(config->tx_cfg[i].fifo_len,
|
|
|
|
lst_per_page);
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].tx_curr_put_info.offset = 0;
|
|
|
|
mac_control->fifos[i].tx_curr_put_info.fifo_len =
|
2005-04-16 22:20:36 +00:00
|
|
|
config->tx_cfg[i].fifo_len - 1;
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].tx_curr_get_info.offset = 0;
|
|
|
|
mac_control->fifos[i].tx_curr_get_info.fifo_len =
|
2005-04-16 22:20:36 +00:00
|
|
|
config->tx_cfg[i].fifo_len - 1;
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].fifo_no = i;
|
|
|
|
mac_control->fifos[i].nic = nic;
|
2005-11-14 20:25:08 +00:00
|
|
|
mac_control->fifos[i].max_txds = MAX_SKB_FRAGS + 2;
|
2008-02-20 21:44:07 +00:00
|
|
|
mac_control->fifos[i].dev = dev;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
for (j = 0; j < page_num; j++) {
|
|
|
|
int k = 0;
|
|
|
|
dma_addr_t tmp_p;
|
|
|
|
void *tmp_v;
|
|
|
|
tmp_v = pci_alloc_consistent(nic->pdev,
|
|
|
|
PAGE_SIZE, &tmp_p);
|
|
|
|
if (!tmp_v) {
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG,
|
2005-04-16 22:20:36 +00:00
|
|
|
"pci_alloc_consistent ");
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "failed for TxDL\n");
|
2005-04-16 22:20:36 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2005-09-07 04:36:56 +00:00
|
|
|
/* If we got a zero DMA address(can happen on
|
|
|
|
* certain platforms like PPC), reallocate.
|
|
|
|
* Store virtual address of page we don't want,
|
|
|
|
* to be freed later.
|
|
|
|
*/
|
|
|
|
if (!tmp_p) {
|
|
|
|
mac_control->zerodma_virt_addr = tmp_v;
|
2006-09-13 17:24:59 +00:00
|
|
|
DBG_PRINT(INIT_DBG,
|
2005-09-07 04:36:56 +00:00
|
|
|
"%s: Zero DMA address for TxDL. ", dev->name);
|
2006-09-13 17:24:59 +00:00
|
|
|
DBG_PRINT(INIT_DBG,
|
2005-09-13 06:21:55 +00:00
|
|
|
"Virtual address %p\n", tmp_v);
|
2005-09-07 04:36:56 +00:00
|
|
|
tmp_v = pci_alloc_consistent(nic->pdev,
|
|
|
|
PAGE_SIZE, &tmp_p);
|
|
|
|
if (!tmp_v) {
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG,
|
2005-09-07 04:36:56 +00:00
|
|
|
"pci_alloc_consistent ");
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "failed for TxDL\n");
|
2005-09-07 04:36:56 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_allocated += PAGE_SIZE;
|
2005-09-07 04:36:56 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
while (k < lst_per_page) {
|
|
|
|
int l = (j * lst_per_page) + k;
|
|
|
|
if (l == config->tx_cfg[i].fifo_len)
|
2005-08-03 19:24:33 +00:00
|
|
|
break;
|
|
|
|
mac_control->fifos[i].list_info[l].list_virt_addr =
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_v + (k * lst_size);
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].list_info[l].list_phy_addr =
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_p + (k * lst_size);
|
|
|
|
k++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
size = config->tx_cfg[i].fifo_len;
|
|
|
|
mac_control->fifos[i].ufo_in_band_v
|
|
|
|
= kcalloc(size, sizeof(u64), GFP_KERNEL);
|
|
|
|
if (!mac_control->fifos[i].ufo_in_band_v)
|
|
|
|
return -ENOMEM;
|
|
|
|
mem_allocated += (size * sizeof(u64));
|
|
|
|
}
|
2005-11-14 20:25:08 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Allocation and initialization of RXDs in Rings */
|
|
|
|
size = 0;
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2005-10-31 21:55:31 +00:00
|
|
|
if (config->rx_cfg[i].num_rxd %
|
|
|
|
(rxd_count[nic->rxd_mode] + 1)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: RxD count of ", dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "Ring%d is not a multiple of ",
|
|
|
|
i);
|
|
|
|
DBG_PRINT(ERR_DBG, "RxDs per Block");
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
size += config->rx_cfg[i].num_rxd;
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].block_count =
|
2005-10-31 21:55:31 +00:00
|
|
|
config->rx_cfg[i].num_rxd /
|
|
|
|
(rxd_count[nic->rxd_mode] + 1 );
|
|
|
|
mac_control->rings[i].pkt_cnt = config->rx_cfg[i].num_rxd -
|
|
|
|
mac_control->rings[i].block_count;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-10-31 21:55:31 +00:00
|
|
|
if (nic->rxd_mode == RXD_MODE_1)
|
2007-01-31 19:09:29 +00:00
|
|
|
size = (size * (sizeof(struct RxD1)));
|
2005-10-31 21:55:31 +00:00
|
|
|
else
|
2007-01-31 19:09:29 +00:00
|
|
|
size = (size * (sizeof(struct RxD3)));
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].rx_curr_get_info.block_index = 0;
|
|
|
|
mac_control->rings[i].rx_curr_get_info.offset = 0;
|
|
|
|
mac_control->rings[i].rx_curr_get_info.ring_len =
|
2005-04-16 22:20:36 +00:00
|
|
|
config->rx_cfg[i].num_rxd - 1;
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].rx_curr_put_info.block_index = 0;
|
|
|
|
mac_control->rings[i].rx_curr_put_info.offset = 0;
|
|
|
|
mac_control->rings[i].rx_curr_put_info.ring_len =
|
2005-04-16 22:20:36 +00:00
|
|
|
config->rx_cfg[i].num_rxd - 1;
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].nic = nic;
|
|
|
|
mac_control->rings[i].ring_no = i;
|
2008-04-29 01:08:45 +00:00
|
|
|
mac_control->rings[i].lro = lro_enable;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
blk_cnt = config->rx_cfg[i].num_rxd /
|
|
|
|
(rxd_count[nic->rxd_mode] + 1);
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Allocating all the Rx blocks */
|
|
|
|
for (j = 0; j < blk_cnt; j++) {
|
2007-01-31 19:09:29 +00:00
|
|
|
struct rx_block_info *rx_blocks;
|
2005-10-31 21:55:31 +00:00
|
|
|
int l;
|
|
|
|
|
|
|
|
rx_blocks = &mac_control->rings[i].rx_blocks[j];
|
|
|
|
size = SIZE_OF_BLOCK; //size is always page size
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_v_addr = pci_alloc_consistent(nic->pdev, size,
|
|
|
|
&tmp_p_addr);
|
|
|
|
if (tmp_v_addr == NULL) {
|
|
|
|
/*
|
2005-08-03 19:24:33 +00:00
|
|
|
* In case of failure, free_shared_mem()
|
|
|
|
* is called, which should free any
|
|
|
|
* memory that was alloced till the
|
2005-04-16 22:20:36 +00:00
|
|
|
* failure happened.
|
|
|
|
*/
|
2005-10-31 21:55:31 +00:00
|
|
|
rx_blocks->block_virt_addr = tmp_v_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_allocated += size;
|
2005-04-16 22:20:36 +00:00
|
|
|
memset(tmp_v_addr, 0, size);
|
2005-10-31 21:55:31 +00:00
|
|
|
rx_blocks->block_virt_addr = tmp_v_addr;
|
|
|
|
rx_blocks->block_dma_addr = tmp_p_addr;
|
2007-01-31 19:09:29 +00:00
|
|
|
rx_blocks->rxds = kmalloc(sizeof(struct rxd_info)*
|
2005-10-31 21:55:31 +00:00
|
|
|
rxd_count[nic->rxd_mode],
|
|
|
|
GFP_KERNEL);
|
2007-01-31 18:32:57 +00:00
|
|
|
if (!rx_blocks->rxds)
|
|
|
|
return -ENOMEM;
|
2007-09-18 22:14:20 +00:00
|
|
|
mem_allocated +=
|
2007-05-10 08:22:25 +00:00
|
|
|
(sizeof(struct rxd_info)* rxd_count[nic->rxd_mode]);
|
2005-10-31 21:55:31 +00:00
|
|
|
for (l=0; l<rxd_count[nic->rxd_mode];l++) {
|
|
|
|
rx_blocks->rxds[l].virt_addr =
|
|
|
|
rx_blocks->block_virt_addr +
|
|
|
|
(rxd_size[nic->rxd_mode] * l);
|
|
|
|
rx_blocks->rxds[l].dma_addr =
|
|
|
|
rx_blocks->block_dma_addr +
|
|
|
|
(rxd_size[nic->rxd_mode] * l);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
/* Interlinking all Rx Blocks */
|
|
|
|
for (j = 0; j < blk_cnt; j++) {
|
2005-08-03 19:24:33 +00:00
|
|
|
tmp_v_addr =
|
|
|
|
mac_control->rings[i].rx_blocks[j].block_virt_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_v_addr_next =
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].rx_blocks[(j + 1) %
|
2005-04-16 22:20:36 +00:00
|
|
|
blk_cnt].block_virt_addr;
|
2005-08-03 19:24:33 +00:00
|
|
|
tmp_p_addr =
|
|
|
|
mac_control->rings[i].rx_blocks[j].block_dma_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_p_addr_next =
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].rx_blocks[(j + 1) %
|
2005-04-16 22:20:36 +00:00
|
|
|
blk_cnt].block_dma_addr;
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
pre_rxd_blk = (struct RxD_block *) tmp_v_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
pre_rxd_blk->reserved_2_pNext_RxD_block =
|
|
|
|
(unsigned long) tmp_v_addr_next;
|
|
|
|
pre_rxd_blk->pNext_RxD_Blk_physical =
|
|
|
|
(u64) tmp_p_addr_next;
|
|
|
|
}
|
|
|
|
}
|
2007-07-23 06:20:51 +00:00
|
|
|
if (nic->rxd_mode == RXD_MODE_3B) {
|
2005-10-31 21:55:31 +00:00
|
|
|
/*
|
|
|
|
* Allocation of Storages for buffer addresses in 2BUFF mode
|
|
|
|
* and the buffers as well.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
blk_cnt = config->rx_cfg[i].num_rxd /
|
|
|
|
(rxd_count[nic->rxd_mode]+ 1);
|
|
|
|
mac_control->rings[i].ba =
|
2007-01-31 19:09:29 +00:00
|
|
|
kmalloc((sizeof(struct buffAdd *) * blk_cnt),
|
2005-04-16 22:20:36 +00:00
|
|
|
GFP_KERNEL);
|
2005-10-31 21:55:31 +00:00
|
|
|
if (!mac_control->rings[i].ba)
|
2005-04-16 22:20:36 +00:00
|
|
|
return -ENOMEM;
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_allocated +=(sizeof(struct buffAdd *) * blk_cnt);
|
2005-10-31 21:55:31 +00:00
|
|
|
for (j = 0; j < blk_cnt; j++) {
|
|
|
|
int k = 0;
|
|
|
|
mac_control->rings[i].ba[j] =
|
2007-01-31 19:09:29 +00:00
|
|
|
kmalloc((sizeof(struct buffAdd) *
|
2005-10-31 21:55:31 +00:00
|
|
|
(rxd_count[nic->rxd_mode] + 1)),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (!mac_control->rings[i].ba[j])
|
2005-04-16 22:20:36 +00:00
|
|
|
return -ENOMEM;
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_allocated += (sizeof(struct buffAdd) * \
|
|
|
|
(rxd_count[nic->rxd_mode] + 1));
|
2005-10-31 21:55:31 +00:00
|
|
|
while (k != rxd_count[nic->rxd_mode]) {
|
|
|
|
ba = &mac_control->rings[i].ba[j][k];
|
|
|
|
|
|
|
|
ba->ba_0_org = (void *) kmalloc
|
|
|
|
(BUF0_LEN + ALIGN_SIZE, GFP_KERNEL);
|
|
|
|
if (!ba->ba_0_org)
|
|
|
|
return -ENOMEM;
|
2007-09-18 22:14:20 +00:00
|
|
|
mem_allocated +=
|
2007-05-10 08:22:25 +00:00
|
|
|
(BUF0_LEN + ALIGN_SIZE);
|
2005-10-31 21:55:31 +00:00
|
|
|
tmp = (unsigned long)ba->ba_0_org;
|
|
|
|
tmp += ALIGN_SIZE;
|
|
|
|
tmp &= ~((unsigned long) ALIGN_SIZE);
|
|
|
|
ba->ba_0 = (void *) tmp;
|
|
|
|
|
|
|
|
ba->ba_1_org = (void *) kmalloc
|
|
|
|
(BUF1_LEN + ALIGN_SIZE, GFP_KERNEL);
|
|
|
|
if (!ba->ba_1_org)
|
|
|
|
return -ENOMEM;
|
2007-09-18 22:14:20 +00:00
|
|
|
mem_allocated
|
2007-05-10 08:22:25 +00:00
|
|
|
+= (BUF1_LEN + ALIGN_SIZE);
|
2005-10-31 21:55:31 +00:00
|
|
|
tmp = (unsigned long) ba->ba_1_org;
|
|
|
|
tmp += ALIGN_SIZE;
|
|
|
|
tmp &= ~((unsigned long) ALIGN_SIZE);
|
|
|
|
ba->ba_1 = (void *) tmp;
|
|
|
|
k++;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Allocation and initialization of Statistics block */
|
2007-01-31 19:09:29 +00:00
|
|
|
size = sizeof(struct stat_block);
|
2005-04-16 22:20:36 +00:00
|
|
|
mac_control->stats_mem = pci_alloc_consistent
|
|
|
|
(nic->pdev, size, &mac_control->stats_mem_phy);
|
|
|
|
|
|
|
|
if (!mac_control->stats_mem) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* In case of failure, free_shared_mem() is called, which
|
|
|
|
* should free any memory that was alloced till the
|
2005-04-16 22:20:36 +00:00
|
|
|
* failure happened.
|
|
|
|
*/
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_allocated += size;
|
2005-04-16 22:20:36 +00:00
|
|
|
mac_control->stats_mem_sz = size;
|
|
|
|
|
|
|
|
tmp_v_addr = mac_control->stats_mem;
|
2007-01-31 19:09:29 +00:00
|
|
|
mac_control->stats_info = (struct stat_block *) tmp_v_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
memset(tmp_v_addr, 0, size);
|
|
|
|
DBG_PRINT(INIT_DBG, "%s:Ring Mem PHY: 0x%llx\n", dev->name,
|
|
|
|
(unsigned long long) tmp_p_addr);
|
2007-05-10 08:22:25 +00:00
|
|
|
mac_control->stats_info->sw_stat.mem_allocated += mem_allocated;
|
2005-04-16 22:20:36 +00:00
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* free_shared_mem - Free the allocated Memory
|
2005-04-16 22:20:36 +00:00
|
|
|
* @nic: Device private variable.
|
|
|
|
* Description: This function is to free all memory locations allocated by
|
|
|
|
* the init_shared_mem() function and return it to the kernel.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void free_shared_mem(struct s2io_nic *nic)
|
|
|
|
{
|
|
|
|
int i, j, blk_cnt, size;
|
|
|
|
void *tmp_v_addr;
|
|
|
|
dma_addr_t tmp_p_addr;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
|
|
|
int lst_size, lst_per_page;
|
2007-07-09 03:29:04 +00:00
|
|
|
struct net_device *dev;
|
2007-05-10 08:22:25 +00:00
|
|
|
int page_num = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (!nic)
|
|
|
|
return;
|
|
|
|
|
2007-07-09 03:29:04 +00:00
|
|
|
dev = nic->dev;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
lst_size = (sizeof(struct TxD) * config->max_txds);
|
2005-04-16 22:20:36 +00:00
|
|
|
lst_per_page = PAGE_SIZE / lst_size;
|
|
|
|
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
2007-05-10 08:22:25 +00:00
|
|
|
page_num = TXD_MEM_PAGE_CNT(config->tx_cfg[i].fifo_len,
|
|
|
|
lst_per_page);
|
2005-04-16 22:20:36 +00:00
|
|
|
for (j = 0; j < page_num; j++) {
|
|
|
|
int mem_blks = (j * lst_per_page);
|
2005-09-07 04:36:56 +00:00
|
|
|
if (!mac_control->fifos[i].list_info)
|
2006-09-13 17:24:59 +00:00
|
|
|
return;
|
2005-09-07 04:36:56 +00:00
|
|
|
if (!mac_control->fifos[i].list_info[mem_blks].
|
|
|
|
list_virt_addr)
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
pci_free_consistent(nic->pdev, PAGE_SIZE,
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].
|
|
|
|
list_info[mem_blks].
|
2005-04-16 22:20:36 +00:00
|
|
|
list_virt_addr,
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].
|
|
|
|
list_info[mem_blks].
|
2005-04-16 22:20:36 +00:00
|
|
|
list_phy_addr);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
2007-05-10 08:22:25 +00:00
|
|
|
+= PAGE_SIZE;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-09-07 04:36:56 +00:00
|
|
|
/* If we got a zero DMA address during allocation,
|
|
|
|
* free the page now
|
|
|
|
*/
|
|
|
|
if (mac_control->zerodma_virt_addr) {
|
|
|
|
pci_free_consistent(nic->pdev, PAGE_SIZE,
|
|
|
|
mac_control->zerodma_virt_addr,
|
|
|
|
(dma_addr_t)0);
|
2006-09-13 17:24:59 +00:00
|
|
|
DBG_PRINT(INIT_DBG,
|
2005-09-13 06:21:55 +00:00
|
|
|
"%s: Freeing TxDL with zero DMA addr. ",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(INIT_DBG, "Virtual address %p\n",
|
|
|
|
mac_control->zerodma_virt_addr);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
2007-05-10 08:22:25 +00:00
|
|
|
+= PAGE_SIZE;
|
2005-09-07 04:36:56 +00:00
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
kfree(mac_control->fifos[i].list_info);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed +=
|
2007-05-10 08:22:25 +00:00
|
|
|
(nic->config.tx_cfg[i].fifo_len *sizeof(struct list_info_hold));
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
size = SIZE_OF_BLOCK;
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2005-08-03 19:24:33 +00:00
|
|
|
blk_cnt = mac_control->rings[i].block_count;
|
2005-04-16 22:20:36 +00:00
|
|
|
for (j = 0; j < blk_cnt; j++) {
|
2005-08-03 19:24:33 +00:00
|
|
|
tmp_v_addr = mac_control->rings[i].rx_blocks[j].
|
|
|
|
block_virt_addr;
|
|
|
|
tmp_p_addr = mac_control->rings[i].rx_blocks[j].
|
|
|
|
block_dma_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
if (tmp_v_addr == NULL)
|
|
|
|
break;
|
|
|
|
pci_free_consistent(nic->pdev, size,
|
|
|
|
tmp_v_addr, tmp_p_addr);
|
2007-05-10 08:22:25 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed += size;
|
2005-10-31 21:55:31 +00:00
|
|
|
kfree(mac_control->rings[i].rx_blocks[j].rxds);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed +=
|
2007-05-10 08:22:25 +00:00
|
|
|
( sizeof(struct rxd_info)* rxd_count[nic->rxd_mode]);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-07-23 06:20:51 +00:00
|
|
|
if (nic->rxd_mode == RXD_MODE_3B) {
|
2005-10-31 21:55:31 +00:00
|
|
|
/* Freeing buffer storage addresses in 2BUFF mode. */
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
blk_cnt = config->rx_cfg[i].num_rxd /
|
|
|
|
(rxd_count[nic->rxd_mode] + 1);
|
|
|
|
for (j = 0; j < blk_cnt; j++) {
|
|
|
|
int k = 0;
|
|
|
|
if (!mac_control->rings[i].ba[j])
|
|
|
|
continue;
|
|
|
|
while (k != rxd_count[nic->rxd_mode]) {
|
2007-01-31 19:09:29 +00:00
|
|
|
struct buffAdd *ba =
|
2005-10-31 21:55:31 +00:00
|
|
|
&mac_control->rings[i].ba[j][k];
|
|
|
|
kfree(ba->ba_0_org);
|
2007-05-10 08:22:25 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.\
|
|
|
|
mem_freed += (BUF0_LEN + ALIGN_SIZE);
|
2005-10-31 21:55:31 +00:00
|
|
|
kfree(ba->ba_1_org);
|
2007-05-10 08:22:25 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.\
|
|
|
|
mem_freed += (BUF1_LEN + ALIGN_SIZE);
|
2005-10-31 21:55:31 +00:00
|
|
|
k++;
|
|
|
|
}
|
|
|
|
kfree(mac_control->rings[i].ba[j]);
|
2007-09-06 10:21:54 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed +=
|
|
|
|
(sizeof(struct buffAdd) *
|
|
|
|
(rxd_count[nic->rxd_mode] + 1));
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-10-31 21:55:31 +00:00
|
|
|
kfree(mac_control->rings[i].ba);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed +=
|
2007-05-10 08:22:25 +00:00
|
|
|
(sizeof(struct buffAdd *) * blk_cnt);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
for (i = 0; i < nic->config.tx_fifo_num; i++) {
|
|
|
|
if (mac_control->fifos[i].ufo_in_band_v) {
|
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
|
|
|
+= (config->tx_cfg[i].fifo_len * sizeof(u64));
|
|
|
|
kfree(mac_control->fifos[i].ufo_in_band_v);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (mac_control->stats_mem) {
|
2008-01-24 10:08:59 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed +=
|
|
|
|
mac_control->stats_mem_sz;
|
2005-04-16 22:20:36 +00:00
|
|
|
pci_free_consistent(nic->pdev,
|
|
|
|
mac_control->stats_mem_sz,
|
|
|
|
mac_control->stats_mem,
|
|
|
|
mac_control->stats_mem_phy);
|
2007-05-10 08:22:25 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
/**
|
|
|
|
* s2io_verify_pci_mode -
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_verify_pci_mode(struct s2io_nic *nic)
|
2005-08-03 19:36:55 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-08-03 19:36:55 +00:00
|
|
|
register u64 val64 = 0;
|
|
|
|
int mode;
|
|
|
|
|
|
|
|
val64 = readq(&bar0->pci_mode);
|
|
|
|
mode = (u8)GET_PCI_MODE(val64);
|
|
|
|
|
|
|
|
if ( val64 & PCI_MODE_UNKNOWN_MODE)
|
|
|
|
return -1; /* Unknown PCI mode */
|
|
|
|
return mode;
|
|
|
|
}
|
|
|
|
|
2006-04-21 23:18:03 +00:00
|
|
|
#define NEC_VENID 0x1033
|
|
|
|
#define NEC_DEVID 0x0125
|
|
|
|
static int s2io_on_nec_bridge(struct pci_dev *s2io_pdev)
|
|
|
|
{
|
|
|
|
struct pci_dev *tdev = NULL;
|
2006-09-15 14:22:51 +00:00
|
|
|
while ((tdev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, tdev)) != NULL) {
|
|
|
|
if (tdev->vendor == NEC_VENID && tdev->device == NEC_DEVID) {
|
2008-05-13 11:16:54 +00:00
|
|
|
if (tdev->bus == s2io_pdev->bus->parent) {
|
2006-09-15 14:22:51 +00:00
|
|
|
pci_dev_put(tdev);
|
2006-04-21 23:18:03 +00:00
|
|
|
return 1;
|
2008-05-13 11:16:54 +00:00
|
|
|
}
|
2006-04-21 23:18:03 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2005-08-03 19:36:55 +00:00
|
|
|
|
2006-05-16 15:30:50 +00:00
|
|
|
static int bus_speed[8] = {33, 133, 133, 200, 266, 133, 200, 266};
|
2005-08-03 19:36:55 +00:00
|
|
|
/**
|
|
|
|
* s2io_print_pci_mode -
|
|
|
|
*/
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_print_pci_mode(struct s2io_nic *nic)
|
2005-08-03 19:36:55 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-08-03 19:36:55 +00:00
|
|
|
register u64 val64 = 0;
|
|
|
|
int mode;
|
|
|
|
struct config_param *config = &nic->config;
|
|
|
|
|
|
|
|
val64 = readq(&bar0->pci_mode);
|
|
|
|
mode = (u8)GET_PCI_MODE(val64);
|
|
|
|
|
|
|
|
if ( val64 & PCI_MODE_UNKNOWN_MODE)
|
|
|
|
return -1; /* Unknown PCI mode */
|
|
|
|
|
2006-04-21 23:18:03 +00:00
|
|
|
config->bus_speed = bus_speed[mode];
|
|
|
|
|
|
|
|
if (s2io_on_nec_bridge(nic->pdev)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Device is on PCI-E bus\n",
|
|
|
|
nic->dev->name);
|
|
|
|
return mode;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
if (val64 & PCI_MODE_32_BITS) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Device is on 32 bit ", nic->dev->name);
|
|
|
|
} else {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Device is on 64 bit ", nic->dev->name);
|
|
|
|
}
|
|
|
|
|
|
|
|
switch(mode) {
|
|
|
|
case PCI_MODE_PCI_33:
|
|
|
|
DBG_PRINT(ERR_DBG, "33MHz PCI bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCI_66:
|
|
|
|
DBG_PRINT(ERR_DBG, "66MHz PCI bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCIX_M1_66:
|
|
|
|
DBG_PRINT(ERR_DBG, "66MHz PCIX(M1) bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCIX_M1_100:
|
|
|
|
DBG_PRINT(ERR_DBG, "100MHz PCIX(M1) bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCIX_M1_133:
|
|
|
|
DBG_PRINT(ERR_DBG, "133MHz PCIX(M1) bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCIX_M2_66:
|
|
|
|
DBG_PRINT(ERR_DBG, "133MHz PCIX(M2) bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCIX_M2_100:
|
|
|
|
DBG_PRINT(ERR_DBG, "200MHz PCIX(M2) bus\n");
|
|
|
|
break;
|
|
|
|
case PCI_MODE_PCIX_M2_133:
|
|
|
|
DBG_PRINT(ERR_DBG, "266MHz PCIX(M2) bus\n");
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return -1; /* Unsupported bus speed */
|
|
|
|
}
|
|
|
|
|
|
|
|
return mode;
|
|
|
|
}
|
|
|
|
|
2007-12-17 19:40:15 +00:00
|
|
|
/**
|
|
|
|
* init_tti - Initialization transmit traffic interrupt scheme
|
|
|
|
* @nic: device private variable
|
|
|
|
* @link: link status (UP/DOWN) used to enable/disable continuous
|
|
|
|
* transmit interrupts
|
|
|
|
* Description: The function configures transmit traffic interrupts
|
|
|
|
* Return Value: SUCCESS on success and
|
|
|
|
* '-1' on failure
|
|
|
|
*/
|
|
|
|
|
2008-03-04 23:19:22 +00:00
|
|
|
static int init_tti(struct s2io_nic *nic, int link)
|
2007-12-17 19:40:15 +00:00
|
|
|
{
|
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
|
|
|
register u64 val64 = 0;
|
|
|
|
int i;
|
|
|
|
struct config_param *config;
|
|
|
|
|
|
|
|
config = &nic->config;
|
|
|
|
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
/*
|
|
|
|
* TTI Initialization. Default Tx timer gets us about
|
|
|
|
* 250 interrupts per sec. Continuous interrupts are enabled
|
|
|
|
* by default.
|
|
|
|
*/
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE) {
|
|
|
|
int count = (nic->config.bus_speed * 125)/2;
|
|
|
|
val64 = TTI_DATA1_MEM_TX_TIMER_VAL(count);
|
|
|
|
} else
|
|
|
|
val64 = TTI_DATA1_MEM_TX_TIMER_VAL(0x2078);
|
|
|
|
|
|
|
|
val64 |= TTI_DATA1_MEM_TX_URNG_A(0xA) |
|
|
|
|
TTI_DATA1_MEM_TX_URNG_B(0x10) |
|
|
|
|
TTI_DATA1_MEM_TX_URNG_C(0x30) |
|
|
|
|
TTI_DATA1_MEM_TX_TIMER_AC_EN;
|
2008-05-12 17:41:32 +00:00
|
|
|
if (i == 0)
|
|
|
|
if (use_continuous_tx_intrs && (link == LINK_UP))
|
|
|
|
val64 |= TTI_DATA1_MEM_TX_TIMER_CI_EN;
|
2007-12-17 19:40:15 +00:00
|
|
|
writeq(val64, &bar0->tti_data1_mem);
|
|
|
|
|
2008-05-12 17:41:32 +00:00
|
|
|
if (nic->config.intr_type == MSI_X) {
|
|
|
|
val64 = TTI_DATA2_MEM_TX_UFC_A(0x10) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_B(0x100) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_C(0x200) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_D(0x300);
|
|
|
|
} else {
|
|
|
|
if ((nic->config.tx_steering_type ==
|
|
|
|
TX_DEFAULT_STEERING) &&
|
|
|
|
(config->tx_fifo_num > 1) &&
|
|
|
|
(i >= nic->udp_fifo_idx) &&
|
|
|
|
(i < (nic->udp_fifo_idx +
|
|
|
|
nic->total_udp_fifos)))
|
|
|
|
val64 = TTI_DATA2_MEM_TX_UFC_A(0x50) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_B(0x80) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_C(0x100) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_D(0x120);
|
|
|
|
else
|
|
|
|
val64 = TTI_DATA2_MEM_TX_UFC_A(0x10) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_B(0x20) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_C(0x40) |
|
|
|
|
TTI_DATA2_MEM_TX_UFC_D(0x80);
|
|
|
|
}
|
2007-12-17 19:40:15 +00:00
|
|
|
|
|
|
|
writeq(val64, &bar0->tti_data2_mem);
|
|
|
|
|
|
|
|
val64 = TTI_CMD_MEM_WE | TTI_CMD_MEM_STROBE_NEW_CMD |
|
|
|
|
TTI_CMD_MEM_OFFSET(i);
|
|
|
|
writeq(val64, &bar0->tti_command_mem);
|
|
|
|
|
|
|
|
if (wait_for_cmd_complete(&bar0->tti_command_mem,
|
|
|
|
TTI_CMD_MEM_STROBE_NEW_CMD, S2IO_BIT_RESET) != SUCCESS)
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* init_nic - Initialization of hardware
|
2007-12-17 19:40:15 +00:00
|
|
|
* @nic: device private variable
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: The function sequentially configures every block
|
|
|
|
* of the H/W from their reset values.
|
|
|
|
* Return Value: SUCCESS on success and
|
2005-04-16 22:20:36 +00:00
|
|
|
* '-1' on failure (endian settings incorrect).
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int init_nic(struct s2io_nic *nic)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct net_device *dev = nic->dev;
|
|
|
|
register u64 val64 = 0;
|
|
|
|
void __iomem *add;
|
|
|
|
u32 time;
|
|
|
|
int i, j;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2006-04-21 23:18:03 +00:00
|
|
|
int dtx_cnt = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
unsigned long long mem_share;
|
2005-08-03 19:24:33 +00:00
|
|
|
int mem_size;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
2005-08-03 19:27:09 +00:00
|
|
|
/* to set the swapper controle on the card */
|
2005-08-03 19:24:33 +00:00
|
|
|
if(s2io_set_swapper(nic)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG,"ERROR: Setting Swapper failed\n");
|
2007-11-30 06:46:08 +00:00
|
|
|
return -EIO;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
/*
|
|
|
|
* Herc requires EOI to be removed from reset before XGXS, so..
|
|
|
|
*/
|
|
|
|
if (nic->device_type & XFRAME_II_DEVICE) {
|
|
|
|
val64 = 0xA500000000ULL;
|
|
|
|
writeq(val64, &bar0->sw_reset);
|
|
|
|
msleep(500);
|
|
|
|
val64 = readq(&bar0->sw_reset);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Remove XGXS from reset state */
|
|
|
|
val64 = 0;
|
|
|
|
writeq(val64, &bar0->sw_reset);
|
|
|
|
msleep(500);
|
2005-08-03 19:24:33 +00:00
|
|
|
val64 = readq(&bar0->sw_reset);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-12-06 04:59:28 +00:00
|
|
|
/* Ensure that it's safe to access registers by checking
|
|
|
|
* RIC_RUNNING bit is reset. Check is valid only for XframeII.
|
|
|
|
*/
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE) {
|
|
|
|
for (i = 0; i < 50; i++) {
|
|
|
|
val64 = readq(&bar0->adapter_status);
|
|
|
|
if (!(val64 & ADAPTER_STATUS_RIC_RUNNING))
|
|
|
|
break;
|
|
|
|
msleep(10);
|
|
|
|
}
|
|
|
|
if (i == 50)
|
|
|
|
return -ENODEV;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Enable Receiving broadcasts */
|
|
|
|
add = &bar0->mac_cfg;
|
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
val64 |= MAC_RMAC_BCAST_ENABLE;
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) val64, add);
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64 >> 32), (add + 4));
|
|
|
|
|
|
|
|
/* Read registers in all blocks */
|
|
|
|
val64 = readq(&bar0->mac_int_mask);
|
|
|
|
val64 = readq(&bar0->mc_int_mask);
|
|
|
|
val64 = readq(&bar0->xgxs_int_mask);
|
|
|
|
|
|
|
|
/* Set MTU */
|
|
|
|
val64 = dev->mtu;
|
|
|
|
writeq(vBIT(val64, 2, 14), &bar0->rmac_max_pyld_len);
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
if (nic->device_type & XFRAME_II_DEVICE) {
|
|
|
|
while (herc_act_dtx_cfg[dtx_cnt] != END_SIGN) {
|
2005-08-03 19:41:38 +00:00
|
|
|
SPECIAL_REG_WRITE(herc_act_dtx_cfg[dtx_cnt],
|
2005-04-16 22:20:36 +00:00
|
|
|
&bar0->dtx_control, UF);
|
2005-08-03 19:36:55 +00:00
|
|
|
if (dtx_cnt & 0x1)
|
|
|
|
msleep(1); /* Necessary!! */
|
2005-04-16 22:20:36 +00:00
|
|
|
dtx_cnt++;
|
|
|
|
}
|
2005-08-03 19:36:55 +00:00
|
|
|
} else {
|
2006-04-21 23:18:03 +00:00
|
|
|
while (xena_dtx_cfg[dtx_cnt] != END_SIGN) {
|
|
|
|
SPECIAL_REG_WRITE(xena_dtx_cfg[dtx_cnt],
|
|
|
|
&bar0->dtx_control, UF);
|
|
|
|
val64 = readq(&bar0->dtx_control);
|
|
|
|
dtx_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Tx DMA Initialization */
|
|
|
|
val64 = 0;
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_0);
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_1);
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_2);
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_3);
|
|
|
|
|
|
|
|
|
|
|
|
for (i = 0, j = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
val64 |=
|
2007-12-17 19:40:15 +00:00
|
|
|
vBIT(config->tx_cfg[i].fifo_len - 1, ((j * 32) + 19),
|
2005-04-16 22:20:36 +00:00
|
|
|
13) | vBIT(config->tx_cfg[i].fifo_priority,
|
2007-12-17 19:40:15 +00:00
|
|
|
((j * 32) + 5), 3);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (i == (config->tx_fifo_num - 1)) {
|
|
|
|
if (i % 2 == 0)
|
|
|
|
i++;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (i) {
|
|
|
|
case 1:
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_0);
|
|
|
|
val64 = 0;
|
2007-12-17 19:40:15 +00:00
|
|
|
j = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
case 3:
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_1);
|
|
|
|
val64 = 0;
|
2007-12-17 19:40:15 +00:00
|
|
|
j = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
case 5:
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_2);
|
|
|
|
val64 = 0;
|
2007-12-17 19:40:15 +00:00
|
|
|
j = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
case 7:
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0;
|
|
|
|
j = 0;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
j++;
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:27:09 +00:00
|
|
|
/*
|
|
|
|
* Disable 4 PCCs for Xena1, 2 and 3 as per H/W bug
|
|
|
|
* SXE-008 TRANSMIT DMA ARBITRATION ISSUE.
|
|
|
|
*/
|
2005-08-03 19:36:55 +00:00
|
|
|
if ((nic->device_type == XFRAME_I_DEVICE) &&
|
2007-06-08 22:46:36 +00:00
|
|
|
(nic->pdev->revision < 4))
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(PCC_ENABLE_FOUR, &bar0->pcc_enable);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->tx_fifo_partition_0);
|
|
|
|
DBG_PRINT(INIT_DBG, "Fifo partition at: 0x%p is: 0x%llx\n",
|
|
|
|
&bar0->tx_fifo_partition_0, (unsigned long long) val64);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Initialization of Tx_PA_CONFIG register to ignore packet
|
2005-04-16 22:20:36 +00:00
|
|
|
* integrity checking.
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->tx_pa_cfg);
|
|
|
|
val64 |= TX_PA_CFG_IGNORE_FRM_ERR | TX_PA_CFG_IGNORE_SNAP_OUI |
|
|
|
|
TX_PA_CFG_IGNORE_LLC_CTRL | TX_PA_CFG_IGNORE_L2_ERR;
|
|
|
|
writeq(val64, &bar0->tx_pa_cfg);
|
|
|
|
|
|
|
|
/* Rx DMA intialization. */
|
|
|
|
val64 = 0;
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
val64 |=
|
|
|
|
vBIT(config->rx_cfg[i].ring_priority, (5 + (i * 8)),
|
|
|
|
3);
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->rx_queue_priority);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Allocating equal share of memory to all the
|
2005-04-16 22:20:36 +00:00
|
|
|
* configured Rings.
|
|
|
|
*/
|
|
|
|
val64 = 0;
|
2005-08-03 19:36:55 +00:00
|
|
|
if (nic->device_type & XFRAME_II_DEVICE)
|
|
|
|
mem_size = 32;
|
|
|
|
else
|
|
|
|
mem_size = 64;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
switch (i) {
|
|
|
|
case 0:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num +
|
|
|
|
mem_size % config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q0_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 1:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q1_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 2:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q2_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 3:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q3_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 4:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q4_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 5:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q5_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 6:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q6_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
case 7:
|
2005-08-03 19:24:33 +00:00
|
|
|
mem_share = (mem_size / config->rx_ring_num);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 |= RX_QUEUE_CFG_Q7_SZ(mem_share);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->rx_queue_cfg);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-08-03 19:27:09 +00:00
|
|
|
* Filling Tx round robin registers
|
2007-12-17 19:40:15 +00:00
|
|
|
* as per the number of FIFOs for equal scheduling priority
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2005-08-03 19:27:09 +00:00
|
|
|
switch (config->tx_fifo_num) {
|
|
|
|
case 1:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 2:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001000100010001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001000100000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 3:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020001020001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0200010200010200ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0102000102000102ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020001020001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0200010200000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 4:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020300010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020300000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 5:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020304000102ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0304000102030400ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0102030400010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0400010203040001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0203040000000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 6:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020304050001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0203040500010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0405000102030405ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020304050001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0203040500000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 7:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020304050600ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0102030405060001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0203040506000102ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0304050600010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0405060000000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
case 8:
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020304050607ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->tx_w_round_robin_3);
|
2007-12-17 19:40:15 +00:00
|
|
|
val64 = 0x0001020300000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->tx_w_round_robin_4);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2006-07-24 23:52:49 +00:00
|
|
|
/* Enable all configured Tx FIFO partitions */
|
2006-04-21 23:23:26 +00:00
|
|
|
val64 = readq(&bar0->tx_fifo_partition_0);
|
|
|
|
val64 |= (TX_FIFO_PARTITION_EN);
|
|
|
|
writeq(val64, &bar0->tx_fifo_partition_0);
|
|
|
|
|
2005-08-03 19:27:09 +00:00
|
|
|
/* Filling the Rx round robin registers as per the
|
2008-04-29 01:08:45 +00:00
|
|
|
* number of Rings and steering based on QoS with
|
|
|
|
* equal priority.
|
|
|
|
*/
|
2005-08-03 19:27:09 +00:00
|
|
|
switch (config->rx_ring_num) {
|
|
|
|
case 1:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0;
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
2005-08-03 19:27:09 +00:00
|
|
|
val64 = 0x8080808080808080ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 2:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001000100010001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001000100000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8080808040404040ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 3:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020001020001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0200010200010200ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0102000102000102ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020001020001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0200010200000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8080804040402020ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 4:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020300010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020300000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8080404020201010ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 5:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020304000102ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0304000102030400ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0102030400010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0400010203040001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0203040000000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8080404020201008ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 6:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020304050001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0203040500010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0405000102030405ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020304050001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0203040500000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8080404020100804ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 7:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020304050600ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0102030405060001ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0203040506000102ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0304050600010203ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0405060000000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8080402010080402ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
case 8:
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020304050607ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_0);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_1);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_2);
|
|
|
|
writeq(val64, &bar0->rx_w_round_robin_3);
|
2008-04-29 01:08:45 +00:00
|
|
|
val64 = 0x0001020300000000ULL;
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(val64, &bar0->rx_w_round_robin_4);
|
|
|
|
|
|
|
|
val64 = 0x8040201008040201ULL;
|
|
|
|
writeq(val64, &bar0->rts_qos_steering);
|
|
|
|
break;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* UDP Fix */
|
|
|
|
val64 = 0;
|
2005-08-03 19:24:33 +00:00
|
|
|
for (i = 0; i < 8; i++)
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->rts_frm_len_n[i]);
|
|
|
|
|
2005-08-03 19:27:09 +00:00
|
|
|
/* Set the default rts frame length for the rings configured */
|
|
|
|
val64 = MAC_RTS_FRM_LEN_SET(dev->mtu+22);
|
|
|
|
for (i = 0 ; i < config->rx_ring_num ; i++)
|
|
|
|
writeq(val64, &bar0->rts_frm_len_n[i]);
|
|
|
|
|
|
|
|
/* Set the frame length for the configured rings
|
|
|
|
* desired by the user
|
|
|
|
*/
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
/* If rts_frm_len[i] == 0 then it is assumed that user not
|
|
|
|
* specified frame length steering.
|
|
|
|
* If the user provides the frame length then program
|
|
|
|
* the rts_frm_len register for those values or else
|
|
|
|
* leave it as it is.
|
|
|
|
*/
|
|
|
|
if (rts_frm_len[i] != 0) {
|
|
|
|
writeq(MAC_RTS_FRM_LEN_SET(rts_frm_len[i]),
|
|
|
|
&bar0->rts_frm_len_n[i]);
|
|
|
|
}
|
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
|
2007-02-24 06:57:32 +00:00
|
|
|
/* Disable differentiated services steering logic */
|
|
|
|
for (i = 0; i < 64; i++) {
|
|
|
|
if (rts_ds_steer(nic, i, 0) == FAILURE) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: failed rts ds steering",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "set on codepoint %d\n", i);
|
2007-11-30 06:46:08 +00:00
|
|
|
return -ENODEV;
|
2007-02-24 06:57:32 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/* Program statistics memory */
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(mac_control->stats_mem_phy, &bar0->stat_addr);
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
if (nic->device_type == XFRAME_II_DEVICE) {
|
|
|
|
val64 = STAT_BC(0x320);
|
|
|
|
writeq(val64, &bar0->stat_byte_cnt);
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Initializing the sampling rate for the device to calculate the
|
|
|
|
* bandwidth utilization.
|
|
|
|
*/
|
|
|
|
val64 = MAC_TX_LINK_UTIL_VAL(tmac_util_period) |
|
|
|
|
MAC_RX_LINK_UTIL_VAL(rmac_util_period);
|
|
|
|
writeq(val64, &bar0->mac_link_util);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Initializing the Transmit and Receive Traffic Interrupt
|
2005-04-16 22:20:36 +00:00
|
|
|
* Scheme.
|
|
|
|
*/
|
|
|
|
|
2007-12-17 19:40:15 +00:00
|
|
|
/* Initialize TTI */
|
|
|
|
if (SUCCESS != init_tti(nic, nic->last_link_state))
|
|
|
|
return -ENODEV;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-09-18 22:14:20 +00:00
|
|
|
/* RTI Initialization */
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE) {
|
2005-08-03 19:36:55 +00:00
|
|
|
/*
|
2007-09-18 22:14:20 +00:00
|
|
|
* Programmed to generate Apprx 500 Intrs per
|
|
|
|
* second
|
|
|
|
*/
|
|
|
|
int count = (nic->config.bus_speed * 125)/4;
|
|
|
|
val64 = RTI_DATA1_MEM_RX_TIMER_VAL(count);
|
|
|
|
} else
|
|
|
|
val64 = RTI_DATA1_MEM_RX_TIMER_VAL(0xFFF);
|
|
|
|
val64 |= RTI_DATA1_MEM_RX_URNG_A(0xA) |
|
|
|
|
RTI_DATA1_MEM_RX_URNG_B(0x10) |
|
|
|
|
RTI_DATA1_MEM_RX_URNG_C(0x30) | RTI_DATA1_MEM_RX_TIMER_AC_EN;
|
|
|
|
|
|
|
|
writeq(val64, &bar0->rti_data1_mem);
|
|
|
|
|
|
|
|
val64 = RTI_DATA2_MEM_RX_UFC_A(0x1) |
|
|
|
|
RTI_DATA2_MEM_RX_UFC_B(0x2) ;
|
|
|
|
if (nic->config.intr_type == MSI_X)
|
|
|
|
val64 |= (RTI_DATA2_MEM_RX_UFC_C(0x20) | \
|
|
|
|
RTI_DATA2_MEM_RX_UFC_D(0x40));
|
|
|
|
else
|
|
|
|
val64 |= (RTI_DATA2_MEM_RX_UFC_C(0x40) | \
|
|
|
|
RTI_DATA2_MEM_RX_UFC_D(0x80));
|
|
|
|
writeq(val64, &bar0->rti_data2_mem);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-09-18 22:14:20 +00:00
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
val64 = RTI_CMD_MEM_WE | RTI_CMD_MEM_STROBE_NEW_CMD
|
|
|
|
| RTI_CMD_MEM_OFFSET(i);
|
|
|
|
writeq(val64, &bar0->rti_command_mem);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-09-18 22:14:20 +00:00
|
|
|
/*
|
|
|
|
* Once the operation completes, the Strobe bit of the
|
|
|
|
* command register will be reset. We poll for this
|
|
|
|
* particular condition. We wait for a maximum of 500ms
|
|
|
|
* for the operation to complete, if it's not complete
|
|
|
|
* by then we return error.
|
|
|
|
*/
|
|
|
|
time = 0;
|
|
|
|
while (TRUE) {
|
|
|
|
val64 = readq(&bar0->rti_command_mem);
|
|
|
|
if (!(val64 & RTI_CMD_MEM_STROBE_NEW_CMD))
|
|
|
|
break;
|
2005-08-03 19:38:01 +00:00
|
|
|
|
2007-09-18 22:14:20 +00:00
|
|
|
if (time > 10) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: RTI init Failed\n",
|
|
|
|
dev->name);
|
2007-11-30 06:46:08 +00:00
|
|
|
return -ENODEV;
|
2005-08-03 19:38:01 +00:00
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
time++;
|
|
|
|
msleep(50);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Initializing proper values as Pause threshold into all
|
2005-04-16 22:20:36 +00:00
|
|
|
* the 8 Queues on Rx side.
|
|
|
|
*/
|
|
|
|
writeq(0xffbbffbbffbbffbbULL, &bar0->mc_pause_thresh_q0q3);
|
|
|
|
writeq(0xffbbffbbffbbffbbULL, &bar0->mc_pause_thresh_q4q7);
|
|
|
|
|
|
|
|
/* Disable RMAC PAD STRIPPING */
|
2005-09-05 02:25:58 +00:00
|
|
|
add = &bar0->mac_cfg;
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
val64 &= ~(MAC_CFG_RMAC_STRIP_PAD);
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64), add);
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64 >> 32), (add + 4));
|
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/* Enable FCS stripping by adapter */
|
|
|
|
add = &bar0->mac_cfg;
|
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
val64 |= MAC_CFG_RMAC_STRIP_FCS;
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE)
|
|
|
|
writeq(val64, &bar0->mac_cfg);
|
|
|
|
else {
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64), add);
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64 >> 32), (add + 4));
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Set the time value to be inserted in the pause frame
|
2005-04-16 22:20:36 +00:00
|
|
|
* generated by xena.
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->rmac_pause_cfg);
|
|
|
|
val64 &= ~(RMAC_PAUSE_HG_PTIME(0xffff));
|
|
|
|
val64 |= RMAC_PAUSE_HG_PTIME(nic->mac_control.rmac_pause_time);
|
|
|
|
writeq(val64, &bar0->rmac_pause_cfg);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Set the Threshold Limit for Generating the pause frame
|
|
|
|
* If the amount of data in any Queue exceeds ratio of
|
|
|
|
* (mac_control.mc_pause_threshold_q0q3 or q4q7)/256
|
|
|
|
* pause frame is generated
|
|
|
|
*/
|
|
|
|
val64 = 0;
|
|
|
|
for (i = 0; i < 4; i++) {
|
|
|
|
val64 |=
|
|
|
|
(((u64) 0xFF00 | nic->mac_control.
|
|
|
|
mc_pause_threshold_q0q3)
|
|
|
|
<< (i * 2 * 8));
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->mc_pause_thresh_q0q3);
|
|
|
|
|
|
|
|
val64 = 0;
|
|
|
|
for (i = 0; i < 4; i++) {
|
|
|
|
val64 |=
|
|
|
|
(((u64) 0xFF00 | nic->mac_control.
|
|
|
|
mc_pause_threshold_q4q7)
|
|
|
|
<< (i * 2 * 8));
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->mc_pause_thresh_q4q7);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* TxDMA will stop Read request if the number of read split has
|
2005-04-16 22:20:36 +00:00
|
|
|
* exceeded the limit pointed by shared_splits
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->pic_control);
|
|
|
|
val64 |= PIC_CNTL_SHARED_SPLITS(shared_splits);
|
|
|
|
writeq(val64, &bar0->pic_control);
|
|
|
|
|
2006-04-21 23:03:13 +00:00
|
|
|
if (nic->config.bus_speed == 266) {
|
|
|
|
writeq(TXREQTO_VAL(0x7f) | TXREQTO_EN, &bar0->txreqtimeout);
|
|
|
|
writeq(0x0, &bar0->read_retry_delay);
|
|
|
|
writeq(0x0, &bar0->write_retry_delay);
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
/*
|
|
|
|
* Programming the Herc to split every write transaction
|
|
|
|
* that does not start on an ADB to reduce disconnects.
|
|
|
|
*/
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE) {
|
2007-01-31 18:30:49 +00:00
|
|
|
val64 = FAULT_BEHAVIOUR | EXT_REQ_EN |
|
|
|
|
MISC_LINK_STABILITY_PRD(3);
|
2006-04-21 23:03:13 +00:00
|
|
|
writeq(val64, &bar0->misc_control);
|
|
|
|
val64 = readq(&bar0->pic_control2);
|
2007-10-19 06:40:29 +00:00
|
|
|
val64 &= ~(s2BIT(13)|s2BIT(14)|s2BIT(15));
|
2006-04-21 23:03:13 +00:00
|
|
|
writeq(val64, &bar0->pic_control2);
|
2005-08-03 19:36:55 +00:00
|
|
|
}
|
2006-04-21 23:18:03 +00:00
|
|
|
if (strstr(nic->product_name, "CX4")) {
|
|
|
|
val64 = TMAC_AVG_IPG(0x17);
|
|
|
|
writeq(val64, &bar0->tmac_avg_ipg);
|
2005-08-03 19:38:59 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return SUCCESS;
|
|
|
|
}
|
2005-08-03 19:38:59 +00:00
|
|
|
#define LINK_UP_DOWN_INTERRUPT 1
|
|
|
|
#define MAC_RMAC_ERR_TIMER 2
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_link_fault_indication(struct s2io_nic *nic)
|
2005-08-03 19:38:59 +00:00
|
|
|
{
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE)
|
|
|
|
return LINK_UP_DOWN_INTERRUPT;
|
|
|
|
else
|
|
|
|
return MAC_RMAC_ERR_TIMER;
|
|
|
|
}
|
2007-09-17 20:05:35 +00:00
|
|
|
|
2007-09-06 10:21:54 +00:00
|
|
|
/**
|
|
|
|
* do_s2io_write_bits - update alarm bits in alarm register
|
|
|
|
* @value: alarm bits
|
|
|
|
* @flag: interrupt status
|
|
|
|
* @addr: address value
|
|
|
|
* Description: update alarm bits in alarm register
|
|
|
|
* Return Value:
|
|
|
|
* NONE.
|
|
|
|
*/
|
|
|
|
static void do_s2io_write_bits(u64 value, int flag, void __iomem *addr)
|
|
|
|
{
|
|
|
|
u64 temp64;
|
|
|
|
|
|
|
|
temp64 = readq(addr);
|
|
|
|
|
|
|
|
if(flag == ENABLE_INTRS)
|
|
|
|
temp64 &= ~((u64) value);
|
|
|
|
else
|
|
|
|
temp64 |= ((u64) value);
|
|
|
|
writeq(temp64, addr);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-10-05 19:39:21 +00:00
|
|
|
static void en_dis_err_alarms(struct s2io_nic *nic, u16 mask, int flag)
|
2007-09-06 10:21:54 +00:00
|
|
|
{
|
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
|
|
|
register u64 gen_int_mask = 0;
|
2008-07-10 03:49:21 +00:00
|
|
|
u64 interruptible;
|
2007-09-06 10:21:54 +00:00
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
writeq(DISABLE_ALL_INTRS, &bar0->general_int_mask);
|
2007-09-06 10:21:54 +00:00
|
|
|
if (mask & TX_DMA_INTR) {
|
|
|
|
|
|
|
|
gen_int_mask |= TXDMA_INT_M;
|
|
|
|
|
|
|
|
do_s2io_write_bits(TXDMA_TDA_INT | TXDMA_PFC_INT |
|
|
|
|
TXDMA_PCC_INT | TXDMA_TTI_INT |
|
|
|
|
TXDMA_LSO_INT | TXDMA_TPA_INT |
|
|
|
|
TXDMA_SM_INT, flag, &bar0->txdma_int_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(PFC_ECC_DB_ERR | PFC_SM_ERR_ALARM |
|
|
|
|
PFC_MISC_0_ERR | PFC_MISC_1_ERR |
|
|
|
|
PFC_PCIX_ERR | PFC_ECC_SG_ERR, flag,
|
|
|
|
&bar0->pfc_err_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(TDA_Fn_ECC_DB_ERR | TDA_SM0_ERR_ALARM |
|
|
|
|
TDA_SM1_ERR_ALARM | TDA_Fn_ECC_SG_ERR |
|
|
|
|
TDA_PCIX_ERR, flag, &bar0->tda_err_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(PCC_FB_ECC_DB_ERR | PCC_TXB_ECC_DB_ERR |
|
|
|
|
PCC_SM_ERR_ALARM | PCC_WR_ERR_ALARM |
|
|
|
|
PCC_N_SERR | PCC_6_COF_OV_ERR |
|
|
|
|
PCC_7_COF_OV_ERR | PCC_6_LSO_OV_ERR |
|
|
|
|
PCC_7_LSO_OV_ERR | PCC_FB_ECC_SG_ERR |
|
|
|
|
PCC_TXB_ECC_SG_ERR, flag, &bar0->pcc_err_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(TTI_SM_ERR_ALARM | TTI_ECC_SG_ERR |
|
|
|
|
TTI_ECC_DB_ERR, flag, &bar0->tti_err_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(LSO6_ABORT | LSO7_ABORT |
|
|
|
|
LSO6_SM_ERR_ALARM | LSO7_SM_ERR_ALARM |
|
|
|
|
LSO6_SEND_OFLOW | LSO7_SEND_OFLOW,
|
|
|
|
flag, &bar0->lso_err_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(TPA_SM_ERR_ALARM | TPA_TX_FRM_DROP,
|
|
|
|
flag, &bar0->tpa_err_mask);
|
|
|
|
|
|
|
|
do_s2io_write_bits(SM_SM_ERR_ALARM, flag, &bar0->sm_err_mask);
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mask & TX_MAC_INTR) {
|
|
|
|
gen_int_mask |= TXMAC_INT_M;
|
|
|
|
do_s2io_write_bits(MAC_INT_STATUS_TMAC_INT, flag,
|
|
|
|
&bar0->mac_int_mask);
|
|
|
|
do_s2io_write_bits(TMAC_TX_BUF_OVRN | TMAC_TX_SM_ERR |
|
|
|
|
TMAC_ECC_SG_ERR | TMAC_ECC_DB_ERR |
|
|
|
|
TMAC_DESC_ECC_SG_ERR | TMAC_DESC_ECC_DB_ERR,
|
|
|
|
flag, &bar0->mac_tmac_err_mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mask & TX_XGXS_INTR) {
|
|
|
|
gen_int_mask |= TXXGXS_INT_M;
|
|
|
|
do_s2io_write_bits(XGXS_INT_STATUS_TXGXS, flag,
|
|
|
|
&bar0->xgxs_int_mask);
|
|
|
|
do_s2io_write_bits(TXGXS_ESTORE_UFLOW | TXGXS_TX_SM_ERR |
|
|
|
|
TXGXS_ECC_SG_ERR | TXGXS_ECC_DB_ERR,
|
|
|
|
flag, &bar0->xgxs_txgxs_err_mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mask & RX_DMA_INTR) {
|
|
|
|
gen_int_mask |= RXDMA_INT_M;
|
|
|
|
do_s2io_write_bits(RXDMA_INT_RC_INT_M | RXDMA_INT_RPA_INT_M |
|
|
|
|
RXDMA_INT_RDA_INT_M | RXDMA_INT_RTI_INT_M,
|
|
|
|
flag, &bar0->rxdma_int_mask);
|
|
|
|
do_s2io_write_bits(RC_PRCn_ECC_DB_ERR | RC_FTC_ECC_DB_ERR |
|
|
|
|
RC_PRCn_SM_ERR_ALARM | RC_FTC_SM_ERR_ALARM |
|
|
|
|
RC_PRCn_ECC_SG_ERR | RC_FTC_ECC_SG_ERR |
|
|
|
|
RC_RDA_FAIL_WR_Rn, flag, &bar0->rc_err_mask);
|
|
|
|
do_s2io_write_bits(PRC_PCI_AB_RD_Rn | PRC_PCI_AB_WR_Rn |
|
|
|
|
PRC_PCI_AB_F_WR_Rn | PRC_PCI_DP_RD_Rn |
|
|
|
|
PRC_PCI_DP_WR_Rn | PRC_PCI_DP_F_WR_Rn, flag,
|
|
|
|
&bar0->prc_pcix_err_mask);
|
|
|
|
do_s2io_write_bits(RPA_SM_ERR_ALARM | RPA_CREDIT_ERR |
|
|
|
|
RPA_ECC_SG_ERR | RPA_ECC_DB_ERR, flag,
|
|
|
|
&bar0->rpa_err_mask);
|
|
|
|
do_s2io_write_bits(RDA_RXDn_ECC_DB_ERR | RDA_FRM_ECC_DB_N_AERR |
|
|
|
|
RDA_SM1_ERR_ALARM | RDA_SM0_ERR_ALARM |
|
|
|
|
RDA_RXD_ECC_DB_SERR | RDA_RXDn_ECC_SG_ERR |
|
|
|
|
RDA_FRM_ECC_SG_ERR | RDA_MISC_ERR|RDA_PCIX_ERR,
|
|
|
|
flag, &bar0->rda_err_mask);
|
|
|
|
do_s2io_write_bits(RTI_SM_ERR_ALARM |
|
|
|
|
RTI_ECC_SG_ERR | RTI_ECC_DB_ERR,
|
|
|
|
flag, &bar0->rti_err_mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mask & RX_MAC_INTR) {
|
|
|
|
gen_int_mask |= RXMAC_INT_M;
|
|
|
|
do_s2io_write_bits(MAC_INT_STATUS_RMAC_INT, flag,
|
|
|
|
&bar0->mac_int_mask);
|
2008-07-10 03:49:21 +00:00
|
|
|
interruptible = RMAC_RX_BUFF_OVRN | RMAC_RX_SM_ERR |
|
2007-09-06 10:21:54 +00:00
|
|
|
RMAC_UNUSED_INT | RMAC_SINGLE_ECC_ERR |
|
2008-07-10 03:49:21 +00:00
|
|
|
RMAC_DOUBLE_ECC_ERR;
|
|
|
|
if (s2io_link_fault_indication(nic) == MAC_RMAC_ERR_TIMER)
|
|
|
|
interruptible |= RMAC_LINK_STATE_CHANGE_INT;
|
|
|
|
do_s2io_write_bits(interruptible,
|
2007-09-06 10:21:54 +00:00
|
|
|
flag, &bar0->mac_rmac_err_mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mask & RX_XGXS_INTR)
|
|
|
|
{
|
|
|
|
gen_int_mask |= RXXGXS_INT_M;
|
|
|
|
do_s2io_write_bits(XGXS_INT_STATUS_RXGXS, flag,
|
|
|
|
&bar0->xgxs_int_mask);
|
|
|
|
do_s2io_write_bits(RXGXS_ESTORE_OFLOW | RXGXS_RX_SM_ERR, flag,
|
|
|
|
&bar0->xgxs_rxgxs_err_mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (mask & MC_INTR) {
|
|
|
|
gen_int_mask |= MC_INT_M;
|
|
|
|
do_s2io_write_bits(MC_INT_MASK_MC_INT, flag, &bar0->mc_int_mask);
|
|
|
|
do_s2io_write_bits(MC_ERR_REG_SM_ERR | MC_ERR_REG_ECC_ALL_SNG |
|
|
|
|
MC_ERR_REG_ECC_ALL_DBL | PLL_LOCK_N, flag,
|
|
|
|
&bar0->mc_err_mask);
|
|
|
|
}
|
|
|
|
nic->general_int_mask = gen_int_mask;
|
|
|
|
|
|
|
|
/* Remove this line when alarm interrupts are enabled */
|
|
|
|
nic->general_int_mask = 0;
|
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* en_dis_able_nic_intrs - Enable or Disable the interrupts
|
2005-04-16 22:20:36 +00:00
|
|
|
* @nic: device private variable,
|
|
|
|
* @mask: A mask indicating which Intr block must be modified and,
|
|
|
|
* @flag: A flag indicating whether to enable or disable the Intrs.
|
|
|
|
* Description: This function will either disable or enable the interrupts
|
2005-08-03 19:24:33 +00:00
|
|
|
* depending on the flag argument. The mask argument can be used to
|
|
|
|
* enable/disable any Intr block.
|
2005-04-16 22:20:36 +00:00
|
|
|
* Return Value: NONE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void en_dis_able_nic_intrs(struct s2io_nic *nic, u16 mask, int flag)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2007-09-06 10:21:54 +00:00
|
|
|
register u64 temp64 = 0, intr_mask = 0;
|
|
|
|
|
|
|
|
intr_mask = nic->general_int_mask;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Top level interrupt classification */
|
|
|
|
/* PIC Interrupts */
|
2007-09-06 10:21:54 +00:00
|
|
|
if (mask & TX_PIC_INTR) {
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Enable PIC Intrs in the general intr mask register */
|
2007-09-06 10:21:54 +00:00
|
|
|
intr_mask |= TXPIC_INT_M;
|
2005-04-16 22:20:36 +00:00
|
|
|
if (flag == ENABLE_INTRS) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-08-03 19:38:59 +00:00
|
|
|
* If Hercules adapter enable GPIO otherwise
|
2006-07-24 23:52:49 +00:00
|
|
|
* disable all PCIX, Flash, MDIO, IIC and GPIO
|
2005-08-03 19:24:33 +00:00
|
|
|
* interrupts for now.
|
|
|
|
* TODO
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2005-08-03 19:38:59 +00:00
|
|
|
if (s2io_link_fault_indication(nic) ==
|
|
|
|
LINK_UP_DOWN_INTERRUPT ) {
|
2007-09-06 10:21:54 +00:00
|
|
|
do_s2io_write_bits(PIC_INT_GPIO, flag,
|
|
|
|
&bar0->pic_int_mask);
|
|
|
|
do_s2io_write_bits(GPIO_INT_MASK_LINK_UP, flag,
|
|
|
|
&bar0->gpio_int_mask);
|
|
|
|
} else
|
2005-08-03 19:38:59 +00:00
|
|
|
writeq(DISABLE_ALL_INTRS, &bar0->pic_int_mask);
|
2005-04-16 22:20:36 +00:00
|
|
|
} else if (flag == DISABLE_INTRS) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Disable PIC Intrs in the general
|
|
|
|
* intr mask register
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
writeq(DISABLE_ALL_INTRS, &bar0->pic_int_mask);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Tx traffic interrupts */
|
|
|
|
if (mask & TX_TRAFFIC_INTR) {
|
2007-09-06 10:21:54 +00:00
|
|
|
intr_mask |= TXTRAFFIC_INT_M;
|
2005-04-16 22:20:36 +00:00
|
|
|
if (flag == ENABLE_INTRS) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Enable all the Tx side interrupts
|
2005-08-03 19:24:33 +00:00
|
|
|
* writing 0 Enables all 64 TX interrupt levels
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
writeq(0x0, &bar0->tx_traffic_mask);
|
|
|
|
} else if (flag == DISABLE_INTRS) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Disable Tx Traffic Intrs in the general intr mask
|
2005-04-16 22:20:36 +00:00
|
|
|
* register.
|
|
|
|
*/
|
|
|
|
writeq(DISABLE_ALL_INTRS, &bar0->tx_traffic_mask);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Rx traffic interrupts */
|
|
|
|
if (mask & RX_TRAFFIC_INTR) {
|
2007-09-06 10:21:54 +00:00
|
|
|
intr_mask |= RXTRAFFIC_INT_M;
|
2005-04-16 22:20:36 +00:00
|
|
|
if (flag == ENABLE_INTRS) {
|
|
|
|
/* writing 0 Enables all 8 RX interrupt levels */
|
|
|
|
writeq(0x0, &bar0->rx_traffic_mask);
|
|
|
|
} else if (flag == DISABLE_INTRS) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Disable Rx Traffic Intrs in the general intr mask
|
2005-04-16 22:20:36 +00:00
|
|
|
* register.
|
|
|
|
*/
|
|
|
|
writeq(DISABLE_ALL_INTRS, &bar0->rx_traffic_mask);
|
|
|
|
}
|
|
|
|
}
|
2007-09-06 10:21:54 +00:00
|
|
|
|
|
|
|
temp64 = readq(&bar0->general_int_mask);
|
|
|
|
if (flag == ENABLE_INTRS)
|
|
|
|
temp64 &= ~((u64) intr_mask);
|
|
|
|
else
|
|
|
|
temp64 = DISABLE_ALL_INTRS;
|
|
|
|
writeq(temp64, &bar0->general_int_mask);
|
|
|
|
|
|
|
|
nic->general_int_mask = readq(&bar0->general_int_mask);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
/**
|
|
|
|
* verify_pcc_quiescent- Checks for PCC quiescent state
|
|
|
|
* Return: 1 If PCC is quiescence
|
|
|
|
* 0 If PCC is not quiescence
|
|
|
|
*/
|
2007-01-31 19:09:29 +00:00
|
|
|
static int verify_pcc_quiescent(struct s2io_nic *sp, int flag)
|
2005-08-03 19:24:33 +00:00
|
|
|
{
|
2007-01-31 18:30:49 +00:00
|
|
|
int ret = 0, herc;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2007-01-31 18:30:49 +00:00
|
|
|
u64 val64 = readq(&bar0->adapter_status);
|
2007-09-18 22:14:20 +00:00
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
herc = (sp->device_type == XFRAME_II_DEVICE);
|
2005-08-03 19:24:33 +00:00
|
|
|
|
|
|
|
if (flag == FALSE) {
|
2007-06-08 22:46:36 +00:00
|
|
|
if ((!herc && (sp->pdev->revision >= 4)) || herc) {
|
2007-01-31 18:30:49 +00:00
|
|
|
if (!(val64 & ADAPTER_STATUS_RMAC_PCC_IDLE))
|
2005-08-03 19:27:09 +00:00
|
|
|
ret = 1;
|
2007-01-31 18:30:49 +00:00
|
|
|
} else {
|
|
|
|
if (!(val64 & ADAPTER_STATUS_RMAC_PCC_FOUR_IDLE))
|
2005-08-03 19:27:09 +00:00
|
|
|
ret = 1;
|
2005-08-03 19:24:33 +00:00
|
|
|
}
|
|
|
|
} else {
|
2007-06-08 22:46:36 +00:00
|
|
|
if ((!herc && (sp->pdev->revision >= 4)) || herc) {
|
2005-08-03 19:27:09 +00:00
|
|
|
if (((val64 & ADAPTER_STATUS_RMAC_PCC_IDLE) ==
|
2007-01-31 18:30:49 +00:00
|
|
|
ADAPTER_STATUS_RMAC_PCC_IDLE))
|
2005-08-03 19:27:09 +00:00
|
|
|
ret = 1;
|
|
|
|
} else {
|
|
|
|
if (((val64 & ADAPTER_STATUS_RMAC_PCC_FOUR_IDLE) ==
|
2007-01-31 18:30:49 +00:00
|
|
|
ADAPTER_STATUS_RMAC_PCC_FOUR_IDLE))
|
2005-08-03 19:27:09 +00:00
|
|
|
ret = 1;
|
2005-08-03 19:24:33 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
/**
|
|
|
|
* verify_xena_quiescence - Checks whether the H/W is ready
|
2005-04-16 22:20:36 +00:00
|
|
|
* Description: Returns whether the H/W is ready to go or not. Depending
|
2005-08-03 19:24:33 +00:00
|
|
|
* on whether adapter enable bit was written or not the comparison
|
2005-04-16 22:20:36 +00:00
|
|
|
* differs and the calling function passes the input argument flag to
|
|
|
|
* indicate this.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Return: 1 If xena is quiescence
|
2005-04-16 22:20:36 +00:00
|
|
|
* 0 If Xena is not quiescence
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int verify_xena_quiescence(struct s2io_nic *sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 18:30:49 +00:00
|
|
|
int mode;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2007-01-31 18:30:49 +00:00
|
|
|
u64 val64 = readq(&bar0->adapter_status);
|
|
|
|
mode = s2io_verify_pci_mode(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
if (!(val64 & ADAPTER_STATUS_TDMA_READY)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "TDMA is not ready!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_RDMA_READY)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "RDMA is not ready!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_PFC_READY)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "PFC is not ready!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_TMAC_BUF_EMPTY)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "TMAC BUF is not empty!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_PIC_QUIESCENT)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "PIC is not QUIESCENT!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_MC_DRAM_READY)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "MC_DRAM is not ready!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_MC_QUEUES_READY)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "MC_QUEUES is not ready!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!(val64 & ADAPTER_STATUS_M_PLL_LOCK)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "M_PLL is not locked!");
|
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
/*
|
|
|
|
* In PCI 33 mode, the P_PLL is not used, and therefore,
|
|
|
|
* the the P_PLL_LOCK bit in the adapter_status register will
|
|
|
|
* not be asserted.
|
|
|
|
*/
|
|
|
|
if (!(val64 & ADAPTER_STATUS_P_PLL_LOCK) &&
|
|
|
|
sp->device_type == XFRAME_II_DEVICE && mode !=
|
|
|
|
PCI_MODE_PCI_33) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "P_PLL is not locked!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
if (!((val64 & ADAPTER_STATUS_RC_PRC_QUIESCENT) ==
|
|
|
|
ADAPTER_STATUS_RC_PRC_QUIESCENT)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s", "RC_PRC is not QUIESCENT!");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
return 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* fix_mac_address - Fix for Mac addr problem on Alpha platforms
|
|
|
|
* @sp: Pointer to device specifc structure
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description :
|
2005-04-16 22:20:36 +00:00
|
|
|
* New procedure to clear mac address reading problems on Alpha platforms
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void fix_mac_address(struct s2io_nic * sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64;
|
|
|
|
int i = 0;
|
|
|
|
|
|
|
|
while (fix_mac[i] != END_SIGN) {
|
|
|
|
writeq(fix_mac[i++], &bar0->gpio_control);
|
2005-08-03 19:24:33 +00:00
|
|
|
udelay(10);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* start_nic - Turns the device on
|
2005-04-16 22:20:36 +00:00
|
|
|
* @nic : device private variable.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
|
|
|
* This function actually turns the device on. Before this function is
|
|
|
|
* called,all Registers are configured from their reset states
|
|
|
|
* and shared memory is allocated but the NIC is still quiescent. On
|
2005-04-16 22:20:36 +00:00
|
|
|
* calling this function, the device interrupts are cleared and the NIC is
|
|
|
|
* literally switched on by writing into the adapter control register.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Return Value:
|
2005-04-16 22:20:36 +00:00
|
|
|
* SUCCESS on success and -1 on failure.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int start_nic(struct s2io_nic *nic)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct net_device *dev = nic->dev;
|
|
|
|
register u64 val64 = 0;
|
2005-08-03 19:24:33 +00:00
|
|
|
u16 subid, i;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
|
|
|
|
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
|
|
|
/* PRC Initialization and configuration */
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2005-08-03 19:24:33 +00:00
|
|
|
writeq((u64) mac_control->rings[i].rx_blocks[0].block_dma_addr,
|
2005-04-16 22:20:36 +00:00
|
|
|
&bar0->prc_rxd0_n[i]);
|
|
|
|
|
|
|
|
val64 = readq(&bar0->prc_ctrl_n[i]);
|
2005-10-31 21:55:31 +00:00
|
|
|
if (nic->rxd_mode == RXD_MODE_1)
|
|
|
|
val64 |= PRC_CTRL_RC_ENABLED;
|
|
|
|
else
|
|
|
|
val64 |= PRC_CTRL_RC_ENABLED | PRC_CTRL_RING_MODE_3;
|
2006-04-21 23:03:13 +00:00
|
|
|
if (nic->device_type == XFRAME_II_DEVICE)
|
|
|
|
val64 |= PRC_CTRL_GROUP_READS;
|
|
|
|
val64 &= ~PRC_CTRL_RXD_BACKOFF_INTERVAL(0xFFFFFF);
|
|
|
|
val64 |= PRC_CTRL_RXD_BACKOFF_INTERVAL(0x1000);
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->prc_ctrl_n[i]);
|
|
|
|
}
|
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
if (nic->rxd_mode == RXD_MODE_3B) {
|
|
|
|
/* Enabling 2 buffer mode by writing into Rx_pa_cfg reg. */
|
|
|
|
val64 = readq(&bar0->rx_pa_cfg);
|
|
|
|
val64 |= RX_PA_CFG_IGNORE_L2_ERR;
|
|
|
|
writeq(val64, &bar0->rx_pa_cfg);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-02-24 06:59:39 +00:00
|
|
|
if (vlan_tag_strip == 0) {
|
|
|
|
val64 = readq(&bar0->rx_pa_cfg);
|
|
|
|
val64 &= ~RX_PA_CFG_STRIP_VLAN_TAG;
|
|
|
|
writeq(val64, &bar0->rx_pa_cfg);
|
2008-09-04 20:52:54 +00:00
|
|
|
nic->vlan_strip_flag = 0;
|
2007-02-24 06:59:39 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Enabling MC-RLDRAM. After enabling the device, we timeout
|
|
|
|
* for around 100ms, which is approximately the time required
|
|
|
|
* for the device to be ready for operation.
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->mc_rldram_mrs);
|
|
|
|
val64 |= MC_RLDRAM_QUEUE_SIZE_ENABLE | MC_RLDRAM_MRS_ENABLE;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->mc_rldram_mrs, UF);
|
|
|
|
val64 = readq(&bar0->mc_rldram_mrs);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
msleep(100); /* Delay by around 100 ms. */
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Enabling ECC Protection. */
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 &= ~ADAPTER_ECC_EN;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Verify if the device is ready to be enabled, if so enable
|
2005-04-16 22:20:36 +00:00
|
|
|
* it.
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->adapter_status);
|
2007-01-31 18:30:49 +00:00
|
|
|
if (!verify_xena_quiescence(nic)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: device is not ready, ", dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "Adapter status reads: 0x%llx\n",
|
|
|
|
(unsigned long long) val64);
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* With some switches, link might be already up at this point.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Because of this weird behavior, when we enable laser,
|
|
|
|
* we may not get link. We need to handle this. We cannot
|
|
|
|
* figure out which switch is misbehaving. So we are forced to
|
|
|
|
* make a global change.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
/* Enabling Laser. */
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 |= ADAPTER_EOI_TX_ON;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
|
2006-04-21 23:18:03 +00:00
|
|
|
if (s2io_link_fault_indication(nic) == MAC_RMAC_ERR_TIMER) {
|
|
|
|
/*
|
|
|
|
* Dont see link state interrupts initally on some switches,
|
|
|
|
* so directly scheduling the link state task here.
|
|
|
|
*/
|
|
|
|
schedule_work(&nic->set_link_task);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
/* SXE-002: Initialize link and activity LED */
|
|
|
|
subid = nic->pdev->subsystem_device;
|
2005-08-03 19:36:55 +00:00
|
|
|
if (((subid & 0xFF) >= 0x07) &&
|
|
|
|
(nic->device_type == XFRAME_I_DEVICE)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
val64 |= 0x0000800000000000ULL;
|
|
|
|
writeq(val64, &bar0->gpio_control);
|
|
|
|
val64 = 0x0411040400000000ULL;
|
2005-09-05 02:25:58 +00:00
|
|
|
writeq(val64, (void __iomem *)bar0 + 0x2700);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
2005-11-14 20:25:08 +00:00
|
|
|
/**
|
|
|
|
* s2io_txdl_getskb - Get the skb from txdl, unmap and return skb
|
|
|
|
*/
|
2007-01-31 19:09:29 +00:00
|
|
|
static struct sk_buff *s2io_txdl_getskb(struct fifo_info *fifo_data, struct \
|
|
|
|
TxD *txdlp, int get_off)
|
2005-11-14 20:25:08 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *nic = fifo_data->nic;
|
2005-11-14 20:25:08 +00:00
|
|
|
struct sk_buff *skb;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct TxD *txds;
|
2005-11-14 20:25:08 +00:00
|
|
|
u16 j, frg_cnt;
|
|
|
|
|
|
|
|
txds = txdlp;
|
2008-01-24 10:08:59 +00:00
|
|
|
if (txds->Host_Control == (u64)(long)fifo_data->ufo_in_band_v) {
|
2005-11-14 20:25:08 +00:00
|
|
|
pci_unmap_single(nic->pdev, (dma_addr_t)
|
|
|
|
txds->Buffer_Pointer, sizeof(u64),
|
|
|
|
PCI_DMA_TODEVICE);
|
|
|
|
txds++;
|
|
|
|
}
|
|
|
|
|
|
|
|
skb = (struct sk_buff *) ((unsigned long)
|
|
|
|
txds->Host_Control);
|
|
|
|
if (!skb) {
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(txdlp, 0, (sizeof(struct TxD) * fifo_data->max_txds));
|
2005-11-14 20:25:08 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
pci_unmap_single(nic->pdev, (dma_addr_t)
|
|
|
|
txds->Buffer_Pointer,
|
|
|
|
skb->len - skb->data_len,
|
|
|
|
PCI_DMA_TODEVICE);
|
|
|
|
frg_cnt = skb_shinfo(skb)->nr_frags;
|
|
|
|
if (frg_cnt) {
|
|
|
|
txds++;
|
|
|
|
for (j = 0; j < frg_cnt; j++, txds++) {
|
|
|
|
skb_frag_t *frag = &skb_shinfo(skb)->frags[j];
|
|
|
|
if (!txds->Buffer_Pointer)
|
|
|
|
break;
|
2006-09-13 17:24:59 +00:00
|
|
|
pci_unmap_page(nic->pdev, (dma_addr_t)
|
2005-11-14 20:25:08 +00:00
|
|
|
txds->Buffer_Pointer,
|
|
|
|
frag->size, PCI_DMA_TODEVICE);
|
|
|
|
}
|
|
|
|
}
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(txdlp,0, (sizeof(struct TxD) * fifo_data->max_txds));
|
2005-11-14 20:25:08 +00:00
|
|
|
return(skb);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* free_tx_buffers - Free all queued Tx buffers
|
2005-04-16 22:20:36 +00:00
|
|
|
* @nic : device private variable.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
2005-04-16 22:20:36 +00:00
|
|
|
* Free all queued Tx buffers.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Return Value: void
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
static void free_tx_buffers(struct s2io_nic *nic)
|
|
|
|
{
|
|
|
|
struct net_device *dev = nic->dev;
|
|
|
|
struct sk_buff *skb;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct TxD *txdp;
|
2005-04-16 22:20:36 +00:00
|
|
|
int i, j;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2005-11-14 20:25:08 +00:00
|
|
|
int cnt = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
2008-01-24 10:08:59 +00:00
|
|
|
unsigned long flags;
|
|
|
|
spin_lock_irqsave(&mac_control->fifos[i].tx_lock, flags);
|
2008-04-23 17:28:08 +00:00
|
|
|
for (j = 0; j < config->tx_cfg[i].fifo_len; j++) {
|
2007-05-10 08:22:25 +00:00
|
|
|
txdp = (struct TxD *) \
|
|
|
|
mac_control->fifos[i].list_info[j].list_virt_addr;
|
2005-11-14 20:25:08 +00:00
|
|
|
skb = s2io_txdl_getskb(&mac_control->fifos[i], txdp, j);
|
|
|
|
if (skb) {
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
2007-05-10 08:22:25 +00:00
|
|
|
+= skb->truesize;
|
2005-11-14 20:25:08 +00:00
|
|
|
dev_kfree_skb(skb);
|
|
|
|
cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
DBG_PRINT(INTR_DBG,
|
|
|
|
"%s:forcibly freeing %d skbs on FIFO%d\n",
|
|
|
|
dev->name, cnt, i);
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->fifos[i].tx_curr_get_info.offset = 0;
|
|
|
|
mac_control->fifos[i].tx_curr_put_info.offset = 0;
|
2008-01-24 10:08:59 +00:00
|
|
|
spin_unlock_irqrestore(&mac_control->fifos[i].tx_lock, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* stop_nic - To stop the nic
|
2005-04-16 22:20:36 +00:00
|
|
|
* @nic ; device private variable.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
|
|
|
* This function does exactly the opposite of what the start_nic()
|
2005-04-16 22:20:36 +00:00
|
|
|
* function does. This function is called to stop the device.
|
|
|
|
* Return Value:
|
|
|
|
* void.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void stop_nic(struct s2io_nic *nic)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
register u64 val64 = 0;
|
2006-04-21 23:23:26 +00:00
|
|
|
u16 interruptible;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
|
|
|
|
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
|
|
|
/* Disable all interrupts */
|
2007-09-06 10:21:54 +00:00
|
|
|
en_dis_err_alarms(nic, ENA_ALL_INTRS, DISABLE_INTRS);
|
2005-08-12 17:15:59 +00:00
|
|
|
interruptible = TX_TRAFFIC_INTR | RX_TRAFFIC_INTR;
|
2007-09-06 10:21:54 +00:00
|
|
|
interruptible |= TX_PIC_INTR;
|
2005-04-16 22:20:36 +00:00
|
|
|
en_dis_able_nic_intrs(nic, interruptible, DISABLE_INTRS);
|
|
|
|
|
2006-04-21 23:23:26 +00:00
|
|
|
/* Clearing Adapter_En bit of ADAPTER_CONTROL Register */
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 &= ~(ADAPTER_CNTL_EN);
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* fill_rx_buffers - Allocates the Rx side skbs
|
2008-04-29 01:08:45 +00:00
|
|
|
* @ring_info: per ring structure
|
2008-07-10 03:47:46 +00:00
|
|
|
* @from_card_up: If this is true, we will map the buffer to get
|
|
|
|
* the dma address for buf0 and buf1 to give it to the card.
|
|
|
|
* Else we will sync the already mapped buffer to give it to the card.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
2005-04-16 22:20:36 +00:00
|
|
|
* The function allocates Rx side skbs and puts the physical
|
|
|
|
* address of these buffers into the RxD buffer pointers, so that the NIC
|
|
|
|
* can DMA the received frame into these locations.
|
|
|
|
* The NIC supports 3 receive modes, viz
|
|
|
|
* 1. single buffer,
|
|
|
|
* 2. three buffer and
|
|
|
|
* 3. Five buffer modes.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Each mode defines how many fragments the received frame will be split
|
|
|
|
* up into by the NIC. The frame is split into L3 header, L4 Header,
|
2005-04-16 22:20:36 +00:00
|
|
|
* L4 payload in three buffer mode and in 5 buffer mode, L4 payload itself
|
|
|
|
* is split into 3 fragments. As of now only single buffer mode is
|
|
|
|
* supported.
|
|
|
|
* Return Value:
|
|
|
|
* SUCCESS on success or an appropriate -ve value on failure.
|
|
|
|
*/
|
2008-07-26 02:44:49 +00:00
|
|
|
static int fill_rx_buffers(struct s2io_nic *nic, struct ring_info *ring,
|
|
|
|
int from_card_up)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct sk_buff *skb;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct RxD_t *rxdp;
|
2008-04-29 01:08:45 +00:00
|
|
|
int off, size, block_no, block_no1;
|
2005-04-16 22:20:36 +00:00
|
|
|
u32 alloc_tab = 0;
|
2005-08-03 19:24:33 +00:00
|
|
|
u32 alloc_cnt;
|
|
|
|
u64 tmp;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct buffAdd *ba;
|
|
|
|
struct RxD_t *first_rxdp = NULL;
|
2007-03-07 01:01:00 +00:00
|
|
|
u64 Buffer0_ptr = 0, Buffer1_ptr = 0;
|
2008-04-29 01:08:45 +00:00
|
|
|
int rxd_index = 0;
|
2007-07-23 06:20:51 +00:00
|
|
|
struct RxD1 *rxdp1;
|
|
|
|
struct RxD3 *rxdp3;
|
2008-04-29 01:08:45 +00:00
|
|
|
struct swStat *stats = &ring->nic->mac_control.stats_info->sw_stat;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
alloc_cnt = ring->pkt_cnt - ring->rx_bufs_left;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
block_no1 = ring->rx_curr_get_info.block_index;
|
2005-04-16 22:20:36 +00:00
|
|
|
while (alloc_tab < alloc_cnt) {
|
2008-04-29 01:08:45 +00:00
|
|
|
block_no = ring->rx_curr_put_info.block_index;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
off = ring->rx_curr_put_info.offset;
|
|
|
|
|
|
|
|
rxdp = ring->rx_blocks[block_no].rxds[off].virt_addr;
|
|
|
|
|
|
|
|
rxd_index = off + 1;
|
|
|
|
if (block_no)
|
|
|
|
rxd_index += (block_no * ring->rxd_count);
|
2005-10-31 21:55:31 +00:00
|
|
|
|
2008-05-13 05:41:58 +00:00
|
|
|
if ((block_no == block_no1) &&
|
2008-04-29 01:08:45 +00:00
|
|
|
(off == ring->rx_curr_get_info.offset) &&
|
|
|
|
(rxdp->Host_Control)) {
|
2005-10-31 21:55:31 +00:00
|
|
|
DBG_PRINT(INTR_DBG, "%s: Get and Put",
|
2008-04-29 01:08:45 +00:00
|
|
|
ring->dev->name);
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(INTR_DBG, " info equated\n");
|
|
|
|
goto end;
|
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
if (off && (off == ring->rxd_count)) {
|
|
|
|
ring->rx_curr_put_info.block_index++;
|
|
|
|
if (ring->rx_curr_put_info.block_index ==
|
|
|
|
ring->block_count)
|
|
|
|
ring->rx_curr_put_info.block_index = 0;
|
|
|
|
block_no = ring->rx_curr_put_info.block_index;
|
|
|
|
off = 0;
|
|
|
|
ring->rx_curr_put_info.offset = off;
|
|
|
|
rxdp = ring->rx_blocks[block_no].block_virt_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(INTR_DBG, "%s: Next block at: %p\n",
|
2008-04-29 01:08:45 +00:00
|
|
|
ring->dev->name, rxdp);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-04-23 17:31:33 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
if ((rxdp->Control_1 & RXD_OWN_XENA) &&
|
2008-04-29 01:08:45 +00:00
|
|
|
((ring->rxd_mode == RXD_MODE_3B) &&
|
2007-10-19 06:40:29 +00:00
|
|
|
(rxdp->Control_2 & s2BIT(0)))) {
|
2008-04-29 01:08:45 +00:00
|
|
|
ring->rx_curr_put_info.offset = off;
|
2005-04-16 22:20:36 +00:00
|
|
|
goto end;
|
|
|
|
}
|
2005-10-31 21:55:31 +00:00
|
|
|
/* calculate size of skb based on ring mode */
|
2008-04-29 01:08:45 +00:00
|
|
|
size = ring->mtu + HEADER_ETHERNET_II_802_3_SIZE +
|
2005-10-31 21:55:31 +00:00
|
|
|
HEADER_802_2_SIZE + HEADER_SNAP_SIZE;
|
2008-04-29 01:08:45 +00:00
|
|
|
if (ring->rxd_mode == RXD_MODE_1)
|
2005-10-31 21:55:31 +00:00
|
|
|
size += NET_IP_ALIGN;
|
|
|
|
else
|
2008-04-29 01:08:45 +00:00
|
|
|
size = ring->mtu + ALIGN_SIZE + BUF0_LEN + 4;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
/* allocate skb */
|
|
|
|
skb = dev_alloc_skb(size);
|
|
|
|
if(!skb) {
|
2008-04-29 01:08:45 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: Out of ", ring->dev->name);
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "memory to allocate SKBs\n");
|
2005-08-03 19:41:38 +00:00
|
|
|
if (first_rxdp) {
|
|
|
|
wmb();
|
|
|
|
first_rxdp->Control_1 |= RXD_OWN_XENA;
|
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
stats->mem_alloc_fail_cnt++;
|
2008-05-13 05:41:58 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
return -ENOMEM ;
|
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
stats->mem_allocated += skb->truesize;
|
|
|
|
|
|
|
|
if (ring->rxd_mode == RXD_MODE_1) {
|
2005-10-31 21:55:31 +00:00
|
|
|
/* 1 buffer mode - normal operation mode */
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1 = (struct RxD1*)rxdp;
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(rxdp, 0, sizeof(struct RxD1));
|
2005-10-31 21:55:31 +00:00
|
|
|
skb_reserve(skb, NET_IP_ALIGN);
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1->Buffer0_ptr = pci_map_single
|
2008-04-29 01:08:45 +00:00
|
|
|
(ring->pdev, skb->data, size - NET_IP_ALIGN,
|
2006-04-21 23:03:13 +00:00
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(nic->pdev,
|
|
|
|
rxdp1->Buffer0_ptr))
|
2007-07-23 06:37:14 +00:00
|
|
|
goto pci_map_failed;
|
|
|
|
|
2007-09-18 22:14:20 +00:00
|
|
|
rxdp->Control_2 =
|
2007-05-10 08:22:25 +00:00
|
|
|
SET_BUFFER0_SIZE_1(size - NET_IP_ALIGN);
|
2008-04-29 01:08:45 +00:00
|
|
|
rxdp->Host_Control = (unsigned long) (skb);
|
|
|
|
} else if (ring->rxd_mode == RXD_MODE_3B) {
|
2005-10-31 21:55:31 +00:00
|
|
|
/*
|
2007-07-23 06:20:51 +00:00
|
|
|
* 2 buffer mode -
|
|
|
|
* 2 buffer mode provides 128
|
2005-10-31 21:55:31 +00:00
|
|
|
* byte aligned receive buffers.
|
|
|
|
*/
|
|
|
|
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3 = (struct RxD3*)rxdp;
|
2007-05-10 08:22:25 +00:00
|
|
|
/* save buffer pointers to avoid frequent dma mapping */
|
2007-07-23 06:20:51 +00:00
|
|
|
Buffer0_ptr = rxdp3->Buffer0_ptr;
|
|
|
|
Buffer1_ptr = rxdp3->Buffer1_ptr;
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(rxdp, 0, sizeof(struct RxD3));
|
2007-03-07 01:01:00 +00:00
|
|
|
/* restore the buffer pointers for dma sync*/
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer0_ptr = Buffer0_ptr;
|
|
|
|
rxdp3->Buffer1_ptr = Buffer1_ptr;
|
2007-03-07 01:01:00 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
ba = &ring->ba[block_no][off];
|
2005-10-31 21:55:31 +00:00
|
|
|
skb_reserve(skb, BUF0_LEN);
|
|
|
|
tmp = (u64)(unsigned long) skb->data;
|
|
|
|
tmp += ALIGN_SIZE;
|
|
|
|
tmp &= ~ALIGN_SIZE;
|
|
|
|
skb->data = (void *) (unsigned long)tmp;
|
2007-04-20 03:29:13 +00:00
|
|
|
skb_reset_tail_pointer(skb);
|
2005-10-31 21:55:31 +00:00
|
|
|
|
2008-07-10 03:47:46 +00:00
|
|
|
if (from_card_up) {
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer0_ptr =
|
2008-04-29 01:08:45 +00:00
|
|
|
pci_map_single(ring->pdev, ba->ba_0,
|
|
|
|
BUF0_LEN, PCI_DMA_FROMDEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(nic->pdev,
|
|
|
|
rxdp3->Buffer0_ptr))
|
2008-07-10 03:47:46 +00:00
|
|
|
goto pci_map_failed;
|
|
|
|
} else
|
2008-04-29 01:08:45 +00:00
|
|
|
pci_dma_sync_single_for_device(ring->pdev,
|
2007-07-23 06:20:51 +00:00
|
|
|
(dma_addr_t) rxdp3->Buffer0_ptr,
|
2006-07-24 23:55:09 +00:00
|
|
|
BUF0_LEN, PCI_DMA_FROMDEVICE);
|
2007-07-23 06:37:14 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
rxdp->Control_2 = SET_BUFFER0_SIZE_3(BUF0_LEN);
|
2008-04-29 01:08:45 +00:00
|
|
|
if (ring->rxd_mode == RXD_MODE_3B) {
|
2005-10-31 21:55:31 +00:00
|
|
|
/* Two buffer mode */
|
|
|
|
|
|
|
|
/*
|
2006-09-13 17:24:59 +00:00
|
|
|
* Buffer2 will have L3/L4 header plus
|
2005-10-31 21:55:31 +00:00
|
|
|
* L4 payload
|
|
|
|
*/
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer2_ptr = pci_map_single
|
2008-04-29 01:08:45 +00:00
|
|
|
(ring->pdev, skb->data, ring->mtu + 4,
|
2005-10-31 21:55:31 +00:00
|
|
|
PCI_DMA_FROMDEVICE);
|
|
|
|
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(nic->pdev,
|
|
|
|
rxdp3->Buffer2_ptr))
|
2007-07-23 06:37:14 +00:00
|
|
|
goto pci_map_failed;
|
|
|
|
|
2008-07-10 03:47:46 +00:00
|
|
|
if (from_card_up) {
|
2008-04-29 01:08:45 +00:00
|
|
|
rxdp3->Buffer1_ptr =
|
|
|
|
pci_map_single(ring->pdev,
|
2006-07-24 23:55:09 +00:00
|
|
|
ba->ba_1, BUF1_LEN,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-04-29 01:08:45 +00:00
|
|
|
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(nic->pdev,
|
|
|
|
rxdp3->Buffer1_ptr)) {
|
2008-07-10 03:47:46 +00:00
|
|
|
pci_unmap_single
|
|
|
|
(ring->pdev,
|
|
|
|
(dma_addr_t)(unsigned long)
|
|
|
|
skb->data,
|
|
|
|
ring->mtu + 4,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
|
|
|
goto pci_map_failed;
|
|
|
|
}
|
2006-07-24 23:55:09 +00:00
|
|
|
}
|
2005-10-31 21:55:31 +00:00
|
|
|
rxdp->Control_2 |= SET_BUFFER1_SIZE_3(1);
|
|
|
|
rxdp->Control_2 |= SET_BUFFER2_SIZE_3
|
2008-04-29 01:08:45 +00:00
|
|
|
(ring->mtu + 4);
|
2005-10-31 21:55:31 +00:00
|
|
|
}
|
2007-10-19 06:40:29 +00:00
|
|
|
rxdp->Control_2 |= s2BIT(0);
|
2008-04-29 01:08:45 +00:00
|
|
|
rxdp->Host_Control = (unsigned long) (skb);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-08-03 19:41:38 +00:00
|
|
|
if (alloc_tab & ((1 << rxsync_frequency) - 1))
|
|
|
|
rxdp->Control_1 |= RXD_OWN_XENA;
|
2005-04-16 22:20:36 +00:00
|
|
|
off++;
|
2008-04-29 01:08:45 +00:00
|
|
|
if (off == (ring->rxd_count + 1))
|
2005-10-31 21:55:31 +00:00
|
|
|
off = 0;
|
2008-04-29 01:08:45 +00:00
|
|
|
ring->rx_curr_put_info.offset = off;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
rxdp->Control_2 |= SET_RXD_MARKER;
|
2005-08-03 19:41:38 +00:00
|
|
|
if (!(alloc_tab & ((1 << rxsync_frequency) - 1))) {
|
|
|
|
if (first_rxdp) {
|
|
|
|
wmb();
|
|
|
|
first_rxdp->Control_1 |= RXD_OWN_XENA;
|
|
|
|
}
|
|
|
|
first_rxdp = rxdp;
|
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
ring->rx_bufs_left += 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
alloc_tab++;
|
|
|
|
}
|
|
|
|
|
|
|
|
end:
|
2005-08-03 19:41:38 +00:00
|
|
|
/* Transfer ownership of first descriptor to adapter just before
|
|
|
|
* exiting. Before that, use memory barrier so that ownership
|
|
|
|
* and other fields are seen by adapter correctly.
|
|
|
|
*/
|
|
|
|
if (first_rxdp) {
|
|
|
|
wmb();
|
|
|
|
first_rxdp->Control_1 |= RXD_OWN_XENA;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return SUCCESS;
|
2007-07-23 06:37:14 +00:00
|
|
|
pci_map_failed:
|
|
|
|
stats->pci_map_fail_cnt++;
|
|
|
|
stats->mem_freed += skb->truesize;
|
|
|
|
dev_kfree_skb_irq(skb);
|
|
|
|
return -ENOMEM;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
static void free_rxd_blk(struct s2io_nic *sp, int ring_no, int blk)
|
|
|
|
{
|
|
|
|
struct net_device *dev = sp->dev;
|
|
|
|
int j;
|
|
|
|
struct sk_buff *skb;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct RxD_t *rxdp;
|
|
|
|
struct mac_info *mac_control;
|
|
|
|
struct buffAdd *ba;
|
2007-07-23 06:20:51 +00:00
|
|
|
struct RxD1 *rxdp1;
|
|
|
|
struct RxD3 *rxdp3;
|
2005-10-31 21:55:31 +00:00
|
|
|
|
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
for (j = 0 ; j < rxd_count[sp->rxd_mode]; j++) {
|
|
|
|
rxdp = mac_control->rings[ring_no].
|
|
|
|
rx_blocks[blk].rxds[j].virt_addr;
|
|
|
|
skb = (struct sk_buff *)
|
|
|
|
((unsigned long) rxdp->Host_Control);
|
|
|
|
if (!skb) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (sp->rxd_mode == RXD_MODE_1) {
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1 = (struct RxD1*)rxdp;
|
2005-10-31 21:55:31 +00:00
|
|
|
pci_unmap_single(sp->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1->Buffer0_ptr,
|
|
|
|
dev->mtu +
|
|
|
|
HEADER_ETHERNET_II_802_3_SIZE
|
|
|
|
+ HEADER_802_2_SIZE +
|
|
|
|
HEADER_SNAP_SIZE,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(rxdp, 0, sizeof(struct RxD1));
|
2005-10-31 21:55:31 +00:00
|
|
|
} else if(sp->rxd_mode == RXD_MODE_3B) {
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3 = (struct RxD3*)rxdp;
|
2005-10-31 21:55:31 +00:00
|
|
|
ba = &mac_control->rings[ring_no].
|
|
|
|
ba[blk][j];
|
|
|
|
pci_unmap_single(sp->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer0_ptr,
|
|
|
|
BUF0_LEN,
|
2005-10-31 21:55:31 +00:00
|
|
|
PCI_DMA_FROMDEVICE);
|
|
|
|
pci_unmap_single(sp->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer1_ptr,
|
|
|
|
BUF1_LEN,
|
2005-10-31 21:55:31 +00:00
|
|
|
PCI_DMA_FROMDEVICE);
|
|
|
|
pci_unmap_single(sp->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer2_ptr,
|
|
|
|
dev->mtu + 4,
|
2005-10-31 21:55:31 +00:00
|
|
|
PCI_DMA_FROMDEVICE);
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(rxdp, 0, sizeof(struct RxD3));
|
2005-10-31 21:55:31 +00:00
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_freed += skb->truesize;
|
2005-10-31 21:55:31 +00:00
|
|
|
dev_kfree_skb(skb);
|
2008-04-29 01:08:45 +00:00
|
|
|
mac_control->rings[ring_no].rx_bufs_left -= 1;
|
2005-10-31 21:55:31 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* free_rx_buffers - Frees all Rx buffers
|
2005-04-16 22:20:36 +00:00
|
|
|
* @sp: device private variable.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
2005-04-16 22:20:36 +00:00
|
|
|
* This function will free all Rx buffers allocated by host.
|
|
|
|
* Return Value:
|
|
|
|
* NONE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void free_rx_buffers(struct s2io_nic *sp)
|
|
|
|
{
|
|
|
|
struct net_device *dev = sp->dev;
|
2005-10-31 21:55:31 +00:00
|
|
|
int i, blk = 0, buf_cnt = 0;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
|
|
|
|
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
config = &sp->config;
|
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2005-10-31 21:55:31 +00:00
|
|
|
for (blk = 0; blk < rx_ring_sz[i]; blk++)
|
|
|
|
free_rxd_blk(sp,i,blk);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
mac_control->rings[i].rx_curr_put_info.block_index = 0;
|
|
|
|
mac_control->rings[i].rx_curr_get_info.block_index = 0;
|
|
|
|
mac_control->rings[i].rx_curr_put_info.offset = 0;
|
|
|
|
mac_control->rings[i].rx_curr_get_info.offset = 0;
|
2008-04-29 01:08:45 +00:00
|
|
|
mac_control->rings[i].rx_bufs_left = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(INIT_DBG, "%s:Freed 0x%x Rx Buffers on ring%d\n",
|
|
|
|
dev->name, buf_cnt, i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-07-26 02:44:49 +00:00
|
|
|
static int s2io_chk_rx_buffers(struct s2io_nic *nic, struct ring_info *ring)
|
2008-05-12 17:42:17 +00:00
|
|
|
{
|
2008-07-26 02:44:49 +00:00
|
|
|
if (fill_rx_buffers(nic, ring, 0) == -ENOMEM) {
|
2008-05-12 17:42:17 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s:Out of memory", ring->dev->name);
|
|
|
|
DBG_PRINT(INFO_DBG, " in Rx Intr!!\n");
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* s2io_poll - Rx interrupt handler for NAPI support
|
[NET]: Make NAPI polling independent of struct net_device objects.
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-10-03 23:41:36 +00:00
|
|
|
* @napi : pointer to the napi structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @budget : The number of packets that were budgeted to be processed
|
2005-04-16 22:20:36 +00:00
|
|
|
* during one pass through the 'Poll" function.
|
|
|
|
* Description:
|
|
|
|
* Comes into picture only if NAPI support has been incorporated. It does
|
|
|
|
* the same thing that rx_intr_handler does, but not in a interrupt context
|
|
|
|
* also It will process only a given number of packets.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success and 1 if there are No Rx packets to be processed.
|
|
|
|
*/
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
static int s2io_poll_msix(struct napi_struct *napi, int budget)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-05-12 17:42:17 +00:00
|
|
|
struct ring_info *ring = container_of(napi, struct ring_info, napi);
|
|
|
|
struct net_device *dev = ring->dev;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2008-05-12 17:42:17 +00:00
|
|
|
struct mac_info *mac_control;
|
|
|
|
int pkts_processed = 0;
|
2008-06-02 09:59:02 +00:00
|
|
|
u8 __iomem *addr = NULL;
|
|
|
|
u8 val8 = 0;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *nic = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2008-05-12 17:42:17 +00:00
|
|
|
int budget_org = budget;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
config = &nic->config;
|
2008-05-12 17:42:17 +00:00
|
|
|
mac_control = &nic->mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (unlikely(!is_s2io_card_up(nic)))
|
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
pkts_processed = rx_intr_handler(ring, budget);
|
2008-07-26 02:44:49 +00:00
|
|
|
s2io_chk_rx_buffers(nic, ring);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (pkts_processed < budget_org) {
|
2009-01-20 00:43:59 +00:00
|
|
|
napi_complete(napi);
|
2008-05-12 17:42:17 +00:00
|
|
|
/*Re Enable MSI-Rx Vector*/
|
2008-06-02 09:59:02 +00:00
|
|
|
addr = (u8 __iomem *)&bar0->xmsi_mask_reg;
|
2008-05-12 17:42:17 +00:00
|
|
|
addr += 7 - ring->ring_no;
|
|
|
|
val8 = (ring->ring_no == 0) ? 0x3f : 0xbf;
|
|
|
|
writeb(val8, addr);
|
|
|
|
val8 = readb(addr);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-05-12 17:42:17 +00:00
|
|
|
return pkts_processed;
|
|
|
|
}
|
|
|
|
static int s2io_poll_inta(struct napi_struct *napi, int budget)
|
|
|
|
{
|
|
|
|
struct s2io_nic *nic = container_of(napi, struct s2io_nic, napi);
|
|
|
|
struct ring_info *ring;
|
|
|
|
struct config_param *config;
|
|
|
|
struct mac_info *mac_control;
|
|
|
|
int pkts_processed = 0;
|
|
|
|
int ring_pkts_processed, i;
|
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
|
|
|
int budget_org = budget;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
config = &nic->config;
|
|
|
|
mac_control = &nic->mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (unlikely(!is_s2io_card_up(nic)))
|
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2008-05-12 17:42:17 +00:00
|
|
|
ring = &mac_control->rings[i];
|
|
|
|
ring_pkts_processed = rx_intr_handler(ring, budget);
|
2008-07-26 02:44:49 +00:00
|
|
|
s2io_chk_rx_buffers(nic, ring);
|
2008-05-12 17:42:17 +00:00
|
|
|
pkts_processed += ring_pkts_processed;
|
|
|
|
budget -= ring_pkts_processed;
|
|
|
|
if (budget <= 0)
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
}
|
2008-05-12 17:42:17 +00:00
|
|
|
if (pkts_processed < budget_org) {
|
2009-01-20 00:43:59 +00:00
|
|
|
napi_complete(napi);
|
2008-05-12 17:42:17 +00:00
|
|
|
/* Re enable the Rx interrupts for the ring */
|
|
|
|
writeq(0, &bar0->rx_traffic_mask);
|
|
|
|
readl(&bar0->rx_traffic_mask);
|
|
|
|
}
|
|
|
|
return pkts_processed;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2006-07-24 23:52:49 +00:00
|
|
|
#ifdef CONFIG_NET_POLL_CONTROLLER
|
2006-06-15 18:36:36 +00:00
|
|
|
/**
|
2006-07-24 23:52:49 +00:00
|
|
|
* s2io_netpoll - netpoll event handler entry point
|
2006-06-15 18:36:36 +00:00
|
|
|
* @dev : pointer to the device structure.
|
|
|
|
* Description:
|
2006-07-24 23:52:49 +00:00
|
|
|
* This function will be called by upper layer to check for events on the
|
|
|
|
* interface in situations where interrupts are disabled. It is used for
|
|
|
|
* specific in-kernel networking tasks, such as remote consoles and kernel
|
|
|
|
* debugging over the network (example netdump in RedHat).
|
2006-06-15 18:36:36 +00:00
|
|
|
*/
|
|
|
|
static void s2io_netpoll(struct net_device *dev)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *nic = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2006-06-15 18:36:36 +00:00
|
|
|
struct config_param *config;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2006-07-24 23:52:49 +00:00
|
|
|
u64 val64 = 0xFFFFFFFFFFFFFFFFULL;
|
2006-06-15 18:36:36 +00:00
|
|
|
int i;
|
|
|
|
|
2007-05-14 23:37:30 +00:00
|
|
|
if (pci_channel_offline(nic->pdev))
|
|
|
|
return;
|
|
|
|
|
2006-06-15 18:36:36 +00:00
|
|
|
disable_irq(dev->irq);
|
|
|
|
|
|
|
|
mac_control = &nic->mac_control;
|
|
|
|
config = &nic->config;
|
|
|
|
|
|
|
|
writeq(val64, &bar0->rx_traffic_int);
|
2006-07-24 23:52:49 +00:00
|
|
|
writeq(val64, &bar0->tx_traffic_int);
|
|
|
|
|
2006-09-13 17:24:59 +00:00
|
|
|
/* we need to free up the transmitted skbufs or else netpoll will
|
2006-07-24 23:52:49 +00:00
|
|
|
* run out of skbs and will fail and eventually netpoll application such
|
|
|
|
* as netdump will fail.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++)
|
|
|
|
tx_intr_handler(&mac_control->fifos[i]);
|
2006-06-15 18:36:36 +00:00
|
|
|
|
2006-07-24 23:52:49 +00:00
|
|
|
/* check for received packet and indicate up to network */
|
2006-06-15 18:36:36 +00:00
|
|
|
for (i = 0; i < config->rx_ring_num; i++)
|
2008-05-12 17:42:17 +00:00
|
|
|
rx_intr_handler(&mac_control->rings[i], 0);
|
2006-06-15 18:36:36 +00:00
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2008-07-26 02:44:49 +00:00
|
|
|
if (fill_rx_buffers(nic, &mac_control->rings[i], 0) ==
|
|
|
|
-ENOMEM) {
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s:Out of memory", dev->name);
|
|
|
|
DBG_PRINT(INFO_DBG, " in Rx Netpoll!!\n");
|
2006-06-15 18:36:36 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
enable_irq(dev->irq);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
2005-04-16 22:20:36 +00:00
|
|
|
* rx_intr_handler - Rx interrupt handler
|
2008-05-12 17:42:17 +00:00
|
|
|
* @ring_info: per ring structure.
|
|
|
|
* @budget: budget for napi processing.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
|
|
|
* If the interrupt is because of a received frame or if the
|
2005-04-16 22:20:36 +00:00
|
|
|
* receive ring contains fresh as yet un-processed frames,this function is
|
2005-08-03 19:24:33 +00:00
|
|
|
* called. It picks out the RxD at which place the last Rx processing had
|
|
|
|
* stopped and sends the skb to the OSM's Rx handler and then increments
|
2005-04-16 22:20:36 +00:00
|
|
|
* the offset.
|
|
|
|
* Return Value:
|
2008-05-12 17:42:17 +00:00
|
|
|
* No. of napi packets processed.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2008-05-12 17:42:17 +00:00
|
|
|
static int rx_intr_handler(struct ring_info *ring_data, int budget)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-04-23 17:31:33 +00:00
|
|
|
int get_block, put_block;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct rx_curr_get_info get_info, put_info;
|
|
|
|
struct RxD_t *rxdp;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct sk_buff *skb;
|
2008-05-12 17:42:17 +00:00
|
|
|
int pkt_cnt = 0, napi_pkts = 0;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
int i;
|
2007-07-23 06:20:51 +00:00
|
|
|
struct RxD1* rxdp1;
|
|
|
|
struct RxD3* rxdp3;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
get_info = ring_data->rx_curr_get_info;
|
|
|
|
get_block = get_info.block_index;
|
2007-01-31 19:09:29 +00:00
|
|
|
memcpy(&put_info, &ring_data->rx_curr_put_info, sizeof(put_info));
|
2005-08-03 19:24:33 +00:00
|
|
|
put_block = put_info.block_index;
|
2005-10-31 21:55:31 +00:00
|
|
|
rxdp = ring_data->rx_blocks[get_block].rxds[get_info.offset].virt_addr;
|
2007-01-31 18:28:08 +00:00
|
|
|
|
2005-10-31 21:55:31 +00:00
|
|
|
while (RXD_IS_UP2DT(rxdp)) {
|
2007-01-31 18:28:08 +00:00
|
|
|
/*
|
|
|
|
* If your are next to put index then it's
|
|
|
|
* FIFO full condition
|
|
|
|
*/
|
2005-10-31 21:55:31 +00:00
|
|
|
if ((get_block == put_block) &&
|
|
|
|
(get_info.offset + 1) == put_info.offset) {
|
2008-04-29 01:08:45 +00:00
|
|
|
DBG_PRINT(INTR_DBG, "%s: Ring Full\n",
|
|
|
|
ring_data->dev->name);
|
2005-10-31 21:55:31 +00:00
|
|
|
break;
|
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
skb = (struct sk_buff *) ((unsigned long)rxdp->Host_Control);
|
|
|
|
if (skb == NULL) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: The skb is ",
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->dev->name);
|
2005-08-03 19:24:33 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "Null in Rx Intr\n");
|
2008-05-12 17:42:17 +00:00
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
if (ring_data->rxd_mode == RXD_MODE_1) {
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1 = (struct RxD1*)rxdp;
|
2008-04-29 01:08:45 +00:00
|
|
|
pci_unmap_single(ring_data->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1->Buffer0_ptr,
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->mtu +
|
2007-07-23 06:20:51 +00:00
|
|
|
HEADER_ETHERNET_II_802_3_SIZE +
|
|
|
|
HEADER_802_2_SIZE +
|
|
|
|
HEADER_SNAP_SIZE,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-04-29 01:08:45 +00:00
|
|
|
} else if (ring_data->rxd_mode == RXD_MODE_3B) {
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3 = (struct RxD3*)rxdp;
|
2008-04-29 01:08:45 +00:00
|
|
|
pci_dma_sync_single_for_cpu(ring_data->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer0_ptr,
|
|
|
|
BUF0_LEN, PCI_DMA_FROMDEVICE);
|
2008-04-29 01:08:45 +00:00
|
|
|
pci_unmap_single(ring_data->pdev, (dma_addr_t)
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer2_ptr,
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->mtu + 4,
|
2007-07-23 06:20:51 +00:00
|
|
|
PCI_DMA_FROMDEVICE);
|
2005-10-31 21:55:31 +00:00
|
|
|
}
|
2006-04-21 23:03:13 +00:00
|
|
|
prefetch(skb->data);
|
2005-08-03 19:24:33 +00:00
|
|
|
rx_osm_handler(ring_data, rxdp);
|
|
|
|
get_info.offset++;
|
2005-10-31 21:55:31 +00:00
|
|
|
ring_data->rx_curr_get_info.offset = get_info.offset;
|
|
|
|
rxdp = ring_data->rx_blocks[get_block].
|
|
|
|
rxds[get_info.offset].virt_addr;
|
2008-04-29 01:08:45 +00:00
|
|
|
if (get_info.offset == rxd_count[ring_data->rxd_mode]) {
|
2005-08-03 19:24:33 +00:00
|
|
|
get_info.offset = 0;
|
2005-10-31 21:55:31 +00:00
|
|
|
ring_data->rx_curr_get_info.offset = get_info.offset;
|
2005-08-03 19:24:33 +00:00
|
|
|
get_block++;
|
2005-10-31 21:55:31 +00:00
|
|
|
if (get_block == ring_data->block_count)
|
|
|
|
get_block = 0;
|
|
|
|
ring_data->rx_curr_get_info.block_index = get_block;
|
2005-08-03 19:24:33 +00:00
|
|
|
rxdp = ring_data->rx_blocks[get_block].block_virt_addr;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (ring_data->nic->config.napi) {
|
|
|
|
budget--;
|
|
|
|
napi_pkts++;
|
|
|
|
if (!budget)
|
2008-04-29 01:08:45 +00:00
|
|
|
break;
|
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
pkt_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
if ((indicate_max_pkts) && (pkt_cnt > indicate_max_pkts))
|
|
|
|
break;
|
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
if (ring_data->lro) {
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/* Clear all LRO sessions before exiting */
|
|
|
|
for (i=0; i<MAX_LRO_SESSIONS; i++) {
|
2008-04-29 01:08:45 +00:00
|
|
|
struct lro *lro = &ring_data->lro0_n[i];
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (lro->in_use) {
|
2008-04-29 01:08:45 +00:00
|
|
|
update_L3L4_header(ring_data->nic, lro);
|
2008-02-20 22:09:15 +00:00
|
|
|
queue_rx_frame(lro->parent, lro->vlan_tag);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
clear_lro_session(lro);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2008-05-12 17:42:17 +00:00
|
|
|
return(napi_pkts);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
|
|
|
|
/**
|
2005-04-16 22:20:36 +00:00
|
|
|
* tx_intr_handler - Transmit interrupt handler
|
|
|
|
* @nic : device private variable
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
|
|
|
* If an interrupt was raised to indicate DMA complete of the
|
|
|
|
* Tx packet, this function is called. It identifies the last TxD
|
|
|
|
* whose buffer was freed and frees all skbs whose data have already
|
2005-04-16 22:20:36 +00:00
|
|
|
* DMA'ed into the NICs internal memory.
|
|
|
|
* Return Value:
|
|
|
|
* NONE
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void tx_intr_handler(struct fifo_info *fifo_data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *nic = fifo_data->nic;
|
|
|
|
struct tx_curr_get_info get_info, put_info;
|
2008-02-20 21:44:07 +00:00
|
|
|
struct sk_buff *skb = NULL;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct TxD *txdlp;
|
2008-02-20 21:44:07 +00:00
|
|
|
int pkt_cnt = 0;
|
2008-01-24 10:08:59 +00:00
|
|
|
unsigned long flags = 0;
|
2007-06-19 20:41:10 +00:00
|
|
|
u8 err_mask;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
if (!spin_trylock_irqsave(&fifo_data->tx_lock, flags))
|
|
|
|
return;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
get_info = fifo_data->tx_curr_get_info;
|
2007-01-31 19:09:29 +00:00
|
|
|
memcpy(&put_info, &fifo_data->tx_curr_put_info, sizeof(put_info));
|
|
|
|
txdlp = (struct TxD *) fifo_data->list_info[get_info.offset].
|
2005-08-03 19:24:33 +00:00
|
|
|
list_virt_addr;
|
|
|
|
while ((!(txdlp->Control_1 & TXD_LIST_OWN_XENA)) &&
|
|
|
|
(get_info.offset != put_info.offset) &&
|
|
|
|
(txdlp->Host_Control)) {
|
|
|
|
/* Check for TxD errors */
|
|
|
|
if (txdlp->Control_1 & TXD_T_CODE) {
|
|
|
|
unsigned long long err;
|
|
|
|
err = txdlp->Control_1 & TXD_T_CODE;
|
2006-04-21 23:20:22 +00:00
|
|
|
if (err & 0x1) {
|
|
|
|
nic->mac_control.stats_info->sw_stat.
|
|
|
|
parity_err_cnt++;
|
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
|
|
|
|
/* update t_code statistics */
|
2007-06-19 20:41:10 +00:00
|
|
|
err_mask = err >> 48;
|
|
|
|
switch(err_mask) {
|
2007-05-10 08:22:25 +00:00
|
|
|
case 2:
|
|
|
|
nic->mac_control.stats_info->sw_stat.
|
|
|
|
tx_buf_abort_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 3:
|
|
|
|
nic->mac_control.stats_info->sw_stat.
|
|
|
|
tx_desc_abort_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 7:
|
|
|
|
nic->mac_control.stats_info->sw_stat.
|
|
|
|
tx_parity_err_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 10:
|
|
|
|
nic->mac_control.stats_info->sw_stat.
|
|
|
|
tx_link_loss_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 15:
|
|
|
|
nic->mac_control.stats_info->sw_stat.
|
|
|
|
tx_list_proc_err_cnt++;
|
|
|
|
break;
|
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-11-14 20:25:08 +00:00
|
|
|
skb = s2io_txdl_getskb(fifo_data, txdlp, get_info.offset);
|
2005-08-03 19:24:33 +00:00
|
|
|
if (skb == NULL) {
|
2008-01-24 10:08:59 +00:00
|
|
|
spin_unlock_irqrestore(&fifo_data->tx_lock, flags);
|
2005-08-03 19:24:33 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Null skb ",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
2005-08-03 19:24:33 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "in Tx Free Intr\n");
|
|
|
|
return;
|
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
pkt_cnt++;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
|
|
|
/* Updating the statistics block */
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
nic->dev->stats.tx_bytes += skb->len;
|
2007-05-10 08:22:25 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed += skb->truesize;
|
2005-08-03 19:24:33 +00:00
|
|
|
dev_kfree_skb_irq(skb);
|
|
|
|
|
|
|
|
get_info.offset++;
|
2006-04-21 23:03:13 +00:00
|
|
|
if (get_info.offset == get_info.fifo_len + 1)
|
|
|
|
get_info.offset = 0;
|
2007-01-31 19:09:29 +00:00
|
|
|
txdlp = (struct TxD *) fifo_data->list_info
|
2005-08-03 19:24:33 +00:00
|
|
|
[get_info.offset].list_virt_addr;
|
|
|
|
fifo_data->tx_curr_get_info.offset =
|
|
|
|
get_info.offset;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_wake_tx_queue(fifo_data, pkt_cnt, nic->config.multiq);
|
2008-01-24 10:08:59 +00:00
|
|
|
|
|
|
|
spin_unlock_irqrestore(&fifo_data->tx_lock, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2006-04-21 23:20:22 +00:00
|
|
|
/**
|
|
|
|
* s2io_mdio_write - Function to write in to MDIO registers
|
|
|
|
* @mmd_type : MMD type value (PMA/PMD/WIS/PCS/PHYXS)
|
|
|
|
* @addr : address value
|
|
|
|
* @value : data value
|
|
|
|
* @dev : pointer to net_device structure
|
|
|
|
* Description:
|
|
|
|
* This function is used to write values to the MDIO registers
|
|
|
|
* NONE
|
|
|
|
*/
|
|
|
|
static void s2io_mdio_write(u32 mmd_type, u64 addr, u16 value, struct net_device *dev)
|
|
|
|
{
|
|
|
|
u64 val64 = 0x0;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2006-04-21 23:20:22 +00:00
|
|
|
|
|
|
|
//address transaction
|
|
|
|
val64 = val64 | MDIO_MMD_INDX_ADDR(addr)
|
|
|
|
| MDIO_MMD_DEV_ADDR(mmd_type)
|
|
|
|
| MDIO_MMS_PRT_ADDR(0x0);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
val64 = val64 | MDIO_CTRL_START_TRANS(0xE);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
udelay(100);
|
|
|
|
|
|
|
|
//Data transaction
|
|
|
|
val64 = 0x0;
|
|
|
|
val64 = val64 | MDIO_MMD_INDX_ADDR(addr)
|
|
|
|
| MDIO_MMD_DEV_ADDR(mmd_type)
|
|
|
|
| MDIO_MMS_PRT_ADDR(0x0)
|
|
|
|
| MDIO_MDIO_DATA(value)
|
|
|
|
| MDIO_OP(MDIO_OP_WRITE_TRANS);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
val64 = val64 | MDIO_CTRL_START_TRANS(0xE);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
udelay(100);
|
|
|
|
|
|
|
|
val64 = 0x0;
|
|
|
|
val64 = val64 | MDIO_MMD_INDX_ADDR(addr)
|
|
|
|
| MDIO_MMD_DEV_ADDR(mmd_type)
|
|
|
|
| MDIO_MMS_PRT_ADDR(0x0)
|
|
|
|
| MDIO_OP(MDIO_OP_READ_TRANS);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
val64 = val64 | MDIO_CTRL_START_TRANS(0xE);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
udelay(100);
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_mdio_read - Function to write in to MDIO registers
|
|
|
|
* @mmd_type : MMD type value (PMA/PMD/WIS/PCS/PHYXS)
|
|
|
|
* @addr : address value
|
|
|
|
* @dev : pointer to net_device structure
|
|
|
|
* Description:
|
|
|
|
* This function is used to read values to the MDIO registers
|
|
|
|
* NONE
|
|
|
|
*/
|
|
|
|
static u64 s2io_mdio_read(u32 mmd_type, u64 addr, struct net_device *dev)
|
|
|
|
{
|
|
|
|
u64 val64 = 0x0;
|
|
|
|
u64 rval64 = 0x0;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2006-04-21 23:20:22 +00:00
|
|
|
|
|
|
|
/* address transaction */
|
|
|
|
val64 = val64 | MDIO_MMD_INDX_ADDR(addr)
|
|
|
|
| MDIO_MMD_DEV_ADDR(mmd_type)
|
|
|
|
| MDIO_MMS_PRT_ADDR(0x0);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
val64 = val64 | MDIO_CTRL_START_TRANS(0xE);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
udelay(100);
|
|
|
|
|
|
|
|
/* Data transaction */
|
|
|
|
val64 = 0x0;
|
|
|
|
val64 = val64 | MDIO_MMD_INDX_ADDR(addr)
|
|
|
|
| MDIO_MMD_DEV_ADDR(mmd_type)
|
|
|
|
| MDIO_MMS_PRT_ADDR(0x0)
|
|
|
|
| MDIO_OP(MDIO_OP_READ_TRANS);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
val64 = val64 | MDIO_CTRL_START_TRANS(0xE);
|
|
|
|
writeq(val64, &bar0->mdio_control);
|
|
|
|
udelay(100);
|
|
|
|
|
|
|
|
/* Read the value from regs */
|
|
|
|
rval64 = readq(&bar0->mdio_control);
|
|
|
|
rval64 = rval64 & 0xFFFF0000;
|
|
|
|
rval64 = rval64 >> 16;
|
|
|
|
return rval64;
|
|
|
|
}
|
|
|
|
/**
|
|
|
|
* s2io_chk_xpak_counter - Function to check the status of the xpak counters
|
|
|
|
* @counter : couter value to be updated
|
|
|
|
* @flag : flag to indicate the status
|
|
|
|
* @type : counter type
|
|
|
|
* Description:
|
|
|
|
* This function is to check the status of the xpak counters value
|
|
|
|
* NONE
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void s2io_chk_xpak_counter(u64 *counter, u64 * regs_stat, u32 index, u16 flag, u16 type)
|
|
|
|
{
|
|
|
|
u64 mask = 0x3;
|
|
|
|
u64 val64;
|
|
|
|
int i;
|
|
|
|
for(i = 0; i <index; i++)
|
|
|
|
mask = mask << 0x2;
|
|
|
|
|
|
|
|
if(flag > 0)
|
|
|
|
{
|
|
|
|
*counter = *counter + 1;
|
|
|
|
val64 = *regs_stat & mask;
|
|
|
|
val64 = val64 >> (index * 0x2);
|
|
|
|
val64 = val64 + 1;
|
|
|
|
if(val64 == 3)
|
|
|
|
{
|
|
|
|
switch(type)
|
|
|
|
{
|
|
|
|
case 1:
|
|
|
|
DBG_PRINT(ERR_DBG, "Take Xframe NIC out of "
|
|
|
|
"service. Excessive temperatures may "
|
|
|
|
"result in premature transceiver "
|
|
|
|
"failure \n");
|
|
|
|
break;
|
|
|
|
case 2:
|
|
|
|
DBG_PRINT(ERR_DBG, "Take Xframe NIC out of "
|
|
|
|
"service Excessive bias currents may "
|
|
|
|
"indicate imminent laser diode "
|
|
|
|
"failure \n");
|
|
|
|
break;
|
|
|
|
case 3:
|
|
|
|
DBG_PRINT(ERR_DBG, "Take Xframe NIC out of "
|
|
|
|
"service Excessive laser output "
|
|
|
|
"power may saturate far-end "
|
|
|
|
"receiver\n");
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DBG_PRINT(ERR_DBG, "Incorrect XPAK Alarm "
|
|
|
|
"type \n");
|
|
|
|
}
|
|
|
|
val64 = 0x0;
|
|
|
|
}
|
|
|
|
val64 = val64 << (index * 0x2);
|
|
|
|
*regs_stat = (*regs_stat & (~mask)) | (val64);
|
|
|
|
|
|
|
|
} else {
|
|
|
|
*regs_stat = *regs_stat & (~mask);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_updt_xpak_counter - Function to update the xpak counters
|
|
|
|
* @dev : pointer to net_device struct
|
|
|
|
* Description:
|
|
|
|
* This function is to upate the status of the xpak counters value
|
|
|
|
* NONE
|
|
|
|
*/
|
|
|
|
static void s2io_updt_xpak_counter(struct net_device *dev)
|
|
|
|
{
|
|
|
|
u16 flag = 0x0;
|
|
|
|
u16 type = 0x0;
|
|
|
|
u16 val16 = 0x0;
|
|
|
|
u64 val64 = 0x0;
|
|
|
|
u64 addr = 0x0;
|
|
|
|
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct stat_block *stat_info = sp->mac_control.stats_info;
|
2006-04-21 23:20:22 +00:00
|
|
|
|
|
|
|
/* Check the communication with the MDIO slave */
|
|
|
|
addr = 0x0000;
|
|
|
|
val64 = 0x0;
|
|
|
|
val64 = s2io_mdio_read(MDIO_MMD_PMA_DEV_ADDR, addr, dev);
|
|
|
|
if((val64 == 0xFFFF) || (val64 == 0x0000))
|
|
|
|
{
|
|
|
|
DBG_PRINT(ERR_DBG, "ERR: MDIO slave access failed - "
|
|
|
|
"Returned %llx\n", (unsigned long long)val64);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Check for the expecte value of 2040 at PMA address 0x0000 */
|
|
|
|
if(val64 != 0x2040)
|
|
|
|
{
|
|
|
|
DBG_PRINT(ERR_DBG, "Incorrect value at PMA address 0x0000 - ");
|
|
|
|
DBG_PRINT(ERR_DBG, "Returned: %llx- Expected: 0x2040\n",
|
|
|
|
(unsigned long long)val64);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Loading the DOM register to MDIO register */
|
|
|
|
addr = 0xA100;
|
|
|
|
s2io_mdio_write(MDIO_MMD_PMA_DEV_ADDR, addr, val16, dev);
|
|
|
|
val64 = s2io_mdio_read(MDIO_MMD_PMA_DEV_ADDR, addr, dev);
|
|
|
|
|
|
|
|
/* Reading the Alarm flags */
|
|
|
|
addr = 0xA070;
|
|
|
|
val64 = 0x0;
|
|
|
|
val64 = s2io_mdio_read(MDIO_MMD_PMA_DEV_ADDR, addr, dev);
|
|
|
|
|
|
|
|
flag = CHECKBIT(val64, 0x7);
|
|
|
|
type = 1;
|
|
|
|
s2io_chk_xpak_counter(&stat_info->xpak_stat.alarm_transceiver_temp_high,
|
|
|
|
&stat_info->xpak_stat.xpak_regs_stat,
|
|
|
|
0x0, flag, type);
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x6))
|
|
|
|
stat_info->xpak_stat.alarm_transceiver_temp_low++;
|
|
|
|
|
|
|
|
flag = CHECKBIT(val64, 0x3);
|
|
|
|
type = 2;
|
|
|
|
s2io_chk_xpak_counter(&stat_info->xpak_stat.alarm_laser_bias_current_high,
|
|
|
|
&stat_info->xpak_stat.xpak_regs_stat,
|
|
|
|
0x2, flag, type);
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x2))
|
|
|
|
stat_info->xpak_stat.alarm_laser_bias_current_low++;
|
|
|
|
|
|
|
|
flag = CHECKBIT(val64, 0x1);
|
|
|
|
type = 3;
|
|
|
|
s2io_chk_xpak_counter(&stat_info->xpak_stat.alarm_laser_output_power_high,
|
|
|
|
&stat_info->xpak_stat.xpak_regs_stat,
|
|
|
|
0x4, flag, type);
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x0))
|
|
|
|
stat_info->xpak_stat.alarm_laser_output_power_low++;
|
|
|
|
|
|
|
|
/* Reading the Warning flags */
|
|
|
|
addr = 0xA074;
|
|
|
|
val64 = 0x0;
|
|
|
|
val64 = s2io_mdio_read(MDIO_MMD_PMA_DEV_ADDR, addr, dev);
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x7))
|
|
|
|
stat_info->xpak_stat.warn_transceiver_temp_high++;
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x6))
|
|
|
|
stat_info->xpak_stat.warn_transceiver_temp_low++;
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x3))
|
|
|
|
stat_info->xpak_stat.warn_laser_bias_current_high++;
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x2))
|
|
|
|
stat_info->xpak_stat.warn_laser_bias_current_low++;
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x1))
|
|
|
|
stat_info->xpak_stat.warn_laser_output_power_high++;
|
|
|
|
|
|
|
|
if(CHECKBIT(val64, 0x0))
|
|
|
|
stat_info->xpak_stat.warn_laser_output_power_low++;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
2005-04-16 22:20:36 +00:00
|
|
|
* wait_for_cmd_complete - waits for a command to complete.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: Function that waits for a command to Write into RMAC
|
|
|
|
* ADDR DATA registers to be completed and returns either success or
|
|
|
|
* error depending on whether the command was complete or not.
|
2005-04-16 22:20:36 +00:00
|
|
|
* Return value:
|
|
|
|
* SUCCESS on success and FAILURE on failure.
|
|
|
|
*/
|
|
|
|
|
2007-02-24 06:57:32 +00:00
|
|
|
static int wait_for_cmd_complete(void __iomem *addr, u64 busy_bit,
|
|
|
|
int bit_state)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-02-24 06:57:32 +00:00
|
|
|
int ret = FAILURE, cnt = 0, delay = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64;
|
|
|
|
|
2007-02-24 06:57:32 +00:00
|
|
|
if ((bit_state != S2IO_BIT_RESET) && (bit_state != S2IO_BIT_SET))
|
|
|
|
return FAILURE;
|
|
|
|
|
|
|
|
do {
|
2006-04-21 23:18:03 +00:00
|
|
|
val64 = readq(addr);
|
2007-02-24 06:57:32 +00:00
|
|
|
if (bit_state == S2IO_BIT_RESET) {
|
|
|
|
if (!(val64 & busy_bit)) {
|
|
|
|
ret = SUCCESS;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (!(val64 & busy_bit)) {
|
|
|
|
ret = SUCCESS;
|
|
|
|
break;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2006-04-21 23:18:03 +00:00
|
|
|
|
|
|
|
if(in_interrupt())
|
2007-02-24 06:57:32 +00:00
|
|
|
mdelay(delay);
|
2006-04-21 23:18:03 +00:00
|
|
|
else
|
2007-02-24 06:57:32 +00:00
|
|
|
msleep(delay);
|
2006-04-21 23:18:03 +00:00
|
|
|
|
2007-02-24 06:57:32 +00:00
|
|
|
if (++cnt >= 10)
|
|
|
|
delay = 50;
|
|
|
|
} while (cnt < 20);
|
2005-04-16 22:20:36 +00:00
|
|
|
return ret;
|
|
|
|
}
|
2007-01-31 18:30:49 +00:00
|
|
|
/*
|
|
|
|
* check_pci_device_id - Checks if the device id is supported
|
|
|
|
* @id : device id
|
|
|
|
* Description: Function to check if the pci device id is supported by driver.
|
|
|
|
* Return value: Actual device id if supported else PCI_ANY_ID
|
|
|
|
*/
|
|
|
|
static u16 check_pci_device_id(u16 id)
|
|
|
|
{
|
|
|
|
switch (id) {
|
|
|
|
case PCI_DEVICE_ID_HERC_WIN:
|
|
|
|
case PCI_DEVICE_ID_HERC_UNI:
|
|
|
|
return XFRAME_II_DEVICE;
|
|
|
|
case PCI_DEVICE_ID_S2IO_UNI:
|
|
|
|
case PCI_DEVICE_ID_S2IO_WIN:
|
|
|
|
return XFRAME_I_DEVICE;
|
|
|
|
default:
|
|
|
|
return PCI_ANY_ID;
|
|
|
|
}
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* s2io_reset - Resets the card.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @sp : private member of the device structure.
|
|
|
|
* Description: Function to Reset the card. This function then also
|
2005-08-03 19:24:33 +00:00
|
|
|
* restores the previously saved PCI configuration space registers as
|
2005-04-16 22:20:36 +00:00
|
|
|
* the card reset also resets the configuration space.
|
|
|
|
* Return value:
|
|
|
|
* void.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_reset(struct s2io_nic * sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64;
|
2005-08-03 19:27:09 +00:00
|
|
|
u16 subid, pci_cmd;
|
2007-01-31 18:30:49 +00:00
|
|
|
int i;
|
|
|
|
u16 val16;
|
2007-05-10 08:22:25 +00:00
|
|
|
unsigned long long up_cnt, down_cnt, up_time, down_time, reset_cnt;
|
|
|
|
unsigned long long mem_alloc_cnt, mem_free_cnt, watchdog_cnt;
|
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
DBG_PRINT(INIT_DBG,"%s - Resetting XFrame card %s\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__, sp->dev->name);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:39:56 +00:00
|
|
|
/* Back up the PCI-X CMD reg, dont want to lose MMRBC, OST settings */
|
2005-08-12 17:15:59 +00:00
|
|
|
pci_read_config_word(sp->pdev, PCIX_COMMAND_REGISTER, &(pci_cmd));
|
2005-08-03 19:39:56 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = SW_RESET_ALL;
|
|
|
|
writeq(val64, &bar0->sw_reset);
|
2006-04-21 23:18:03 +00:00
|
|
|
if (strstr(sp->product_name, "CX4")) {
|
|
|
|
msleep(750);
|
|
|
|
}
|
2007-01-31 18:30:49 +00:00
|
|
|
msleep(250);
|
|
|
|
for (i = 0; i < S2IO_MAX_PCI_CONFIG_SPACE_REINIT; i++) {
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
/* Restore the PCI state saved during initialization. */
|
|
|
|
pci_restore_state(sp->pdev);
|
|
|
|
pci_read_config_word(sp->pdev, 0x2, &val16);
|
|
|
|
if (check_pci_device_id(val16) != (u16)PCI_ANY_ID)
|
|
|
|
break;
|
|
|
|
msleep(200);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
if (check_pci_device_id(val16) == (u16)PCI_ANY_ID) {
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(ERR_DBG,"%s SW_Reset failed!\n", __func__);
|
2007-01-31 18:30:49 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
pci_write_config_word(sp->pdev, PCIX_COMMAND_REGISTER, pci_cmd);
|
|
|
|
|
|
|
|
s2io_init_pci(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/* Set swapper to enable I/O register access */
|
|
|
|
s2io_set_swapper(sp);
|
|
|
|
|
2008-01-24 09:45:43 +00:00
|
|
|
/* restore mac_addr entries */
|
|
|
|
do_s2io_restore_unicast_mc(sp);
|
|
|
|
|
2005-10-04 10:41:24 +00:00
|
|
|
/* Restore the MSIX table entries from local variables */
|
|
|
|
restore_xmsi_data(sp);
|
|
|
|
|
2005-08-03 19:27:09 +00:00
|
|
|
/* Clear certain PCI/PCI-X fields after reset */
|
2005-08-03 19:41:38 +00:00
|
|
|
if (sp->device_type == XFRAME_II_DEVICE) {
|
2006-07-24 23:52:49 +00:00
|
|
|
/* Clear "detected parity error" bit */
|
2005-08-03 19:41:38 +00:00
|
|
|
pci_write_config_word(sp->pdev, PCI_STATUS, 0x8000);
|
2005-08-03 19:27:09 +00:00
|
|
|
|
2005-08-03 19:41:38 +00:00
|
|
|
/* Clearing PCIX Ecc status register */
|
|
|
|
pci_write_config_dword(sp->pdev, 0x68, 0x7C);
|
2005-08-03 19:27:09 +00:00
|
|
|
|
2005-08-03 19:41:38 +00:00
|
|
|
/* Clearing PCI_STATUS error reflected here */
|
2007-10-19 06:40:29 +00:00
|
|
|
writeq(s2BIT(62), &bar0->txpic_int_reg);
|
2005-08-03 19:41:38 +00:00
|
|
|
}
|
2005-08-03 19:27:09 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/* Reset device statistics maintained by OS */
|
|
|
|
memset(&sp->stats, 0, sizeof (struct net_device_stats));
|
2007-09-18 22:14:20 +00:00
|
|
|
|
2007-05-10 08:22:25 +00:00
|
|
|
up_cnt = sp->mac_control.stats_info->sw_stat.link_up_cnt;
|
|
|
|
down_cnt = sp->mac_control.stats_info->sw_stat.link_down_cnt;
|
|
|
|
up_time = sp->mac_control.stats_info->sw_stat.link_up_time;
|
|
|
|
down_time = sp->mac_control.stats_info->sw_stat.link_down_time;
|
2007-03-07 01:01:00 +00:00
|
|
|
reset_cnt = sp->mac_control.stats_info->sw_stat.soft_reset_cnt;
|
2007-05-10 08:22:25 +00:00
|
|
|
mem_alloc_cnt = sp->mac_control.stats_info->sw_stat.mem_allocated;
|
|
|
|
mem_free_cnt = sp->mac_control.stats_info->sw_stat.mem_freed;
|
|
|
|
watchdog_cnt = sp->mac_control.stats_info->sw_stat.watchdog_timer_cnt;
|
|
|
|
/* save link up/down time/cnt, reset/memory/watchdog cnt */
|
2007-03-07 01:01:00 +00:00
|
|
|
memset(sp->mac_control.stats_info, 0, sizeof(struct stat_block));
|
2007-05-10 08:22:25 +00:00
|
|
|
/* restore link up/down time/cnt, reset/memory/watchdog cnt */
|
|
|
|
sp->mac_control.stats_info->sw_stat.link_up_cnt = up_cnt;
|
|
|
|
sp->mac_control.stats_info->sw_stat.link_down_cnt = down_cnt;
|
|
|
|
sp->mac_control.stats_info->sw_stat.link_up_time = up_time;
|
|
|
|
sp->mac_control.stats_info->sw_stat.link_down_time = down_time;
|
2007-03-07 01:01:00 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.soft_reset_cnt = reset_cnt;
|
2007-05-10 08:22:25 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_allocated = mem_alloc_cnt;
|
|
|
|
sp->mac_control.stats_info->sw_stat.mem_freed = mem_free_cnt;
|
|
|
|
sp->mac_control.stats_info->sw_stat.watchdog_timer_cnt = watchdog_cnt;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* SXE-002: Configure link and activity LED to turn it off */
|
|
|
|
subid = sp->pdev->subsystem_device;
|
2005-08-03 19:36:55 +00:00
|
|
|
if (((subid & 0xFF) >= 0x07) &&
|
|
|
|
(sp->device_type == XFRAME_I_DEVICE)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
val64 |= 0x0000800000000000ULL;
|
|
|
|
writeq(val64, &bar0->gpio_control);
|
|
|
|
val64 = 0x0411040400000000ULL;
|
2005-09-05 02:25:58 +00:00
|
|
|
writeq(val64, (void __iomem *)bar0 + 0x2700);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
/*
|
|
|
|
* Clear spurious ECC interrupts that would have occured on
|
|
|
|
* XFRAME II cards after reset.
|
|
|
|
*/
|
|
|
|
if (sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
val64 = readq(&bar0->pcc_err_reg);
|
|
|
|
writeq(val64, &bar0->pcc_err_reg);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
sp->device_enabled_once = FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_set_swapper - to set the swapper controle on the card
|
|
|
|
* @sp : private member of the device structure,
|
2005-04-16 22:20:36 +00:00
|
|
|
* pointer to the s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: Function to set the swapper control on the card
|
2005-04-16 22:20:36 +00:00
|
|
|
* correctly depending on the 'endianness' of the system.
|
|
|
|
* Return value:
|
|
|
|
* SUCCESS on success and FAILURE on failure.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_set_swapper(struct s2io_nic * sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = sp->dev;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64, valt, valr;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Set proper endian settings and verify the same by reading
|
|
|
|
* the PIF Feed-back register.
|
|
|
|
*/
|
|
|
|
|
|
|
|
val64 = readq(&bar0->pif_rd_swapper_fb);
|
|
|
|
if (val64 != 0x0123456789ABCDEFULL) {
|
|
|
|
int i = 0;
|
|
|
|
u64 value[] = { 0xC30000C3C30000C3ULL, /* FE=1, SE=1 */
|
|
|
|
0x8100008181000081ULL, /* FE=1, SE=0 */
|
|
|
|
0x4200004242000042ULL, /* FE=0, SE=1 */
|
|
|
|
0}; /* FE=0, SE=0 */
|
|
|
|
|
|
|
|
while(i<4) {
|
|
|
|
writeq(value[i], &bar0->swapper_ctrl);
|
|
|
|
val64 = readq(&bar0->pif_rd_swapper_fb);
|
|
|
|
if (val64 == 0x0123456789ABCDEFULL)
|
|
|
|
break;
|
|
|
|
i++;
|
|
|
|
}
|
|
|
|
if (i == 4) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Endian settings are wrong, ",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "feedback read %llx\n",
|
|
|
|
(unsigned long long) val64);
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
valr = value[i];
|
|
|
|
} else {
|
|
|
|
valr = readq(&bar0->swapper_ctrl);
|
|
|
|
}
|
|
|
|
|
|
|
|
valt = 0x0123456789ABCDEFULL;
|
|
|
|
writeq(valt, &bar0->xmsi_address);
|
|
|
|
val64 = readq(&bar0->xmsi_address);
|
|
|
|
|
|
|
|
if(val64 != valt) {
|
|
|
|
int i = 0;
|
|
|
|
u64 value[] = { 0x00C3C30000C3C300ULL, /* FE=1, SE=1 */
|
|
|
|
0x0081810000818100ULL, /* FE=1, SE=0 */
|
|
|
|
0x0042420000424200ULL, /* FE=0, SE=1 */
|
|
|
|
0}; /* FE=0, SE=0 */
|
|
|
|
|
|
|
|
while(i<4) {
|
|
|
|
writeq((value[i] | valr), &bar0->swapper_ctrl);
|
|
|
|
writeq(valt, &bar0->xmsi_address);
|
|
|
|
val64 = readq(&bar0->xmsi_address);
|
|
|
|
if(val64 == valt)
|
|
|
|
break;
|
|
|
|
i++;
|
|
|
|
}
|
|
|
|
if(i == 4) {
|
2005-08-03 19:24:33 +00:00
|
|
|
unsigned long long x = val64;
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "Write failed, Xmsi_addr ");
|
2005-08-03 19:24:33 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "reads:0x%llx\n", x);
|
2005-04-16 22:20:36 +00:00
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
val64 = readq(&bar0->swapper_ctrl);
|
|
|
|
val64 &= 0xFFFF000000000000ULL;
|
|
|
|
|
|
|
|
#ifdef __BIG_ENDIAN
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* The device by default set to a big endian format, so a
|
2005-04-16 22:20:36 +00:00
|
|
|
* big endian driver need not set anything.
|
|
|
|
*/
|
|
|
|
val64 |= (SWAPPER_CTRL_TXP_FE |
|
|
|
|
SWAPPER_CTRL_TXP_SE |
|
|
|
|
SWAPPER_CTRL_TXD_R_FE |
|
|
|
|
SWAPPER_CTRL_TXD_W_FE |
|
|
|
|
SWAPPER_CTRL_TXF_R_FE |
|
|
|
|
SWAPPER_CTRL_RXD_R_FE |
|
|
|
|
SWAPPER_CTRL_RXD_W_FE |
|
|
|
|
SWAPPER_CTRL_RXF_W_FE |
|
|
|
|
SWAPPER_CTRL_XMSI_FE |
|
|
|
|
SWAPPER_CTRL_STATS_FE | SWAPPER_CTRL_STATS_SE);
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == INTA)
|
2005-10-04 10:41:24 +00:00
|
|
|
val64 |= SWAPPER_CTRL_XMSI_SE;
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->swapper_ctrl);
|
|
|
|
#else
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Initially we enable all bits to make it accessible by the
|
2005-08-03 19:24:33 +00:00
|
|
|
* driver, then we selectively enable only those bits that
|
2005-04-16 22:20:36 +00:00
|
|
|
* we want to set.
|
|
|
|
*/
|
|
|
|
val64 |= (SWAPPER_CTRL_TXP_FE |
|
|
|
|
SWAPPER_CTRL_TXP_SE |
|
|
|
|
SWAPPER_CTRL_TXD_R_FE |
|
|
|
|
SWAPPER_CTRL_TXD_R_SE |
|
|
|
|
SWAPPER_CTRL_TXD_W_FE |
|
|
|
|
SWAPPER_CTRL_TXD_W_SE |
|
|
|
|
SWAPPER_CTRL_TXF_R_FE |
|
|
|
|
SWAPPER_CTRL_RXD_R_FE |
|
|
|
|
SWAPPER_CTRL_RXD_R_SE |
|
|
|
|
SWAPPER_CTRL_RXD_W_FE |
|
|
|
|
SWAPPER_CTRL_RXD_W_SE |
|
|
|
|
SWAPPER_CTRL_RXF_W_FE |
|
|
|
|
SWAPPER_CTRL_XMSI_FE |
|
|
|
|
SWAPPER_CTRL_STATS_FE | SWAPPER_CTRL_STATS_SE);
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == INTA)
|
2005-10-04 10:41:24 +00:00
|
|
|
val64 |= SWAPPER_CTRL_XMSI_SE;
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->swapper_ctrl);
|
|
|
|
#endif
|
|
|
|
val64 = readq(&bar0->swapper_ctrl);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Verifying if endian settings are accurate by reading a
|
2005-04-16 22:20:36 +00:00
|
|
|
* feedback register.
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->pif_rd_swapper_fb);
|
|
|
|
if (val64 != 0x0123456789ABCDEFULL) {
|
|
|
|
/* Endian settings are incorrect, calls for another dekko. */
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Endian settings are wrong, ",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "feedback read %llx\n",
|
|
|
|
(unsigned long long) val64);
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int wait_for_msix_trans(struct s2io_nic *nic, int i)
|
2005-10-04 10:41:24 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-10-04 10:41:24 +00:00
|
|
|
u64 val64;
|
|
|
|
int ret = 0, cnt = 0;
|
|
|
|
|
|
|
|
do {
|
|
|
|
val64 = readq(&bar0->xmsi_access);
|
2007-10-19 06:40:29 +00:00
|
|
|
if (!(val64 & s2BIT(15)))
|
2005-10-04 10:41:24 +00:00
|
|
|
break;
|
|
|
|
mdelay(1);
|
|
|
|
cnt++;
|
|
|
|
} while(cnt < 5);
|
|
|
|
if (cnt == 5) {
|
|
|
|
DBG_PRINT(ERR_DBG, "XMSI # %d Access failed\n", i);
|
|
|
|
ret = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void restore_xmsi_data(struct s2io_nic *nic)
|
2005-10-04 10:41:24 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-10-04 10:41:24 +00:00
|
|
|
u64 val64;
|
2008-05-12 17:42:17 +00:00
|
|
|
int i, msix_index;
|
|
|
|
|
|
|
|
|
|
|
|
if (nic->device_type == XFRAME_I_DEVICE)
|
|
|
|
return;
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2006-07-24 23:55:09 +00:00
|
|
|
for (i=0; i < MAX_REQUESTED_MSI_X; i++) {
|
2008-05-12 17:42:17 +00:00
|
|
|
msix_index = (i) ? ((i-1) * 8 + 1): 0;
|
2005-10-04 10:41:24 +00:00
|
|
|
writeq(nic->msix_info[i].addr, &bar0->xmsi_address);
|
|
|
|
writeq(nic->msix_info[i].data, &bar0->xmsi_data);
|
2008-05-12 17:42:17 +00:00
|
|
|
val64 = (s2BIT(7) | s2BIT(15) | vBIT(msix_index, 26, 6));
|
2005-10-04 10:41:24 +00:00
|
|
|
writeq(val64, &bar0->xmsi_access);
|
2008-05-12 17:42:17 +00:00
|
|
|
if (wait_for_msix_trans(nic, msix_index)) {
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "failed in %s\n", __func__);
|
2005-10-04 10:41:24 +00:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void store_xmsi_data(struct s2io_nic *nic)
|
2005-10-04 10:41:24 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-10-04 10:41:24 +00:00
|
|
|
u64 val64, addr, data;
|
2008-05-12 17:42:17 +00:00
|
|
|
int i, msix_index;
|
|
|
|
|
|
|
|
if (nic->device_type == XFRAME_I_DEVICE)
|
|
|
|
return;
|
2005-10-04 10:41:24 +00:00
|
|
|
|
|
|
|
/* Store and display */
|
2006-07-24 23:55:09 +00:00
|
|
|
for (i=0; i < MAX_REQUESTED_MSI_X; i++) {
|
2008-05-12 17:42:17 +00:00
|
|
|
msix_index = (i) ? ((i-1) * 8 + 1): 0;
|
|
|
|
val64 = (s2BIT(15) | vBIT(msix_index, 26, 6));
|
2005-10-04 10:41:24 +00:00
|
|
|
writeq(val64, &bar0->xmsi_access);
|
2008-05-12 17:42:17 +00:00
|
|
|
if (wait_for_msix_trans(nic, msix_index)) {
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "failed in %s\n", __func__);
|
2005-10-04 10:41:24 +00:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
addr = readq(&bar0->xmsi_address);
|
|
|
|
data = readq(&bar0->xmsi_data);
|
|
|
|
if (addr && data) {
|
|
|
|
nic->msix_info[i].addr = addr;
|
|
|
|
nic->msix_info[i].data = data;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_enable_msi_x(struct s2io_nic *nic)
|
2005-10-04 10:41:24 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2008-05-12 17:41:32 +00:00
|
|
|
u64 rx_mat;
|
2005-10-04 10:41:24 +00:00
|
|
|
u16 msi_control; /* Temp variable */
|
|
|
|
int ret, i, j, msix_indx = 1;
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
nic->entries = kmalloc(nic->num_entries * sizeof(struct msix_entry),
|
2005-10-04 10:41:24 +00:00
|
|
|
GFP_KERNEL);
|
2007-09-14 11:28:50 +00:00
|
|
|
if (!nic->entries) {
|
2007-05-10 08:22:25 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: Memory allocation failed\n", \
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
2007-05-10 08:18:54 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_alloc_fail_cnt++;
|
2005-10-04 10:41:24 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_allocated
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (nic->num_entries * sizeof(struct msix_entry));
|
|
|
|
|
|
|
|
memset(nic->entries, 0, nic->num_entries * sizeof(struct msix_entry));
|
2005-10-04 10:41:24 +00:00
|
|
|
|
|
|
|
nic->s2io_entries =
|
2008-05-12 17:42:17 +00:00
|
|
|
kmalloc(nic->num_entries * sizeof(struct s2io_msix_entry),
|
2005-10-04 10:41:24 +00:00
|
|
|
GFP_KERNEL);
|
2007-09-14 11:28:50 +00:00
|
|
|
if (!nic->s2io_entries) {
|
2007-09-18 22:14:20 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: Memory allocation failed\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
2007-05-10 08:18:54 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_alloc_fail_cnt++;
|
2005-10-04 10:41:24 +00:00
|
|
|
kfree(nic->entries);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (nic->num_entries * sizeof(struct msix_entry));
|
2005-10-04 10:41:24 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_allocated
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (nic->num_entries * sizeof(struct s2io_msix_entry));
|
|
|
|
memset(nic->s2io_entries, 0,
|
|
|
|
nic->num_entries * sizeof(struct s2io_msix_entry));
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2008-05-12 17:41:32 +00:00
|
|
|
nic->entries[0].entry = 0;
|
|
|
|
nic->s2io_entries[0].entry = 0;
|
|
|
|
nic->s2io_entries[0].in_use = MSIX_FLG;
|
|
|
|
nic->s2io_entries[0].type = MSIX_ALARM_TYPE;
|
|
|
|
nic->s2io_entries[0].arg = &nic->mac_control.fifos;
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
for (i = 1; i < nic->num_entries; i++) {
|
|
|
|
nic->entries[i].entry = ((i - 1) * 8) + 1;
|
|
|
|
nic->s2io_entries[i].entry = ((i - 1) * 8) + 1;
|
2005-10-04 10:41:24 +00:00
|
|
|
nic->s2io_entries[i].arg = NULL;
|
|
|
|
nic->s2io_entries[i].in_use = 0;
|
|
|
|
}
|
|
|
|
|
2007-09-18 22:14:20 +00:00
|
|
|
rx_mat = readq(&bar0->rx_mat);
|
2008-05-12 17:42:17 +00:00
|
|
|
for (j = 0; j < nic->config.rx_ring_num; j++) {
|
2007-09-18 22:14:20 +00:00
|
|
|
rx_mat |= RX_MAT_SET(j, msix_indx);
|
2008-05-12 17:42:17 +00:00
|
|
|
nic->s2io_entries[j+1].arg = &nic->mac_control.rings[j];
|
|
|
|
nic->s2io_entries[j+1].type = MSIX_RING_TYPE;
|
|
|
|
nic->s2io_entries[j+1].in_use = MSIX_FLG;
|
|
|
|
msix_indx += 8;
|
2005-10-04 10:41:24 +00:00
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
writeq(rx_mat, &bar0->rx_mat);
|
2008-05-12 17:42:17 +00:00
|
|
|
readq(&bar0->rx_mat);
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
ret = pci_enable_msix(nic->pdev, nic->entries, nic->num_entries);
|
2006-04-21 23:18:03 +00:00
|
|
|
/* We fail init if error or we get less vectors than min required */
|
2005-10-04 10:41:24 +00:00
|
|
|
if (ret) {
|
2009-02-03 23:15:15 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Enabling MSI-X failed\n");
|
2005-10-04 10:41:24 +00:00
|
|
|
kfree(nic->entries);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (nic->num_entries * sizeof(struct msix_entry));
|
2005-10-04 10:41:24 +00:00
|
|
|
kfree(nic->s2io_entries);
|
2007-09-18 22:14:20 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (nic->num_entries * sizeof(struct s2io_msix_entry));
|
2005-10-04 10:41:24 +00:00
|
|
|
nic->entries = NULL;
|
|
|
|
nic->s2io_entries = NULL;
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* To enable MSI-X, MSI also needs to be enabled, due to a bug
|
|
|
|
* in the herc NIC. (Temp change, needs to be removed later)
|
|
|
|
*/
|
|
|
|
pci_read_config_word(nic->pdev, 0x42, &msi_control);
|
|
|
|
msi_control |= 0x1; /* Enable MSI */
|
|
|
|
pci_write_config_word(nic->pdev, 0x42, msi_control);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-09-15 20:11:34 +00:00
|
|
|
/* Handle software interrupt used during MSI(X) test */
|
2007-12-11 22:23:06 +00:00
|
|
|
static irqreturn_t s2io_test_intr(int irq, void *dev_id)
|
2007-09-15 20:11:34 +00:00
|
|
|
{
|
|
|
|
struct s2io_nic *sp = dev_id;
|
|
|
|
|
|
|
|
sp->msi_detected = 1;
|
|
|
|
wake_up(&sp->msi_wait);
|
|
|
|
|
|
|
|
return IRQ_HANDLED;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Test interrupt path by forcing a a software IRQ */
|
2007-12-11 22:23:06 +00:00
|
|
|
static int s2io_test_msi(struct s2io_nic *sp)
|
2007-09-15 20:11:34 +00:00
|
|
|
{
|
|
|
|
struct pci_dev *pdev = sp->pdev;
|
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
|
|
|
int err;
|
|
|
|
u64 val64, saved64;
|
|
|
|
|
|
|
|
err = request_irq(sp->entries[1].vector, s2io_test_intr, 0,
|
|
|
|
sp->name, sp);
|
|
|
|
if (err) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: PCI %s: cannot assign irq %d\n",
|
|
|
|
sp->dev->name, pci_name(pdev), pdev->irq);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
init_waitqueue_head (&sp->msi_wait);
|
|
|
|
sp->msi_detected = 0;
|
|
|
|
|
|
|
|
saved64 = val64 = readq(&bar0->scheduled_int_ctrl);
|
|
|
|
val64 |= SCHED_INT_CTRL_ONE_SHOT;
|
|
|
|
val64 |= SCHED_INT_CTRL_TIMER_EN;
|
|
|
|
val64 |= SCHED_INT_CTRL_INT2MSI(1);
|
|
|
|
writeq(val64, &bar0->scheduled_int_ctrl);
|
|
|
|
|
|
|
|
wait_event_timeout(sp->msi_wait, sp->msi_detected, HZ/10);
|
|
|
|
|
|
|
|
if (!sp->msi_detected) {
|
|
|
|
/* MSI(X) test failed, go back to INTx mode */
|
2007-11-20 01:48:28 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: PCI %s: No interrupt was generated "
|
2007-09-15 20:11:34 +00:00
|
|
|
"using MSI(X) during test\n", sp->dev->name,
|
|
|
|
pci_name(pdev));
|
|
|
|
|
|
|
|
err = -EOPNOTSUPP;
|
|
|
|
}
|
|
|
|
|
|
|
|
free_irq(sp->entries[1].vector, sp);
|
|
|
|
|
|
|
|
writeq(saved64, &bar0->scheduled_int_ctrl);
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
2007-11-14 09:41:06 +00:00
|
|
|
|
|
|
|
static void remove_msix_isr(struct s2io_nic *sp)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
u16 msi_control;
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
for (i = 0; i < sp->num_entries; i++) {
|
2007-11-14 09:41:06 +00:00
|
|
|
if (sp->s2io_entries[i].in_use ==
|
|
|
|
MSIX_REGISTERED_SUCCESS) {
|
|
|
|
int vector = sp->entries[i].vector;
|
|
|
|
void *arg = sp->s2io_entries[i].arg;
|
|
|
|
free_irq(vector, arg);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
kfree(sp->entries);
|
|
|
|
kfree(sp->s2io_entries);
|
|
|
|
sp->entries = NULL;
|
|
|
|
sp->s2io_entries = NULL;
|
|
|
|
|
|
|
|
pci_read_config_word(sp->pdev, 0x42, &msi_control);
|
|
|
|
msi_control &= 0xFFFE; /* Disable MSI */
|
|
|
|
pci_write_config_word(sp->pdev, 0x42, msi_control);
|
|
|
|
|
|
|
|
pci_disable_msix(sp->pdev);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void remove_inta_isr(struct s2io_nic *sp)
|
|
|
|
{
|
|
|
|
struct net_device *dev = sp->dev;
|
|
|
|
|
|
|
|
free_irq(sp->pdev->irq, dev);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* ********************************************************* *
|
|
|
|
* Functions defined below concern the OS part of the driver *
|
|
|
|
* ********************************************************* */
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_open - open entry point of the driver
|
|
|
|
* @dev : pointer to the device structure.
|
|
|
|
* Description:
|
|
|
|
* This function is the open entry point of the driver. It mainly calls a
|
|
|
|
* function to allocate Rx buffers and inserts them into the buffer
|
2005-08-03 19:24:33 +00:00
|
|
|
* descriptors and then enables the Rx part of the NIC.
|
2005-04-16 22:20:36 +00:00
|
|
|
* Return value:
|
|
|
|
* 0 on success and an appropriate (-)ve integer as defined in errno.h
|
|
|
|
* file on failure.
|
|
|
|
*/
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_open(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
int err = 0;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Make sure you have link off by default every time
|
2005-04-16 22:20:36 +00:00
|
|
|
* Nic is initialized
|
|
|
|
*/
|
|
|
|
netif_carrier_off(dev);
|
2005-08-03 19:39:56 +00:00
|
|
|
sp->last_link_state = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Initialize H/W and enable interrupts */
|
2006-04-21 23:18:03 +00:00
|
|
|
err = s2io_card_up(sp);
|
|
|
|
if (err) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: H/W initialization failed\n",
|
|
|
|
dev->name);
|
2006-07-07 06:58:23 +00:00
|
|
|
goto hw_init_failed;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-09-14 11:39:19 +00:00
|
|
|
if (do_s2io_prog_unicast(dev, dev->dev_addr) == FAILURE) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "Set Mac Address Failed\n");
|
2006-07-07 06:58:23 +00:00
|
|
|
s2io_card_down(sp);
|
2005-08-03 19:24:33 +00:00
|
|
|
err = -ENODEV;
|
2006-07-07 06:58:23 +00:00
|
|
|
goto hw_init_failed;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_start_all_tx_queue(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
|
|
|
hw_init_failed:
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == MSI_X) {
|
2007-05-10 08:22:25 +00:00
|
|
|
if (sp->entries) {
|
2005-10-04 10:41:24 +00:00
|
|
|
kfree(sp->entries);
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_freed
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (sp->num_entries * sizeof(struct msix_entry));
|
2007-05-10 08:22:25 +00:00
|
|
|
}
|
|
|
|
if (sp->s2io_entries) {
|
2005-10-04 10:41:24 +00:00
|
|
|
kfree(sp->s2io_entries);
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_freed
|
2008-05-12 17:42:17 +00:00
|
|
|
+= (sp->num_entries * sizeof(struct s2io_msix_entry));
|
2007-05-10 08:22:25 +00:00
|
|
|
}
|
2005-10-04 10:41:24 +00:00
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
return err;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_close -close entry point of the driver
|
|
|
|
* @dev : device pointer.
|
|
|
|
* Description:
|
|
|
|
* This is the stop entry point of the driver. It needs to undo exactly
|
|
|
|
* whatever was done by the open entry point,thus it's usually referred to
|
|
|
|
* as the close function.Among other things this function mainly stops the
|
|
|
|
* Rx side of the NIC and frees all the Rx buffers in the Rx rings.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success and an appropriate (-)ve integer as defined in errno.h
|
|
|
|
* file on failure.
|
|
|
|
*/
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_close(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2008-01-24 09:45:43 +00:00
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
u64 tmp64;
|
|
|
|
int offset;
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2007-11-30 06:46:08 +00:00
|
|
|
/* Return if the device is already closed *
|
|
|
|
* Can happen when s2io_card_up failed in change_mtu *
|
|
|
|
*/
|
|
|
|
if (!is_s2io_card_up(sp))
|
|
|
|
return 0;
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_all_tx_queue(sp);
|
2008-01-24 09:45:43 +00:00
|
|
|
/* delete all populated mac entries */
|
|
|
|
for (offset = 1; offset < config->max_mc_addr; offset++) {
|
|
|
|
tmp64 = do_s2io_read_unicast_mc(sp, offset);
|
|
|
|
if (tmp64 != S2IO_DISABLE_MAC_ENTRY)
|
|
|
|
do_s2io_delete_unicast_mc(sp, tmp64);
|
|
|
|
}
|
|
|
|
|
2006-07-07 06:58:23 +00:00
|
|
|
s2io_card_down(sp);
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_xmit - Tx entry point of te driver
|
|
|
|
* @skb : the socket buffer containing the Tx data.
|
|
|
|
* @dev : device pointer.
|
|
|
|
* Description :
|
|
|
|
* This function is the Tx entry point of the driver. S2IO NIC supports
|
|
|
|
* certain protocol assist features on Tx side, namely CSO, S/G, LSO.
|
|
|
|
* NOTE: when device cant queue the pkt,just the trans_start variable will
|
|
|
|
* not be upadted.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success & 1 on failure.
|
|
|
|
*/
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_xmit(struct sk_buff *skb, struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
u16 frg_cnt, frg_len, i, queue, queue_len, put_off, get_off;
|
|
|
|
register u64 val64;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct TxD *txdp;
|
|
|
|
struct TxFIFO_element __iomem *tx_fifo;
|
2008-01-24 10:08:59 +00:00
|
|
|
unsigned long flags = 0;
|
2005-08-03 19:35:55 +00:00
|
|
|
u16 vlan_tag = 0;
|
2008-01-24 10:08:59 +00:00
|
|
|
struct fifo_info *fifo = NULL;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2008-02-20 22:07:51 +00:00
|
|
|
int do_spin_lock = 1;
|
2006-07-24 23:55:09 +00:00
|
|
|
int offload_type;
|
2008-02-20 22:07:51 +00:00
|
|
|
int enable_per_list_interrupt = 0;
|
2007-07-23 06:37:14 +00:00
|
|
|
struct swStat *stats = &sp->mac_control.stats_info->sw_stat;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
config = &sp->config;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
DBG_PRINT(TX_DBG, "%s: In Neterion Tx routine\n", dev->name);
|
2007-05-10 08:22:25 +00:00
|
|
|
|
|
|
|
if (unlikely(skb->len <= 0)) {
|
|
|
|
DBG_PRINT(TX_DBG, "%s:Buffer has no data..\n", dev->name);
|
|
|
|
dev_kfree_skb_any(skb);
|
|
|
|
return 0;
|
2008-01-24 10:08:59 +00:00
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
|
2007-09-06 10:51:14 +00:00
|
|
|
if (!is_s2io_card_up(sp)) {
|
2005-08-03 19:24:33 +00:00
|
|
|
DBG_PRINT(TX_DBG, "%s: Card going down for reset\n",
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->name);
|
2005-08-03 19:24:33 +00:00
|
|
|
dev_kfree_skb(skb);
|
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
queue = 0;
|
2008-02-20 21:44:07 +00:00
|
|
|
if (sp->vlgrp && vlan_tx_tag_present(skb))
|
2005-08-03 19:35:55 +00:00
|
|
|
vlan_tag = vlan_tx_tag_get(skb);
|
2008-02-20 22:07:51 +00:00
|
|
|
if (sp->config.tx_steering_type == TX_DEFAULT_STEERING) {
|
|
|
|
if (skb->protocol == htons(ETH_P_IP)) {
|
|
|
|
struct iphdr *ip;
|
|
|
|
struct tcphdr *th;
|
|
|
|
ip = ip_hdr(skb);
|
|
|
|
|
|
|
|
if ((ip->frag_off & htons(IP_OFFSET|IP_MF)) == 0) {
|
|
|
|
th = (struct tcphdr *)(((unsigned char *)ip) +
|
|
|
|
ip->ihl*4);
|
|
|
|
|
|
|
|
if (ip->protocol == IPPROTO_TCP) {
|
|
|
|
queue_len = sp->total_tcp_fifos;
|
|
|
|
queue = (ntohs(th->source) +
|
|
|
|
ntohs(th->dest)) &
|
|
|
|
sp->fifo_selector[queue_len - 1];
|
|
|
|
if (queue >= queue_len)
|
|
|
|
queue = queue_len - 1;
|
|
|
|
} else if (ip->protocol == IPPROTO_UDP) {
|
|
|
|
queue_len = sp->total_udp_fifos;
|
|
|
|
queue = (ntohs(th->source) +
|
|
|
|
ntohs(th->dest)) &
|
|
|
|
sp->fifo_selector[queue_len - 1];
|
|
|
|
if (queue >= queue_len)
|
|
|
|
queue = queue_len - 1;
|
|
|
|
queue += sp->udp_fifo_idx;
|
|
|
|
if (skb->len > 1024)
|
|
|
|
enable_per_list_interrupt = 1;
|
|
|
|
do_spin_lock = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (sp->config.tx_steering_type == TX_PRIORITY_STEERING)
|
|
|
|
/* get fifo number based on skb->priority value */
|
|
|
|
queue = config->fifo_mapping
|
|
|
|
[skb->priority & (MAX_TX_FIFOS - 1)];
|
|
|
|
fifo = &mac_control->fifos[queue];
|
2008-02-20 21:44:07 +00:00
|
|
|
|
2008-02-20 22:07:51 +00:00
|
|
|
if (do_spin_lock)
|
|
|
|
spin_lock_irqsave(&fifo->tx_lock, flags);
|
|
|
|
else {
|
|
|
|
if (unlikely(!spin_trylock_irqsave(&fifo->tx_lock, flags)))
|
|
|
|
return NETDEV_TX_LOCKED;
|
|
|
|
}
|
2005-08-03 19:35:55 +00:00
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
if (sp->config.multiq) {
|
|
|
|
if (__netif_subqueue_stopped(dev, fifo->fifo_no)) {
|
|
|
|
spin_unlock_irqrestore(&fifo->tx_lock, flags);
|
|
|
|
return NETDEV_TX_BUSY;
|
|
|
|
}
|
2008-07-09 06:14:24 +00:00
|
|
|
} else if (unlikely(fifo->queue_state == FIFO_QUEUE_STOP)) {
|
2008-02-20 21:44:07 +00:00
|
|
|
if (netif_queue_stopped(dev)) {
|
|
|
|
spin_unlock_irqrestore(&fifo->tx_lock, flags);
|
|
|
|
return NETDEV_TX_BUSY;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
put_off = (u16) fifo->tx_curr_put_info.offset;
|
|
|
|
get_off = (u16) fifo->tx_curr_get_info.offset;
|
|
|
|
txdp = (struct TxD *) fifo->list_info[put_off].list_virt_addr;
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
queue_len = fifo->tx_curr_put_info.fifo_len + 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Avoid "put" pointer going beyond "get" pointer */
|
2006-04-21 23:03:13 +00:00
|
|
|
if (txdp->Host_Control ||
|
|
|
|
((put_off+1) == queue_len ? 0 : (put_off+1)) == get_off) {
|
2005-09-07 04:36:56 +00:00
|
|
|
DBG_PRINT(TX_DBG, "Error in xmit, No free TXDs.\n");
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_tx_queue(sp, fifo->fifo_no);
|
2005-04-16 22:20:36 +00:00
|
|
|
dev_kfree_skb(skb);
|
2008-01-24 10:08:59 +00:00
|
|
|
spin_unlock_irqrestore(&fifo->tx_lock, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
|
|
|
}
|
2005-08-03 19:39:56 +00:00
|
|
|
|
2006-07-24 23:55:09 +00:00
|
|
|
offload_type = s2io_offload_type(skb);
|
|
|
|
if (offload_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
txdp->Control_1 |= TXD_TCP_LSO_EN;
|
2006-07-24 23:55:09 +00:00
|
|
|
txdp->Control_1 |= TXD_TCP_LSO_MSS(s2io_tcp_mss(skb));
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2006-08-29 23:44:56 +00:00
|
|
|
if (skb->ip_summed == CHECKSUM_PARTIAL) {
|
2005-04-16 22:20:36 +00:00
|
|
|
txdp->Control_2 |=
|
|
|
|
(TXD_TX_CKO_IPV4_EN | TXD_TX_CKO_TCP_EN |
|
|
|
|
TXD_TX_CKO_UDP_EN);
|
|
|
|
}
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp->Control_1 |= TXD_GATHER_CODE_FIRST;
|
|
|
|
txdp->Control_1 |= TXD_LIST_OWN_XENA;
|
2008-01-24 10:08:59 +00:00
|
|
|
txdp->Control_2 |= TXD_INT_NUMBER(fifo->fifo_no);
|
2008-02-20 22:07:51 +00:00
|
|
|
if (enable_per_list_interrupt)
|
|
|
|
if (put_off & (queue_len >> 5))
|
|
|
|
txdp->Control_2 |= TXD_INT_TYPE_PER_LIST;
|
2008-02-20 21:44:07 +00:00
|
|
|
if (vlan_tag) {
|
2005-08-03 19:35:55 +00:00
|
|
|
txdp->Control_2 |= TXD_VLAN_ENABLE;
|
|
|
|
txdp->Control_2 |= TXD_VLAN_TAG(vlan_tag);
|
|
|
|
}
|
|
|
|
|
2005-11-14 20:25:08 +00:00
|
|
|
frg_len = skb->len - skb->data_len;
|
2006-07-24 23:55:09 +00:00
|
|
|
if (offload_type == SKB_GSO_UDP) {
|
2005-11-14 20:25:08 +00:00
|
|
|
int ufo_size;
|
|
|
|
|
2006-07-24 23:55:09 +00:00
|
|
|
ufo_size = s2io_udp_mss(skb);
|
2005-11-14 20:25:08 +00:00
|
|
|
ufo_size &= ~7;
|
|
|
|
txdp->Control_1 |= TXD_UFO_EN;
|
|
|
|
txdp->Control_1 |= TXD_UFO_MSS(ufo_size);
|
|
|
|
txdp->Control_1 |= TXD_BUFFER0_SIZE(8);
|
|
|
|
#ifdef __BIG_ENDIAN
|
2008-03-16 22:23:14 +00:00
|
|
|
/* both variants do cpu_to_be64(be32_to_cpu(...)) */
|
2008-01-24 10:08:59 +00:00
|
|
|
fifo->ufo_in_band_v[put_off] =
|
2008-03-16 22:23:14 +00:00
|
|
|
(__force u64)skb_shinfo(skb)->ip6_frag_id;
|
2005-11-14 20:25:08 +00:00
|
|
|
#else
|
2008-01-24 10:08:59 +00:00
|
|
|
fifo->ufo_in_band_v[put_off] =
|
2008-03-16 22:23:14 +00:00
|
|
|
(__force u64)skb_shinfo(skb)->ip6_frag_id << 32;
|
2005-11-14 20:25:08 +00:00
|
|
|
#endif
|
2008-01-24 10:08:59 +00:00
|
|
|
txdp->Host_Control = (unsigned long)fifo->ufo_in_band_v;
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp->Buffer_Pointer = pci_map_single(sp->pdev,
|
2008-01-24 10:08:59 +00:00
|
|
|
fifo->ufo_in_band_v,
|
2005-11-14 20:25:08 +00:00
|
|
|
sizeof(u64), PCI_DMA_TODEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(sp->pdev, txdp->Buffer_Pointer))
|
2007-07-23 06:37:14 +00:00
|
|
|
goto pci_map_failed;
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp++;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp->Buffer_Pointer = pci_map_single
|
|
|
|
(sp->pdev, skb->data, frg_len, PCI_DMA_TODEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(sp->pdev, txdp->Buffer_Pointer))
|
2007-07-23 06:37:14 +00:00
|
|
|
goto pci_map_failed;
|
|
|
|
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp->Host_Control = (unsigned long) skb;
|
|
|
|
txdp->Control_1 |= TXD_BUFFER0_SIZE(frg_len);
|
2006-07-24 23:55:09 +00:00
|
|
|
if (offload_type == SKB_GSO_UDP)
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp->Control_1 |= TXD_UFO_EN;
|
|
|
|
|
|
|
|
frg_cnt = skb_shinfo(skb)->nr_frags;
|
2005-04-16 22:20:36 +00:00
|
|
|
/* For fragmented SKB. */
|
|
|
|
for (i = 0; i < frg_cnt; i++) {
|
|
|
|
skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
|
2005-08-03 19:39:56 +00:00
|
|
|
/* A '0' length fragment will be ignored */
|
|
|
|
if (!frag->size)
|
|
|
|
continue;
|
2005-04-16 22:20:36 +00:00
|
|
|
txdp++;
|
|
|
|
txdp->Buffer_Pointer = (u64) pci_map_page
|
|
|
|
(sp->pdev, frag->page, frag->page_offset,
|
|
|
|
frag->size, PCI_DMA_TODEVICE);
|
2006-01-19 19:11:54 +00:00
|
|
|
txdp->Control_1 = TXD_BUFFER0_SIZE(frag->size);
|
2006-07-24 23:55:09 +00:00
|
|
|
if (offload_type == SKB_GSO_UDP)
|
2005-11-14 20:25:08 +00:00
|
|
|
txdp->Control_1 |= TXD_UFO_EN;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
txdp->Control_1 |= TXD_GATHER_CODE_LAST;
|
|
|
|
|
2006-07-24 23:55:09 +00:00
|
|
|
if (offload_type == SKB_GSO_UDP)
|
2005-11-14 20:25:08 +00:00
|
|
|
frg_cnt++; /* as Txd0 was used for inband header */
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
tx_fifo = mac_control->tx_FIFO_start[queue];
|
2008-01-24 10:08:59 +00:00
|
|
|
val64 = fifo->list_info[put_off].list_phy_addr;
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &tx_fifo->TxDL_Pointer);
|
|
|
|
|
|
|
|
val64 = (TX_FIFO_LAST_TXD_NUM(frg_cnt) | TX_FIFO_FIRST_LIST |
|
|
|
|
TX_FIFO_LAST_LIST);
|
2006-07-24 23:55:09 +00:00
|
|
|
if (offload_type)
|
2005-11-14 20:25:08 +00:00
|
|
|
val64 |= TX_FIFO_SPECIAL_FUNC;
|
2006-07-24 23:55:09 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &tx_fifo->List_Control);
|
|
|
|
|
2005-08-03 19:41:38 +00:00
|
|
|
mmiowb();
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
put_off++;
|
2008-01-24 10:08:59 +00:00
|
|
|
if (put_off == fifo->tx_curr_put_info.fifo_len + 1)
|
2006-04-21 23:03:13 +00:00
|
|
|
put_off = 0;
|
2008-01-24 10:08:59 +00:00
|
|
|
fifo->tx_curr_put_info.offset = put_off;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Avoid "put" pointer going beyond "get" pointer */
|
2006-04-21 23:03:13 +00:00
|
|
|
if (((put_off+1) == queue_len ? 0 : (put_off+1)) == get_off) {
|
2006-04-21 23:20:22 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.fifo_full_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(TX_DBG,
|
|
|
|
"No free TxDs for xmit, Put: 0x%x Get:0x%x\n",
|
|
|
|
put_off, get_off);
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_tx_queue(sp, fifo->fifo_no);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
mac_control->stats_info->sw_stat.mem_allocated += skb->truesize;
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->trans_start = jiffies;
|
2008-01-24 10:08:59 +00:00
|
|
|
spin_unlock_irqrestore(&fifo->tx_lock, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-03-25 19:11:56 +00:00
|
|
|
if (sp->config.intr_type == MSI_X)
|
|
|
|
tx_intr_handler(fifo);
|
|
|
|
|
2007-07-23 06:37:14 +00:00
|
|
|
return 0;
|
|
|
|
pci_map_failed:
|
|
|
|
stats->pci_map_fail_cnt++;
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_tx_queue(sp, fifo->fifo_no);
|
2007-07-23 06:37:14 +00:00
|
|
|
stats->mem_freed += skb->truesize;
|
|
|
|
dev_kfree_skb(skb);
|
2008-01-24 10:08:59 +00:00
|
|
|
spin_unlock_irqrestore(&fifo->tx_lock, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:34:11 +00:00
|
|
|
static void
|
|
|
|
s2io_alarm_handle(unsigned long data)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *sp = (struct s2io_nic *)data;
|
2007-09-17 20:05:35 +00:00
|
|
|
struct net_device *dev = sp->dev;
|
2005-08-03 19:34:11 +00:00
|
|
|
|
2007-09-17 20:05:35 +00:00
|
|
|
s2io_handle_errors(dev);
|
2005-08-03 19:34:11 +00:00
|
|
|
mod_timer(&sp->alarm_timer, jiffies + HZ / 2);
|
|
|
|
}
|
|
|
|
|
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers
Maintain a per-CPU global "struct pt_regs *" variable which can be used instead
of passing regs around manually through all ~1800 interrupt handlers in the
Linux kernel.
The regs pointer is used in few places, but it potentially costs both stack
space and code to pass it around. On the FRV arch, removing the regs parameter
from all the genirq function results in a 20% speed up of the IRQ exit path
(ie: from leaving timer_interrupt() to leaving do_IRQ()).
Where appropriate, an arch may override the generic storage facility and do
something different with the variable. On FRV, for instance, the address is
maintained in GR28 at all times inside the kernel as part of general exception
handling.
Having looked over the code, it appears that the parameter may be handed down
through up to twenty or so layers of functions. Consider a USB character
device attached to a USB hub, attached to a USB controller that posts its
interrupts through a cascaded auxiliary interrupt controller. A character
device driver may want to pass regs to the sysrq handler through the input
layer which adds another few layers of parameter passing.
I've build this code with allyesconfig for x86_64 and i386. I've runtested the
main part of the code on FRV and i386, though I can't test most of the drivers.
I've also done partial conversion for powerpc and MIPS - these at least compile
with minimal configurations.
This will affect all archs. Mostly the changes should be relatively easy.
Take do_IRQ(), store the regs pointer at the beginning, saving the old one:
struct pt_regs *old_regs = set_irq_regs(regs);
And put the old one back at the end:
set_irq_regs(old_regs);
Don't pass regs through to generic_handle_irq() or __do_IRQ().
In timer_interrupt(), this sort of change will be necessary:
- update_process_times(user_mode(regs));
- profile_tick(CPU_PROFILING, regs);
+ update_process_times(user_mode(get_irq_regs()));
+ profile_tick(CPU_PROFILING);
I'd like to move update_process_times()'s use of get_irq_regs() into itself,
except that i386, alone of the archs, uses something other than user_mode().
Some notes on the interrupt handling in the drivers:
(*) input_dev() is now gone entirely. The regs pointer is no longer stored in
the input_dev struct.
(*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does
something different depending on whether it's been supplied with a regs
pointer or not.
(*) Various IRQ handler function pointers have been moved to type
irq_handler_t.
Signed-Off-By: David Howells <dhowells@redhat.com>
(cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
|
|
|
static irqreturn_t s2io_msix_ring_handle(int irq, void *dev_id)
|
2005-10-04 10:41:24 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct ring_info *ring = (struct ring_info *)dev_id;
|
|
|
|
struct s2io_nic *sp = ring->nic;
|
2008-05-12 17:42:17 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (unlikely(!is_s2io_card_up(sp)))
|
2007-09-06 10:51:14 +00:00
|
|
|
return IRQ_HANDLED;
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (sp->config.napi) {
|
2008-06-02 09:59:02 +00:00
|
|
|
u8 __iomem *addr = NULL;
|
|
|
|
u8 val8 = 0;
|
2008-05-12 17:42:17 +00:00
|
|
|
|
2008-06-02 09:59:02 +00:00
|
|
|
addr = (u8 __iomem *)&bar0->xmsi_mask_reg;
|
2008-05-12 17:42:17 +00:00
|
|
|
addr += (7 - ring->ring_no);
|
|
|
|
val8 = (ring->ring_no == 0) ? 0x7f : 0xff;
|
|
|
|
writeb(val8, addr);
|
|
|
|
val8 = readb(addr);
|
2009-01-20 00:43:59 +00:00
|
|
|
napi_schedule(&ring->napi);
|
2008-05-12 17:42:17 +00:00
|
|
|
} else {
|
|
|
|
rx_intr_handler(ring, 0);
|
2008-07-26 02:44:49 +00:00
|
|
|
s2io_chk_rx_buffers(sp, ring);
|
2008-05-12 17:42:17 +00:00
|
|
|
}
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
2005-10-04 10:41:24 +00:00
|
|
|
return IRQ_HANDLED;
|
|
|
|
}
|
|
|
|
|
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers
Maintain a per-CPU global "struct pt_regs *" variable which can be used instead
of passing regs around manually through all ~1800 interrupt handlers in the
Linux kernel.
The regs pointer is used in few places, but it potentially costs both stack
space and code to pass it around. On the FRV arch, removing the regs parameter
from all the genirq function results in a 20% speed up of the IRQ exit path
(ie: from leaving timer_interrupt() to leaving do_IRQ()).
Where appropriate, an arch may override the generic storage facility and do
something different with the variable. On FRV, for instance, the address is
maintained in GR28 at all times inside the kernel as part of general exception
handling.
Having looked over the code, it appears that the parameter may be handed down
through up to twenty or so layers of functions. Consider a USB character
device attached to a USB hub, attached to a USB controller that posts its
interrupts through a cascaded auxiliary interrupt controller. A character
device driver may want to pass regs to the sysrq handler through the input
layer which adds another few layers of parameter passing.
I've build this code with allyesconfig for x86_64 and i386. I've runtested the
main part of the code on FRV and i386, though I can't test most of the drivers.
I've also done partial conversion for powerpc and MIPS - these at least compile
with minimal configurations.
This will affect all archs. Mostly the changes should be relatively easy.
Take do_IRQ(), store the regs pointer at the beginning, saving the old one:
struct pt_regs *old_regs = set_irq_regs(regs);
And put the old one back at the end:
set_irq_regs(old_regs);
Don't pass regs through to generic_handle_irq() or __do_IRQ().
In timer_interrupt(), this sort of change will be necessary:
- update_process_times(user_mode(regs));
- profile_tick(CPU_PROFILING, regs);
+ update_process_times(user_mode(get_irq_regs()));
+ profile_tick(CPU_PROFILING);
I'd like to move update_process_times()'s use of get_irq_regs() into itself,
except that i386, alone of the archs, uses something other than user_mode().
Some notes on the interrupt handling in the drivers:
(*) input_dev() is now gone entirely. The regs pointer is no longer stored in
the input_dev struct.
(*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does
something different depending on whether it's been supplied with a regs
pointer or not.
(*) Various IRQ handler function pointers have been moved to type
irq_handler_t.
Signed-Off-By: David Howells <dhowells@redhat.com>
(cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
|
|
|
static irqreturn_t s2io_msix_fifo_handle(int irq, void *dev_id)
|
2005-10-04 10:41:24 +00:00
|
|
|
{
|
2008-05-12 17:41:32 +00:00
|
|
|
int i;
|
|
|
|
struct fifo_info *fifos = (struct fifo_info *)dev_id;
|
|
|
|
struct s2io_nic *sp = fifos->nic;
|
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
u64 reason;
|
2005-10-04 10:41:24 +00:00
|
|
|
|
2008-05-12 17:41:32 +00:00
|
|
|
if (unlikely(!is_s2io_card_up(sp)))
|
|
|
|
return IRQ_NONE;
|
|
|
|
|
|
|
|
reason = readq(&bar0->general_int_status);
|
|
|
|
if (unlikely(reason == S2IO_MINUS_ONE))
|
|
|
|
/* Nothing much can be done. Get out */
|
2007-09-06 10:51:14 +00:00
|
|
|
return IRQ_HANDLED;
|
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
if (reason & (GEN_INTR_TXPIC | GEN_INTR_TXTRAFFIC)) {
|
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->general_int_mask);
|
2008-05-12 17:41:32 +00:00
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
if (reason & GEN_INTR_TXPIC)
|
|
|
|
s2io_txpic_intr_handle(sp);
|
2008-05-12 17:41:32 +00:00
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
if (reason & GEN_INTR_TXTRAFFIC)
|
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->tx_traffic_int);
|
2008-05-12 17:41:32 +00:00
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
for (i = 0; i < config->tx_fifo_num; i++)
|
|
|
|
tx_intr_handler(&fifos[i]);
|
2008-05-12 17:41:32 +00:00
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
writeq(sp->general_int_mask, &bar0->general_int_mask);
|
|
|
|
readl(&bar0->general_int_status);
|
|
|
|
return IRQ_HANDLED;
|
|
|
|
}
|
|
|
|
/* The interrupt was not raised by us */
|
|
|
|
return IRQ_NONE;
|
2005-10-04 10:41:24 +00:00
|
|
|
}
|
2008-05-12 17:41:32 +00:00
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_txpic_intr_handle(struct s2io_nic *sp)
|
2005-08-03 19:38:59 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-08-03 19:38:59 +00:00
|
|
|
u64 val64;
|
|
|
|
|
|
|
|
val64 = readq(&bar0->pic_int_status);
|
|
|
|
if (val64 & PIC_INT_GPIO) {
|
|
|
|
val64 = readq(&bar0->gpio_int_reg);
|
|
|
|
if ((val64 & GPIO_INT_REG_LINK_DOWN) &&
|
|
|
|
(val64 & GPIO_INT_REG_LINK_UP)) {
|
2006-04-21 23:18:03 +00:00
|
|
|
/*
|
|
|
|
* This is unstable state so clear both up/down
|
|
|
|
* interrupt and adapter to re-evaluate the link state.
|
|
|
|
*/
|
2005-08-03 19:38:59 +00:00
|
|
|
val64 |= GPIO_INT_REG_LINK_DOWN;
|
|
|
|
val64 |= GPIO_INT_REG_LINK_UP;
|
|
|
|
writeq(val64, &bar0->gpio_int_reg);
|
|
|
|
val64 = readq(&bar0->gpio_int_mask);
|
2006-04-21 23:18:03 +00:00
|
|
|
val64 &= ~(GPIO_INT_MASK_LINK_UP |
|
|
|
|
GPIO_INT_MASK_LINK_DOWN);
|
2005-08-03 19:38:59 +00:00
|
|
|
writeq(val64, &bar0->gpio_int_mask);
|
|
|
|
}
|
2006-04-21 23:18:03 +00:00
|
|
|
else if (val64 & GPIO_INT_REG_LINK_UP) {
|
|
|
|
val64 = readq(&bar0->adapter_status);
|
|
|
|
/* Enable Adapter */
|
2007-01-31 18:30:49 +00:00
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 |= ADAPTER_CNTL_EN;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
val64 |= ADAPTER_LED_ON;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
if (!sp->device_enabled_once)
|
|
|
|
sp->device_enabled_once = 1;
|
2006-04-21 23:18:03 +00:00
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
s2io_link(sp, LINK_UP);
|
|
|
|
/*
|
|
|
|
* unmask link down interrupt and mask link-up
|
|
|
|
* intr
|
|
|
|
*/
|
|
|
|
val64 = readq(&bar0->gpio_int_mask);
|
|
|
|
val64 &= ~GPIO_INT_MASK_LINK_DOWN;
|
|
|
|
val64 |= GPIO_INT_MASK_LINK_UP;
|
|
|
|
writeq(val64, &bar0->gpio_int_mask);
|
2006-04-21 23:18:03 +00:00
|
|
|
|
|
|
|
}else if (val64 & GPIO_INT_REG_LINK_DOWN) {
|
|
|
|
val64 = readq(&bar0->adapter_status);
|
2007-01-31 18:30:49 +00:00
|
|
|
s2io_link(sp, LINK_DOWN);
|
|
|
|
/* Link is down so unmaks link up interrupt */
|
|
|
|
val64 = readq(&bar0->gpio_int_mask);
|
|
|
|
val64 &= ~GPIO_INT_MASK_LINK_UP;
|
|
|
|
val64 |= GPIO_INT_MASK_LINK_DOWN;
|
|
|
|
writeq(val64, &bar0->gpio_int_mask);
|
2007-02-24 07:01:31 +00:00
|
|
|
|
|
|
|
/* turn off LED */
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 = val64 &(~ADAPTER_LED_ON);
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
2005-08-03 19:38:59 +00:00
|
|
|
}
|
|
|
|
}
|
2006-04-21 23:18:03 +00:00
|
|
|
val64 = readq(&bar0->gpio_int_mask);
|
2005-08-03 19:38:59 +00:00
|
|
|
}
|
|
|
|
|
2007-09-17 20:05:35 +00:00
|
|
|
/**
|
|
|
|
* do_s2io_chk_alarm_bit - Check for alarm and incrment the counter
|
|
|
|
* @value: alarm bits
|
|
|
|
* @addr: address value
|
|
|
|
* @cnt: counter variable
|
|
|
|
* Description: Check for alarm and increment the counter
|
|
|
|
* Return Value:
|
|
|
|
* 1 - if alarm bit set
|
|
|
|
* 0 - if alarm bit is not set
|
|
|
|
*/
|
2007-10-05 19:39:21 +00:00
|
|
|
static int do_s2io_chk_alarm_bit(u64 value, void __iomem * addr,
|
2007-09-17 20:05:35 +00:00
|
|
|
unsigned long long *cnt)
|
|
|
|
{
|
|
|
|
u64 val64;
|
|
|
|
val64 = readq(addr);
|
|
|
|
if ( val64 & value ) {
|
|
|
|
writeq(val64, addr);
|
|
|
|
(*cnt)++;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_handle_errors - Xframe error indication handler
|
|
|
|
* @nic: device private variable
|
|
|
|
* Description: Handle alarms such as loss of link, single or
|
|
|
|
* double ECC errors, critical and serious errors.
|
|
|
|
* Return Value:
|
|
|
|
* NONE
|
|
|
|
*/
|
|
|
|
static void s2io_handle_errors(void * dev_id)
|
|
|
|
{
|
|
|
|
struct net_device *dev = (struct net_device *) dev_id;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-09-17 20:05:35 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
|
|
|
u64 temp64 = 0,val64=0;
|
|
|
|
int i = 0;
|
|
|
|
|
|
|
|
struct swStat *sw_stat = &sp->mac_control.stats_info->sw_stat;
|
|
|
|
struct xpakStat *stats = &sp->mac_control.stats_info->xpak_stat;
|
|
|
|
|
2007-09-06 10:51:14 +00:00
|
|
|
if (!is_s2io_card_up(sp))
|
2007-09-17 20:05:35 +00:00
|
|
|
return;
|
|
|
|
|
|
|
|
if (pci_channel_offline(sp->pdev))
|
|
|
|
return;
|
|
|
|
|
|
|
|
memset(&sw_stat->ring_full_cnt, 0,
|
|
|
|
sizeof(sw_stat->ring_full_cnt));
|
|
|
|
|
|
|
|
/* Handling the XPAK counters update */
|
|
|
|
if(stats->xpak_timer_count < 72000) {
|
|
|
|
/* waiting for an hour */
|
|
|
|
stats->xpak_timer_count++;
|
|
|
|
} else {
|
|
|
|
s2io_updt_xpak_counter(dev);
|
|
|
|
/* reset the count to zero */
|
|
|
|
stats->xpak_timer_count = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Handling link status change error Intr */
|
|
|
|
if (s2io_link_fault_indication(sp) == MAC_RMAC_ERR_TIMER) {
|
|
|
|
val64 = readq(&bar0->mac_rmac_err_reg);
|
|
|
|
writeq(val64, &bar0->mac_rmac_err_reg);
|
|
|
|
if (val64 & RMAC_LINK_STATE_CHANGE_INT)
|
|
|
|
schedule_work(&sp->set_link_task);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* In case of a serious error, the device will be Reset. */
|
|
|
|
if (do_s2io_chk_alarm_bit(SERR_SOURCE_ANY, &bar0->serr_source,
|
|
|
|
&sw_stat->serious_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
|
|
|
|
/* Check for data parity error */
|
|
|
|
if (do_s2io_chk_alarm_bit(GPIO_INT_REG_DP_ERR_INT, &bar0->gpio_int_reg,
|
|
|
|
&sw_stat->parity_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
|
|
|
|
/* Check for ring full counter */
|
|
|
|
if (sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
val64 = readq(&bar0->ring_bump_counter1);
|
|
|
|
for (i=0; i<4; i++) {
|
|
|
|
temp64 = ( val64 & vBIT(0xFFFF,(i*16),16));
|
|
|
|
temp64 >>= 64 - ((i+1)*16);
|
|
|
|
sw_stat->ring_full_cnt[i] += temp64;
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->ring_bump_counter2);
|
|
|
|
for (i=0; i<4; i++) {
|
|
|
|
temp64 = ( val64 & vBIT(0xFFFF,(i*16),16));
|
|
|
|
temp64 >>= 64 - ((i+1)*16);
|
|
|
|
sw_stat->ring_full_cnt[i+4] += temp64;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->txdma_int_status);
|
|
|
|
/*check for pfc_err*/
|
|
|
|
if (val64 & TXDMA_PFC_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(PFC_ECC_DB_ERR | PFC_SM_ERR_ALARM|
|
|
|
|
PFC_MISC_0_ERR | PFC_MISC_1_ERR|
|
|
|
|
PFC_PCIX_ERR, &bar0->pfc_err_reg,
|
|
|
|
&sw_stat->pfc_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(PFC_ECC_SG_ERR, &bar0->pfc_err_reg,
|
|
|
|
&sw_stat->pfc_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*check for tda_err*/
|
|
|
|
if (val64 & TXDMA_TDA_INT) {
|
|
|
|
if(do_s2io_chk_alarm_bit(TDA_Fn_ECC_DB_ERR | TDA_SM0_ERR_ALARM |
|
|
|
|
TDA_SM1_ERR_ALARM, &bar0->tda_err_reg,
|
|
|
|
&sw_stat->tda_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(TDA_Fn_ECC_SG_ERR | TDA_PCIX_ERR,
|
|
|
|
&bar0->tda_err_reg, &sw_stat->tda_err_cnt);
|
|
|
|
}
|
|
|
|
/*check for pcc_err*/
|
|
|
|
if (val64 & TXDMA_PCC_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(PCC_SM_ERR_ALARM | PCC_WR_ERR_ALARM
|
|
|
|
| PCC_N_SERR | PCC_6_COF_OV_ERR
|
|
|
|
| PCC_7_COF_OV_ERR | PCC_6_LSO_OV_ERR
|
|
|
|
| PCC_7_LSO_OV_ERR | PCC_FB_ECC_DB_ERR
|
|
|
|
| PCC_TXB_ECC_DB_ERR, &bar0->pcc_err_reg,
|
|
|
|
&sw_stat->pcc_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(PCC_FB_ECC_SG_ERR | PCC_TXB_ECC_SG_ERR,
|
|
|
|
&bar0->pcc_err_reg, &sw_stat->pcc_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*check for tti_err*/
|
|
|
|
if (val64 & TXDMA_TTI_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(TTI_SM_ERR_ALARM, &bar0->tti_err_reg,
|
|
|
|
&sw_stat->tti_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(TTI_ECC_SG_ERR | TTI_ECC_DB_ERR,
|
|
|
|
&bar0->tti_err_reg, &sw_stat->tti_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*check for lso_err*/
|
|
|
|
if (val64 & TXDMA_LSO_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(LSO6_ABORT | LSO7_ABORT
|
|
|
|
| LSO6_SM_ERR_ALARM | LSO7_SM_ERR_ALARM,
|
|
|
|
&bar0->lso_err_reg, &sw_stat->lso_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(LSO6_SEND_OFLOW | LSO7_SEND_OFLOW,
|
|
|
|
&bar0->lso_err_reg, &sw_stat->lso_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*check for tpa_err*/
|
|
|
|
if (val64 & TXDMA_TPA_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(TPA_SM_ERR_ALARM, &bar0->tpa_err_reg,
|
|
|
|
&sw_stat->tpa_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(TPA_TX_FRM_DROP, &bar0->tpa_err_reg,
|
|
|
|
&sw_stat->tpa_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*check for sm_err*/
|
|
|
|
if (val64 & TXDMA_SM_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(SM_SM_ERR_ALARM, &bar0->sm_err_reg,
|
|
|
|
&sw_stat->sm_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->mac_int_status);
|
|
|
|
if (val64 & MAC_INT_STATUS_TMAC_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(TMAC_TX_BUF_OVRN | TMAC_TX_SM_ERR,
|
|
|
|
&bar0->mac_tmac_err_reg,
|
|
|
|
&sw_stat->mac_tmac_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(TMAC_ECC_SG_ERR | TMAC_ECC_DB_ERR
|
|
|
|
| TMAC_DESC_ECC_SG_ERR | TMAC_DESC_ECC_DB_ERR,
|
|
|
|
&bar0->mac_tmac_err_reg,
|
|
|
|
&sw_stat->mac_tmac_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->xgxs_int_status);
|
|
|
|
if (val64 & XGXS_INT_STATUS_TXGXS) {
|
|
|
|
if (do_s2io_chk_alarm_bit(TXGXS_ESTORE_UFLOW | TXGXS_TX_SM_ERR,
|
|
|
|
&bar0->xgxs_txgxs_err_reg,
|
|
|
|
&sw_stat->xgxs_txgxs_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(TXGXS_ECC_SG_ERR | TXGXS_ECC_DB_ERR,
|
|
|
|
&bar0->xgxs_txgxs_err_reg,
|
|
|
|
&sw_stat->xgxs_txgxs_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->rxdma_int_status);
|
|
|
|
if (val64 & RXDMA_INT_RC_INT_M) {
|
|
|
|
if (do_s2io_chk_alarm_bit(RC_PRCn_ECC_DB_ERR | RC_FTC_ECC_DB_ERR
|
|
|
|
| RC_PRCn_SM_ERR_ALARM |RC_FTC_SM_ERR_ALARM,
|
|
|
|
&bar0->rc_err_reg, &sw_stat->rc_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(RC_PRCn_ECC_SG_ERR | RC_FTC_ECC_SG_ERR
|
|
|
|
| RC_RDA_FAIL_WR_Rn, &bar0->rc_err_reg,
|
|
|
|
&sw_stat->rc_err_cnt);
|
|
|
|
if (do_s2io_chk_alarm_bit(PRC_PCI_AB_RD_Rn | PRC_PCI_AB_WR_Rn
|
|
|
|
| PRC_PCI_AB_F_WR_Rn, &bar0->prc_pcix_err_reg,
|
|
|
|
&sw_stat->prc_pcix_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(PRC_PCI_DP_RD_Rn | PRC_PCI_DP_WR_Rn
|
|
|
|
| PRC_PCI_DP_F_WR_Rn, &bar0->prc_pcix_err_reg,
|
|
|
|
&sw_stat->prc_pcix_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (val64 & RXDMA_INT_RPA_INT_M) {
|
|
|
|
if (do_s2io_chk_alarm_bit(RPA_SM_ERR_ALARM | RPA_CREDIT_ERR,
|
|
|
|
&bar0->rpa_err_reg, &sw_stat->rpa_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(RPA_ECC_SG_ERR | RPA_ECC_DB_ERR,
|
|
|
|
&bar0->rpa_err_reg, &sw_stat->rpa_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (val64 & RXDMA_INT_RDA_INT_M) {
|
|
|
|
if (do_s2io_chk_alarm_bit(RDA_RXDn_ECC_DB_ERR
|
|
|
|
| RDA_FRM_ECC_DB_N_AERR | RDA_SM1_ERR_ALARM
|
|
|
|
| RDA_SM0_ERR_ALARM | RDA_RXD_ECC_DB_SERR,
|
|
|
|
&bar0->rda_err_reg, &sw_stat->rda_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(RDA_RXDn_ECC_SG_ERR | RDA_FRM_ECC_SG_ERR
|
|
|
|
| RDA_MISC_ERR | RDA_PCIX_ERR,
|
|
|
|
&bar0->rda_err_reg, &sw_stat->rda_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (val64 & RXDMA_INT_RTI_INT_M) {
|
|
|
|
if (do_s2io_chk_alarm_bit(RTI_SM_ERR_ALARM, &bar0->rti_err_reg,
|
|
|
|
&sw_stat->rti_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(RTI_ECC_SG_ERR | RTI_ECC_DB_ERR,
|
|
|
|
&bar0->rti_err_reg, &sw_stat->rti_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->mac_int_status);
|
|
|
|
if (val64 & MAC_INT_STATUS_RMAC_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(RMAC_RX_BUFF_OVRN | RMAC_RX_SM_ERR,
|
|
|
|
&bar0->mac_rmac_err_reg,
|
|
|
|
&sw_stat->mac_rmac_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
do_s2io_chk_alarm_bit(RMAC_UNUSED_INT|RMAC_SINGLE_ECC_ERR|
|
|
|
|
RMAC_DOUBLE_ECC_ERR, &bar0->mac_rmac_err_reg,
|
|
|
|
&sw_stat->mac_rmac_err_cnt);
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->xgxs_int_status);
|
|
|
|
if (val64 & XGXS_INT_STATUS_RXGXS) {
|
|
|
|
if (do_s2io_chk_alarm_bit(RXGXS_ESTORE_OFLOW | RXGXS_RX_SM_ERR,
|
|
|
|
&bar0->xgxs_rxgxs_err_reg,
|
|
|
|
&sw_stat->xgxs_rxgxs_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->mc_int_status);
|
|
|
|
if(val64 & MC_INT_STATUS_MC_INT) {
|
|
|
|
if (do_s2io_chk_alarm_bit(MC_ERR_REG_SM_ERR, &bar0->mc_err_reg,
|
|
|
|
&sw_stat->mc_err_cnt))
|
|
|
|
goto reset;
|
|
|
|
|
|
|
|
/* Handling Ecc errors */
|
|
|
|
if (val64 & (MC_ERR_REG_ECC_ALL_SNG | MC_ERR_REG_ECC_ALL_DBL)) {
|
|
|
|
writeq(val64, &bar0->mc_err_reg);
|
|
|
|
if (val64 & MC_ERR_REG_ECC_ALL_DBL) {
|
|
|
|
sw_stat->double_ecc_errs++;
|
|
|
|
if (sp->device_type != XFRAME_II_DEVICE) {
|
|
|
|
/*
|
|
|
|
* Reset XframeI only if critical error
|
|
|
|
*/
|
|
|
|
if (val64 &
|
|
|
|
(MC_ERR_REG_MIRI_ECC_DB_ERR_0 |
|
|
|
|
MC_ERR_REG_MIRI_ECC_DB_ERR_1))
|
|
|
|
goto reset;
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
sw_stat->single_ecc_errs++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
|
|
|
|
reset:
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_all_tx_queue(sp);
|
2007-09-17 20:05:35 +00:00
|
|
|
schedule_work(&sp->rst_timer_task);
|
|
|
|
sw_stat->soft_reset_cnt++;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* s2io_isr - ISR handler of the device .
|
|
|
|
* @irq: the irq of the device.
|
|
|
|
* @dev_id: a void pointer to the dev structure of the NIC.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: This function is the ISR handler of the device. It
|
|
|
|
* identifies the reason for the interrupt and calls the relevant
|
|
|
|
* service routines. As a contongency measure, this ISR allocates the
|
2005-04-16 22:20:36 +00:00
|
|
|
* recv buffers, if their numbers are below the panic value which is
|
|
|
|
* presently set to 25% of the original number of rcv buffers allocated.
|
|
|
|
* Return value:
|
2005-08-03 19:24:33 +00:00
|
|
|
* IRQ_HANDLED: will be returned if IRQ was handled by this routine
|
2005-04-16 22:20:36 +00:00
|
|
|
* IRQ_NONE: will be returned if interrupt is not from our device
|
|
|
|
*/
|
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers
Maintain a per-CPU global "struct pt_regs *" variable which can be used instead
of passing regs around manually through all ~1800 interrupt handlers in the
Linux kernel.
The regs pointer is used in few places, but it potentially costs both stack
space and code to pass it around. On the FRV arch, removing the regs parameter
from all the genirq function results in a 20% speed up of the IRQ exit path
(ie: from leaving timer_interrupt() to leaving do_IRQ()).
Where appropriate, an arch may override the generic storage facility and do
something different with the variable. On FRV, for instance, the address is
maintained in GR28 at all times inside the kernel as part of general exception
handling.
Having looked over the code, it appears that the parameter may be handed down
through up to twenty or so layers of functions. Consider a USB character
device attached to a USB hub, attached to a USB controller that posts its
interrupts through a cascaded auxiliary interrupt controller. A character
device driver may want to pass regs to the sysrq handler through the input
layer which adds another few layers of parameter passing.
I've build this code with allyesconfig for x86_64 and i386. I've runtested the
main part of the code on FRV and i386, though I can't test most of the drivers.
I've also done partial conversion for powerpc and MIPS - these at least compile
with minimal configurations.
This will affect all archs. Mostly the changes should be relatively easy.
Take do_IRQ(), store the regs pointer at the beginning, saving the old one:
struct pt_regs *old_regs = set_irq_regs(regs);
And put the old one back at the end:
set_irq_regs(old_regs);
Don't pass regs through to generic_handle_irq() or __do_IRQ().
In timer_interrupt(), this sort of change will be necessary:
- update_process_times(user_mode(regs));
- profile_tick(CPU_PROFILING, regs);
+ update_process_times(user_mode(get_irq_regs()));
+ profile_tick(CPU_PROFILING);
I'd like to move update_process_times()'s use of get_irq_regs() into itself,
except that i386, alone of the archs, uses something other than user_mode().
Some notes on the interrupt handling in the drivers:
(*) input_dev() is now gone entirely. The regs pointer is no longer stored in
the input_dev struct.
(*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does
something different depending on whether it's been supplied with a regs
pointer or not.
(*) Various IRQ handler function pointers have been moved to type
irq_handler_t.
Signed-Off-By: David Howells <dhowells@redhat.com>
(cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 13:55:46 +00:00
|
|
|
static irqreturn_t s2io_isr(int irq, void *dev_id)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = (struct net_device *) dev_id;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-08-03 19:24:33 +00:00
|
|
|
int i;
|
2007-01-31 18:30:49 +00:00
|
|
|
u64 reason = 0;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
|
|
|
|
2007-05-14 23:37:30 +00:00
|
|
|
/* Pretend we handled any irq's from a disconnected card */
|
|
|
|
if (pci_channel_offline(sp->pdev))
|
|
|
|
return IRQ_NONE;
|
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
if (!is_s2io_card_up(sp))
|
2007-09-06 10:51:14 +00:00
|
|
|
return IRQ_NONE;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
config = &sp->config;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Identify the cause for interrupt and call the appropriate
|
|
|
|
* interrupt handler. Causes for the interrupt could be;
|
|
|
|
* 1. Rx of packet.
|
|
|
|
* 2. Tx complete.
|
|
|
|
* 3. Link down.
|
|
|
|
*/
|
|
|
|
reason = readq(&bar0->general_int_status);
|
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
if (unlikely(reason == S2IO_MINUS_ONE) ) {
|
|
|
|
/* Nothing much can be done. Get out */
|
|
|
|
return IRQ_HANDLED;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2006-04-21 23:23:26 +00:00
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
if (reason & (GEN_INTR_RXTRAFFIC |
|
|
|
|
GEN_INTR_TXTRAFFIC | GEN_INTR_TXPIC))
|
|
|
|
{
|
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->general_int_mask);
|
|
|
|
|
|
|
|
if (config->napi) {
|
|
|
|
if (reason & GEN_INTR_RXTRAFFIC) {
|
2009-01-20 00:43:59 +00:00
|
|
|
napi_schedule(&sp->napi);
|
2008-05-12 17:42:17 +00:00
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->rx_traffic_mask);
|
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->rx_traffic_int);
|
|
|
|
readl(&bar0->rx_traffic_int);
|
2007-01-31 18:28:08 +00:00
|
|
|
}
|
2007-09-15 21:24:03 +00:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* rx_traffic_int reg is an R1 register, writing all 1's
|
|
|
|
* will ensure that the actual interrupt causing bit
|
|
|
|
* get's cleared and hence a read can be avoided.
|
|
|
|
*/
|
|
|
|
if (reason & GEN_INTR_RXTRAFFIC)
|
2007-01-31 18:30:49 +00:00
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->rx_traffic_int);
|
2007-09-15 21:24:03 +00:00
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++)
|
2008-05-12 17:42:17 +00:00
|
|
|
rx_intr_handler(&mac_control->rings[i], 0);
|
2007-01-31 18:28:08 +00:00
|
|
|
}
|
2007-09-15 21:24:03 +00:00
|
|
|
|
2007-01-31 18:28:08 +00:00
|
|
|
/*
|
2007-09-15 21:24:03 +00:00
|
|
|
* tx_traffic_int reg is an R1 register, writing all 1's
|
2007-01-31 18:28:08 +00:00
|
|
|
* will ensure that the actual interrupt causing bit get's
|
|
|
|
* cleared and hence a read can be avoided.
|
|
|
|
*/
|
2007-09-15 21:24:03 +00:00
|
|
|
if (reason & GEN_INTR_TXTRAFFIC)
|
|
|
|
writeq(S2IO_MINUS_ONE, &bar0->tx_traffic_int);
|
2007-01-31 18:30:49 +00:00
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
for (i = 0; i < config->tx_fifo_num; i++)
|
|
|
|
tx_intr_handler(&mac_control->fifos[i]);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
if (reason & GEN_INTR_TXPIC)
|
|
|
|
s2io_txpic_intr_handle(sp);
|
2005-08-03 19:32:00 +00:00
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
/*
|
|
|
|
* Reallocate the buffers from the interrupt handler itself.
|
|
|
|
*/
|
|
|
|
if (!config->napi) {
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++)
|
2008-07-26 02:44:49 +00:00
|
|
|
s2io_chk_rx_buffers(sp, &mac_control->rings[i]);
|
2007-09-15 21:24:03 +00:00
|
|
|
}
|
|
|
|
writeq(sp->general_int_mask, &bar0->general_int_mask);
|
|
|
|
readl(&bar0->general_int_status);
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
return IRQ_HANDLED;
|
2007-01-31 18:28:08 +00:00
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
}
|
|
|
|
else if (!reason) {
|
|
|
|
/* The interrupt was not raised by us */
|
|
|
|
return IRQ_NONE;
|
|
|
|
}
|
2007-01-31 18:28:08 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return IRQ_HANDLED;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:29:20 +00:00
|
|
|
/**
|
|
|
|
* s2io_updt_stats -
|
|
|
|
*/
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_updt_stats(struct s2io_nic *sp)
|
2005-08-03 19:29:20 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-08-03 19:29:20 +00:00
|
|
|
u64 val64;
|
|
|
|
int cnt = 0;
|
|
|
|
|
2007-09-06 10:51:14 +00:00
|
|
|
if (is_s2io_card_up(sp)) {
|
2005-08-03 19:29:20 +00:00
|
|
|
/* Apprx 30us on a 133 MHz bus */
|
|
|
|
val64 = SET_UPDT_CLICKS(10) |
|
|
|
|
STAT_CFG_ONE_SHOT_EN | STAT_CFG_STAT_EN;
|
|
|
|
writeq(val64, &bar0->stat_cfg);
|
|
|
|
do {
|
|
|
|
udelay(100);
|
|
|
|
val64 = readq(&bar0->stat_cfg);
|
2007-10-19 06:40:29 +00:00
|
|
|
if (!(val64 & s2BIT(0)))
|
2005-08-03 19:29:20 +00:00
|
|
|
break;
|
|
|
|
cnt++;
|
|
|
|
if (cnt == 5)
|
|
|
|
break; /* Updt failed */
|
|
|
|
} while(1);
|
2007-09-18 22:14:20 +00:00
|
|
|
}
|
2005-08-03 19:29:20 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_get_stats - Updates the device statistics structure.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @dev : pointer to the device structure.
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* This function updates the device statistics structure in the s2io_nic
|
2005-04-16 22:20:36 +00:00
|
|
|
* structure and returns a pointer to the same.
|
|
|
|
* Return value:
|
|
|
|
* pointer to the updated net_device_stats structure.
|
|
|
|
*/
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static struct net_device_stats *s2io_get_stats(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2008-04-29 01:08:45 +00:00
|
|
|
int i;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
config = &sp->config;
|
|
|
|
|
2005-08-03 19:29:20 +00:00
|
|
|
/* Configure Stats for immediate updt */
|
|
|
|
s2io_updt_stats(sp);
|
|
|
|
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
/* Using sp->stats as a staging area, because reset (due to mtu
|
|
|
|
change, for example) will clear some hardware counters */
|
|
|
|
dev->stats.tx_packets +=
|
|
|
|
le32_to_cpu(mac_control->stats_info->tmac_frms) -
|
|
|
|
sp->stats.tx_packets;
|
2005-08-03 19:29:20 +00:00
|
|
|
sp->stats.tx_packets =
|
|
|
|
le32_to_cpu(mac_control->stats_info->tmac_frms);
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.tx_errors +=
|
|
|
|
le32_to_cpu(mac_control->stats_info->tmac_any_err_frms) -
|
|
|
|
sp->stats.tx_errors;
|
2005-08-03 19:24:33 +00:00
|
|
|
sp->stats.tx_errors =
|
|
|
|
le32_to_cpu(mac_control->stats_info->tmac_any_err_frms);
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.rx_errors +=
|
|
|
|
le64_to_cpu(mac_control->stats_info->rmac_drop_frms) -
|
|
|
|
sp->stats.rx_errors;
|
2005-08-03 19:24:33 +00:00
|
|
|
sp->stats.rx_errors =
|
2006-09-23 00:28:17 +00:00
|
|
|
le64_to_cpu(mac_control->stats_info->rmac_drop_frms);
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.multicast =
|
|
|
|
le32_to_cpu(mac_control->stats_info->rmac_vld_mcst_frms) -
|
|
|
|
sp->stats.multicast;
|
2005-08-03 19:24:33 +00:00
|
|
|
sp->stats.multicast =
|
|
|
|
le32_to_cpu(mac_control->stats_info->rmac_vld_mcst_frms);
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.rx_length_errors =
|
|
|
|
le64_to_cpu(mac_control->stats_info->rmac_long_frms) -
|
|
|
|
sp->stats.rx_length_errors;
|
2005-04-16 22:20:36 +00:00
|
|
|
sp->stats.rx_length_errors =
|
2006-09-23 00:28:17 +00:00
|
|
|
le64_to_cpu(mac_control->stats_info->rmac_long_frms);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
/* collect per-ring rx_packets and rx_bytes */
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.rx_packets = dev->stats.rx_bytes = 0;
|
2008-04-29 01:08:45 +00:00
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.rx_packets += mac_control->rings[i].rx_packets;
|
|
|
|
dev->stats.rx_bytes += mac_control->rings[i].rx_bytes;
|
2008-04-29 01:08:45 +00:00
|
|
|
}
|
|
|
|
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
return (&dev->stats);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_set_multicast - entry point for multicast address enable/disable.
|
|
|
|
* @dev : pointer to the device structure
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* This function is a driver entry point which gets called by the kernel
|
|
|
|
* whenever multicast addresses must be enabled/disabled. This also gets
|
2005-04-16 22:20:36 +00:00
|
|
|
* called to set/reset promiscuous mode. Depending on the deivce flag, we
|
|
|
|
* determine, if multicast address must be enabled or if promiscuous mode
|
|
|
|
* is to be disabled etc.
|
|
|
|
* Return value:
|
|
|
|
* void.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void s2io_set_multicast(struct net_device *dev)
|
|
|
|
{
|
|
|
|
int i, j, prev_cnt;
|
|
|
|
struct dev_mc_list *mclist;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64 = 0, multi_mac = 0x010203040506ULL, mask =
|
|
|
|
0xfeffffffffffULL;
|
2008-01-24 09:45:43 +00:00
|
|
|
u64 dis_addr = S2IO_DISABLE_MAC_ENTRY, mac_addr = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
void __iomem *add;
|
2008-01-24 09:45:43 +00:00
|
|
|
struct config_param *config = &sp->config;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if ((dev->flags & IFF_ALLMULTI) && (!sp->m_cast_flg)) {
|
|
|
|
/* Enable all Multicast addresses */
|
|
|
|
writeq(RMAC_ADDR_DATA0_MEM_ADDR(multi_mac),
|
|
|
|
&bar0->rmac_addr_data0_mem);
|
|
|
|
writeq(RMAC_ADDR_DATA1_MEM_MASK(mask),
|
|
|
|
&bar0->rmac_addr_data1_mem);
|
|
|
|
val64 = RMAC_ADDR_CMD_MEM_WE |
|
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
2008-01-24 09:45:43 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET(config->max_mc_addr - 1);
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
|
|
|
/* Wait till command completes */
|
2006-04-21 23:18:03 +00:00
|
|
|
wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
2007-02-24 06:57:32 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING,
|
|
|
|
S2IO_BIT_RESET);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
sp->m_cast_flg = 1;
|
2008-01-24 09:45:43 +00:00
|
|
|
sp->all_multi_pos = config->max_mc_addr - 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
} else if ((dev->flags & IFF_ALLMULTI) && (sp->m_cast_flg)) {
|
|
|
|
/* Disable all Multicast addresses */
|
|
|
|
writeq(RMAC_ADDR_DATA0_MEM_ADDR(dis_addr),
|
|
|
|
&bar0->rmac_addr_data0_mem);
|
2005-08-03 19:27:09 +00:00
|
|
|
writeq(RMAC_ADDR_DATA1_MEM_MASK(0x0),
|
|
|
|
&bar0->rmac_addr_data1_mem);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = RMAC_ADDR_CMD_MEM_WE |
|
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET(sp->all_multi_pos);
|
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
|
|
|
/* Wait till command completes */
|
2006-04-21 23:18:03 +00:00
|
|
|
wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
2007-02-24 06:57:32 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING,
|
|
|
|
S2IO_BIT_RESET);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
sp->m_cast_flg = 0;
|
|
|
|
sp->all_multi_pos = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((dev->flags & IFF_PROMISC) && (!sp->promisc_flg)) {
|
|
|
|
/* Put the NIC into promiscuous mode */
|
|
|
|
add = &bar0->mac_cfg;
|
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
val64 |= MAC_CFG_RMAC_PROM_ENABLE;
|
|
|
|
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) val64, add);
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64 >> 32), (add + 4));
|
|
|
|
|
2007-02-24 06:59:39 +00:00
|
|
|
if (vlan_tag_strip != 1) {
|
|
|
|
val64 = readq(&bar0->rx_pa_cfg);
|
|
|
|
val64 &= ~RX_PA_CFG_STRIP_VLAN_TAG;
|
|
|
|
writeq(val64, &bar0->rx_pa_cfg);
|
2008-09-04 20:52:54 +00:00
|
|
|
sp->vlan_strip_flag = 0;
|
2007-02-24 06:59:39 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
sp->promisc_flg = 1;
|
2005-09-07 04:36:56 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: entered promiscuous mode\n",
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->name);
|
|
|
|
} else if (!(dev->flags & IFF_PROMISC) && (sp->promisc_flg)) {
|
|
|
|
/* Remove the NIC from promiscuous mode */
|
|
|
|
add = &bar0->mac_cfg;
|
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
val64 &= ~MAC_CFG_RMAC_PROM_ENABLE;
|
|
|
|
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) val64, add);
|
|
|
|
writeq(RMAC_CFG_KEY(0x4C0D), &bar0->rmac_cfg_key);
|
|
|
|
writel((u32) (val64 >> 32), (add + 4));
|
|
|
|
|
2007-02-24 06:59:39 +00:00
|
|
|
if (vlan_tag_strip != 0) {
|
|
|
|
val64 = readq(&bar0->rx_pa_cfg);
|
|
|
|
val64 |= RX_PA_CFG_STRIP_VLAN_TAG;
|
|
|
|
writeq(val64, &bar0->rx_pa_cfg);
|
2008-09-04 20:52:54 +00:00
|
|
|
sp->vlan_strip_flag = 1;
|
2007-02-24 06:59:39 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->mac_cfg);
|
|
|
|
sp->promisc_flg = 0;
|
2005-09-07 04:36:56 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: left promiscuous mode\n",
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->name);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Update individual M_CAST address list */
|
|
|
|
if ((!sp->m_cast_flg) && dev->mc_count) {
|
|
|
|
if (dev->mc_count >
|
2008-01-24 09:45:43 +00:00
|
|
|
(config->max_mc_addr - config->max_mac_addr)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: No more Rx filters ",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "can be added, please enable ");
|
|
|
|
DBG_PRINT(ERR_DBG, "ALL_MULTI instead\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
prev_cnt = sp->mc_addr_count;
|
|
|
|
sp->mc_addr_count = dev->mc_count;
|
|
|
|
|
|
|
|
/* Clear out the previous list of Mc in the H/W. */
|
|
|
|
for (i = 0; i < prev_cnt; i++) {
|
|
|
|
writeq(RMAC_ADDR_DATA0_MEM_ADDR(dis_addr),
|
|
|
|
&bar0->rmac_addr_data0_mem);
|
|
|
|
writeq(RMAC_ADDR_DATA1_MEM_MASK(0ULL),
|
2005-08-03 19:24:33 +00:00
|
|
|
&bar0->rmac_addr_data1_mem);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = RMAC_ADDR_CMD_MEM_WE |
|
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET
|
2008-01-24 09:45:43 +00:00
|
|
|
(config->mc_start_offset + i);
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
|
|
|
|
|
|
|
/* Wait for command completes */
|
2006-04-21 23:18:03 +00:00
|
|
|
if (wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
2007-02-24 06:57:32 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING,
|
|
|
|
S2IO_BIT_RESET)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Adding ",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "Multicasts failed\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Create the new Rx filter list and update the same in H/W. */
|
|
|
|
for (i = 0, mclist = dev->mc_list; i < dev->mc_count;
|
|
|
|
i++, mclist = mclist->next) {
|
|
|
|
memcpy(sp->usr_addrs[i].addr, mclist->dmi_addr,
|
|
|
|
ETH_ALEN);
|
2006-03-04 17:06:51 +00:00
|
|
|
mac_addr = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
for (j = 0; j < ETH_ALEN; j++) {
|
|
|
|
mac_addr |= mclist->dmi_addr[j];
|
|
|
|
mac_addr <<= 8;
|
|
|
|
}
|
|
|
|
mac_addr >>= 8;
|
|
|
|
writeq(RMAC_ADDR_DATA0_MEM_ADDR(mac_addr),
|
|
|
|
&bar0->rmac_addr_data0_mem);
|
|
|
|
writeq(RMAC_ADDR_DATA1_MEM_MASK(0ULL),
|
2005-08-03 19:24:33 +00:00
|
|
|
&bar0->rmac_addr_data1_mem);
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = RMAC_ADDR_CMD_MEM_WE |
|
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET
|
2008-01-24 09:45:43 +00:00
|
|
|
(i + config->mc_start_offset);
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
|
|
|
|
|
|
|
/* Wait for command completes */
|
2006-04-21 23:18:03 +00:00
|
|
|
if (wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
2007-02-24 06:57:32 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING,
|
|
|
|
S2IO_BIT_RESET)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Adding ",
|
|
|
|
dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "Multicasts failed\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-01-24 09:45:43 +00:00
|
|
|
/* read from CAM unicast & multicast addresses and store it in
|
|
|
|
* def_mac_addr structure
|
|
|
|
*/
|
2008-12-26 07:56:45 +00:00
|
|
|
static void do_s2io_store_unicast_mc(struct s2io_nic *sp)
|
2008-01-24 09:45:43 +00:00
|
|
|
{
|
|
|
|
int offset;
|
|
|
|
u64 mac_addr = 0x0;
|
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
|
|
|
|
/* store unicast & multicast mac addresses */
|
|
|
|
for (offset = 0; offset < config->max_mc_addr; offset++) {
|
|
|
|
mac_addr = do_s2io_read_unicast_mc(sp, offset);
|
|
|
|
/* if read fails disable the entry */
|
|
|
|
if (mac_addr == FAILURE)
|
|
|
|
mac_addr = S2IO_DISABLE_MAC_ENTRY;
|
|
|
|
do_s2io_copy_mac_addr(sp, offset, mac_addr);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* restore unicast & multicast MAC to CAM from def_mac_addr structure */
|
|
|
|
static void do_s2io_restore_unicast_mc(struct s2io_nic *sp)
|
|
|
|
{
|
|
|
|
int offset;
|
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
/* restore unicast mac address */
|
|
|
|
for (offset = 0; offset < config->max_mac_addr; offset++)
|
|
|
|
do_s2io_prog_unicast(sp->dev,
|
|
|
|
sp->def_mac_addr[offset].mac_addr);
|
|
|
|
|
|
|
|
/* restore multicast mac address */
|
|
|
|
for (offset = config->mc_start_offset;
|
|
|
|
offset < config->max_mc_addr; offset++)
|
|
|
|
do_s2io_add_mc(sp, sp->def_mac_addr[offset].mac_addr);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* add a multicast MAC address to CAM */
|
|
|
|
static int do_s2io_add_mc(struct s2io_nic *sp, u8 *addr)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
u64 mac_addr = 0;
|
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
|
|
|
|
for (i = 0; i < ETH_ALEN; i++) {
|
|
|
|
mac_addr <<= 8;
|
|
|
|
mac_addr |= addr[i];
|
|
|
|
}
|
|
|
|
if ((0ULL == mac_addr) || (mac_addr == S2IO_DISABLE_MAC_ENTRY))
|
|
|
|
return SUCCESS;
|
|
|
|
|
|
|
|
/* check if the multicast mac already preset in CAM */
|
|
|
|
for (i = config->mc_start_offset; i < config->max_mc_addr; i++) {
|
|
|
|
u64 tmp64;
|
|
|
|
tmp64 = do_s2io_read_unicast_mc(sp, i);
|
|
|
|
if (tmp64 == S2IO_DISABLE_MAC_ENTRY) /* CAM entry is empty */
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (tmp64 == mac_addr)
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
if (i == config->max_mc_addr) {
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"CAM full no space left for multicast MAC\n");
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
/* Update the internal structure with this new mac address */
|
|
|
|
do_s2io_copy_mac_addr(sp, i, mac_addr);
|
|
|
|
|
|
|
|
return (do_s2io_add_mac(sp, mac_addr, i));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* add MAC address to CAM */
|
|
|
|
static int do_s2io_add_mac(struct s2io_nic *sp, u64 addr, int off)
|
2007-09-14 11:39:19 +00:00
|
|
|
{
|
|
|
|
u64 val64;
|
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
|
|
|
|
|
|
|
writeq(RMAC_ADDR_DATA0_MEM_ADDR(addr),
|
|
|
|
&bar0->rmac_addr_data0_mem);
|
|
|
|
|
|
|
|
val64 =
|
|
|
|
RMAC_ADDR_CMD_MEM_WE | RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET(off);
|
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
|
|
|
|
|
|
|
/* Wait till command completes */
|
|
|
|
if (wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING,
|
|
|
|
S2IO_BIT_RESET)) {
|
2008-01-24 09:45:43 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "do_s2io_add_mac failed\n");
|
2007-09-14 11:39:19 +00:00
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
2008-01-24 09:45:43 +00:00
|
|
|
/* deletes a specified unicast/multicast mac entry from CAM */
|
|
|
|
static int do_s2io_delete_unicast_mc(struct s2io_nic *sp, u64 addr)
|
|
|
|
{
|
|
|
|
int offset;
|
|
|
|
u64 dis_addr = S2IO_DISABLE_MAC_ENTRY, tmp64;
|
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
|
|
|
|
for (offset = 1;
|
|
|
|
offset < config->max_mc_addr; offset++) {
|
|
|
|
tmp64 = do_s2io_read_unicast_mc(sp, offset);
|
|
|
|
if (tmp64 == addr) {
|
|
|
|
/* disable the entry by writing 0xffffffffffffULL */
|
|
|
|
if (do_s2io_add_mac(sp, dis_addr, offset) == FAILURE)
|
|
|
|
return FAILURE;
|
|
|
|
/* store the new mac list from CAM */
|
|
|
|
do_s2io_store_unicast_mc(sp);
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
DBG_PRINT(ERR_DBG, "MAC address 0x%llx not found in CAM\n",
|
|
|
|
(unsigned long long)addr);
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* read mac entries from CAM */
|
|
|
|
static u64 do_s2io_read_unicast_mc(struct s2io_nic *sp, int offset)
|
|
|
|
{
|
|
|
|
u64 tmp64 = 0xffffffffffff0000ULL, val64;
|
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
|
|
|
|
|
|
|
/* read mac addr */
|
|
|
|
val64 =
|
|
|
|
RMAC_ADDR_CMD_MEM_RD | RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET(offset);
|
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
|
|
|
|
|
|
|
/* Wait till command completes */
|
|
|
|
if (wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING,
|
|
|
|
S2IO_BIT_RESET)) {
|
|
|
|
DBG_PRINT(INFO_DBG, "do_s2io_read_unicast_mc failed\n");
|
|
|
|
return FAILURE;
|
|
|
|
}
|
|
|
|
tmp64 = readq(&bar0->rmac_addr_data0_mem);
|
|
|
|
return (tmp64 >> 16);
|
|
|
|
}
|
2007-09-14 11:39:19 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_set_mac_addr driver entry point
|
|
|
|
*/
|
2008-01-24 09:45:43 +00:00
|
|
|
|
2007-09-14 11:39:19 +00:00
|
|
|
static int s2io_set_mac_addr(struct net_device *dev, void *p)
|
|
|
|
{
|
|
|
|
struct sockaddr *addr = p;
|
|
|
|
|
|
|
|
if (!is_valid_ether_addr(addr->sa_data))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
|
|
|
|
|
|
|
|
/* store the MAC address in CAM */
|
|
|
|
return (do_s2io_prog_unicast(dev, dev->dev_addr));
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
2007-09-14 11:39:19 +00:00
|
|
|
* do_s2io_prog_unicast - Programs the Xframe mac address
|
2005-04-16 22:20:36 +00:00
|
|
|
* @dev : pointer to the device structure.
|
|
|
|
* @addr: a uchar pointer to the new mac address which is to be set.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description : This procedure will program the Xframe to receive
|
2005-04-16 22:20:36 +00:00
|
|
|
* frames with new Mac Address
|
2005-08-03 19:24:33 +00:00
|
|
|
* Return value: SUCCESS on success and an appropriate (-)ve integer
|
2005-04-16 22:20:36 +00:00
|
|
|
* as defined in errno.h file on failure.
|
|
|
|
*/
|
2008-01-24 09:45:43 +00:00
|
|
|
|
2007-09-14 11:39:19 +00:00
|
|
|
static int do_s2io_prog_unicast(struct net_device *dev, u8 *addr)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-09-14 11:39:19 +00:00
|
|
|
register u64 mac_addr = 0, perm_addr = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
int i;
|
2008-01-24 09:45:43 +00:00
|
|
|
u64 tmp64;
|
|
|
|
struct config_param *config = &sp->config;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2007-09-14 11:39:19 +00:00
|
|
|
* Set the new MAC address as the new unicast filter and reflect this
|
|
|
|
* change on the device address registered with the OS. It will be
|
|
|
|
* at offset 0.
|
|
|
|
*/
|
2005-04-16 22:20:36 +00:00
|
|
|
for (i = 0; i < ETH_ALEN; i++) {
|
|
|
|
mac_addr <<= 8;
|
|
|
|
mac_addr |= addr[i];
|
2007-09-14 11:39:19 +00:00
|
|
|
perm_addr <<= 8;
|
|
|
|
perm_addr |= sp->def_mac_addr[0].mac_addr[i];
|
2007-02-24 07:04:24 +00:00
|
|
|
}
|
|
|
|
|
2007-09-14 11:39:19 +00:00
|
|
|
/* check if the dev_addr is different than perm_addr */
|
|
|
|
if (mac_addr == perm_addr)
|
2007-02-24 07:04:24 +00:00
|
|
|
return SUCCESS;
|
|
|
|
|
2008-01-24 09:45:43 +00:00
|
|
|
/* check if the mac already preset in CAM */
|
|
|
|
for (i = 1; i < config->max_mac_addr; i++) {
|
|
|
|
tmp64 = do_s2io_read_unicast_mc(sp, i);
|
|
|
|
if (tmp64 == S2IO_DISABLE_MAC_ENTRY) /* CAM entry is empty */
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (tmp64 == mac_addr) {
|
|
|
|
DBG_PRINT(INFO_DBG,
|
|
|
|
"MAC addr:0x%llx already present in CAM\n",
|
|
|
|
(unsigned long long)mac_addr);
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (i == config->max_mac_addr) {
|
|
|
|
DBG_PRINT(ERR_DBG, "CAM full no space left for Unicast MAC\n");
|
|
|
|
return FAILURE;
|
|
|
|
}
|
2007-02-24 07:04:24 +00:00
|
|
|
/* Update the internal structure with this new mac address */
|
2008-01-24 09:45:43 +00:00
|
|
|
do_s2io_copy_mac_addr(sp, i, mac_addr);
|
|
|
|
return (do_s2io_add_mac(sp, mac_addr, i));
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_ethtool_sset - Sets different link parameters.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the * s2io_nic structure.
|
|
|
|
* @info: pointer to the structure with parameters given by ethtool to set
|
|
|
|
* link information.
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* The function sets different link parameters provided by the user onto
|
2005-04-16 22:20:36 +00:00
|
|
|
* the NIC.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int s2io_ethtool_sset(struct net_device *dev,
|
|
|
|
struct ethtool_cmd *info)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
if ((info->autoneg == AUTONEG_ENABLE) ||
|
|
|
|
(info->speed != SPEED_10000) || (info->duplex != DUPLEX_FULL))
|
|
|
|
return -EINVAL;
|
|
|
|
else {
|
|
|
|
s2io_close(sp->dev);
|
|
|
|
s2io_open(sp->dev);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_ethtol_gset - Return link specific information.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @sp : private member of the device structure, pointer to the
|
|
|
|
* s2io_nic structure.
|
|
|
|
* @info : pointer to the structure with parameters given by ethtool
|
|
|
|
* to return link information.
|
|
|
|
* Description:
|
|
|
|
* Returns link specific information like speed, duplex etc.. to ethtool.
|
|
|
|
* Return value :
|
|
|
|
* return 0 on success.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int s2io_ethtool_gset(struct net_device *dev, struct ethtool_cmd *info)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
info->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
|
|
|
|
info->advertising = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
|
|
|
|
info->port = PORT_FIBRE;
|
2007-09-14 11:43:16 +00:00
|
|
|
|
|
|
|
/* info->transceiver */
|
|
|
|
info->transceiver = XCVR_EXTERNAL;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (netif_carrier_ok(sp->dev)) {
|
|
|
|
info->speed = 10000;
|
|
|
|
info->duplex = DUPLEX_FULL;
|
|
|
|
} else {
|
|
|
|
info->speed = -1;
|
|
|
|
info->duplex = -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
info->autoneg = AUTONEG_DISABLE;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_ethtool_gdrvinfo - Returns driver specific information.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
|
|
|
* @info : pointer to the structure with parameters given by ethtool to
|
|
|
|
* return driver information.
|
|
|
|
* Description:
|
|
|
|
* Returns driver specefic information like name, version etc.. to ethtool.
|
|
|
|
* Return value:
|
|
|
|
* void
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void s2io_ethtool_gdrvinfo(struct net_device *dev,
|
|
|
|
struct ethtool_drvinfo *info)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-09-28 21:50:51 +00:00
|
|
|
strncpy(info->driver, s2io_driver_name, sizeof(info->driver));
|
|
|
|
strncpy(info->version, s2io_driver_version, sizeof(info->version));
|
|
|
|
strncpy(info->fw_version, "", sizeof(info->fw_version));
|
|
|
|
strncpy(info->bus_info, pci_name(sp->pdev), sizeof(info->bus_info));
|
2005-04-16 22:20:36 +00:00
|
|
|
info->regdump_len = XENA_REG_SPACE;
|
|
|
|
info->eedump_len = XENA_EEPROM_SPACE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_ethtool_gregs - dumps the entire space of Xfame into the buffer.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @sp: private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @regs : pointer to the structure with parameters given by ethtool for
|
2005-04-16 22:20:36 +00:00
|
|
|
* dumping the registers.
|
|
|
|
* @reg_space: The input argumnet into which all the registers are dumped.
|
|
|
|
* Description:
|
|
|
|
* Dumps the entire register space of xFrame NIC into the user given
|
|
|
|
* buffer area.
|
|
|
|
* Return value :
|
|
|
|
* void .
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void s2io_ethtool_gregs(struct net_device *dev,
|
|
|
|
struct ethtool_regs *regs, void *space)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
u64 reg;
|
|
|
|
u8 *reg_space = (u8 *) space;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
regs->len = XENA_REG_SPACE;
|
|
|
|
regs->version = sp->pdev->subsystem_device;
|
|
|
|
|
|
|
|
for (i = 0; i < regs->len; i += 8) {
|
|
|
|
reg = readq(sp->bar0 + i);
|
|
|
|
memcpy((reg_space + i), ®, 8);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_phy_id - timer function that alternates adapter LED.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @data : address of the private member of the device structure, which
|
2005-04-16 22:20:36 +00:00
|
|
|
* is a pointer to the s2io_nic structure, provided as an u32.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: This is actually the timer function that alternates the
|
|
|
|
* adapter LED bit of the adapter control bit to set/reset every time on
|
|
|
|
* invocation. The timer is set for 1/2 a second, hence tha NIC blinks
|
2005-04-16 22:20:36 +00:00
|
|
|
* once every second.
|
|
|
|
*/
|
|
|
|
static void s2io_phy_id(unsigned long data)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *sp = (struct s2io_nic *) data;
|
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64 = 0;
|
|
|
|
u16 subid;
|
|
|
|
|
|
|
|
subid = sp->pdev->subsystem_device;
|
2005-08-03 19:36:55 +00:00
|
|
|
if ((sp->device_type == XFRAME_II_DEVICE) ||
|
|
|
|
((subid & 0xFF) >= 0x07)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
val64 ^= GPIO_CTRL_GPIO_0;
|
|
|
|
writeq(val64, &bar0->gpio_control);
|
|
|
|
} else {
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 ^= ADAPTER_LED_ON;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
}
|
|
|
|
|
|
|
|
mod_timer(&sp->id_timer, jiffies + HZ / 2);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_ethtool_idnic - To physically identify the nic on the system.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @id : pointer to the structure with identification parameters given by
|
2005-04-16 22:20:36 +00:00
|
|
|
* ethtool.
|
|
|
|
* Description: Used to physically identify the NIC on the system.
|
2005-08-03 19:24:33 +00:00
|
|
|
* The Link LED will blink for a time specified by the user for
|
2005-04-16 22:20:36 +00:00
|
|
|
* identification.
|
2005-08-03 19:24:33 +00:00
|
|
|
* NOTE: The Link has to be Up to be able to blink the LED. Hence
|
2005-04-16 22:20:36 +00:00
|
|
|
* identification is possible only if it's link is up.
|
|
|
|
* Return value:
|
|
|
|
* int , returns 0 on success
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int s2io_ethtool_idnic(struct net_device *dev, u32 data)
|
|
|
|
{
|
|
|
|
u64 val64 = 0, last_gpio_ctrl_val;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u16 subid;
|
|
|
|
|
|
|
|
subid = sp->pdev->subsystem_device;
|
|
|
|
last_gpio_ctrl_val = readq(&bar0->gpio_control);
|
2005-08-03 19:36:55 +00:00
|
|
|
if ((sp->device_type == XFRAME_I_DEVICE) &&
|
|
|
|
((subid & 0xFF) < 0x07)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
if (!(val64 & ADAPTER_CNTL_EN)) {
|
|
|
|
printk(KERN_ERR
|
|
|
|
"Adapter Link down, cannot blink LED\n");
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (sp->id_timer.function == NULL) {
|
|
|
|
init_timer(&sp->id_timer);
|
|
|
|
sp->id_timer.function = s2io_phy_id;
|
|
|
|
sp->id_timer.data = (unsigned long) sp;
|
|
|
|
}
|
|
|
|
mod_timer(&sp->id_timer, jiffies);
|
|
|
|
if (data)
|
2005-08-03 19:24:33 +00:00
|
|
|
msleep_interruptible(data * HZ);
|
2005-04-16 22:20:36 +00:00
|
|
|
else
|
2005-08-03 19:24:33 +00:00
|
|
|
msleep_interruptible(MAX_FLICKER_TIME);
|
2005-04-16 22:20:36 +00:00
|
|
|
del_timer_sync(&sp->id_timer);
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
if (CARDS_WITH_FAULTY_LINK_INDICATORS(sp->device_type, subid)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(last_gpio_ctrl_val, &bar0->gpio_control);
|
|
|
|
last_gpio_ctrl_val = readq(&bar0->gpio_control);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-05-10 08:06:28 +00:00
|
|
|
static void s2io_ethtool_gringparam(struct net_device *dev,
|
|
|
|
struct ethtool_ringparam *ering)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-05-10 08:06:28 +00:00
|
|
|
int i,tx_desc_count=0,rx_desc_count=0;
|
|
|
|
|
|
|
|
if (sp->rxd_mode == RXD_MODE_1)
|
|
|
|
ering->rx_max_pending = MAX_RX_DESC_1;
|
|
|
|
else if (sp->rxd_mode == RXD_MODE_3B)
|
|
|
|
ering->rx_max_pending = MAX_RX_DESC_2;
|
|
|
|
|
|
|
|
ering->tx_max_pending = MAX_TX_DESC;
|
2007-09-18 22:14:20 +00:00
|
|
|
for (i = 0 ; i < sp->config.tx_fifo_num ; i++)
|
2007-05-10 08:06:28 +00:00
|
|
|
tx_desc_count += sp->config.tx_cfg[i].fifo_len;
|
2007-09-18 22:14:20 +00:00
|
|
|
|
2007-05-10 08:06:28 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"\nmax txds : %d\n",sp->config.max_txds);
|
|
|
|
ering->tx_pending = tx_desc_count;
|
|
|
|
rx_desc_count = 0;
|
2007-09-18 22:14:20 +00:00
|
|
|
for (i = 0 ; i < sp->config.rx_ring_num ; i++)
|
2007-05-10 08:06:28 +00:00
|
|
|
rx_desc_count += sp->config.rx_cfg[i].num_rxd;
|
2007-07-23 06:39:43 +00:00
|
|
|
|
2007-05-10 08:06:28 +00:00
|
|
|
ering->rx_pending = rx_desc_count;
|
|
|
|
|
|
|
|
ering->rx_mini_max_pending = 0;
|
|
|
|
ering->rx_mini_pending = 0;
|
|
|
|
if(sp->rxd_mode == RXD_MODE_1)
|
|
|
|
ering->rx_jumbo_max_pending = MAX_RX_DESC_1;
|
|
|
|
else if (sp->rxd_mode == RXD_MODE_3B)
|
|
|
|
ering->rx_jumbo_max_pending = MAX_RX_DESC_2;
|
|
|
|
ering->rx_jumbo_pending = rx_desc_count;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* s2io_ethtool_getpause_data -Pause frame frame generation and reception.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @ep : pointer to the structure with pause parameters given by ethtool.
|
|
|
|
* Description:
|
|
|
|
* Returns the Pause frame generation and reception capability of the NIC.
|
|
|
|
* Return value:
|
|
|
|
* void
|
|
|
|
*/
|
|
|
|
static void s2io_ethtool_getpause_data(struct net_device *dev,
|
|
|
|
struct ethtool_pauseparam *ep)
|
|
|
|
{
|
|
|
|
u64 val64;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
val64 = readq(&bar0->rmac_pause_cfg);
|
|
|
|
if (val64 & RMAC_PAUSE_GEN_ENABLE)
|
|
|
|
ep->tx_pause = TRUE;
|
|
|
|
if (val64 & RMAC_PAUSE_RX_ENABLE)
|
|
|
|
ep->rx_pause = TRUE;
|
|
|
|
ep->autoneg = FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_ethtool_setpause_data - set/reset pause frame generation.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
|
|
|
* @ep : pointer to the structure with pause parameters given by ethtool.
|
|
|
|
* Description:
|
|
|
|
* It can be used to set or reset Pause frame generation or reception
|
|
|
|
* support of the NIC.
|
|
|
|
* Return value:
|
|
|
|
* int, returns 0 on Success
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int s2io_ethtool_setpause_data(struct net_device *dev,
|
2005-08-03 19:24:33 +00:00
|
|
|
struct ethtool_pauseparam *ep)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
u64 val64;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
val64 = readq(&bar0->rmac_pause_cfg);
|
|
|
|
if (ep->tx_pause)
|
|
|
|
val64 |= RMAC_PAUSE_GEN_ENABLE;
|
|
|
|
else
|
|
|
|
val64 &= ~RMAC_PAUSE_GEN_ENABLE;
|
|
|
|
if (ep->rx_pause)
|
|
|
|
val64 |= RMAC_PAUSE_RX_ENABLE;
|
|
|
|
else
|
|
|
|
val64 &= ~RMAC_PAUSE_RX_ENABLE;
|
|
|
|
writeq(val64, &bar0->rmac_pause_cfg);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* read_eeprom - reads 4 bytes of data from user given offset.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
|
|
|
* @off : offset at which the data must be written
|
|
|
|
* @data : Its an output parameter where the data read at the given
|
2005-08-03 19:24:33 +00:00
|
|
|
* offset is stored.
|
2005-04-16 22:20:36 +00:00
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* Will read 4 bytes of data from the user given offset and return the
|
2005-04-16 22:20:36 +00:00
|
|
|
* read data.
|
|
|
|
* NOTE: Will allow to read only part of the EEPROM visible through the
|
|
|
|
* I2C bus.
|
|
|
|
* Return value:
|
|
|
|
* -1 on failure and 0 on success.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define S2IO_DEV_ID 5
|
2007-01-31 19:09:29 +00:00
|
|
|
static int read_eeprom(struct s2io_nic * sp, int off, u64 * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
int ret = -1;
|
|
|
|
u32 exit_cnt = 0;
|
|
|
|
u64 val64;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_I_DEVICE) {
|
|
|
|
val64 = I2C_CONTROL_DEV_ID(S2IO_DEV_ID) | I2C_CONTROL_ADDR(off) |
|
|
|
|
I2C_CONTROL_BYTE_CNT(0x3) | I2C_CONTROL_READ |
|
|
|
|
I2C_CONTROL_CNTL_START;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->i2c_control, LF);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
while (exit_cnt < 5) {
|
|
|
|
val64 = readq(&bar0->i2c_control);
|
|
|
|
if (I2C_CONTROL_CNTL_END(val64)) {
|
|
|
|
*data = I2C_CONTROL_GET_DATA(val64);
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
msleep(50);
|
|
|
|
exit_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
val64 = SPI_CONTROL_KEY(0x9) | SPI_CONTROL_SEL1 |
|
2006-09-13 17:24:59 +00:00
|
|
|
SPI_CONTROL_BYTECNT(0x3) |
|
2005-10-18 01:26:20 +00:00
|
|
|
SPI_CONTROL_CMD(0x3) | SPI_CONTROL_ADDR(off);
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->spi_control, LF);
|
|
|
|
val64 |= SPI_CONTROL_REQ;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->spi_control, LF);
|
|
|
|
while (exit_cnt < 5) {
|
|
|
|
val64 = readq(&bar0->spi_control);
|
|
|
|
if (val64 & SPI_CONTROL_NACK) {
|
|
|
|
ret = 1;
|
|
|
|
break;
|
|
|
|
} else if (val64 & SPI_CONTROL_DONE) {
|
|
|
|
*data = readq(&bar0->spi_data);
|
|
|
|
*data &= 0xffffff;
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
msleep(50);
|
|
|
|
exit_cnt++;
|
|
|
|
}
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* write_eeprom - actually writes the relevant part of the data value.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
|
|
|
* @off : offset at which the data must be written
|
|
|
|
* @data : The data that is to be written
|
2005-08-03 19:24:33 +00:00
|
|
|
* @cnt : Number of bytes of the data that are actually to be written into
|
2005-04-16 22:20:36 +00:00
|
|
|
* the Eeprom. (max of 3)
|
|
|
|
* Description:
|
|
|
|
* Actually writes the relevant part of the data value into the Eeprom
|
|
|
|
* through the I2C bus.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success, -1 on failure.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int write_eeprom(struct s2io_nic * sp, int off, u64 data, int cnt)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
int exit_cnt = 0, ret = -1;
|
|
|
|
u64 val64;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_I_DEVICE) {
|
|
|
|
val64 = I2C_CONTROL_DEV_ID(S2IO_DEV_ID) | I2C_CONTROL_ADDR(off) |
|
|
|
|
I2C_CONTROL_BYTE_CNT(cnt) | I2C_CONTROL_SET_DATA((u32)data) |
|
|
|
|
I2C_CONTROL_CNTL_START;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->i2c_control, LF);
|
|
|
|
|
|
|
|
while (exit_cnt < 5) {
|
|
|
|
val64 = readq(&bar0->i2c_control);
|
|
|
|
if (I2C_CONTROL_CNTL_END(val64)) {
|
|
|
|
if (!(val64 & I2C_CONTROL_NACK))
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
msleep(50);
|
|
|
|
exit_cnt++;
|
|
|
|
}
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
int write_cnt = (cnt == 8) ? 0 : cnt;
|
|
|
|
writeq(SPI_DATA_WRITE(data,(cnt<<3)), &bar0->spi_data);
|
|
|
|
|
|
|
|
val64 = SPI_CONTROL_KEY(0x9) | SPI_CONTROL_SEL1 |
|
2006-09-13 17:24:59 +00:00
|
|
|
SPI_CONTROL_BYTECNT(write_cnt) |
|
2005-10-18 01:26:20 +00:00
|
|
|
SPI_CONTROL_CMD(0x2) | SPI_CONTROL_ADDR(off);
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->spi_control, LF);
|
|
|
|
val64 |= SPI_CONTROL_REQ;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->spi_control, LF);
|
|
|
|
while (exit_cnt < 5) {
|
|
|
|
val64 = readq(&bar0->spi_control);
|
|
|
|
if (val64 & SPI_CONTROL_NACK) {
|
|
|
|
ret = 1;
|
|
|
|
break;
|
|
|
|
} else if (val64 & SPI_CONTROL_DONE) {
|
2005-04-16 22:20:36 +00:00
|
|
|
ret = 0;
|
2005-10-18 01:26:20 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
msleep(50);
|
|
|
|
exit_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_vpd_read(struct s2io_nic *nic)
|
2006-04-21 23:05:41 +00:00
|
|
|
{
|
2006-07-24 23:52:49 +00:00
|
|
|
u8 *vpd_data;
|
|
|
|
u8 data;
|
2006-04-21 23:05:41 +00:00
|
|
|
int i=0, cnt, fail = 0;
|
|
|
|
int vpd_addr = 0x80;
|
|
|
|
|
|
|
|
if (nic->device_type == XFRAME_II_DEVICE) {
|
|
|
|
strcpy(nic->product_name, "Xframe II 10GbE network adapter");
|
|
|
|
vpd_addr = 0x80;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
strcpy(nic->product_name, "Xframe I 10GbE network adapter");
|
|
|
|
vpd_addr = 0x50;
|
|
|
|
}
|
2007-01-31 18:30:49 +00:00
|
|
|
strcpy(nic->serial_num, "NOT AVAILABLE");
|
2006-04-21 23:05:41 +00:00
|
|
|
|
2006-07-24 23:52:49 +00:00
|
|
|
vpd_data = kmalloc(256, GFP_KERNEL);
|
2007-05-10 08:18:54 +00:00
|
|
|
if (!vpd_data) {
|
|
|
|
nic->mac_control.stats_info->sw_stat.mem_alloc_fail_cnt++;
|
2006-07-24 23:52:49 +00:00
|
|
|
return;
|
2007-05-10 08:18:54 +00:00
|
|
|
}
|
2007-05-10 08:22:25 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_allocated += 256;
|
2006-07-24 23:52:49 +00:00
|
|
|
|
2006-04-21 23:05:41 +00:00
|
|
|
for (i = 0; i < 256; i +=4 ) {
|
|
|
|
pci_write_config_byte(nic->pdev, (vpd_addr + 2), i);
|
|
|
|
pci_read_config_byte(nic->pdev, (vpd_addr + 2), &data);
|
|
|
|
pci_write_config_byte(nic->pdev, (vpd_addr + 3), 0);
|
|
|
|
for (cnt = 0; cnt <5; cnt++) {
|
|
|
|
msleep(2);
|
|
|
|
pci_read_config_byte(nic->pdev, (vpd_addr + 3), &data);
|
|
|
|
if (data == 0x80)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (cnt >= 5) {
|
|
|
|
DBG_PRINT(ERR_DBG, "Read of VPD data failed\n");
|
|
|
|
fail = 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
pci_read_config_dword(nic->pdev, (vpd_addr + 4),
|
|
|
|
(u32 *)&vpd_data[i]);
|
|
|
|
}
|
2007-01-31 18:30:49 +00:00
|
|
|
|
|
|
|
if(!fail) {
|
|
|
|
/* read serial number of adapter */
|
|
|
|
for (cnt = 0; cnt < 256; cnt++) {
|
|
|
|
if ((vpd_data[cnt] == 'S') &&
|
|
|
|
(vpd_data[cnt+1] == 'N') &&
|
|
|
|
(vpd_data[cnt+2] < VPD_STRING_LEN)) {
|
|
|
|
memset(nic->serial_num, 0, VPD_STRING_LEN);
|
|
|
|
memcpy(nic->serial_num, &vpd_data[cnt + 3],
|
|
|
|
vpd_data[cnt+2]);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((!fail) && (vpd_data[1] < VPD_STRING_LEN)) {
|
2006-04-21 23:05:41 +00:00
|
|
|
memset(nic->product_name, 0, vpd_data[1]);
|
|
|
|
memcpy(nic->product_name, &vpd_data[3], vpd_data[1]);
|
|
|
|
}
|
2006-07-24 23:52:49 +00:00
|
|
|
kfree(vpd_data);
|
2007-05-10 08:22:25 +00:00
|
|
|
nic->mac_control.stats_info->sw_stat.mem_freed += 256;
|
2006-04-21 23:05:41 +00:00
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* s2io_ethtool_geeprom - reads the value stored in the Eeprom.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the * s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @eeprom : pointer to the user level structure provided by ethtool,
|
2005-04-16 22:20:36 +00:00
|
|
|
* containing all relevant information.
|
|
|
|
* @data_buf : user defined value to be written into Eeprom.
|
|
|
|
* Description: Reads the values stored in the Eeprom at given offset
|
|
|
|
* for a given length. Stores these values int the input argument data
|
|
|
|
* buffer 'data_buf' and returns these to the caller (ethtool.)
|
|
|
|
* Return value:
|
|
|
|
* int 0 on success
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int s2io_ethtool_geeprom(struct net_device *dev,
|
2005-08-03 19:24:33 +00:00
|
|
|
struct ethtool_eeprom *eeprom, u8 * data_buf)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2005-10-18 01:26:20 +00:00
|
|
|
u32 i, valid;
|
|
|
|
u64 data;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
eeprom->magic = sp->pdev->vendor | (sp->pdev->device << 16);
|
|
|
|
|
|
|
|
if ((eeprom->offset + eeprom->len) > (XENA_EEPROM_SPACE))
|
|
|
|
eeprom->len = XENA_EEPROM_SPACE - eeprom->offset;
|
|
|
|
|
|
|
|
for (i = 0; i < eeprom->len; i += 4) {
|
|
|
|
if (read_eeprom(sp, (eeprom->offset + i), &data)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "Read of EEPROM failed\n");
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
valid = INV(data);
|
|
|
|
memcpy((data_buf + i), &valid, 4);
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_ethtool_seeprom - tries to write the user provided value in Eeprom
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @eeprom : pointer to the user level structure provided by ethtool,
|
2005-04-16 22:20:36 +00:00
|
|
|
* containing all relevant information.
|
|
|
|
* @data_buf ; user defined value to be written into Eeprom.
|
|
|
|
* Description:
|
|
|
|
* Tries to write the user provided value in the Eeprom, at the offset
|
|
|
|
* given by the user.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success, -EFAULT on failure.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int s2io_ethtool_seeprom(struct net_device *dev,
|
|
|
|
struct ethtool_eeprom *eeprom,
|
|
|
|
u8 * data_buf)
|
|
|
|
{
|
|
|
|
int len = eeprom->len, cnt = 0;
|
2005-10-18 01:26:20 +00:00
|
|
|
u64 valid = 0, data;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (eeprom->magic != (sp->pdev->vendor | (sp->pdev->device << 16))) {
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"ETHTOOL_WRITE_EEPROM Err: Magic value ");
|
|
|
|
DBG_PRINT(ERR_DBG, "is wrong, Its not 0x%x\n",
|
|
|
|
eeprom->magic);
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
|
|
|
|
while (len) {
|
|
|
|
data = (u32) data_buf[cnt] & 0x000000FF;
|
|
|
|
if (data) {
|
|
|
|
valid = (u32) (data << 24);
|
|
|
|
} else
|
|
|
|
valid = data;
|
|
|
|
|
|
|
|
if (write_eeprom(sp, (eeprom->offset + cnt), valid, 0)) {
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"ETHTOOL_WRITE_EEPROM Err: Cannot ");
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"write into the specified offset\n");
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
cnt++;
|
|
|
|
len--;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_register_test - reads and writes into all clock domains.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
|
|
|
* @data : variable that returns the result of each of the test conducted b
|
|
|
|
* by the driver.
|
|
|
|
* Description:
|
|
|
|
* Read and write into all clock domains. The NIC has 3 clock domains,
|
|
|
|
* see that registers in all the three regions are accessible.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_register_test(struct s2io_nic * sp, uint64_t * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-10-18 01:26:20 +00:00
|
|
|
u64 val64 = 0, exp_val;
|
2005-04-16 22:20:36 +00:00
|
|
|
int fail = 0;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
val64 = readq(&bar0->pif_rd_swapper_fb);
|
|
|
|
if (val64 != 0x123456789abcdefULL) {
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
|
|
|
DBG_PRINT(INFO_DBG, "Read Test level 1 fails\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->rmac_pause_cfg);
|
|
|
|
if (val64 != 0xc000ffff00000000ULL) {
|
|
|
|
fail = 1;
|
|
|
|
DBG_PRINT(INFO_DBG, "Read Test level 2 fails\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->rx_queue_cfg);
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_II_DEVICE)
|
|
|
|
exp_val = 0x0404040404040404ULL;
|
|
|
|
else
|
|
|
|
exp_val = 0x0808080808080808ULL;
|
|
|
|
if (val64 != exp_val) {
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
|
|
|
DBG_PRINT(INFO_DBG, "Read Test level 3 fails\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = readq(&bar0->xgxs_efifo_cfg);
|
|
|
|
if (val64 != 0x000000001923141EULL) {
|
|
|
|
fail = 1;
|
|
|
|
DBG_PRINT(INFO_DBG, "Read Test level 4 fails\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = 0x5A5A5A5A5A5A5A5AULL;
|
|
|
|
writeq(val64, &bar0->xmsi_data);
|
|
|
|
val64 = readq(&bar0->xmsi_data);
|
|
|
|
if (val64 != 0x5A5A5A5A5A5A5A5AULL) {
|
|
|
|
fail = 1;
|
|
|
|
DBG_PRINT(ERR_DBG, "Write Test level 1 fails\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
val64 = 0xA5A5A5A5A5A5A5A5ULL;
|
|
|
|
writeq(val64, &bar0->xmsi_data);
|
|
|
|
val64 = readq(&bar0->xmsi_data);
|
|
|
|
if (val64 != 0xA5A5A5A5A5A5A5A5ULL) {
|
|
|
|
fail = 1;
|
|
|
|
DBG_PRINT(ERR_DBG, "Write Test level 2 fails\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
*data = fail;
|
2005-10-18 01:26:20 +00:00
|
|
|
return fail;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_eeprom_test - to verify that EEprom in the xena can be programmed.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
|
|
|
* @data:variable that returns the result of each of the test conducted by
|
|
|
|
* the driver.
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* Verify that EEPROM in the xena can be programmed using I2C_CONTROL
|
2005-04-16 22:20:36 +00:00
|
|
|
* register.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_eeprom_test(struct s2io_nic * sp, uint64_t * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
int fail = 0;
|
2005-10-18 01:26:20 +00:00
|
|
|
u64 ret_data, org_4F0, org_7F0;
|
|
|
|
u8 saved_4F0 = 0, saved_7F0 = 0;
|
|
|
|
struct net_device *dev = sp->dev;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Test Write Error at offset 0 */
|
2005-10-18 01:26:20 +00:00
|
|
|
/* Note that SPI interface allows write access to all areas
|
|
|
|
* of EEPROM. Hence doing all negative testing only for Xframe I.
|
|
|
|
*/
|
|
|
|
if (sp->device_type == XFRAME_I_DEVICE)
|
|
|
|
if (!write_eeprom(sp, 0, 0, 3))
|
|
|
|
fail = 1;
|
|
|
|
|
|
|
|
/* Save current values at offsets 0x4F0 and 0x7F0 */
|
|
|
|
if (!read_eeprom(sp, 0x4F0, &org_4F0))
|
|
|
|
saved_4F0 = 1;
|
|
|
|
if (!read_eeprom(sp, 0x7F0, &org_7F0))
|
|
|
|
saved_7F0 = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Test Write at offset 4f0 */
|
2005-10-18 01:26:20 +00:00
|
|
|
if (write_eeprom(sp, 0x4F0, 0x012345, 3))
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
|
|
|
if (read_eeprom(sp, 0x4F0, &ret_data))
|
|
|
|
fail = 1;
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (ret_data != 0x012345) {
|
2005-12-15 03:25:23 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: eeprom test error at offset 0x4F0. "
|
|
|
|
"Data written %llx Data read %llx\n",
|
|
|
|
dev->name, (unsigned long long)0x12345,
|
|
|
|
(unsigned long long)ret_data);
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
2005-10-18 01:26:20 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Reset the EEPROM data go FFFF */
|
2005-10-18 01:26:20 +00:00
|
|
|
write_eeprom(sp, 0x4F0, 0xFFFFFF, 3);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Test Write Request Error at offset 0x7c */
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_I_DEVICE)
|
|
|
|
if (!write_eeprom(sp, 0x07C, 0, 3))
|
|
|
|
fail = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
/* Test Write Request at offset 0x7f0 */
|
|
|
|
if (write_eeprom(sp, 0x7F0, 0x012345, 3))
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
2005-10-18 01:26:20 +00:00
|
|
|
if (read_eeprom(sp, 0x7F0, &ret_data))
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (ret_data != 0x012345) {
|
2005-12-15 03:25:23 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: eeprom test error at offset 0x7F0. "
|
|
|
|
"Data written %llx Data read %llx\n",
|
|
|
|
dev->name, (unsigned long long)0x12345,
|
|
|
|
(unsigned long long)ret_data);
|
2005-04-16 22:20:36 +00:00
|
|
|
fail = 1;
|
2005-10-18 01:26:20 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Reset the EEPROM data go FFFF */
|
2005-10-18 01:26:20 +00:00
|
|
|
write_eeprom(sp, 0x7F0, 0xFFFFFF, 3);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
if (sp->device_type == XFRAME_I_DEVICE) {
|
|
|
|
/* Test Write Error at offset 0x80 */
|
|
|
|
if (!write_eeprom(sp, 0x080, 0, 3))
|
|
|
|
fail = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
/* Test Write Error at offset 0xfc */
|
|
|
|
if (!write_eeprom(sp, 0x0FC, 0, 3))
|
|
|
|
fail = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
/* Test Write Error at offset 0x100 */
|
|
|
|
if (!write_eeprom(sp, 0x100, 0, 3))
|
|
|
|
fail = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
/* Test Write Error at offset 4ec */
|
|
|
|
if (!write_eeprom(sp, 0x4EC, 0, 3))
|
|
|
|
fail = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Restore values at offsets 0x4F0 and 0x7F0 */
|
|
|
|
if (saved_4F0)
|
|
|
|
write_eeprom(sp, 0x4F0, org_4F0, 3);
|
|
|
|
if (saved_7F0)
|
|
|
|
write_eeprom(sp, 0x7F0, org_7F0, 3);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
*data = fail;
|
2005-10-18 01:26:20 +00:00
|
|
|
return fail;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_bist_test - invokes the MemBist test of the card .
|
2005-08-03 19:24:33 +00:00
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @data:variable that returns the result of each of the test conducted by
|
2005-04-16 22:20:36 +00:00
|
|
|
* the driver.
|
|
|
|
* Description:
|
|
|
|
* This invokes the MemBist test of the card. We give around
|
|
|
|
* 2 secs time for the Test to complete. If it's still not complete
|
2005-08-03 19:24:33 +00:00
|
|
|
* within this peiod, we consider that the test failed.
|
2005-04-16 22:20:36 +00:00
|
|
|
* Return value:
|
|
|
|
* 0 on success and -1 on failure.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_bist_test(struct s2io_nic * sp, uint64_t * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
u8 bist = 0;
|
|
|
|
int cnt = 0, ret = -1;
|
|
|
|
|
|
|
|
pci_read_config_byte(sp->pdev, PCI_BIST, &bist);
|
|
|
|
bist |= PCI_BIST_START;
|
|
|
|
pci_write_config_word(sp->pdev, PCI_BIST, bist);
|
|
|
|
|
|
|
|
while (cnt < 20) {
|
|
|
|
pci_read_config_byte(sp->pdev, PCI_BIST, &bist);
|
|
|
|
if (!(bist & PCI_BIST_START)) {
|
|
|
|
*data = (bist & PCI_BIST_CODE_MASK);
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
msleep(100);
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io-link_test - verifies the link state of the nic
|
|
|
|
* @sp ; private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
|
|
|
* @data: variable that returns the result of each of the test conducted by
|
|
|
|
* the driver.
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* The function verifies the link state of the NIC and updates the input
|
2005-04-16 22:20:36 +00:00
|
|
|
* argument 'data' appropriately.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_link_test(struct s2io_nic * sp, uint64_t * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64;
|
|
|
|
|
|
|
|
val64 = readq(&bar0->adapter_status);
|
2006-04-21 23:18:03 +00:00
|
|
|
if(!(LINK_IS_UP(val64)))
|
2005-04-16 22:20:36 +00:00
|
|
|
*data = 1;
|
2006-04-21 23:18:03 +00:00
|
|
|
else
|
|
|
|
*data = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-07-24 23:52:49 +00:00
|
|
|
return *data;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_rldram_test - offline test for access to the RldRam chip on the NIC
|
|
|
|
* @sp - private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @data - variable that returns the result of each of the test
|
2005-04-16 22:20:36 +00:00
|
|
|
* conducted by the driver.
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* This is one of the offline test that tests the read and write
|
2005-04-16 22:20:36 +00:00
|
|
|
* access to the RldRam chip on the NIC.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_rldram_test(struct s2io_nic * sp, uint64_t * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
u64 val64;
|
2005-10-18 01:26:20 +00:00
|
|
|
int cnt, iteration = 0, test_fail = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 &= ~ADAPTER_ECC_EN;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
|
|
|
|
val64 = readq(&bar0->mc_rldram_test_ctrl);
|
|
|
|
val64 |= MC_RLDRAM_TEST_MODE;
|
2005-10-18 01:26:20 +00:00
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->mc_rldram_test_ctrl, LF);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
val64 = readq(&bar0->mc_rldram_mrs);
|
|
|
|
val64 |= MC_RLDRAM_QUEUE_SIZE_ENABLE;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->mc_rldram_mrs, UF);
|
|
|
|
|
|
|
|
val64 |= MC_RLDRAM_MRS_ENABLE;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->mc_rldram_mrs, UF);
|
|
|
|
|
|
|
|
while (iteration < 2) {
|
|
|
|
val64 = 0x55555555aaaa0000ULL;
|
|
|
|
if (iteration == 1) {
|
|
|
|
val64 ^= 0xFFFFFFFFFFFF0000ULL;
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->mc_rldram_test_d0);
|
|
|
|
|
|
|
|
val64 = 0xaaaa5a5555550000ULL;
|
|
|
|
if (iteration == 1) {
|
|
|
|
val64 ^= 0xFFFFFFFFFFFF0000ULL;
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->mc_rldram_test_d1);
|
|
|
|
|
|
|
|
val64 = 0x55aaaaaaaa5a0000ULL;
|
|
|
|
if (iteration == 1) {
|
|
|
|
val64 ^= 0xFFFFFFFFFFFF0000ULL;
|
|
|
|
}
|
|
|
|
writeq(val64, &bar0->mc_rldram_test_d2);
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
val64 = (u64) (0x0000003ffffe0100ULL);
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->mc_rldram_test_add);
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
val64 = MC_RLDRAM_TEST_MODE | MC_RLDRAM_TEST_WRITE |
|
|
|
|
MC_RLDRAM_TEST_GO;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->mc_rldram_test_ctrl, LF);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
for (cnt = 0; cnt < 5; cnt++) {
|
|
|
|
val64 = readq(&bar0->mc_rldram_test_ctrl);
|
|
|
|
if (val64 & MC_RLDRAM_TEST_DONE)
|
|
|
|
break;
|
|
|
|
msleep(200);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cnt == 5)
|
|
|
|
break;
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
val64 = MC_RLDRAM_TEST_MODE | MC_RLDRAM_TEST_GO;
|
|
|
|
SPECIAL_REG_WRITE(val64, &bar0->mc_rldram_test_ctrl, LF);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
for (cnt = 0; cnt < 5; cnt++) {
|
|
|
|
val64 = readq(&bar0->mc_rldram_test_ctrl);
|
|
|
|
if (val64 & MC_RLDRAM_TEST_DONE)
|
|
|
|
break;
|
|
|
|
msleep(500);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cnt == 5)
|
|
|
|
break;
|
|
|
|
|
|
|
|
val64 = readq(&bar0->mc_rldram_test_ctrl);
|
2005-10-18 01:26:20 +00:00
|
|
|
if (!(val64 & MC_RLDRAM_TEST_PASS))
|
|
|
|
test_fail = 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
iteration++;
|
|
|
|
}
|
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
*data = test_fail;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-10-18 01:26:20 +00:00
|
|
|
/* Bring the adapter out of test mode */
|
|
|
|
SPECIAL_REG_WRITE(0, &bar0->mc_rldram_test_ctrl, LF);
|
|
|
|
|
|
|
|
return test_fail;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_ethtool_test - conducts 6 tsets to determine the health of card.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
|
|
|
* @ethtest : pointer to a ethtool command specific structure that will be
|
|
|
|
* returned to the user.
|
2005-08-03 19:24:33 +00:00
|
|
|
* @data : variable that returns the result of each of the test
|
2005-04-16 22:20:36 +00:00
|
|
|
* conducted by the driver.
|
|
|
|
* Description:
|
|
|
|
* This function conducts 6 tests ( 4 offline and 2 online) to determine
|
|
|
|
* the health of the card.
|
|
|
|
* Return value:
|
|
|
|
* void
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void s2io_ethtool_test(struct net_device *dev,
|
|
|
|
struct ethtool_test *ethtest,
|
|
|
|
uint64_t * data)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
int orig_state = netif_running(sp->dev);
|
|
|
|
|
|
|
|
if (ethtest->flags == ETH_TEST_FL_OFFLINE) {
|
|
|
|
/* Offline Tests. */
|
2005-08-03 19:24:33 +00:00
|
|
|
if (orig_state)
|
2005-04-16 22:20:36 +00:00
|
|
|
s2io_close(sp->dev);
|
|
|
|
|
|
|
|
if (s2io_register_test(sp, &data[0]))
|
|
|
|
ethtest->flags |= ETH_TEST_FL_FAILED;
|
|
|
|
|
|
|
|
s2io_reset(sp);
|
|
|
|
|
|
|
|
if (s2io_rldram_test(sp, &data[3]))
|
|
|
|
ethtest->flags |= ETH_TEST_FL_FAILED;
|
|
|
|
|
|
|
|
s2io_reset(sp);
|
|
|
|
|
|
|
|
if (s2io_eeprom_test(sp, &data[1]))
|
|
|
|
ethtest->flags |= ETH_TEST_FL_FAILED;
|
|
|
|
|
|
|
|
if (s2io_bist_test(sp, &data[4]))
|
|
|
|
ethtest->flags |= ETH_TEST_FL_FAILED;
|
|
|
|
|
|
|
|
if (orig_state)
|
|
|
|
s2io_open(sp->dev);
|
|
|
|
|
|
|
|
data[2] = 0;
|
|
|
|
} else {
|
|
|
|
/* Online Tests. */
|
|
|
|
if (!orig_state) {
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"%s: is not up, cannot run test\n",
|
|
|
|
dev->name);
|
|
|
|
data[0] = -1;
|
|
|
|
data[1] = -1;
|
|
|
|
data[2] = -1;
|
|
|
|
data[3] = -1;
|
|
|
|
data[4] = -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s2io_link_test(sp, &data[2]))
|
|
|
|
ethtest->flags |= ETH_TEST_FL_FAILED;
|
|
|
|
|
|
|
|
data[0] = 0;
|
|
|
|
data[1] = 0;
|
|
|
|
data[3] = 0;
|
|
|
|
data[4] = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void s2io_get_ethtool_stats(struct net_device *dev,
|
|
|
|
struct ethtool_stats *estats,
|
|
|
|
u64 * tmp_stats)
|
|
|
|
{
|
2007-09-17 20:05:35 +00:00
|
|
|
int i = 0, k;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct stat_block *stat_info = sp->mac_control.stats_info;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:29:20 +00:00
|
|
|
s2io_updt_stats(sp);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_data_octets_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_data_octets);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->tmac_drop_frms);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_mcst_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_mcst_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_bcst_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_bcst_frms);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->tmac_pause_ctrl_frms);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_ttl_octets_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_ttl_octets);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_ucst_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_ucst_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_nucst_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_nucst_frms);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_any_err_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_any_err_frms);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->tmac_ttl_less_fb_octets);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->tmac_vld_ip_octets);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_vld_ip_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_vld_ip);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_drop_ip_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_drop_ip);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_icmp_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_icmp);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->tmac_rst_tcp_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_rst_tcp);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->tmac_tcp);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] = (u64)le32_to_cpu(stat_info->tmac_udp_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->tmac_udp);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_vld_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_vld_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_data_octets_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_data_octets);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_fcs_err_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_drop_frms);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_vld_mcst_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_vld_mcst_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_vld_bcst_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_vld_bcst_frms);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_in_rng_len_err_frms);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_out_rng_len_err_frms);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_long_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_pause_ctrl_frms);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_unsup_ctrl_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_ttl_octets_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_ttl_octets);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_accepted_ucst_frms_oflow)
|
|
|
|
<< 32 | le32_to_cpu(stat_info->rmac_accepted_ucst_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_accepted_nucst_frms_oflow)
|
|
|
|
<< 32 | le32_to_cpu(stat_info->rmac_accepted_nucst_frms);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_discarded_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_discarded_frms);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_drop_events_oflow)
|
|
|
|
<< 32 | le32_to_cpu(stat_info->rmac_drop_events);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_less_fb_octets);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_frms);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_usized_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_usized_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_osized_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_osized_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_frag_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_frag_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_jabber_frms_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_jabber_frms);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_64_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_65_127_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_128_255_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_256_511_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_512_1023_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_1024_1518_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_ip_oflow) << 32 |
|
2005-08-03 19:36:55 +00:00
|
|
|
le32_to_cpu(stat_info->rmac_ip);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ip_octets);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_hdr_err_ip);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_drop_ip_oflow) << 32 |
|
2005-08-03 19:36:55 +00:00
|
|
|
le32_to_cpu(stat_info->rmac_drop_ip);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_icmp_oflow) << 32 |
|
2005-08-03 19:36:55 +00:00
|
|
|
le32_to_cpu(stat_info->rmac_icmp);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_tcp);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_udp_oflow) << 32 |
|
2005-08-03 19:36:55 +00:00
|
|
|
le32_to_cpu(stat_info->rmac_udp);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_err_drp_udp_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_err_drp_udp);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_xgmii_err_sym);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q0);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q1);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q2);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q3);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q4);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q5);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q6);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_frms_q7);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q0);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q1);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q2);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q3);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q4);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q5);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q6);
|
|
|
|
tmp_stats[i++] = le16_to_cpu(stat_info->rmac_full_q7);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_pause_cnt_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_pause_cnt);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_xgmii_data_err_cnt);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_xgmii_ctrl_err_cnt);
|
2005-08-03 19:36:55 +00:00
|
|
|
tmp_stats[i++] =
|
|
|
|
(u64)le32_to_cpu(stat_info->rmac_accepted_ip_oflow) << 32 |
|
|
|
|
le32_to_cpu(stat_info->rmac_accepted_ip);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_err_tcp);
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rd_req_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->new_rd_req_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->new_rd_req_rtry_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rd_rtry_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->wr_rtry_rd_ack_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->wr_req_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->new_wr_req_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->new_wr_req_rtry_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->wr_rtry_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->wr_disc_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rd_rtry_wr_ack_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->txp_wr_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->txd_rd_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->txd_wr_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rxd_rd_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rxd_wr_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->txf_rd_cnt);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rxf_wr_cnt);
|
2007-02-24 07:03:22 +00:00
|
|
|
|
|
|
|
/* Enhanced statistics exist only for Hercules */
|
|
|
|
if(sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
tmp_stats[i++] =
|
|
|
|
le64_to_cpu(stat_info->rmac_ttl_1519_4095_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
le64_to_cpu(stat_info->rmac_ttl_4096_8191_frms);
|
|
|
|
tmp_stats[i++] =
|
|
|
|
le64_to_cpu(stat_info->rmac_ttl_8192_max_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_ttl_gt_max_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_osized_alt_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_jabber_alt_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_gt_max_alt_frms);
|
|
|
|
tmp_stats[i++] = le64_to_cpu(stat_info->rmac_vlan_frms);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_len_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_fcs_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_pf_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_da_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_red_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_rts_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->rmac_ingm_full_discard);
|
|
|
|
tmp_stats[i++] = le32_to_cpu(stat_info->link_fault_cnt);
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:29:20 +00:00
|
|
|
tmp_stats[i++] = 0;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.single_ecc_errs;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.double_ecc_errs;
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.parity_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.serious_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.soft_reset_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.fifo_full_cnt;
|
2007-09-17 20:05:35 +00:00
|
|
|
for (k = 0; k < MAX_RX_RINGS; k++)
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.ring_full_cnt[k];
|
2006-04-21 23:20:22 +00:00
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.alarm_transceiver_temp_high;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.alarm_transceiver_temp_low;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.alarm_laser_bias_current_high;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.alarm_laser_bias_current_low;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.alarm_laser_output_power_high;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.alarm_laser_output_power_low;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.warn_transceiver_temp_high;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.warn_transceiver_temp_low;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.warn_laser_bias_current_high;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.warn_laser_bias_current_low;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.warn_laser_output_power_high;
|
|
|
|
tmp_stats[i++] = stat_info->xpak_stat.warn_laser_output_power_low;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.clubbed_frms_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.sending_both;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.outof_sequence_pkts;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.flush_max_pkts;
|
2006-02-03 09:45:12 +00:00
|
|
|
if (stat_info->sw_stat.num_aggregations) {
|
2006-04-21 23:20:22 +00:00
|
|
|
u64 tmp = stat_info->sw_stat.sum_avg_pkts_aggregated;
|
|
|
|
int count = 0;
|
2006-09-13 17:24:59 +00:00
|
|
|
/*
|
2006-04-21 23:20:22 +00:00
|
|
|
* Since 64-bit divide does not work on all platforms,
|
|
|
|
* do repeated subtraction.
|
|
|
|
*/
|
|
|
|
while (tmp >= stat_info->sw_stat.num_aggregations) {
|
|
|
|
tmp -= stat_info->sw_stat.num_aggregations;
|
|
|
|
count++;
|
|
|
|
}
|
|
|
|
tmp_stats[i++] = count;
|
2006-02-03 09:45:12 +00:00
|
|
|
}
|
2006-04-21 23:20:22 +00:00
|
|
|
else
|
|
|
|
tmp_stats[i++] = 0;
|
2007-05-10 08:18:54 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.mem_alloc_fail_cnt;
|
2007-07-23 06:37:14 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.pci_map_fail_cnt;
|
2007-05-10 08:18:54 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.watchdog_timer_cnt;
|
2007-05-10 08:22:25 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.mem_allocated;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.mem_freed;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.link_up_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.link_down_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.link_up_time;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.link_down_time;
|
|
|
|
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tx_buf_abort_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tx_desc_abort_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tx_parity_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tx_link_loss_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tx_list_proc_err_cnt;
|
|
|
|
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_parity_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_abort_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_parity_abort_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_rda_fail_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_unkn_prot_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_fcs_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_buf_size_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_rxd_corrupt_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rx_unkn_err_cnt;
|
2007-09-17 20:05:35 +00:00
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tda_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.pfc_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.pcc_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tti_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.tpa_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.sm_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.lso_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.mac_tmac_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.mac_rmac_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.xgxs_txgxs_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.xgxs_rxgxs_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rc_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.prc_pcix_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rpa_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rda_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.rti_err_cnt;
|
|
|
|
tmp_stats[i++] = stat_info->sw_stat.mc_err_cnt;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_ethtool_get_regs_len(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
return (XENA_REG_SPACE);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static u32 s2io_ethtool_get_rx_csum(struct net_device * dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
return (sp->rx_csum);
|
|
|
|
}
|
2005-11-06 00:46:47 +00:00
|
|
|
|
|
|
|
static int s2io_ethtool_set_rx_csum(struct net_device *dev, u32 data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (data)
|
|
|
|
sp->rx_csum = 1;
|
|
|
|
else
|
|
|
|
sp->rx_csum = 0;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2005-11-06 00:46:47 +00:00
|
|
|
|
|
|
|
static int s2io_get_eeprom_len(struct net_device *dev)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
return (XENA_EEPROM_SPACE);
|
|
|
|
}
|
|
|
|
|
2007-10-04 01:07:32 +00:00
|
|
|
static int s2io_get_sset_count(struct net_device *dev, int sset)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-10-04 01:07:32 +00:00
|
|
|
|
|
|
|
switch (sset) {
|
|
|
|
case ETH_SS_TEST:
|
|
|
|
return S2IO_TEST_LEN;
|
|
|
|
case ETH_SS_STATS:
|
|
|
|
switch(sp->device_type) {
|
|
|
|
case XFRAME_I_DEVICE:
|
|
|
|
return XFRAME_I_STAT_LEN;
|
|
|
|
case XFRAME_II_DEVICE:
|
|
|
|
return XFRAME_II_STAT_LEN;
|
|
|
|
default:
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2005-11-06 00:46:47 +00:00
|
|
|
|
|
|
|
static void s2io_ethtool_get_strings(struct net_device *dev,
|
|
|
|
u32 stringset, u8 * data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-02-24 07:03:22 +00:00
|
|
|
int stat_size = 0;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-02-24 07:03:22 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
switch (stringset) {
|
|
|
|
case ETH_SS_TEST:
|
|
|
|
memcpy(data, s2io_gstrings, S2IO_STRINGS_LEN);
|
|
|
|
break;
|
|
|
|
case ETH_SS_STATS:
|
2007-02-24 07:03:22 +00:00
|
|
|
stat_size = sizeof(ethtool_xena_stats_keys);
|
|
|
|
memcpy(data, ðtool_xena_stats_keys,stat_size);
|
|
|
|
if(sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
memcpy(data + stat_size,
|
|
|
|
ðtool_enhanced_stats_keys,
|
|
|
|
sizeof(ethtool_enhanced_stats_keys));
|
|
|
|
stat_size += sizeof(ethtool_enhanced_stats_keys);
|
|
|
|
}
|
|
|
|
|
|
|
|
memcpy(data + stat_size, ðtool_driver_stats_keys,
|
|
|
|
sizeof(ethtool_driver_stats_keys));
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_ethtool_op_set_tx_csum(struct net_device *dev, u32 data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
if (data)
|
|
|
|
dev->features |= NETIF_F_IP_CSUM;
|
|
|
|
else
|
|
|
|
dev->features &= ~NETIF_F_IP_CSUM;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-07-24 23:55:09 +00:00
|
|
|
static u32 s2io_ethtool_op_get_tso(struct net_device *dev)
|
|
|
|
{
|
|
|
|
return (dev->features & NETIF_F_TSO) != 0;
|
|
|
|
}
|
|
|
|
static int s2io_ethtool_op_set_tso(struct net_device *dev, u32 data)
|
|
|
|
{
|
|
|
|
if (data)
|
|
|
|
dev->features |= (NETIF_F_TSO | NETIF_F_TSO6);
|
|
|
|
else
|
|
|
|
dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-09-13 18:30:00 +00:00
|
|
|
static const struct ethtool_ops netdev_ethtool_ops = {
|
2005-04-16 22:20:36 +00:00
|
|
|
.get_settings = s2io_ethtool_gset,
|
|
|
|
.set_settings = s2io_ethtool_sset,
|
|
|
|
.get_drvinfo = s2io_ethtool_gdrvinfo,
|
|
|
|
.get_regs_len = s2io_ethtool_get_regs_len,
|
|
|
|
.get_regs = s2io_ethtool_gregs,
|
|
|
|
.get_link = ethtool_op_get_link,
|
|
|
|
.get_eeprom_len = s2io_get_eeprom_len,
|
|
|
|
.get_eeprom = s2io_ethtool_geeprom,
|
|
|
|
.set_eeprom = s2io_ethtool_seeprom,
|
2007-05-10 08:06:28 +00:00
|
|
|
.get_ringparam = s2io_ethtool_gringparam,
|
2005-04-16 22:20:36 +00:00
|
|
|
.get_pauseparam = s2io_ethtool_getpause_data,
|
|
|
|
.set_pauseparam = s2io_ethtool_setpause_data,
|
|
|
|
.get_rx_csum = s2io_ethtool_get_rx_csum,
|
|
|
|
.set_rx_csum = s2io_ethtool_set_rx_csum,
|
|
|
|
.set_tx_csum = s2io_ethtool_op_set_tx_csum,
|
|
|
|
.set_sg = ethtool_op_set_sg,
|
2006-07-24 23:55:09 +00:00
|
|
|
.get_tso = s2io_ethtool_op_get_tso,
|
|
|
|
.set_tso = s2io_ethtool_op_set_tso,
|
2005-11-14 20:25:08 +00:00
|
|
|
.set_ufo = ethtool_op_set_ufo,
|
2005-04-16 22:20:36 +00:00
|
|
|
.self_test = s2io_ethtool_test,
|
|
|
|
.get_strings = s2io_ethtool_get_strings,
|
|
|
|
.phys_id = s2io_ethtool_idnic,
|
2007-10-04 01:07:32 +00:00
|
|
|
.get_ethtool_stats = s2io_get_ethtool_stats,
|
|
|
|
.get_sset_count = s2io_get_sset_count,
|
2005-04-16 22:20:36 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_ioctl - Entry point for the Ioctl
|
2005-04-16 22:20:36 +00:00
|
|
|
* @dev : Device pointer.
|
|
|
|
* @ifr : An IOCTL specefic structure, that can contain a pointer to
|
|
|
|
* a proprietary structure used to pass information to the driver.
|
|
|
|
* @cmd : This is used to distinguish between the different commands that
|
|
|
|
* can be passed to the IOCTL functions.
|
|
|
|
* Description:
|
2005-08-03 19:24:33 +00:00
|
|
|
* Currently there are no special functionality supported in IOCTL, hence
|
|
|
|
* function always return EOPNOTSUPPORTED
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_change_mtu - entry point to change MTU size for the device.
|
|
|
|
* @dev : device pointer.
|
|
|
|
* @new_mtu : the new MTU size for the device.
|
|
|
|
* Description: A driver entry point to change MTU size for the device.
|
|
|
|
* Before changing the MTU the device must be stopped.
|
|
|
|
* Return value:
|
|
|
|
* 0 on success and an appropriate (-)ve integer as defined in errno.h
|
|
|
|
* file on failure.
|
|
|
|
*/
|
|
|
|
|
2005-11-06 00:46:47 +00:00
|
|
|
static int s2io_change_mtu(struct net_device *dev, int new_mtu)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2007-11-30 06:46:08 +00:00
|
|
|
int ret = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if ((new_mtu < MIN_MTU) || (new_mtu > S2IO_JUMBO_SIZE)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: MTU size is invalid.\n",
|
|
|
|
dev->name);
|
|
|
|
return -EPERM;
|
|
|
|
}
|
|
|
|
|
|
|
|
dev->mtu = new_mtu;
|
2005-08-03 19:33:12 +00:00
|
|
|
if (netif_running(dev)) {
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_all_tx_queue(sp);
|
2006-07-07 06:58:23 +00:00
|
|
|
s2io_card_down(sp);
|
2007-11-30 06:46:08 +00:00
|
|
|
ret = s2io_card_up(sp);
|
|
|
|
if (ret) {
|
2005-08-03 19:33:12 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Device bring up failed\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
2007-11-30 06:46:08 +00:00
|
|
|
return ret;
|
2005-08-03 19:33:12 +00:00
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_wake_all_tx_queue(sp);
|
2005-08-03 19:33:12 +00:00
|
|
|
} else { /* Device is down */
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2005-08-03 19:33:12 +00:00
|
|
|
u64 val64 = new_mtu;
|
|
|
|
|
|
|
|
writeq(vBIT(val64, 2, 14), &bar0->rmac_max_pyld_len);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-11-30 06:46:08 +00:00
|
|
|
return ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_set_link - Set the LInk status
|
|
|
|
* @data: long pointer to device private structue
|
|
|
|
* Description: Sets the link status for the adapter
|
|
|
|
*/
|
|
|
|
|
2006-11-22 14:57:56 +00:00
|
|
|
static void s2io_set_link(struct work_struct *work)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *nic = container_of(work, struct s2io_nic, set_link_task);
|
2005-04-16 22:20:36 +00:00
|
|
|
struct net_device *dev = nic->dev;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
2005-04-16 22:20:36 +00:00
|
|
|
register u64 val64;
|
|
|
|
u16 subid;
|
|
|
|
|
2007-02-15 22:37:50 +00:00
|
|
|
rtnl_lock();
|
|
|
|
|
|
|
|
if (!netif_running(dev))
|
|
|
|
goto out_unlock;
|
|
|
|
|
2007-09-06 10:51:14 +00:00
|
|
|
if (test_and_set_bit(__S2IO_STATE_LINK_TASK, &(nic->state))) {
|
2005-04-16 22:20:36 +00:00
|
|
|
/* The card is being reset, no point doing anything */
|
2007-02-15 22:37:50 +00:00
|
|
|
goto out_unlock;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
subid = nic->pdev->subsystem_device;
|
2005-08-03 19:38:59 +00:00
|
|
|
if (s2io_link_fault_indication(nic) == MAC_RMAC_ERR_TIMER) {
|
|
|
|
/*
|
|
|
|
* Allow a small delay for the NICs self initiated
|
|
|
|
* cleanup to complete.
|
|
|
|
*/
|
|
|
|
msleep(100);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
val64 = readq(&bar0->adapter_status);
|
2007-01-31 18:30:49 +00:00
|
|
|
if (LINK_IS_UP(val64)) {
|
|
|
|
if (!(readq(&bar0->adapter_control) & ADAPTER_CNTL_EN)) {
|
|
|
|
if (verify_xena_quiescence(nic)) {
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 |= ADAPTER_CNTL_EN;
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->adapter_control);
|
2007-01-31 18:30:49 +00:00
|
|
|
if (CARDS_WITH_FAULTY_LINK_INDICATORS(
|
|
|
|
nic->device_type, subid)) {
|
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
val64 |= GPIO_CTRL_GPIO_0;
|
|
|
|
writeq(val64, &bar0->gpio_control);
|
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
} else {
|
|
|
|
val64 |= ADAPTER_LED_ON;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
2005-08-03 19:38:59 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
nic->device_enabled_once = TRUE;
|
2007-01-31 18:30:49 +00:00
|
|
|
} else {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Error: ", dev->name);
|
|
|
|
DBG_PRINT(ERR_DBG, "device is not Quiescent\n");
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_all_tx_queue(nic);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2007-01-31 18:30:49 +00:00
|
|
|
}
|
2007-08-06 09:38:19 +00:00
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 |= ADAPTER_LED_ON;
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
|
|
|
s2io_link(nic, LINK_UP);
|
2007-01-31 18:30:49 +00:00
|
|
|
} else {
|
|
|
|
if (CARDS_WITH_FAULTY_LINK_INDICATORS(nic->device_type,
|
|
|
|
subid)) {
|
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
val64 &= ~GPIO_CTRL_GPIO_0;
|
|
|
|
writeq(val64, &bar0->gpio_control);
|
|
|
|
val64 = readq(&bar0->gpio_control);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2007-08-06 09:38:19 +00:00
|
|
|
/* turn off LED */
|
|
|
|
val64 = readq(&bar0->adapter_control);
|
|
|
|
val64 = val64 &(~ADAPTER_LED_ON);
|
|
|
|
writeq(val64, &bar0->adapter_control);
|
2007-01-31 18:30:49 +00:00
|
|
|
s2io_link(nic, LINK_DOWN);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2007-09-06 10:51:14 +00:00
|
|
|
clear_bit(__S2IO_STATE_LINK_TASK, &(nic->state));
|
2007-02-15 22:37:50 +00:00
|
|
|
|
|
|
|
out_unlock:
|
2007-02-24 07:04:24 +00:00
|
|
|
rtnl_unlock();
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int set_rxd_buffer_pointer(struct s2io_nic *sp, struct RxD_t *rxdp,
|
|
|
|
struct buffAdd *ba,
|
|
|
|
struct sk_buff **skb, u64 *temp0, u64 *temp1,
|
|
|
|
u64 *temp2, int size)
|
2006-04-21 23:23:26 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = sp->dev;
|
2007-07-23 06:37:14 +00:00
|
|
|
struct swStat *stats = &sp->mac_control.stats_info->sw_stat;
|
2006-04-21 23:23:26 +00:00
|
|
|
|
|
|
|
if ((sp->rxd_mode == RXD_MODE_1) && (rxdp->Host_Control == 0)) {
|
2007-07-23 06:20:51 +00:00
|
|
|
struct RxD1 *rxdp1 = (struct RxD1 *)rxdp;
|
2006-04-21 23:23:26 +00:00
|
|
|
/* allocate skb */
|
|
|
|
if (*skb) {
|
|
|
|
DBG_PRINT(INFO_DBG, "SKB is not NULL\n");
|
|
|
|
/*
|
|
|
|
* As Rx frame are not going to be processed,
|
|
|
|
* using same mapped address for the Rxd
|
|
|
|
* buffer pointer
|
|
|
|
*/
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1->Buffer0_ptr = *temp0;
|
2006-04-21 23:23:26 +00:00
|
|
|
} else {
|
|
|
|
*skb = dev_alloc_skb(size);
|
|
|
|
if (!(*skb)) {
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: Out of ", dev->name);
|
2007-05-10 08:18:54 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "memory to allocate ");
|
|
|
|
DBG_PRINT(INFO_DBG, "1 buf mode SKBs\n");
|
|
|
|
sp->mac_control.stats_info->sw_stat. \
|
|
|
|
mem_alloc_fail_cnt++;
|
2006-04-21 23:23:26 +00:00
|
|
|
return -ENOMEM ;
|
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_allocated
|
2007-05-10 08:22:25 +00:00
|
|
|
+= (*skb)->truesize;
|
2006-04-21 23:23:26 +00:00
|
|
|
/* storing the mapped addr in a temp variable
|
|
|
|
* such it will be used for next rxd whose
|
|
|
|
* Host Control is NULL
|
|
|
|
*/
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp1->Buffer0_ptr = *temp0 =
|
2006-04-21 23:23:26 +00:00
|
|
|
pci_map_single( sp->pdev, (*skb)->data,
|
|
|
|
size - NET_IP_ALIGN,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(sp->pdev, rxdp1->Buffer0_ptr))
|
2007-07-23 06:37:14 +00:00
|
|
|
goto memalloc_failed;
|
2006-04-21 23:23:26 +00:00
|
|
|
rxdp->Host_Control = (unsigned long) (*skb);
|
|
|
|
}
|
|
|
|
} else if ((sp->rxd_mode == RXD_MODE_3B) && (rxdp->Host_Control == 0)) {
|
2007-07-23 06:20:51 +00:00
|
|
|
struct RxD3 *rxdp3 = (struct RxD3 *)rxdp;
|
2006-04-21 23:23:26 +00:00
|
|
|
/* Two buffer Mode */
|
|
|
|
if (*skb) {
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer2_ptr = *temp2;
|
|
|
|
rxdp3->Buffer0_ptr = *temp0;
|
|
|
|
rxdp3->Buffer1_ptr = *temp1;
|
2006-04-21 23:23:26 +00:00
|
|
|
} else {
|
|
|
|
*skb = dev_alloc_skb(size);
|
2006-10-30 22:19:25 +00:00
|
|
|
if (!(*skb)) {
|
2007-05-10 08:18:54 +00:00
|
|
|
DBG_PRINT(INFO_DBG, "%s: Out of ", dev->name);
|
|
|
|
DBG_PRINT(INFO_DBG, "memory to allocate ");
|
|
|
|
DBG_PRINT(INFO_DBG, "2 buf mode SKBs\n");
|
|
|
|
sp->mac_control.stats_info->sw_stat. \
|
|
|
|
mem_alloc_fail_cnt++;
|
2006-10-30 22:19:25 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_allocated
|
2007-05-10 08:22:25 +00:00
|
|
|
+= (*skb)->truesize;
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer2_ptr = *temp2 =
|
2006-04-21 23:23:26 +00:00
|
|
|
pci_map_single(sp->pdev, (*skb)->data,
|
|
|
|
dev->mtu + 4,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(sp->pdev, rxdp3->Buffer2_ptr))
|
2007-07-23 06:37:14 +00:00
|
|
|
goto memalloc_failed;
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer0_ptr = *temp0 =
|
2006-04-21 23:23:26 +00:00
|
|
|
pci_map_single( sp->pdev, ba->ba_0, BUF0_LEN,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(sp->pdev,
|
|
|
|
rxdp3->Buffer0_ptr)) {
|
2007-07-23 06:37:14 +00:00
|
|
|
pci_unmap_single (sp->pdev,
|
2007-08-02 18:21:30 +00:00
|
|
|
(dma_addr_t)rxdp3->Buffer2_ptr,
|
2007-07-23 06:37:14 +00:00
|
|
|
dev->mtu + 4, PCI_DMA_FROMDEVICE);
|
|
|
|
goto memalloc_failed;
|
|
|
|
}
|
2006-04-21 23:23:26 +00:00
|
|
|
rxdp->Host_Control = (unsigned long) (*skb);
|
|
|
|
|
|
|
|
/* Buffer-1 will be dummy buffer not used */
|
2007-07-23 06:20:51 +00:00
|
|
|
rxdp3->Buffer1_ptr = *temp1 =
|
2006-04-21 23:23:26 +00:00
|
|
|
pci_map_single(sp->pdev, ba->ba_1, BUF1_LEN,
|
|
|
|
PCI_DMA_FROMDEVICE);
|
2008-07-26 02:44:49 +00:00
|
|
|
if (pci_dma_mapping_error(sp->pdev,
|
|
|
|
rxdp3->Buffer1_ptr)) {
|
2007-07-23 06:37:14 +00:00
|
|
|
pci_unmap_single (sp->pdev,
|
2007-08-02 18:21:30 +00:00
|
|
|
(dma_addr_t)rxdp3->Buffer0_ptr,
|
|
|
|
BUF0_LEN, PCI_DMA_FROMDEVICE);
|
|
|
|
pci_unmap_single (sp->pdev,
|
|
|
|
(dma_addr_t)rxdp3->Buffer2_ptr,
|
2007-07-23 06:37:14 +00:00
|
|
|
dev->mtu + 4, PCI_DMA_FROMDEVICE);
|
|
|
|
goto memalloc_failed;
|
|
|
|
}
|
2006-04-21 23:23:26 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
2007-07-23 06:37:14 +00:00
|
|
|
memalloc_failed:
|
|
|
|
stats->pci_map_fail_cnt++;
|
|
|
|
stats->mem_freed += (*skb)->truesize;
|
|
|
|
dev_kfree_skb(*skb);
|
|
|
|
return -ENOMEM;
|
2006-04-21 23:23:26 +00:00
|
|
|
}
|
2007-07-23 06:37:14 +00:00
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void set_rxd_buffer_size(struct s2io_nic *sp, struct RxD_t *rxdp,
|
|
|
|
int size)
|
2006-04-21 23:23:26 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = sp->dev;
|
|
|
|
if (sp->rxd_mode == RXD_MODE_1) {
|
|
|
|
rxdp->Control_2 = SET_BUFFER0_SIZE_1( size - NET_IP_ALIGN);
|
|
|
|
} else if (sp->rxd_mode == RXD_MODE_3B) {
|
|
|
|
rxdp->Control_2 = SET_BUFFER0_SIZE_3(BUF0_LEN);
|
|
|
|
rxdp->Control_2 |= SET_BUFFER1_SIZE_3(1);
|
|
|
|
rxdp->Control_2 |= SET_BUFFER2_SIZE_3( dev->mtu + 4);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int rxd_owner_bit_reset(struct s2io_nic *sp)
|
2006-04-21 23:23:26 +00:00
|
|
|
{
|
|
|
|
int i, j, k, blk_cnt = 0, size;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info * mac_control = &sp->mac_control;
|
2006-04-21 23:23:26 +00:00
|
|
|
struct config_param *config = &sp->config;
|
|
|
|
struct net_device *dev = sp->dev;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct RxD_t *rxdp = NULL;
|
2006-04-21 23:23:26 +00:00
|
|
|
struct sk_buff *skb = NULL;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct buffAdd *ba = NULL;
|
2006-04-21 23:23:26 +00:00
|
|
|
u64 temp0_64 = 0, temp1_64 = 0, temp2_64 = 0;
|
|
|
|
|
|
|
|
/* Calculate the size based on ring mode */
|
|
|
|
size = dev->mtu + HEADER_ETHERNET_II_802_3_SIZE +
|
|
|
|
HEADER_802_2_SIZE + HEADER_SNAP_SIZE;
|
|
|
|
if (sp->rxd_mode == RXD_MODE_1)
|
|
|
|
size += NET_IP_ALIGN;
|
|
|
|
else if (sp->rxd_mode == RXD_MODE_3B)
|
|
|
|
size = dev->mtu + ALIGN_SIZE + BUF0_LEN + 4;
|
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
|
|
|
blk_cnt = config->rx_cfg[i].num_rxd /
|
|
|
|
(rxd_count[sp->rxd_mode] +1);
|
|
|
|
|
|
|
|
for (j = 0; j < blk_cnt; j++) {
|
|
|
|
for (k = 0; k < rxd_count[sp->rxd_mode]; k++) {
|
|
|
|
rxdp = mac_control->rings[i].
|
|
|
|
rx_blocks[j].rxds[k].virt_addr;
|
2007-07-23 06:20:51 +00:00
|
|
|
if(sp->rxd_mode == RXD_MODE_3B)
|
2006-04-21 23:23:26 +00:00
|
|
|
ba = &mac_control->rings[i].ba[j][k];
|
2007-02-24 07:01:31 +00:00
|
|
|
if (set_rxd_buffer_pointer(sp, rxdp, ba,
|
2006-04-21 23:23:26 +00:00
|
|
|
&skb,(u64 *)&temp0_64,
|
|
|
|
(u64 *)&temp1_64,
|
2007-02-24 07:01:31 +00:00
|
|
|
(u64 *)&temp2_64,
|
2008-05-14 23:20:17 +00:00
|
|
|
size) == -ENOMEM) {
|
2007-02-24 07:01:31 +00:00
|
|
|
return 0;
|
|
|
|
}
|
2006-04-21 23:23:26 +00:00
|
|
|
|
|
|
|
set_rxd_buffer_size(sp, rxdp, size);
|
|
|
|
wmb();
|
|
|
|
/* flip the Ownership bit to Hardware */
|
|
|
|
rxdp->Control_1 |= RXD_OWN_XENA;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_add_isr(struct s2io_nic * sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-07-07 06:58:23 +00:00
|
|
|
int ret = 0;
|
2006-04-21 23:18:03 +00:00
|
|
|
struct net_device *dev = sp->dev;
|
2006-07-07 06:58:23 +00:00
|
|
|
int err = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == MSI_X)
|
2006-07-07 06:58:23 +00:00
|
|
|
ret = s2io_enable_msi_x(sp);
|
|
|
|
if (ret) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Defaulting to INTA\n", dev->name);
|
2007-09-15 21:14:22 +00:00
|
|
|
sp->config.intr_type = INTA;
|
2005-08-03 19:24:33 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
/* Store the values of the MSIX table in the struct s2io_nic structure */
|
2006-07-07 06:58:23 +00:00
|
|
|
store_xmsi_data(sp);
|
2006-04-21 23:18:03 +00:00
|
|
|
|
2006-07-07 06:58:23 +00:00
|
|
|
/* After proper initialization of H/W, register ISR */
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == MSI_X) {
|
2008-05-12 17:41:32 +00:00
|
|
|
int i, msix_rx_cnt = 0;
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
for (i = 0; i < sp->num_entries; i++) {
|
|
|
|
if (sp->s2io_entries[i].in_use == MSIX_FLG) {
|
|
|
|
if (sp->s2io_entries[i].type ==
|
2008-05-12 17:41:32 +00:00
|
|
|
MSIX_RING_TYPE) {
|
|
|
|
sprintf(sp->desc[i], "%s:MSI-X-%d-RX",
|
|
|
|
dev->name, i);
|
|
|
|
err = request_irq(sp->entries[i].vector,
|
|
|
|
s2io_msix_ring_handle, 0,
|
|
|
|
sp->desc[i],
|
|
|
|
sp->s2io_entries[i].arg);
|
|
|
|
} else if (sp->s2io_entries[i].type ==
|
|
|
|
MSIX_ALARM_TYPE) {
|
|
|
|
sprintf(sp->desc[i], "%s:MSI-X-%d-TX",
|
2006-07-07 06:58:23 +00:00
|
|
|
dev->name, i);
|
2008-05-12 17:41:32 +00:00
|
|
|
err = request_irq(sp->entries[i].vector,
|
|
|
|
s2io_msix_fifo_handle, 0,
|
|
|
|
sp->desc[i],
|
|
|
|
sp->s2io_entries[i].arg);
|
|
|
|
|
2007-02-24 06:51:50 +00:00
|
|
|
}
|
2008-05-12 17:41:32 +00:00
|
|
|
/* if either data or addr is zero print it. */
|
|
|
|
if (!(sp->msix_info[i].addr &&
|
2007-02-24 06:51:50 +00:00
|
|
|
sp->msix_info[i].data)) {
|
2008-05-12 17:41:32 +00:00
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"%s @Addr:0x%llx Data:0x%llx\n",
|
|
|
|
sp->desc[i],
|
2007-02-24 06:51:50 +00:00
|
|
|
(unsigned long long)
|
|
|
|
sp->msix_info[i].addr,
|
2008-03-16 22:23:14 +00:00
|
|
|
(unsigned long long)
|
2008-05-12 17:41:32 +00:00
|
|
|
ntohl(sp->msix_info[i].data));
|
|
|
|
} else
|
2007-02-24 06:51:50 +00:00
|
|
|
msix_rx_cnt++;
|
2008-05-12 17:41:32 +00:00
|
|
|
if (err) {
|
|
|
|
remove_msix_isr(sp);
|
|
|
|
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"%s:MSI-X-%d registration "
|
|
|
|
"failed\n", dev->name, i);
|
|
|
|
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"%s: Defaulting to INTA\n",
|
|
|
|
dev->name);
|
|
|
|
sp->config.intr_type = INTA;
|
|
|
|
break;
|
2007-02-24 06:51:50 +00:00
|
|
|
}
|
2008-05-12 17:41:32 +00:00
|
|
|
sp->s2io_entries[i].in_use =
|
|
|
|
MSIX_REGISTERED_SUCCESS;
|
2006-04-21 23:18:03 +00:00
|
|
|
}
|
2006-07-07 06:58:23 +00:00
|
|
|
}
|
2007-11-14 09:41:06 +00:00
|
|
|
if (!err) {
|
|
|
|
printk(KERN_INFO "MSI-X-RX %d entries enabled\n",
|
2008-05-12 17:41:32 +00:00
|
|
|
--msix_rx_cnt);
|
|
|
|
DBG_PRINT(INFO_DBG, "MSI-X-TX entries enabled"
|
|
|
|
" through alarm vector\n");
|
2007-11-14 09:41:06 +00:00
|
|
|
}
|
2006-07-07 06:58:23 +00:00
|
|
|
}
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == INTA) {
|
2006-07-07 06:58:23 +00:00
|
|
|
err = request_irq((int) sp->pdev->irq, s2io_isr, IRQF_SHARED,
|
|
|
|
sp->name, dev);
|
|
|
|
if (err) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: ISR registration failed\n",
|
|
|
|
dev->name);
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_rem_isr(struct s2io_nic * sp)
|
2006-07-07 06:58:23 +00:00
|
|
|
{
|
2007-11-14 09:41:06 +00:00
|
|
|
if (sp->config.intr_type == MSI_X)
|
|
|
|
remove_msix_isr(sp);
|
|
|
|
else
|
|
|
|
remove_inta_isr(sp);
|
2006-07-07 06:58:23 +00:00
|
|
|
}
|
|
|
|
|
2007-05-14 23:37:30 +00:00
|
|
|
static void do_s2io_card_down(struct s2io_nic * sp, int do_io)
|
2006-07-07 06:58:23 +00:00
|
|
|
{
|
|
|
|
int cnt = 0;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = sp->bar0;
|
2006-07-07 06:58:23 +00:00
|
|
|
register u64 val64 = 0;
|
2008-01-15 01:23:04 +00:00
|
|
|
struct config_param *config;
|
|
|
|
config = &sp->config;
|
2006-07-07 06:58:23 +00:00
|
|
|
|
2007-11-30 06:46:08 +00:00
|
|
|
if (!is_s2io_card_up(sp))
|
|
|
|
return;
|
|
|
|
|
2006-07-07 06:58:23 +00:00
|
|
|
del_timer_sync(&sp->alarm_timer);
|
|
|
|
/* If s2io_set_link task is executing, wait till it completes. */
|
2007-09-06 10:51:14 +00:00
|
|
|
while (test_and_set_bit(__S2IO_STATE_LINK_TASK, &(sp->state))) {
|
2006-07-07 06:58:23 +00:00
|
|
|
msleep(50);
|
|
|
|
}
|
2007-09-06 10:51:14 +00:00
|
|
|
clear_bit(__S2IO_STATE_CARD_UP, &sp->state);
|
2006-07-07 06:58:23 +00:00
|
|
|
|
2008-01-15 01:23:04 +00:00
|
|
|
/* Disable napi */
|
2008-05-12 17:42:17 +00:00
|
|
|
if (sp->config.napi) {
|
|
|
|
int off = 0;
|
|
|
|
if (config->intr_type == MSI_X) {
|
|
|
|
for (; off < sp->config.rx_ring_num; off++)
|
|
|
|
napi_disable(&sp->mac_control.rings[off].napi);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
napi_disable(&sp->napi);
|
|
|
|
}
|
2008-01-15 01:23:04 +00:00
|
|
|
|
2006-07-07 06:58:23 +00:00
|
|
|
/* disable Tx and Rx traffic on the NIC */
|
2007-05-14 23:37:30 +00:00
|
|
|
if (do_io)
|
|
|
|
stop_nic(sp);
|
2006-07-07 06:58:23 +00:00
|
|
|
|
|
|
|
s2io_rem_isr(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
/* stop the tx queue, indicate link down */
|
|
|
|
s2io_link(sp, LINK_DOWN);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Check if the device is Quiescent and then Reset the NIC */
|
2007-05-14 23:37:30 +00:00
|
|
|
while(do_io) {
|
2006-04-21 23:23:26 +00:00
|
|
|
/* As per the HW requirement we need to replenish the
|
|
|
|
* receive buffer to avoid the ring bump. Since there is
|
|
|
|
* no intention of processing the Rx frame at this pointwe are
|
|
|
|
* just settting the ownership bit of rxd in Each Rx
|
|
|
|
* ring to HW and set the appropriate buffer size
|
|
|
|
* based on the ring mode
|
|
|
|
*/
|
|
|
|
rxd_owner_bit_reset(sp);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
val64 = readq(&bar0->adapter_status);
|
2007-01-31 18:30:49 +00:00
|
|
|
if (verify_xena_quiescence(sp)) {
|
|
|
|
if(verify_pcc_quiescent(sp, sp->device_enabled_once))
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
msleep(50);
|
|
|
|
cnt++;
|
|
|
|
if (cnt == 10) {
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"s2io_close:Device not Quiescent ");
|
|
|
|
DBG_PRINT(ERR_DBG, "adaper status reads 0x%llx\n",
|
|
|
|
(unsigned long long) val64);
|
|
|
|
break;
|
|
|
|
}
|
2007-05-14 23:37:30 +00:00
|
|
|
}
|
|
|
|
if (do_io)
|
|
|
|
s2io_reset(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:29:20 +00:00
|
|
|
/* Free all Tx buffers */
|
2005-04-16 22:20:36 +00:00
|
|
|
free_tx_buffers(sp);
|
2005-08-03 19:29:20 +00:00
|
|
|
|
|
|
|
/* Free all Rx buffers */
|
2005-04-16 22:20:36 +00:00
|
|
|
free_rx_buffers(sp);
|
|
|
|
|
2007-09-06 10:51:14 +00:00
|
|
|
clear_bit(__S2IO_STATE_LINK_TASK, &(sp->state));
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2007-05-14 23:37:30 +00:00
|
|
|
static void s2io_card_down(struct s2io_nic * sp)
|
|
|
|
{
|
|
|
|
do_s2io_card_down(sp, 1);
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int s2io_card_up(struct s2io_nic * sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2005-10-04 10:41:24 +00:00
|
|
|
int i, ret = 0;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
|
|
|
struct net_device *dev = (struct net_device *) sp->dev;
|
2006-07-07 06:58:23 +00:00
|
|
|
u16 interruptible;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Initialize the H/W I/O registers */
|
2007-11-30 06:46:08 +00:00
|
|
|
ret = init_nic(sp);
|
|
|
|
if (ret != 0) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: H/W initialization failed\n",
|
|
|
|
dev->name);
|
2007-11-30 06:46:08 +00:00
|
|
|
if (ret != -EIO)
|
|
|
|
s2io_reset(sp);
|
|
|
|
return ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Initializing the Rx buffers. For now we are considering only 1
|
2005-04-16 22:20:36 +00:00
|
|
|
* Rx ring and initializing buffers into 30 Rx blocks
|
|
|
|
*/
|
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
config = &sp->config;
|
|
|
|
|
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2008-04-29 01:08:45 +00:00
|
|
|
mac_control->rings[i].mtu = dev->mtu;
|
2008-07-26 02:44:49 +00:00
|
|
|
ret = fill_rx_buffers(sp, &mac_control->rings[i], 1);
|
2008-04-29 01:08:45 +00:00
|
|
|
if (ret) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Out of memory in Open\n",
|
|
|
|
dev->name);
|
|
|
|
s2io_reset(sp);
|
|
|
|
free_rx_buffers(sp);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
DBG_PRINT(INFO_DBG, "Buf in ring:%d is %d:\n", i,
|
2008-04-29 01:08:45 +00:00
|
|
|
mac_control->rings[i].rx_bufs_left);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-01-15 01:23:04 +00:00
|
|
|
|
|
|
|
/* Initialise napi */
|
2008-05-12 17:42:17 +00:00
|
|
|
if (config->napi) {
|
|
|
|
if (config->intr_type == MSI_X) {
|
|
|
|
for (i = 0; i < sp->config.rx_ring_num; i++)
|
|
|
|
napi_enable(&sp->mac_control.rings[i].napi);
|
|
|
|
} else {
|
|
|
|
napi_enable(&sp->napi);
|
|
|
|
}
|
|
|
|
}
|
2008-01-15 01:23:04 +00:00
|
|
|
|
2007-01-31 18:30:49 +00:00
|
|
|
/* Maintain the state prior to the open */
|
|
|
|
if (sp->promisc_flg)
|
|
|
|
sp->promisc_flg = 0;
|
|
|
|
if (sp->m_cast_flg) {
|
|
|
|
sp->m_cast_flg = 0;
|
|
|
|
sp->all_multi_pos= 0;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Setting its receive mode */
|
|
|
|
s2io_set_multicast(dev);
|
|
|
|
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (sp->lro) {
|
2006-07-24 23:52:49 +00:00
|
|
|
/* Initialize max aggregatable pkts per session based on MTU */
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
sp->lro_max_aggr_per_sess = ((1<<16) - 1) / dev->mtu;
|
|
|
|
/* Check if we can use(if specified) user provided value */
|
|
|
|
if (lro_max_pkts < sp->lro_max_aggr_per_sess)
|
|
|
|
sp->lro_max_aggr_per_sess = lro_max_pkts;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Enable Rx Traffic and interrupts on the NIC */
|
|
|
|
if (start_nic(sp)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Starting NIC failed\n", dev->name);
|
|
|
|
s2io_reset(sp);
|
2006-07-07 06:58:23 +00:00
|
|
|
free_rx_buffers(sp);
|
|
|
|
return -ENODEV;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Add interrupt service routine */
|
|
|
|
if (s2io_add_isr(sp) != 0) {
|
2007-09-15 21:14:22 +00:00
|
|
|
if (sp->config.intr_type == MSI_X)
|
2006-07-07 06:58:23 +00:00
|
|
|
s2io_rem_isr(sp);
|
|
|
|
s2io_reset(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
free_rx_buffers(sp);
|
|
|
|
return -ENODEV;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:34:11 +00:00
|
|
|
S2IO_TIMER_CONF(sp->alarm_timer, s2io_alarm_handle, sp, (HZ/2));
|
|
|
|
|
2008-07-10 03:49:21 +00:00
|
|
|
set_bit(__S2IO_STATE_CARD_UP, &sp->state);
|
|
|
|
|
2006-07-07 06:58:23 +00:00
|
|
|
/* Enable select interrupts */
|
2007-09-06 10:21:54 +00:00
|
|
|
en_dis_err_alarms(sp, ENA_ALL_INTRS, ENABLE_INTRS);
|
2008-07-10 03:49:21 +00:00
|
|
|
if (sp->config.intr_type != INTA) {
|
|
|
|
interruptible = TX_TRAFFIC_INTR | TX_PIC_INTR;
|
|
|
|
en_dis_able_nic_intrs(sp, interruptible, ENABLE_INTRS);
|
|
|
|
} else {
|
2006-07-07 06:58:23 +00:00
|
|
|
interruptible = TX_TRAFFIC_INTR | RX_TRAFFIC_INTR;
|
2007-09-06 10:21:54 +00:00
|
|
|
interruptible |= TX_PIC_INTR;
|
2006-07-07 06:58:23 +00:00
|
|
|
en_dis_able_nic_intrs(sp, interruptible, ENABLE_INTRS);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_restart_nic - Resets the NIC.
|
|
|
|
* @data : long pointer to the device private structure
|
|
|
|
* Description:
|
|
|
|
* This function is scheduled to be run by the s2io_tx_watchdog
|
2005-08-03 19:24:33 +00:00
|
|
|
* function after 0.5 secs to reset the NIC. The idea is to reduce
|
2005-04-16 22:20:36 +00:00
|
|
|
* the run time of the watch dog routine which is run holding a
|
|
|
|
* spin lock.
|
|
|
|
*/
|
|
|
|
|
2006-11-22 14:57:56 +00:00
|
|
|
static void s2io_restart_nic(struct work_struct *work)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *sp = container_of(work, struct s2io_nic, rst_timer_task);
|
2006-11-22 14:57:56 +00:00
|
|
|
struct net_device *dev = sp->dev;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2007-02-15 22:37:50 +00:00
|
|
|
rtnl_lock();
|
|
|
|
|
|
|
|
if (!netif_running(dev))
|
|
|
|
goto out_unlock;
|
|
|
|
|
2006-07-07 06:58:23 +00:00
|
|
|
s2io_card_down(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (s2io_card_up(sp)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Device bring up failed\n",
|
|
|
|
dev->name);
|
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_wake_all_tx_queue(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: was reset by Tx watchdog timer\n",
|
|
|
|
dev->name);
|
2007-02-15 22:37:50 +00:00
|
|
|
out_unlock:
|
|
|
|
rtnl_unlock();
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* s2io_tx_watchdog - Watchdog for transmit side.
|
2005-04-16 22:20:36 +00:00
|
|
|
* @dev : Pointer to net device structure
|
|
|
|
* Description:
|
|
|
|
* This function is triggered if the Tx Queue is stopped
|
|
|
|
* for a pre-defined amount of time when the Interface is still up.
|
|
|
|
* If the Interface is jammed in such a situation, the hardware is
|
|
|
|
* reset (by s2io_close) and restarted again (by s2io_open) to
|
|
|
|
* overcome any problem that might have been caused in the hardware.
|
|
|
|
* Return value:
|
|
|
|
* void
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void s2io_tx_watchdog(struct net_device *dev)
|
|
|
|
{
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (netif_carrier_ok(dev)) {
|
2007-05-10 08:18:54 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.watchdog_timer_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
schedule_work(&sp->rst_timer_task);
|
2006-04-21 23:20:22 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.soft_reset_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* rx_osm_handler - To perform some OS related operations on SKB.
|
|
|
|
* @sp: private member of the device structure,pointer to s2io_nic structure.
|
|
|
|
* @skb : the socket buffer pointer.
|
|
|
|
* @len : length of the packet
|
|
|
|
* @cksum : FCS checksum of the frame.
|
|
|
|
* @ring_no : the ring from which this RxD was extracted.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description:
|
2006-07-24 23:52:49 +00:00
|
|
|
* This function is called by the Rx interrupt serivce routine to perform
|
2005-04-16 22:20:36 +00:00
|
|
|
* some OS related operations on the SKB before passing it to the upper
|
|
|
|
* layers. It mainly checks if the checksum is OK, if so adds it to the
|
|
|
|
* SKBs cksum variable, increments the Rx packet count and passes the SKB
|
|
|
|
* to the upper layer. If the checksum is wrong, it increments the Rx
|
|
|
|
* packet error count, frees the SKB and returns error.
|
|
|
|
* Return value:
|
|
|
|
* SUCCESS on success and -1 on failure.
|
|
|
|
*/
|
2007-01-31 19:09:29 +00:00
|
|
|
static int rx_osm_handler(struct ring_info *ring_data, struct RxD_t * rxdp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *sp = ring_data->nic;
|
2008-04-29 01:08:45 +00:00
|
|
|
struct net_device *dev = (struct net_device *) ring_data->dev;
|
2005-08-03 19:24:33 +00:00
|
|
|
struct sk_buff *skb = (struct sk_buff *)
|
|
|
|
((unsigned long) rxdp->Host_Control);
|
|
|
|
int ring_no = ring_data->ring_no;
|
2005-04-16 22:20:36 +00:00
|
|
|
u16 l3_csum, l4_csum;
|
2006-04-21 23:03:13 +00:00
|
|
|
unsigned long long err = rxdp->Control_1 & RXD_T_CODE;
|
2008-11-26 00:47:35 +00:00
|
|
|
struct lro *uninitialized_var(lro);
|
2007-06-19 20:41:10 +00:00
|
|
|
u8 err_mask;
|
2005-10-31 21:55:31 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
skb->dev = dev;
|
2006-04-21 23:18:03 +00:00
|
|
|
|
2006-04-21 23:03:13 +00:00
|
|
|
if (err) {
|
2006-04-21 23:20:22 +00:00
|
|
|
/* Check for parity error */
|
|
|
|
if (err & 0x1) {
|
|
|
|
sp->mac_control.stats_info->sw_stat.parity_err_cnt++;
|
|
|
|
}
|
2007-06-19 20:41:10 +00:00
|
|
|
err_mask = err >> 48;
|
|
|
|
switch(err_mask) {
|
2007-05-10 08:22:25 +00:00
|
|
|
case 1:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_parity_err_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 2:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_abort_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 3:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_parity_abort_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 4:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_rda_fail_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 5:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_unkn_prot_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 6:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_fcs_err_cnt++;
|
|
|
|
break;
|
2006-04-21 23:20:22 +00:00
|
|
|
|
2007-05-10 08:22:25 +00:00
|
|
|
case 7:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_buf_size_err_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 8:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_rxd_corrupt_cnt++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case 15:
|
|
|
|
sp->mac_control.stats_info->sw_stat.
|
|
|
|
rx_unkn_err_cnt++;
|
|
|
|
break;
|
|
|
|
}
|
2006-04-21 23:03:13 +00:00
|
|
|
/*
|
|
|
|
* Drop the packet if bad transfer code. Exception being
|
|
|
|
* 0x5, which could be due to unsupported IPv6 extension header.
|
|
|
|
* In this case, we let stack handle the packet.
|
|
|
|
* Note that in this case, since checksum will be incorrect,
|
|
|
|
* stack will validate the same.
|
|
|
|
*/
|
2007-06-19 20:41:10 +00:00
|
|
|
if (err_mask != 0x5) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Rx error Value: 0x%x\n",
|
|
|
|
dev->name, err_mask);
|
S2io: fix statistics flush after a MTU change
On s2io driver, when you change the interface MTU, it invokes a card
reset, which flush some statistics. This patch solves this problem, and
also set the net_device->stats as the default statistics structure,
instead of s2io_nic->stats.
To do that, s2io_nic->stats turned into a staging area, where is saved
statistics of the last hardware statistics query. So, the difference
between the current hardware statistics and s2io_nic->stats, is the
value that should be summed up, in order to get the correct statistics
value, even after a reset.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
2008-07-22 19:27:20 +00:00
|
|
|
dev->stats.rx_crc_errors++;
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_freed
|
2007-05-10 08:22:25 +00:00
|
|
|
+= skb->truesize;
|
2006-04-21 23:03:13 +00:00
|
|
|
dev_kfree_skb(skb);
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->rx_bufs_left -= 1;
|
2006-04-21 23:03:13 +00:00
|
|
|
rxdp->Host_Control = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/* Updating statistics */
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->rx_packets++;
|
2005-08-03 19:24:33 +00:00
|
|
|
rxdp->Host_Control = 0;
|
2005-10-31 21:55:31 +00:00
|
|
|
if (sp->rxd_mode == RXD_MODE_1) {
|
|
|
|
int len = RXD_GET_BUFFER0_SIZE_1(rxdp->Control_2);
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->rx_bytes += len;
|
2005-10-31 21:55:31 +00:00
|
|
|
skb_put(skb, len);
|
|
|
|
|
2007-07-23 06:20:51 +00:00
|
|
|
} else if (sp->rxd_mode == RXD_MODE_3B) {
|
2005-10-31 21:55:31 +00:00
|
|
|
int get_block = ring_data->rx_curr_get_info.block_index;
|
|
|
|
int get_off = ring_data->rx_curr_get_info.offset;
|
|
|
|
int buf0_len = RXD_GET_BUFFER0_SIZE_3(rxdp->Control_2);
|
|
|
|
int buf2_len = RXD_GET_BUFFER2_SIZE_3(rxdp->Control_2);
|
|
|
|
unsigned char *buff = skb_push(skb, buf0_len);
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
struct buffAdd *ba = &ring_data->ba[get_block][get_off];
|
2008-04-29 01:08:45 +00:00
|
|
|
ring_data->rx_bytes += buf0_len + buf2_len;
|
2005-10-31 21:55:31 +00:00
|
|
|
memcpy(buff, ba->ba_0, buf0_len);
|
2007-07-23 06:20:51 +00:00
|
|
|
skb_put(skb, buf2_len);
|
2005-10-31 21:55:31 +00:00
|
|
|
}
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
if ((rxdp->Control_1 & TCP_OR_UDP_FRAME) && ((!ring_data->lro) ||
|
|
|
|
(ring_data->lro && (!(rxdp->Control_1 & RXD_FRAME_IP_FRAG)))) &&
|
2005-08-03 19:24:33 +00:00
|
|
|
(sp->rx_csum)) {
|
|
|
|
l3_csum = RXD_GET_L3_CKSUM(rxdp->Control_1);
|
2005-04-16 22:20:36 +00:00
|
|
|
l4_csum = RXD_GET_L4_CKSUM(rxdp->Control_1);
|
|
|
|
if ((l3_csum == L3_CKSUM_OK) && (l4_csum == L4_CKSUM_OK)) {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* NIC verifies if the Checksum of the received
|
|
|
|
* frame is Ok or not and accordingly returns
|
|
|
|
* a flag in the RxD.
|
|
|
|
*/
|
|
|
|
skb->ip_summed = CHECKSUM_UNNECESSARY;
|
2008-04-29 01:08:45 +00:00
|
|
|
if (ring_data->lro) {
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
u32 tcp_len;
|
|
|
|
u8 *tcp;
|
|
|
|
int ret = 0;
|
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
ret = s2io_club_tcp_session(ring_data,
|
|
|
|
skb->data, &tcp, &tcp_len, &lro,
|
|
|
|
rxdp, sp);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
switch (ret) {
|
|
|
|
case 3: /* Begin anew */
|
|
|
|
lro->parent = skb;
|
|
|
|
goto aggregate;
|
|
|
|
case 1: /* Aggregate */
|
|
|
|
{
|
|
|
|
lro_append_pkt(sp, lro,
|
|
|
|
skb, tcp_len);
|
|
|
|
goto aggregate;
|
|
|
|
}
|
|
|
|
case 4: /* Flush session */
|
|
|
|
{
|
|
|
|
lro_append_pkt(sp, lro,
|
|
|
|
skb, tcp_len);
|
2008-02-20 22:09:15 +00:00
|
|
|
queue_rx_frame(lro->parent,
|
|
|
|
lro->vlan_tag);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
clear_lro_session(lro);
|
|
|
|
sp->mac_control.stats_info->
|
|
|
|
sw_stat.flush_max_pkts++;
|
|
|
|
goto aggregate;
|
|
|
|
}
|
|
|
|
case 2: /* Flush both */
|
|
|
|
lro->parent->data_len =
|
|
|
|
lro->frags_len;
|
|
|
|
sp->mac_control.stats_info->
|
|
|
|
sw_stat.sending_both++;
|
2008-02-20 22:09:15 +00:00
|
|
|
queue_rx_frame(lro->parent,
|
|
|
|
lro->vlan_tag);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
clear_lro_session(lro);
|
|
|
|
goto send_up;
|
|
|
|
case 0: /* sessions exceeded */
|
2006-04-21 23:18:03 +00:00
|
|
|
case -1: /* non-TCP or not
|
|
|
|
* L2 aggregatable
|
|
|
|
*/
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
case 5: /*
|
|
|
|
* First pkt in session not
|
|
|
|
* L3/L4 aggregatable
|
|
|
|
*/
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"%s: Samadhana!!\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
} else {
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Packet with erroneous checksum, let the
|
2005-04-16 22:20:36 +00:00
|
|
|
* upper layers deal with it.
|
|
|
|
*/
|
|
|
|
skb->ip_summed = CHECKSUM_NONE;
|
|
|
|
}
|
2008-02-20 22:09:15 +00:00
|
|
|
} else
|
2005-04-16 22:20:36 +00:00
|
|
|
skb->ip_summed = CHECKSUM_NONE;
|
2008-02-20 22:09:15 +00:00
|
|
|
|
2007-05-10 08:22:25 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.mem_freed += skb->truesize;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
send_up:
|
2009-01-28 00:22:32 +00:00
|
|
|
skb_record_rx_queue(skb, ring_no);
|
2008-02-20 22:09:15 +00:00
|
|
|
queue_rx_frame(skb, RXD_GET_VLAN_TAG(rxdp->Control_2));
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
aggregate:
|
2008-04-29 01:08:45 +00:00
|
|
|
sp->mac_control.rings[ring_no].rx_bufs_left -= 1;
|
2005-04-16 22:20:36 +00:00
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_link - stops/starts the Tx queue.
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
|
|
|
* s2io_nic structure.
|
|
|
|
* @link : inidicates whether link is UP/DOWN.
|
|
|
|
* Description:
|
|
|
|
* This function stops/starts the Tx queue depending on whether the link
|
2005-08-03 19:24:33 +00:00
|
|
|
* status of the NIC is is down or up. This is called by the Alarm
|
|
|
|
* interrupt handler whenever a link change interrupt comes up.
|
2005-04-16 22:20:36 +00:00
|
|
|
* Return value:
|
|
|
|
* void.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_link(struct s2io_nic * sp, int link)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = (struct net_device *) sp->dev;
|
|
|
|
|
|
|
|
if (link != sp->last_link_state) {
|
2007-12-17 19:40:15 +00:00
|
|
|
init_tti(sp, link);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (link == LINK_DOWN) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Link down\n", dev->name);
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_stop_all_tx_queue(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
netif_carrier_off(dev);
|
2007-05-10 08:22:25 +00:00
|
|
|
if(sp->mac_control.stats_info->sw_stat.link_up_cnt)
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.link_up_time =
|
2007-05-10 08:22:25 +00:00
|
|
|
jiffies - sp->start_time;
|
|
|
|
sp->mac_control.stats_info->sw_stat.link_down_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
} else {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Link Up\n", dev->name);
|
2007-05-10 08:22:25 +00:00
|
|
|
if (sp->mac_control.stats_info->sw_stat.link_down_cnt)
|
2007-09-18 22:14:20 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.link_down_time =
|
2007-05-10 08:22:25 +00:00
|
|
|
jiffies - sp->start_time;
|
|
|
|
sp->mac_control.stats_info->sw_stat.link_up_cnt++;
|
2005-04-16 22:20:36 +00:00
|
|
|
netif_carrier_on(dev);
|
2008-02-20 21:44:07 +00:00
|
|
|
s2io_wake_all_tx_queue(sp);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
sp->last_link_state = link;
|
2007-05-10 08:22:25 +00:00
|
|
|
sp->start_time = jiffies;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/**
|
|
|
|
* s2io_init_pci -Initialization of PCI and PCI-X configuration registers .
|
|
|
|
* @sp : private member of the device structure, which is a pointer to the
|
2005-04-16 22:20:36 +00:00
|
|
|
* s2io_nic structure.
|
|
|
|
* Description:
|
|
|
|
* This function initializes a few of the PCI and PCI-X configuration registers
|
|
|
|
* with recommended values.
|
|
|
|
* Return value:
|
|
|
|
* void
|
|
|
|
*/
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void s2io_init_pci(struct s2io_nic * sp)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2005-08-03 19:24:33 +00:00
|
|
|
u16 pci_cmd = 0, pcix_cmd = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Enable Data Parity Error Recovery in PCI-X command register. */
|
|
|
|
pci_read_config_word(sp->pdev, PCIX_COMMAND_REGISTER,
|
2005-08-03 19:24:33 +00:00
|
|
|
&(pcix_cmd));
|
2005-04-16 22:20:36 +00:00
|
|
|
pci_write_config_word(sp->pdev, PCIX_COMMAND_REGISTER,
|
2005-08-03 19:24:33 +00:00
|
|
|
(pcix_cmd | 1));
|
2005-04-16 22:20:36 +00:00
|
|
|
pci_read_config_word(sp->pdev, PCIX_COMMAND_REGISTER,
|
2005-08-03 19:24:33 +00:00
|
|
|
&(pcix_cmd));
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Set the PErr Response bit in PCI command register. */
|
|
|
|
pci_read_config_word(sp->pdev, PCI_COMMAND, &pci_cmd);
|
|
|
|
pci_write_config_word(sp->pdev, PCI_COMMAND,
|
|
|
|
(pci_cmd | PCI_COMMAND_PARITY));
|
|
|
|
pci_read_config_word(sp->pdev, PCI_COMMAND, &pci_cmd);
|
|
|
|
}
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
static int s2io_verify_parm(struct pci_dev *pdev, u8 *dev_intr_type,
|
|
|
|
u8 *dev_multiq)
|
2006-04-21 23:05:41 +00:00
|
|
|
{
|
2008-01-24 10:08:59 +00:00
|
|
|
if ((tx_fifo_num > MAX_TX_FIFOS) ||
|
2008-02-20 22:07:51 +00:00
|
|
|
(tx_fifo_num < 1)) {
|
2008-01-24 10:08:59 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Requested number of tx fifos "
|
|
|
|
"(%d) not supported\n", tx_fifo_num);
|
2008-02-20 22:07:51 +00:00
|
|
|
|
|
|
|
if (tx_fifo_num < 1)
|
|
|
|
tx_fifo_num = 1;
|
|
|
|
else
|
|
|
|
tx_fifo_num = MAX_TX_FIFOS;
|
|
|
|
|
2008-01-24 10:08:59 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Default to %d ", tx_fifo_num);
|
|
|
|
DBG_PRINT(ERR_DBG, "tx fifos\n");
|
2006-04-21 23:05:41 +00:00
|
|
|
}
|
2008-01-24 10:08:59 +00:00
|
|
|
|
2008-02-20 22:07:51 +00:00
|
|
|
if (multiq)
|
2008-02-20 21:44:07 +00:00
|
|
|
*dev_multiq = multiq;
|
2008-02-20 22:07:51 +00:00
|
|
|
|
|
|
|
if (tx_steering_type && (1 == tx_fifo_num)) {
|
|
|
|
if (tx_steering_type != TX_DEFAULT_STEERING)
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"s2io: Tx steering is not supported with "
|
|
|
|
"one fifo. Disabling Tx steering.\n");
|
|
|
|
tx_steering_type = NO_STEERING;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((tx_steering_type < NO_STEERING) ||
|
|
|
|
(tx_steering_type > TX_DEFAULT_STEERING)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Requested transmit steering not "
|
|
|
|
"supported\n");
|
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Disabling transmit steering\n");
|
|
|
|
tx_steering_type = NO_STEERING;
|
2008-02-20 21:44:07 +00:00
|
|
|
}
|
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
if (rx_ring_num > MAX_RX_RINGS) {
|
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Requested number of rx rings not "
|
2006-04-21 23:05:41 +00:00
|
|
|
"supported\n");
|
2008-04-29 01:08:45 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Default to %d rx rings\n",
|
|
|
|
MAX_RX_RINGS);
|
|
|
|
rx_ring_num = MAX_RX_RINGS;
|
2006-04-21 23:05:41 +00:00
|
|
|
}
|
2008-04-29 01:08:45 +00:00
|
|
|
|
2007-07-23 06:23:54 +00:00
|
|
|
if ((*dev_intr_type != INTA) && (*dev_intr_type != MSI_X)) {
|
2006-04-21 23:05:41 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Wrong intr_type requested. "
|
|
|
|
"Defaulting to INTA\n");
|
|
|
|
*dev_intr_type = INTA;
|
|
|
|
}
|
2007-09-15 21:24:03 +00:00
|
|
|
|
2006-04-21 23:05:41 +00:00
|
|
|
if ((*dev_intr_type == MSI_X) &&
|
|
|
|
((pdev->device != PCI_DEVICE_ID_HERC_WIN) &&
|
|
|
|
(pdev->device != PCI_DEVICE_ID_HERC_UNI))) {
|
2006-09-13 17:24:59 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Xframe I does not support MSI_X. "
|
2006-04-21 23:05:41 +00:00
|
|
|
"Defaulting to INTA\n");
|
|
|
|
*dev_intr_type = INTA;
|
|
|
|
}
|
2007-02-24 06:51:50 +00:00
|
|
|
|
2007-07-23 06:20:51 +00:00
|
|
|
if ((rx_ring_mode != 1) && (rx_ring_mode != 2)) {
|
2006-04-21 23:05:41 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Requested ring mode not supported\n");
|
2007-07-23 06:20:51 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "s2io: Defaulting to 1-buffer mode\n");
|
|
|
|
rx_ring_mode = 1;
|
2006-04-21 23:05:41 +00:00
|
|
|
}
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
|
2007-02-24 06:57:32 +00:00
|
|
|
/**
|
|
|
|
* rts_ds_steer - Receive traffic steering based on IPv4 or IPv6 TOS
|
|
|
|
* or Traffic class respectively.
|
2007-12-17 19:40:15 +00:00
|
|
|
* @nic: device private variable
|
2007-02-24 06:57:32 +00:00
|
|
|
* Description: The function configures the receive steering to
|
|
|
|
* desired receive ring.
|
|
|
|
* Return Value: SUCCESS on success and
|
|
|
|
* '-1' on failure (endian settings incorrect).
|
|
|
|
*/
|
|
|
|
static int rts_ds_steer(struct s2io_nic *nic, u8 ds_codepoint, u8 ring)
|
|
|
|
{
|
|
|
|
struct XENA_dev_config __iomem *bar0 = nic->bar0;
|
|
|
|
register u64 val64 = 0;
|
|
|
|
|
|
|
|
if (ds_codepoint > 63)
|
|
|
|
return FAILURE;
|
|
|
|
|
|
|
|
val64 = RTS_DS_MEM_DATA(ring);
|
|
|
|
writeq(val64, &bar0->rts_ds_mem_data);
|
|
|
|
|
|
|
|
val64 = RTS_DS_MEM_CTRL_WE |
|
|
|
|
RTS_DS_MEM_CTRL_STROBE_NEW_CMD |
|
|
|
|
RTS_DS_MEM_CTRL_OFFSET(ds_codepoint);
|
|
|
|
|
|
|
|
writeq(val64, &bar0->rts_ds_mem_ctrl);
|
|
|
|
|
|
|
|
return wait_for_cmd_complete(&bar0->rts_ds_mem_ctrl,
|
|
|
|
RTS_DS_MEM_CTRL_STROBE_CMD_BEING_EXECUTED,
|
|
|
|
S2IO_BIT_RESET);
|
|
|
|
}
|
|
|
|
|
2008-11-22 01:28:55 +00:00
|
|
|
static const struct net_device_ops s2io_netdev_ops = {
|
|
|
|
.ndo_open = s2io_open,
|
|
|
|
.ndo_stop = s2io_close,
|
|
|
|
.ndo_get_stats = s2io_get_stats,
|
|
|
|
.ndo_start_xmit = s2io_xmit,
|
|
|
|
.ndo_validate_addr = eth_validate_addr,
|
|
|
|
.ndo_set_multicast_list = s2io_set_multicast,
|
|
|
|
.ndo_do_ioctl = s2io_ioctl,
|
|
|
|
.ndo_set_mac_address = s2io_set_mac_addr,
|
|
|
|
.ndo_change_mtu = s2io_change_mtu,
|
|
|
|
.ndo_vlan_rx_register = s2io_vlan_rx_register,
|
|
|
|
.ndo_vlan_rx_kill_vid = s2io_vlan_rx_kill_vid,
|
|
|
|
.ndo_tx_timeout = s2io_tx_watchdog,
|
|
|
|
#ifdef CONFIG_NET_POLL_CONTROLLER
|
|
|
|
.ndo_poll_controller = s2io_netpoll,
|
|
|
|
#endif
|
|
|
|
};
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_init_nic - Initialization of the adapter .
|
2005-04-16 22:20:36 +00:00
|
|
|
* @pdev : structure containing the PCI related information of the device.
|
|
|
|
* @pre: List of PCI devices supported by the driver listed in s2io_tbl.
|
|
|
|
* Description:
|
|
|
|
* The function initializes an adapter identified by the pci_dec structure.
|
2005-08-03 19:24:33 +00:00
|
|
|
* All OS related initialization including memory and device structure and
|
|
|
|
* initlaization of the device private variable is done. Also the swapper
|
|
|
|
* control register is initialized to enable read and write into the I/O
|
2005-04-16 22:20:36 +00:00
|
|
|
* registers of the device.
|
|
|
|
* Return value:
|
|
|
|
* returns 0 on success and negative on failure.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int __devinit
|
|
|
|
s2io_init_nic(struct pci_dev *pdev, const struct pci_device_id *pre)
|
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *sp;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct net_device *dev;
|
|
|
|
int i, j, ret;
|
|
|
|
int dma_flag = FALSE;
|
|
|
|
u32 mac_up, mac_down;
|
|
|
|
u64 val64 = 0, tmp64 = 0;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct XENA_dev_config __iomem *bar0 = NULL;
|
2005-04-16 22:20:36 +00:00
|
|
|
u16 subid;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct mac_info *mac_control;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct config_param *config;
|
2005-08-03 19:36:55 +00:00
|
|
|
int mode;
|
2005-10-04 10:41:24 +00:00
|
|
|
u8 dev_intr_type = intr_type;
|
2008-02-20 21:44:07 +00:00
|
|
|
u8 dev_multiq = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
ret = s2io_verify_parm(pdev, &dev_intr_type, &dev_multiq);
|
|
|
|
if (ret)
|
2006-04-21 23:05:41 +00:00
|
|
|
return ret;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if ((ret = pci_enable_device(pdev))) {
|
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"s2io_init_nic: pci_enable_device failed\n");
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2005-06-26 22:22:14 +00:00
|
|
|
if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(INIT_DBG, "s2io_init_nic: Using 64bit DMA\n");
|
|
|
|
dma_flag = TRUE;
|
|
|
|
if (pci_set_consistent_dma_mask
|
2005-06-26 22:22:14 +00:00
|
|
|
(pdev, DMA_64BIT_MASK)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(ERR_DBG,
|
|
|
|
"Unable to obtain 64bit DMA for \
|
|
|
|
consistent allocations\n");
|
|
|
|
pci_disable_device(pdev);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2005-06-26 22:22:14 +00:00
|
|
|
} else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK)) {
|
2005-04-16 22:20:36 +00:00
|
|
|
DBG_PRINT(INIT_DBG, "s2io_init_nic: Using 32bit DMA\n");
|
|
|
|
} else {
|
|
|
|
pci_disable_device(pdev);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2007-07-23 06:23:54 +00:00
|
|
|
if ((ret = pci_request_regions(pdev, s2io_driver_name))) {
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Request Regions failed - %x \n", __func__, ret);
|
2007-07-23 06:23:54 +00:00
|
|
|
pci_disable_device(pdev);
|
|
|
|
return -ENODEV;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
if (dev_multiq)
|
2008-02-20 22:07:51 +00:00
|
|
|
dev = alloc_etherdev_mq(sizeof(struct s2io_nic), tx_fifo_num);
|
2008-02-20 21:44:07 +00:00
|
|
|
else
|
2008-07-09 06:14:24 +00:00
|
|
|
dev = alloc_etherdev(sizeof(struct s2io_nic));
|
2005-04-16 22:20:36 +00:00
|
|
|
if (dev == NULL) {
|
|
|
|
DBG_PRINT(ERR_DBG, "Device allocation failed\n");
|
|
|
|
pci_disable_device(pdev);
|
|
|
|
pci_release_regions(pdev);
|
|
|
|
return -ENODEV;
|
|
|
|
}
|
|
|
|
|
|
|
|
pci_set_master(pdev);
|
|
|
|
pci_set_drvdata(pdev, dev);
|
|
|
|
SET_NETDEV_DEV(dev, &pdev->dev);
|
|
|
|
|
|
|
|
/* Private member variable initialized to s2io NIC structure */
|
2008-11-13 07:38:14 +00:00
|
|
|
sp = netdev_priv(dev);
|
2007-01-31 19:09:29 +00:00
|
|
|
memset(sp, 0, sizeof(struct s2io_nic));
|
2005-04-16 22:20:36 +00:00
|
|
|
sp->dev = dev;
|
|
|
|
sp->pdev = pdev;
|
|
|
|
sp->high_dma_flag = dma_flag;
|
|
|
|
sp->device_enabled_once = FALSE;
|
2005-10-31 21:55:31 +00:00
|
|
|
if (rx_ring_mode == 1)
|
|
|
|
sp->rxd_mode = RXD_MODE_1;
|
|
|
|
if (rx_ring_mode == 2)
|
|
|
|
sp->rxd_mode = RXD_MODE_3B;
|
|
|
|
|
2007-09-15 21:14:22 +00:00
|
|
|
sp->config.intr_type = dev_intr_type;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
if ((pdev->device == PCI_DEVICE_ID_HERC_WIN) ||
|
|
|
|
(pdev->device == PCI_DEVICE_ID_HERC_UNI))
|
|
|
|
sp->device_type = XFRAME_II_DEVICE;
|
|
|
|
else
|
|
|
|
sp->device_type = XFRAME_I_DEVICE;
|
|
|
|
|
2007-10-05 19:39:21 +00:00
|
|
|
sp->lro = lro_enable;
|
2006-09-13 17:24:59 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Initialize some PCI/PCI-X fields of the NIC. */
|
|
|
|
s2io_init_pci(sp);
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
2005-04-16 22:20:36 +00:00
|
|
|
* Setting the device configuration parameters.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Most of these parameters can be specified by the user during
|
|
|
|
* module insertion as they are module loadable parameters. If
|
|
|
|
* these parameters are not not specified during load time, they
|
2005-04-16 22:20:36 +00:00
|
|
|
* are initialized with default values.
|
|
|
|
*/
|
|
|
|
mac_control = &sp->mac_control;
|
|
|
|
config = &sp->config;
|
|
|
|
|
2007-09-15 21:24:03 +00:00
|
|
|
config->napi = napi;
|
2008-02-20 22:07:51 +00:00
|
|
|
config->tx_steering_type = tx_steering_type;
|
2007-09-15 21:24:03 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Tx side parameters. */
|
2008-02-20 22:07:51 +00:00
|
|
|
if (config->tx_steering_type == TX_PRIORITY_STEERING)
|
|
|
|
config->tx_fifo_num = MAX_TX_FIFOS;
|
|
|
|
else
|
|
|
|
config->tx_fifo_num = tx_fifo_num;
|
|
|
|
|
|
|
|
/* Initialize the fifos used for tx steering */
|
|
|
|
if (config->tx_fifo_num < 5) {
|
|
|
|
if (config->tx_fifo_num == 1)
|
|
|
|
sp->total_tcp_fifos = 1;
|
|
|
|
else
|
|
|
|
sp->total_tcp_fifos = config->tx_fifo_num - 1;
|
|
|
|
sp->udp_fifo_idx = config->tx_fifo_num - 1;
|
|
|
|
sp->total_udp_fifos = 1;
|
|
|
|
sp->other_fifo_idx = sp->total_tcp_fifos - 1;
|
|
|
|
} else {
|
|
|
|
sp->total_tcp_fifos = (tx_fifo_num - FIFO_UDP_MAX_NUM -
|
|
|
|
FIFO_OTHER_MAX_NUM);
|
|
|
|
sp->udp_fifo_idx = sp->total_tcp_fifos;
|
|
|
|
sp->total_udp_fifos = FIFO_UDP_MAX_NUM;
|
|
|
|
sp->other_fifo_idx = sp->udp_fifo_idx + FIFO_UDP_MAX_NUM;
|
|
|
|
}
|
|
|
|
|
2008-02-20 21:44:07 +00:00
|
|
|
config->multiq = dev_multiq;
|
2008-02-20 22:07:51 +00:00
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
2005-04-16 22:20:36 +00:00
|
|
|
config->tx_cfg[i].fifo_len = tx_fifo_len[i];
|
|
|
|
config->tx_cfg[i].fifo_priority = i;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/* mapping the QoS priority to the configured fifos */
|
|
|
|
for (i = 0; i < MAX_TX_FIFOS; i++)
|
2008-02-20 21:44:07 +00:00
|
|
|
config->fifo_mapping[i] = fifo_map[config->tx_fifo_num - 1][i];
|
2005-08-03 19:24:33 +00:00
|
|
|
|
2008-02-20 22:07:51 +00:00
|
|
|
/* map the hashing selector table to the configured fifos */
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++)
|
|
|
|
sp->fifo_selector[i] = fifo_selector[i];
|
|
|
|
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
config->tx_intr_type = TXD_INT_TYPE_UTILZ;
|
|
|
|
for (i = 0; i < config->tx_fifo_num; i++) {
|
|
|
|
config->tx_cfg[i].f_no_snoop =
|
|
|
|
(NO_SNOOP_TXD | NO_SNOOP_TXD_BUFFER);
|
|
|
|
if (config->tx_cfg[i].fifo_len < 65) {
|
|
|
|
config->tx_intr_type = TXD_INT_TYPE_PER_LIST;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2005-11-14 20:25:08 +00:00
|
|
|
/* + 2 because one Txd for skb->data and one Txd for UFO */
|
|
|
|
config->max_txds = MAX_SKB_FRAGS + 2;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Rx side parameters. */
|
|
|
|
config->rx_ring_num = rx_ring_num;
|
2008-04-29 01:08:45 +00:00
|
|
|
for (i = 0; i < config->rx_ring_num; i++) {
|
2005-04-16 22:20:36 +00:00
|
|
|
config->rx_cfg[i].num_rxd = rx_ring_sz[i] *
|
2005-10-31 21:55:31 +00:00
|
|
|
(rxd_count[sp->rxd_mode] + 1);
|
2005-04-16 22:20:36 +00:00
|
|
|
config->rx_cfg[i].ring_priority = i;
|
2008-04-29 01:08:45 +00:00
|
|
|
mac_control->rings[i].rx_bufs_left = 0;
|
|
|
|
mac_control->rings[i].rxd_mode = sp->rxd_mode;
|
|
|
|
mac_control->rings[i].rxd_count = rxd_count[sp->rxd_mode];
|
|
|
|
mac_control->rings[i].pdev = sp->pdev;
|
|
|
|
mac_control->rings[i].dev = sp->dev;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < rx_ring_num; i++) {
|
|
|
|
config->rx_cfg[i].ring_org = RING_ORG_BUFF1;
|
|
|
|
config->rx_cfg[i].f_no_snoop =
|
|
|
|
(NO_SNOOP_RXD | NO_SNOOP_RXD_BUFFER);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Setting Mac Control parameters */
|
|
|
|
mac_control->rmac_pause_time = rmac_pause_time;
|
|
|
|
mac_control->mc_pause_threshold_q0q3 = mc_pause_threshold_q0q3;
|
|
|
|
mac_control->mc_pause_threshold_q4q7 = mc_pause_threshold_q4q7;
|
|
|
|
|
|
|
|
|
|
|
|
/* initialize the shared memory used by the NIC and the host */
|
|
|
|
if (init_shared_mem(sp)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Memory allocation failed\n",
|
2006-07-24 23:52:49 +00:00
|
|
|
dev->name);
|
2005-04-16 22:20:36 +00:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto mem_alloc_failed;
|
|
|
|
}
|
|
|
|
|
2008-10-21 04:42:39 +00:00
|
|
|
sp->bar0 = pci_ioremap_bar(pdev, 0);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!sp->bar0) {
|
2007-01-31 18:30:49 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Neterion: cannot remap io mem1\n",
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->name);
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto bar0_remap_failed;
|
|
|
|
}
|
|
|
|
|
2008-10-21 04:42:39 +00:00
|
|
|
sp->bar1 = pci_ioremap_bar(pdev, 2);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!sp->bar1) {
|
2007-01-31 18:30:49 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Neterion: cannot remap io mem2\n",
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->name);
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto bar1_remap_failed;
|
|
|
|
}
|
|
|
|
|
|
|
|
dev->irq = pdev->irq;
|
|
|
|
dev->base_addr = (unsigned long) sp->bar0;
|
|
|
|
|
|
|
|
/* Initializing the BAR1 address as the start of the FIFO pointer. */
|
|
|
|
for (j = 0; j < MAX_TX_FIFOS; j++) {
|
2007-01-31 19:09:29 +00:00
|
|
|
mac_control->tx_FIFO_start[j] = (struct TxFIFO_element __iomem *)
|
2005-04-16 22:20:36 +00:00
|
|
|
(sp->bar1 + (j * 0x00020000));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Driver entry points */
|
2008-11-22 01:28:55 +00:00
|
|
|
dev->netdev_ops = &s2io_netdev_ops;
|
2005-04-16 22:20:36 +00:00
|
|
|
SET_ETHTOOL_OPS(dev, &netdev_ethtool_ops);
|
2005-08-03 19:35:55 +00:00
|
|
|
dev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
|
2006-06-15 18:36:36 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM;
|
|
|
|
if (sp->high_dma_flag == TRUE)
|
|
|
|
dev->features |= NETIF_F_HIGHDMA;
|
|
|
|
dev->features |= NETIF_F_TSO;
|
2006-06-30 20:37:03 +00:00
|
|
|
dev->features |= NETIF_F_TSO6;
|
2007-01-31 18:28:08 +00:00
|
|
|
if ((sp->device_type & XFRAME_II_DEVICE) && (ufo)) {
|
2005-11-14 20:25:08 +00:00
|
|
|
dev->features |= NETIF_F_UFO;
|
|
|
|
dev->features |= NETIF_F_HW_CSUM;
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
dev->watchdog_timeo = WATCH_DOG_TIMEOUT;
|
2006-11-22 14:57:56 +00:00
|
|
|
INIT_WORK(&sp->rst_timer_task, s2io_restart_nic);
|
|
|
|
INIT_WORK(&sp->set_link_task, s2io_set_link);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2005-08-12 17:15:59 +00:00
|
|
|
pci_save_state(sp->pdev);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Setting swapper control on the NIC, for proper reset operation */
|
|
|
|
if (s2io_set_swapper(sp)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "%s:swapper settings are wrong\n",
|
|
|
|
dev->name);
|
|
|
|
ret = -EAGAIN;
|
|
|
|
goto set_swap_failed;
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
/* Verify if the Herc works on the slot its placed into */
|
|
|
|
if (sp->device_type & XFRAME_II_DEVICE) {
|
|
|
|
mode = s2io_verify_pci_mode(sp);
|
|
|
|
if (mode < 0) {
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: ", __func__);
|
2005-08-03 19:36:55 +00:00
|
|
|
DBG_PRINT(ERR_DBG, " Unsupported PCI bus mode\n");
|
|
|
|
ret = -EBADSLT;
|
|
|
|
goto set_swap_failed;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
if (sp->config.intr_type == MSI_X) {
|
|
|
|
sp->num_entries = config->rx_ring_num + 1;
|
|
|
|
ret = s2io_enable_msi_x(sp);
|
|
|
|
|
|
|
|
if (!ret) {
|
|
|
|
ret = s2io_test_msi(sp);
|
|
|
|
/* rollback MSI-X, will re-enable during add_isr() */
|
|
|
|
remove_msix_isr(sp);
|
|
|
|
}
|
|
|
|
if (ret) {
|
|
|
|
|
|
|
|
DBG_PRINT(ERR_DBG,
|
2009-02-03 23:15:15 +00:00
|
|
|
"s2io: MSI-X requested but failed to enable\n");
|
2008-05-12 17:42:17 +00:00
|
|
|
sp->config.intr_type = INTA;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (config->intr_type == MSI_X) {
|
|
|
|
for (i = 0; i < config->rx_ring_num ; i++)
|
|
|
|
netif_napi_add(dev, &mac_control->rings[i].napi,
|
|
|
|
s2io_poll_msix, 64);
|
|
|
|
} else {
|
|
|
|
netif_napi_add(dev, &sp->napi, s2io_poll_inta, 64);
|
|
|
|
}
|
|
|
|
|
2005-08-03 19:36:55 +00:00
|
|
|
/* Not needed for Herc */
|
|
|
|
if (sp->device_type & XFRAME_I_DEVICE) {
|
|
|
|
/*
|
|
|
|
* Fix for all "FFs" MAC address problems observed on
|
|
|
|
* Alpha platforms
|
|
|
|
*/
|
|
|
|
fix_mac_address(sp);
|
|
|
|
s2io_reset(sp);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* MAC address initialization.
|
|
|
|
* For now only one mac address will be read and used.
|
|
|
|
*/
|
|
|
|
bar0 = sp->bar0;
|
|
|
|
val64 = RMAC_ADDR_CMD_MEM_RD | RMAC_ADDR_CMD_MEM_STROBE_NEW_CMD |
|
2008-01-24 09:45:43 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_OFFSET(0 + S2IO_MAC_ADDR_START_OFFSET);
|
2005-04-16 22:20:36 +00:00
|
|
|
writeq(val64, &bar0->rmac_addr_cmd_mem);
|
2006-04-21 23:18:03 +00:00
|
|
|
wait_for_cmd_complete(&bar0->rmac_addr_cmd_mem,
|
2007-02-24 06:57:32 +00:00
|
|
|
RMAC_ADDR_CMD_MEM_STROBE_CMD_EXECUTING, S2IO_BIT_RESET);
|
2005-04-16 22:20:36 +00:00
|
|
|
tmp64 = readq(&bar0->rmac_addr_data0_mem);
|
|
|
|
mac_down = (u32) tmp64;
|
|
|
|
mac_up = (u32) (tmp64 >> 32);
|
|
|
|
|
|
|
|
sp->def_mac_addr[0].mac_addr[3] = (u8) (mac_up);
|
|
|
|
sp->def_mac_addr[0].mac_addr[2] = (u8) (mac_up >> 8);
|
|
|
|
sp->def_mac_addr[0].mac_addr[1] = (u8) (mac_up >> 16);
|
|
|
|
sp->def_mac_addr[0].mac_addr[0] = (u8) (mac_up >> 24);
|
|
|
|
sp->def_mac_addr[0].mac_addr[5] = (u8) (mac_down >> 16);
|
|
|
|
sp->def_mac_addr[0].mac_addr[4] = (u8) (mac_down >> 24);
|
|
|
|
|
|
|
|
/* Set the factory defined MAC address initially */
|
|
|
|
dev->addr_len = ETH_ALEN;
|
|
|
|
memcpy(dev->dev_addr, sp->def_mac_addr, ETH_ALEN);
|
2007-09-14 11:39:19 +00:00
|
|
|
memcpy(dev->perm_addr, dev->dev_addr, ETH_ALEN);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-01-24 09:45:43 +00:00
|
|
|
/* initialize number of multicast & unicast MAC entries variables */
|
|
|
|
if (sp->device_type == XFRAME_I_DEVICE) {
|
|
|
|
config->max_mc_addr = S2IO_XENA_MAX_MC_ADDRESSES;
|
|
|
|
config->max_mac_addr = S2IO_XENA_MAX_MAC_ADDRESSES;
|
|
|
|
config->mc_start_offset = S2IO_XENA_MC_ADDR_START_OFFSET;
|
|
|
|
} else if (sp->device_type == XFRAME_II_DEVICE) {
|
|
|
|
config->max_mc_addr = S2IO_HERC_MAX_MC_ADDRESSES;
|
|
|
|
config->max_mac_addr = S2IO_HERC_MAX_MAC_ADDRESSES;
|
|
|
|
config->mc_start_offset = S2IO_HERC_MC_ADDR_START_OFFSET;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* store mac addresses from CAM to s2io_nic structure */
|
|
|
|
do_s2io_store_unicast_mc(sp);
|
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
/* Configure MSIX vector for number of rings configured plus one */
|
|
|
|
if ((sp->device_type == XFRAME_II_DEVICE) &&
|
|
|
|
(config->intr_type == MSI_X))
|
|
|
|
sp->num_entries = config->rx_ring_num + 1;
|
|
|
|
|
2007-08-06 09:36:28 +00:00
|
|
|
/* Store the values of the MSIX table in the s2io_nic structure */
|
|
|
|
store_xmsi_data(sp);
|
2006-07-24 23:52:49 +00:00
|
|
|
/* reset Nic and bring it to known state */
|
|
|
|
s2io_reset(sp);
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
2008-04-23 17:29:42 +00:00
|
|
|
* Initialize link state flags
|
2005-08-03 19:36:55 +00:00
|
|
|
* and the card state parameter
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2007-09-06 10:51:14 +00:00
|
|
|
sp->state = 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Initialize spinlocks */
|
2008-01-24 10:08:59 +00:00
|
|
|
for (i = 0; i < sp->config.tx_fifo_num; i++)
|
|
|
|
spin_lock_init(&mac_control->fifos[i].tx_lock);
|
2007-01-31 18:28:08 +00:00
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* SXE-002: Configure link and activity LED to init state
|
|
|
|
* on driver load.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
|
|
|
subid = sp->pdev->subsystem_device;
|
|
|
|
if ((subid & 0xFF) >= 0x07) {
|
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
val64 |= 0x0000800000000000ULL;
|
|
|
|
writeq(val64, &bar0->gpio_control);
|
|
|
|
val64 = 0x0411040400000000ULL;
|
|
|
|
writeq(val64, (void __iomem *) bar0 + 0x2700);
|
|
|
|
val64 = readq(&bar0->gpio_control);
|
|
|
|
}
|
|
|
|
|
|
|
|
sp->rx_csum = 1; /* Rx chksum verify enabled by default */
|
|
|
|
|
|
|
|
if (register_netdev(dev)) {
|
|
|
|
DBG_PRINT(ERR_DBG, "Device registration failed\n");
|
|
|
|
ret = -ENODEV;
|
|
|
|
goto register_failed;
|
|
|
|
}
|
2006-04-21 23:05:41 +00:00
|
|
|
s2io_vpd_read(sp);
|
2007-03-10 02:28:32 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "Copyright(c) 2002-2007 Neterion Inc.\n");
|
2006-07-24 23:52:49 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Neterion %s (rev %d)\n",dev->name,
|
2007-06-08 22:46:36 +00:00
|
|
|
sp->product_name, pdev->revision);
|
2006-07-24 23:52:49 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Driver version %s\n", dev->name,
|
|
|
|
s2io_driver_version);
|
2008-10-27 22:59:26 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: MAC ADDR: %pM\n", dev->name, dev->dev_addr);
|
2007-01-31 18:30:49 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "SERIAL NUMBER: %s\n", sp->serial_num);
|
2006-04-21 23:05:41 +00:00
|
|
|
if (sp->device_type & XFRAME_II_DEVICE) {
|
2005-08-03 19:39:56 +00:00
|
|
|
mode = s2io_print_pci_mode(sp);
|
2005-08-03 19:36:55 +00:00
|
|
|
if (mode < 0) {
|
2006-04-21 23:05:41 +00:00
|
|
|
DBG_PRINT(ERR_DBG, " Unsupported PCI bus mode\n");
|
2005-08-03 19:36:55 +00:00
|
|
|
ret = -EBADSLT;
|
2006-04-21 23:05:41 +00:00
|
|
|
unregister_netdev(dev);
|
2005-08-03 19:36:55 +00:00
|
|
|
goto set_swap_failed;
|
|
|
|
}
|
|
|
|
}
|
2006-04-21 23:05:41 +00:00
|
|
|
switch(sp->rxd_mode) {
|
|
|
|
case RXD_MODE_1:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: 1-Buffer receive mode enabled\n",
|
|
|
|
dev->name);
|
|
|
|
break;
|
|
|
|
case RXD_MODE_3B:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: 2-Buffer receive mode enabled\n",
|
|
|
|
dev->name);
|
|
|
|
break;
|
|
|
|
}
|
2007-01-31 18:28:08 +00:00
|
|
|
|
2008-05-12 17:42:17 +00:00
|
|
|
switch (sp->config.napi) {
|
|
|
|
case 0:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: NAPI disabled\n", dev->name);
|
|
|
|
break;
|
|
|
|
case 1:
|
2007-01-31 18:28:08 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: NAPI enabled\n", dev->name);
|
2008-05-12 17:42:17 +00:00
|
|
|
break;
|
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Using %d Tx fifo(s)\n", dev->name,
|
|
|
|
sp->config.tx_fifo_num);
|
|
|
|
|
2008-04-29 01:08:45 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Using %d Rx ring(s)\n", dev->name,
|
|
|
|
sp->config.rx_ring_num);
|
|
|
|
|
2007-09-15 21:14:22 +00:00
|
|
|
switch(sp->config.intr_type) {
|
2006-04-21 23:05:41 +00:00
|
|
|
case INTA:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Interrupt type INTA\n", dev->name);
|
|
|
|
break;
|
|
|
|
case MSI_X:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Interrupt type MSI-X\n", dev->name);
|
|
|
|
break;
|
|
|
|
}
|
2008-02-20 21:44:07 +00:00
|
|
|
if (sp->config.multiq) {
|
2008-08-19 07:36:01 +00:00
|
|
|
for (i = 0; i < sp->config.tx_fifo_num; i++)
|
|
|
|
mac_control->fifos[i].multiq = config->multiq;
|
2008-02-20 21:44:07 +00:00
|
|
|
DBG_PRINT(ERR_DBG, "%s: Multiqueue support enabled\n",
|
|
|
|
dev->name);
|
|
|
|
} else
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Multiqueue support disabled\n",
|
|
|
|
dev->name);
|
|
|
|
|
2008-02-20 22:07:51 +00:00
|
|
|
switch (sp->config.tx_steering_type) {
|
|
|
|
case NO_STEERING:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: No steering enabled for"
|
|
|
|
" transmit\n", dev->name);
|
|
|
|
break;
|
|
|
|
case TX_PRIORITY_STEERING:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Priority steering enabled for"
|
|
|
|
" transmit\n", dev->name);
|
|
|
|
break;
|
|
|
|
case TX_DEFAULT_STEERING:
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Default steering enabled for"
|
|
|
|
" transmit\n", dev->name);
|
|
|
|
}
|
|
|
|
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (sp->lro)
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: Large receive offload enabled\n",
|
2006-04-21 23:05:41 +00:00
|
|
|
dev->name);
|
2007-01-31 18:28:08 +00:00
|
|
|
if (ufo)
|
|
|
|
DBG_PRINT(ERR_DBG, "%s: UDP Fragmentation Offload(UFO)"
|
|
|
|
" enabled\n", dev->name);
|
2005-08-03 19:29:20 +00:00
|
|
|
/* Initialize device name */
|
2006-04-21 23:05:41 +00:00
|
|
|
sprintf(sp->name, "%s Neterion %s", dev->name, sp->product_name);
|
2005-08-03 19:29:20 +00:00
|
|
|
|
2008-09-04 20:52:54 +00:00
|
|
|
if (vlan_tag_strip)
|
|
|
|
sp->vlan_strip_flag = 1;
|
|
|
|
else
|
|
|
|
sp->vlan_strip_flag = 0;
|
|
|
|
|
2005-08-03 19:24:33 +00:00
|
|
|
/*
|
|
|
|
* Make Link state as off at this point, when the Link change
|
|
|
|
* interrupt comes the state will be automatically changed to
|
2005-04-16 22:20:36 +00:00
|
|
|
* the right state.
|
|
|
|
*/
|
|
|
|
netif_carrier_off(dev);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
register_failed:
|
|
|
|
set_swap_failed:
|
|
|
|
iounmap(sp->bar1);
|
|
|
|
bar1_remap_failed:
|
|
|
|
iounmap(sp->bar0);
|
|
|
|
bar0_remap_failed:
|
|
|
|
mem_alloc_failed:
|
|
|
|
free_shared_mem(sp);
|
|
|
|
pci_disable_device(pdev);
|
2007-07-23 06:23:54 +00:00
|
|
|
pci_release_regions(pdev);
|
2005-04-16 22:20:36 +00:00
|
|
|
pci_set_drvdata(pdev, NULL);
|
|
|
|
free_netdev(dev);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_rem_nic - Free the PCI device
|
2005-04-16 22:20:36 +00:00
|
|
|
* @pdev: structure containing the PCI related information of the device.
|
2005-08-03 19:24:33 +00:00
|
|
|
* Description: This function is called by the Pci subsystem to release a
|
2005-04-16 22:20:36 +00:00
|
|
|
* PCI device and free up all resource held up by the device. This could
|
2005-08-03 19:24:33 +00:00
|
|
|
* be in response to a Hot plug event or when the driver is to be removed
|
2005-04-16 22:20:36 +00:00
|
|
|
* from memory.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void __devexit s2io_rem_nic(struct pci_dev *pdev)
|
|
|
|
{
|
|
|
|
struct net_device *dev =
|
|
|
|
(struct net_device *) pci_get_drvdata(pdev);
|
2007-01-31 19:09:29 +00:00
|
|
|
struct s2io_nic *sp;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
if (dev == NULL) {
|
|
|
|
DBG_PRINT(ERR_DBG, "Driver Data is NULL!!\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2007-02-15 22:37:50 +00:00
|
|
|
flush_scheduled_work();
|
|
|
|
|
2008-11-13 07:38:14 +00:00
|
|
|
sp = netdev_priv(dev);
|
2005-04-16 22:20:36 +00:00
|
|
|
unregister_netdev(dev);
|
|
|
|
|
|
|
|
free_shared_mem(sp);
|
|
|
|
iounmap(sp->bar0);
|
|
|
|
iounmap(sp->bar1);
|
2007-07-23 06:23:54 +00:00
|
|
|
pci_release_regions(pdev);
|
2005-04-16 22:20:36 +00:00
|
|
|
pci_set_drvdata(pdev, NULL);
|
|
|
|
free_netdev(dev);
|
2007-01-31 18:30:49 +00:00
|
|
|
pci_disable_device(pdev);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_starter - Entry point for the driver
|
|
|
|
* Description: This function is the entry point for the driver. It verifies
|
|
|
|
* the module loadable parameters and initializes PCI configuration space.
|
|
|
|
*/
|
|
|
|
|
2007-10-05 19:39:21 +00:00
|
|
|
static int __init s2io_starter(void)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-08-19 21:48:59 +00:00
|
|
|
return pci_register_driver(&s2io_driver);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2005-08-03 19:24:33 +00:00
|
|
|
* s2io_closer - Cleanup routine for the driver
|
2005-04-16 22:20:36 +00:00
|
|
|
* Description: This function is the cleanup routine for the driver. It unregist * ers the driver.
|
|
|
|
*/
|
|
|
|
|
2007-01-31 18:32:57 +00:00
|
|
|
static __exit void s2io_closer(void)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
pci_unregister_driver(&s2io_driver);
|
|
|
|
DBG_PRINT(INIT_DBG, "cleanup done\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
module_init(s2io_starter);
|
|
|
|
module_exit(s2io_closer);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
2006-09-13 17:24:59 +00:00
|
|
|
static int check_L2_lro_capable(u8 *buffer, struct iphdr **ip,
|
2008-02-20 22:09:15 +00:00
|
|
|
struct tcphdr **tcp, struct RxD_t *rxdp,
|
|
|
|
struct s2io_nic *sp)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
{
|
|
|
|
int ip_off;
|
|
|
|
u8 l2_type = (u8)((rxdp->Control_1 >> 37) & 0x7), ip_len;
|
|
|
|
|
|
|
|
if (!(rxdp->Control_1 & RXD_FRAME_PROTO_TCP)) {
|
|
|
|
DBG_PRINT(INIT_DBG,"%s: Non-TCP frames not supported for LRO\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2008-02-20 22:09:15 +00:00
|
|
|
/* Checking for DIX type or DIX type with VLAN */
|
|
|
|
if ((l2_type == 0)
|
|
|
|
|| (l2_type == 4)) {
|
|
|
|
ip_off = HEADER_ETHERNET_II_802_3_SIZE;
|
|
|
|
/*
|
|
|
|
* If vlan stripping is disabled and the frame is VLAN tagged,
|
|
|
|
* shift the offset by the VLAN header size bytes.
|
|
|
|
*/
|
2008-09-04 20:52:54 +00:00
|
|
|
if ((!sp->vlan_strip_flag) &&
|
2008-02-20 22:09:15 +00:00
|
|
|
(rxdp->Control_1 & RXD_FRAME_VLAN_TAG))
|
|
|
|
ip_off += HEADER_VLAN_SIZE;
|
|
|
|
} else {
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/* LLC, SNAP etc are considered non-mergeable */
|
2008-02-20 22:09:15 +00:00
|
|
|
return -1;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
*ip = (struct iphdr *)((u8 *)buffer + ip_off);
|
|
|
|
ip_len = (u8)((*ip)->ihl);
|
|
|
|
ip_len <<= 2;
|
|
|
|
*tcp = (struct tcphdr *)((unsigned long)*ip + ip_len);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int check_for_socket_match(struct lro *lro, struct iphdr *ip,
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
struct tcphdr *tcp)
|
|
|
|
{
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"%s: Been here...\n", __func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if ((lro->iph->saddr != ip->saddr) || (lro->iph->daddr != ip->daddr) ||
|
|
|
|
(lro->tcph->source != tcp->source) || (lro->tcph->dest != tcp->dest))
|
|
|
|
return -1;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int get_l4_pyld_length(struct iphdr *ip, struct tcphdr *tcp)
|
|
|
|
{
|
|
|
|
return(ntohs(ip->tot_len) - (ip->ihl << 2) - (tcp->doff << 2));
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void initiate_new_session(struct lro *lro, u8 *l2h,
|
2008-02-20 22:09:15 +00:00
|
|
|
struct iphdr *ip, struct tcphdr *tcp, u32 tcp_pyld_len, u16 vlan_tag)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
{
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"%s: Been here...\n", __func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
lro->l2h = l2h;
|
|
|
|
lro->iph = ip;
|
|
|
|
lro->tcph = tcp;
|
|
|
|
lro->tcp_next_seq = tcp_pyld_len + ntohl(tcp->seq);
|
2008-02-03 12:27:38 +00:00
|
|
|
lro->tcp_ack = tcp->ack_seq;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
lro->sg_num = 1;
|
|
|
|
lro->total_len = ntohs(ip->tot_len);
|
|
|
|
lro->frags_len = 0;
|
2008-02-20 22:09:15 +00:00
|
|
|
lro->vlan_tag = vlan_tag;
|
2006-09-13 17:24:59 +00:00
|
|
|
/*
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
* check if we saw TCP timestamp. Other consistency checks have
|
|
|
|
* already been done.
|
|
|
|
*/
|
|
|
|
if (tcp->doff == 8) {
|
2008-02-03 12:27:38 +00:00
|
|
|
__be32 *ptr;
|
|
|
|
ptr = (__be32 *)(tcp+1);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
lro->saw_ts = 1;
|
2008-02-03 12:27:38 +00:00
|
|
|
lro->cur_tsval = ntohl(*(ptr+1));
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
lro->cur_tsecr = *(ptr+2);
|
|
|
|
}
|
|
|
|
lro->in_use = 1;
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void update_L3L4_header(struct s2io_nic *sp, struct lro *lro)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
{
|
|
|
|
struct iphdr *ip = lro->iph;
|
|
|
|
struct tcphdr *tcp = lro->tcph;
|
2007-02-09 16:40:15 +00:00
|
|
|
__sum16 nchk;
|
2007-01-31 19:09:29 +00:00
|
|
|
struct stat_block *statinfo = sp->mac_control.stats_info;
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"%s: Been here...\n", __func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
|
|
|
/* Update L3 header */
|
|
|
|
ip->tot_len = htons(lro->total_len);
|
|
|
|
ip->check = 0;
|
|
|
|
nchk = ip_fast_csum((u8 *)lro->iph, ip->ihl);
|
|
|
|
ip->check = nchk;
|
|
|
|
|
|
|
|
/* Update L4 header */
|
|
|
|
tcp->ack_seq = lro->tcp_ack;
|
|
|
|
tcp->window = lro->window;
|
|
|
|
|
|
|
|
/* Update tsecr field if this session has timestamps enabled */
|
|
|
|
if (lro->saw_ts) {
|
2008-02-03 12:27:38 +00:00
|
|
|
__be32 *ptr = (__be32 *)(tcp + 1);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
*(ptr+2) = lro->cur_tsecr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Update counters required for calculation of
|
|
|
|
* average no. of packets aggregated.
|
|
|
|
*/
|
|
|
|
statinfo->sw_stat.sum_avg_pkts_aggregated += lro->sg_num;
|
|
|
|
statinfo->sw_stat.num_aggregations++;
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void aggregate_new_rx(struct lro *lro, struct iphdr *ip,
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
struct tcphdr *tcp, u32 l4_pyld)
|
|
|
|
{
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"%s: Been here...\n", __func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
lro->total_len += l4_pyld;
|
|
|
|
lro->frags_len += l4_pyld;
|
|
|
|
lro->tcp_next_seq += l4_pyld;
|
|
|
|
lro->sg_num++;
|
|
|
|
|
|
|
|
/* Update ack seq no. and window ad(from this pkt) in LRO object */
|
|
|
|
lro->tcp_ack = tcp->ack_seq;
|
|
|
|
lro->window = tcp->window;
|
2006-09-13 17:24:59 +00:00
|
|
|
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (lro->saw_ts) {
|
2008-02-03 12:27:38 +00:00
|
|
|
__be32 *ptr;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/* Update tsecr and tsval from this packet */
|
2008-02-03 12:27:38 +00:00
|
|
|
ptr = (__be32 *)(tcp+1);
|
|
|
|
lro->cur_tsval = ntohl(*(ptr+1));
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
lro->cur_tsecr = *(ptr + 2);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static int verify_l3_l4_lro_capable(struct lro *l_lro, struct iphdr *ip,
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
struct tcphdr *tcp, u32 tcp_pyld_len)
|
|
|
|
{
|
|
|
|
u8 *ptr;
|
|
|
|
|
2008-08-20 23:52:04 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"%s: Been here...\n", __func__);
|
2006-02-03 09:45:13 +00:00
|
|
|
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (!tcp_pyld_len) {
|
|
|
|
/* Runt frame or a pure ack */
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ip->ihl != 5) /* IP has options */
|
|
|
|
return -1;
|
|
|
|
|
2006-07-24 23:55:09 +00:00
|
|
|
/* If we see CE codepoint in IP header, packet is not mergeable */
|
|
|
|
if (INET_ECN_is_ce(ipv4_get_dsfield(ip)))
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
/* If we see ECE or CWR flags in TCP header, packet is not mergeable */
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (tcp->urg || tcp->psh || tcp->rst || tcp->syn || tcp->fin ||
|
2006-07-24 23:55:09 +00:00
|
|
|
tcp->ece || tcp->cwr || !tcp->ack) {
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
/*
|
|
|
|
* Currently recognize only the ack control word and
|
|
|
|
* any other control field being set would result in
|
|
|
|
* flushing the LRO session
|
|
|
|
*/
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2006-09-13 17:24:59 +00:00
|
|
|
/*
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
* Allow only one TCP timestamp option. Don't aggregate if
|
|
|
|
* any other options are detected.
|
|
|
|
*/
|
|
|
|
if (tcp->doff != 5 && tcp->doff != 8)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
if (tcp->doff == 8) {
|
2006-09-13 17:24:59 +00:00
|
|
|
ptr = (u8 *)(tcp + 1);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
while (*ptr == TCPOPT_NOP)
|
|
|
|
ptr++;
|
|
|
|
if (*ptr != TCPOPT_TIMESTAMP || *(ptr+1) != TCPOLEN_TIMESTAMP)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
/* Ensure timestamp value increases monotonically */
|
|
|
|
if (l_lro)
|
2008-02-03 12:27:38 +00:00
|
|
|
if (l_lro->cur_tsval > ntohl(*((__be32 *)(ptr+2))))
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
/* timestamp echo reply should be non-zero */
|
2008-02-03 12:27:38 +00:00
|
|
|
if (*((__be32 *)(ptr+6)) == 0)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2008-04-29 01:08:45 +00:00
|
|
|
s2io_club_tcp_session(struct ring_info *ring_data, u8 *buffer, u8 **tcp,
|
|
|
|
u32 *tcp_len, struct lro **lro, struct RxD_t *rxdp,
|
|
|
|
struct s2io_nic *sp)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
{
|
|
|
|
struct iphdr *ip;
|
|
|
|
struct tcphdr *tcph;
|
|
|
|
int ret = 0, i;
|
2008-02-20 22:09:15 +00:00
|
|
|
u16 vlan_tag = 0;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
|
|
|
if (!(ret = check_L2_lro_capable(buffer, &ip, (struct tcphdr **)tcp,
|
2008-02-20 22:09:15 +00:00
|
|
|
rxdp, sp))) {
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
DBG_PRINT(INFO_DBG,"IP Saddr: %x Daddr: %x\n",
|
|
|
|
ip->saddr, ip->daddr);
|
2008-02-20 22:09:15 +00:00
|
|
|
} else
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
return ret;
|
|
|
|
|
2008-02-20 22:09:15 +00:00
|
|
|
vlan_tag = RXD_GET_VLAN_TAG(rxdp->Control_2);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
tcph = (struct tcphdr *)*tcp;
|
|
|
|
*tcp_len = get_l4_pyld_length(ip, tcph);
|
|
|
|
for (i=0; i<MAX_LRO_SESSIONS; i++) {
|
2008-04-29 01:08:45 +00:00
|
|
|
struct lro *l_lro = &ring_data->lro0_n[i];
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (l_lro->in_use) {
|
|
|
|
if (check_for_socket_match(l_lro, ip, tcph))
|
|
|
|
continue;
|
|
|
|
/* Sock pair matched */
|
|
|
|
*lro = l_lro;
|
|
|
|
|
|
|
|
if ((*lro)->tcp_next_seq != ntohl(tcph->seq)) {
|
|
|
|
DBG_PRINT(INFO_DBG, "%s:Out of order. expected "
|
2008-08-20 23:52:04 +00:00
|
|
|
"0x%x, actual 0x%x\n", __func__,
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
(*lro)->tcp_next_seq,
|
|
|
|
ntohl(tcph->seq));
|
|
|
|
|
|
|
|
sp->mac_control.stats_info->
|
|
|
|
sw_stat.outof_sequence_pkts++;
|
|
|
|
ret = 2;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!verify_l3_l4_lro_capable(l_lro, ip, tcph,*tcp_len))
|
|
|
|
ret = 1; /* Aggregate */
|
|
|
|
else
|
|
|
|
ret = 2; /* Flush both */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ret == 0) {
|
|
|
|
/* Before searching for available LRO objects,
|
|
|
|
* check if the pkt is L3/L4 aggregatable. If not
|
|
|
|
* don't create new LRO session. Just send this
|
|
|
|
* packet up.
|
|
|
|
*/
|
|
|
|
if (verify_l3_l4_lro_capable(NULL, ip, tcph, *tcp_len)) {
|
|
|
|
return 5;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i=0; i<MAX_LRO_SESSIONS; i++) {
|
2008-04-29 01:08:45 +00:00
|
|
|
struct lro *l_lro = &ring_data->lro0_n[i];
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
if (!(l_lro->in_use)) {
|
|
|
|
*lro = l_lro;
|
|
|
|
ret = 3; /* Begin anew */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ret == 0) { /* sessions exceeded */
|
|
|
|
DBG_PRINT(INFO_DBG,"%s:All LRO sessions already in use\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
*lro = NULL;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (ret) {
|
|
|
|
case 3:
|
2008-02-20 22:09:15 +00:00
|
|
|
initiate_new_session(*lro, buffer, ip, tcph, *tcp_len,
|
|
|
|
vlan_tag);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
break;
|
|
|
|
case 2:
|
|
|
|
update_L3L4_header(sp, *lro);
|
|
|
|
break;
|
|
|
|
case 1:
|
|
|
|
aggregate_new_rx(*lro, ip, tcph, *tcp_len);
|
|
|
|
if ((*lro)->sg_num == sp->lro_max_aggr_per_sess) {
|
|
|
|
update_L3L4_header(sp, *lro);
|
|
|
|
ret = 4; /* Flush the LRO */
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DBG_PRINT(ERR_DBG,"%s:Dont know, can't say!!\n",
|
2008-08-20 23:52:04 +00:00
|
|
|
__func__);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void clear_lro_session(struct lro *lro)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
{
|
2007-01-31 19:09:29 +00:00
|
|
|
static u16 lro_struct_size = sizeof(struct lro);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
|
|
|
memset(lro, 0, lro_struct_size);
|
|
|
|
}
|
|
|
|
|
2008-02-20 22:09:15 +00:00
|
|
|
static void queue_rx_frame(struct sk_buff *skb, u16 vlan_tag)
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
{
|
|
|
|
struct net_device *dev = skb->dev;
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(dev);
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
|
|
|
skb->protocol = eth_type_trans(skb, dev);
|
2008-02-20 22:09:15 +00:00
|
|
|
if (sp->vlgrp && vlan_tag
|
2008-09-04 20:52:54 +00:00
|
|
|
&& (sp->vlan_strip_flag)) {
|
2008-02-20 22:09:15 +00:00
|
|
|
/* Queueing the vlan frame to the upper layer */
|
|
|
|
if (sp->config.napi)
|
|
|
|
vlan_hwaccel_receive_skb(skb, sp->vlgrp, vlan_tag);
|
|
|
|
else
|
|
|
|
vlan_hwaccel_rx(skb, sp->vlgrp, vlan_tag);
|
|
|
|
} else {
|
|
|
|
if (sp->config.napi)
|
|
|
|
netif_receive_skb(skb);
|
|
|
|
else
|
|
|
|
netif_rx(skb);
|
|
|
|
}
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
}
|
|
|
|
|
2007-01-31 19:09:29 +00:00
|
|
|
static void lro_append_pkt(struct s2io_nic *sp, struct lro *lro,
|
|
|
|
struct sk_buff *skb,
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
u32 tcp_len)
|
|
|
|
{
|
2006-07-24 23:55:09 +00:00
|
|
|
struct sk_buff *first = lro->parent;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
|
|
|
|
first->len += tcp_len;
|
|
|
|
first->data_len = lro->frags_len;
|
|
|
|
skb_pull(skb, (skb->len - tcp_len));
|
2006-07-24 23:55:09 +00:00
|
|
|
if (skb_shinfo(first)->frag_list)
|
|
|
|
lro->last_frag->next = skb;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
else
|
|
|
|
skb_shinfo(first)->frag_list = skb;
|
2007-01-31 18:32:57 +00:00
|
|
|
first->truesize += skb->truesize;
|
2006-07-24 23:55:09 +00:00
|
|
|
lro->last_frag = skb;
|
[PATCH] S2io: Large Receive Offload (LRO) feature(v2) for Neterion (s2io) 10GbE Xframe PCI-X and PCI-E NICs
Hi,
Below is a patch for the Large Receive Offload feature.
Please review and let us know your comments.
LRO algorithm was described in an OLS 2005 presentation, located at
ftp.s2io.com
user: linuxdocs
password: HALdocs
The same ftp site has Programming Manual for Xframe-I ASIC.
LRO feature is supported on Neterion Xframe-I, Xframe-II and
Xframe-Express 10GbE NICs.
Brief description:
The Large Receive Offload(LRO) feature is a stateless offload
that is complementary to TSO feature but on the receive path.
The idea is to combine and collapse(upto 64K maximum) in the
driver, in-sequence TCP packets belonging to the same session.
It is mainly designed to improve 1500 mtu receive performance,
since Jumbo frame performance is already close to 10GbE line
rate. Some performance numbers are attached below.
Implementation details:
1. Handle packet chains from multiple sessions(current default
MAX_LRO_SESSSIONS=32).
2. Examine each packet for eligiblity to aggregate. A packet is
considered eligible if it meets all the below criteria.
a. It is a TCP/IP packet and L2 type is not LLC or SNAP.
b. The packet has no checksum errors(L3 and L4).
c. There are no IP options. The only TCP option supported is timestamps.
d. Search and locate the LRO object corresponding to this
socket and ensure packet is in TCP sequence.
e. It's not a special packet(SYN, FIN, RST, URG, PSH etc. flags are not set).
f. TCP payload is non-zero(It's not a pure ACK).
g. It's not an IP-fragmented packet.
3. If a packet is found eligible, the LRO object is updated with
information such as next sequence number expected, current length
of aggregated packet and so on. If not eligible or max packets
reached, update IP and TCP headers of first packet in the chain
and pass it up to stack.
4. The frag_list in skb structure is used to chain packets into one
large packet.
Kernel changes required: None
Performance results:
Main focus of the initial testing was on 1500 mtu receiver, since this
is a bottleneck not covered by the existing stateless offloads.
There are couple disclaimers about the performance results below:
1. Your mileage will vary!!!! We initially concentrated on couple pci-x
2.0 platforms that are powerful enough to push 10 GbE NIC and do not
have bottlenecks other than cpu%; testing on other platforms is still
in progress. On some lower end systems we are seeing lower gains.
2. Current LRO implementation is still (for the most part) software based,
and therefore performance potential of the feature is far from being realized.
Full hw implementation of LRO is expected in the next version of Xframe ASIC.
Performance delta(with MTU=1500) going from LRO disabled to enabled:
IBM 2-way Xeon (x366) : 3.5 to 7.1 Gbps
2-way Opteron : 4.5 to 6.1 Gbps
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
2006-01-25 19:53:07 +00:00
|
|
|
sp->mac_control.stats_info->sw_stat.clubbed_frms_cnt++;
|
|
|
|
return;
|
|
|
|
}
|
2007-05-14 23:37:30 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_io_error_detected - called when PCI error is detected
|
|
|
|
* @pdev: Pointer to PCI device
|
2007-07-10 09:58:02 +00:00
|
|
|
* @state: The current pci connection state
|
2007-05-14 23:37:30 +00:00
|
|
|
*
|
|
|
|
* This function is called after a PCI bus error affecting
|
|
|
|
* this device has been detected.
|
|
|
|
*/
|
|
|
|
static pci_ers_result_t s2io_io_error_detected(struct pci_dev *pdev,
|
|
|
|
pci_channel_state_t state)
|
|
|
|
{
|
|
|
|
struct net_device *netdev = pci_get_drvdata(pdev);
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(netdev);
|
2007-05-14 23:37:30 +00:00
|
|
|
|
|
|
|
netif_device_detach(netdev);
|
|
|
|
|
|
|
|
if (netif_running(netdev)) {
|
|
|
|
/* Bring down the card, while avoiding PCI I/O */
|
|
|
|
do_s2io_card_down(sp, 0);
|
|
|
|
}
|
|
|
|
pci_disable_device(pdev);
|
|
|
|
|
|
|
|
return PCI_ERS_RESULT_NEED_RESET;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_io_slot_reset - called after the pci bus has been reset.
|
|
|
|
* @pdev: Pointer to PCI device
|
|
|
|
*
|
|
|
|
* Restart the card from scratch, as if from a cold-boot.
|
|
|
|
* At this point, the card has exprienced a hard reset,
|
|
|
|
* followed by fixups by BIOS, and has its config space
|
|
|
|
* set up identically to what it was at cold boot.
|
|
|
|
*/
|
|
|
|
static pci_ers_result_t s2io_io_slot_reset(struct pci_dev *pdev)
|
|
|
|
{
|
|
|
|
struct net_device *netdev = pci_get_drvdata(pdev);
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(netdev);
|
2007-05-14 23:37:30 +00:00
|
|
|
|
|
|
|
if (pci_enable_device(pdev)) {
|
|
|
|
printk(KERN_ERR "s2io: "
|
|
|
|
"Cannot re-enable PCI device after reset.\n");
|
|
|
|
return PCI_ERS_RESULT_DISCONNECT;
|
|
|
|
}
|
|
|
|
|
|
|
|
pci_set_master(pdev);
|
|
|
|
s2io_reset(sp);
|
|
|
|
|
|
|
|
return PCI_ERS_RESULT_RECOVERED;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* s2io_io_resume - called when traffic can start flowing again.
|
|
|
|
* @pdev: Pointer to PCI device
|
|
|
|
*
|
|
|
|
* This callback is called when the error recovery driver tells
|
|
|
|
* us that its OK to resume normal operation.
|
|
|
|
*/
|
|
|
|
static void s2io_io_resume(struct pci_dev *pdev)
|
|
|
|
{
|
|
|
|
struct net_device *netdev = pci_get_drvdata(pdev);
|
2008-11-13 07:38:14 +00:00
|
|
|
struct s2io_nic *sp = netdev_priv(netdev);
|
2007-05-14 23:37:30 +00:00
|
|
|
|
|
|
|
if (netif_running(netdev)) {
|
|
|
|
if (s2io_card_up(sp)) {
|
|
|
|
printk(KERN_ERR "s2io: "
|
|
|
|
"Can't bring device back up after reset.\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s2io_set_mac_addr(netdev, netdev->dev_addr) == FAILURE) {
|
|
|
|
s2io_card_down(sp);
|
|
|
|
printk(KERN_ERR "s2io: "
|
|
|
|
"Can't resetore mac addr after reset.\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
netif_device_attach(netdev);
|
2008-07-17 08:56:23 +00:00
|
|
|
netif_tx_wake_all_queues(netdev);
|
2007-05-14 23:37:30 +00:00
|
|
|
}
|