When using bnx2 in a high transmit load, bnx2_tx_int() cost is pretty high.
There are two reasons.
One is an expensive call to bnx2_get_hw_tx_cons(bnapi) for each freed skb
One is cpu stalls when accessing skb_is_gso(skb) / skb_shinfo(skb)->nr_frags
because of two cache line misses.
(One to get skb->end/head to compute skb_shinfo(skb),
one to get is_gso/nr_frags)
This patch :
1) avoids calling bnx2_get_hw_tx_cons(bnapi) too many times.
2) makes bnx2_start_xmit() cache is_gso & nr_frags into sw_tx_bd descriptor.
This uses a litle bit more ram (256 longs per device on x86), but helps a lot.
3) uses a prefetch(&skb->end) to speedup dev_kfree_skb(), bringing
cache line that will be needed in skb_release_data()
result is 5 % bandwidth increase in benchmarks, involving UDP or TCP receive
& transmits, when a cpu is dedicated to ksoftirqd for bnx2.
bnx2_tx_int going from 3.33 % cpu to 0.5 % cpu in oprofile
Note : skb_dma_unmap() still very expensive but this is for another patch,
not related to bnx2 (2.9 % of cpu, while it does nothing on x86_32)
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When TCP frees up write buffer space, avoid waking up tasks that have
done a poll() or select() on the same socket specifying read-side
events.
This is an extension of a read-side patch by Eric Dumazet.
Signed-off-by: John Dykstra <john.dykstra1@gmail.com>
Acked-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
netdev->dev_addr changed from being an array to being a pointer, so we
should not take its address for memcpy().
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds FCoE related statistics to 82599, including number Rx-ed and Tx-ed
FCoE packets, number of Rx-ed and Tx-ed FCoE packets in dwords, number of bad
Fiber Channel CRCs detected in FCoE packets, and number of FCoE packets dropped
on the Rx side.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements the FCoE Rx side offload feature in ixgbe_main.c
to 82599 using the Rx offload infrastructure code added in the previous
patch. The large receive offload by Direct Data Placement (DDP) for
FCoE is achieved by implementing the ndo_fcoe_ddp_setup and ndo_fcoe_ddp_done
in net_device_ops via netdev. It is up to the ULD, i.e., fcoe and libfc
to query and setup large receive offload accordingly through the corresponding
netdev upon creating fcoe instances.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds infrastructure code for FCoE Rx side offload feature to
82599, which provides large receive offload for FCoE by Direct
Data Placement (DDP). The ixgbe_fcoe_ddp_get() and ixgbe_fcoe_ddp_put()
pair corresponds to the netdev support to FCoE by the function pointers
provided in net_device_ops as ndo_fcoe_ddp_setup and ndo_fcoe_ddp_done.
The implementation of these in ixgbe is shown in the next patch.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements the FCoE Tx side offload features in ixgbe_main.c
to 82599 using the Tx offload infrastructure code added in the previous
patch. This is achieved by the calling the FCoE Sequence Offload (FSO)
function ixgbe_fso() on the transmit path of ixgbe.
This patch also includes an EEPROM check to make sure the NIC we're loading
on is an offload-enabled SKU.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds infrastructure code for FCoE Tx side offload feature to
82599, including Fiber Channel CRC calculation, auto insertion of
the start of frame (SOF) and end of frame (EOF) of FCoE packets,
and large send by FCoE Sequence Offload (FSO).
Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds the FCoE feature code ixgbe_fcoe.c to 82599. For a start, this patch
only adds ixgbe_configure_fcoe() to configure related register for FCoE to 82599.
In patches that follow, I will be adding more functions to ixgbe_fcoe.c to add
support of FCoE offload features to 82599.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds the FCoE feature header ixgbe_fcoe.h to 82599. This header includes
the defines and structures required by the ixgbe driver to support various
offload features in 82599 for Fiber Channel over Ethernet (FCoE). These offloads
features include Fiber Channel CRC calculation, FCoE SOF/EOF auto insertion,
FCoE Sequence Offload (FSO) for large send, and Direct Data Placement (DDP)
for large receive.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds FCoE related register defines to 82599.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
PPC_MULTIPLATFORM was killed in commit 28794d3 but this stale occurrence was
hiding the mv643xx_eth driver in some cases (e.g. Pegasos II)
Signed-off-by: Emil Medve <Emilian.Medve@Freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
i386 allmodconfig:
drivers/net/82596.c: In function 'init_rx_bufs':
drivers/net/82596.c:544: warning: cast to pointer from integer of different size
drivers/net/82596.c:545: warning: cast to pointer from integer of different size
drivers/net/82596.c:548: warning: cast to pointer from integer of different size
drivers/net/82596.c:557: warning: cast to pointer from integer of different size
drivers/net/82596.c:565: warning: cast to pointer from integer of different size
drivers/net/82596.c:569: warning: cast to pointer from integer of different size
drivers/net/82596.c:575: warning: cast to pointer from integer of different size
drivers/net/82596.c: In function 'rebuild_rx_bufs':
drivers/net/82596.c:606: warning: cast to pointer from integer of different size
drivers/net/82596.c:608: warning: cast to pointer from integer of different size
drivers/net/82596.c: In function 'init_i596_mem':
drivers/net/82596.c:680: warning: cast to pointer from integer of different size
drivers/net/82596.c:681: warning: cast to pointer from integer of different size
drivers/net/82596.c: In function 'i596_rx':
drivers/net/82596.c:818: warning: cast to pointer from integer of different size
drivers/net/82596.c: In function 'i596_add_cmd':
drivers/net/82596.c:975: warning: cast to pointer from integer of different size
drivers/net/82596.c:979: warning: cast to pointer from integer of different size
drivers/net/82596.c: In function 'i596_start_xmit':
drivers/net/82596.c:1088: warning: cast to pointer from integer of different size
drivers/net/82596.c:1099: warning: cast to pointer from integer of different size
drivers/net/82596.c: In function 'i596_interrupt':
drivers/net/82596.c:1404: warning: cast to pointer from integer of different size
(ugh)
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change error message when alloc_cpumask_var fails.
Repairs "cpumask: convert drivers/net/sfc".
Signed-off-by: Mike Travis <travis@sgi.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can remove this lock here, since we are in cgroup write handler and
thus the cgrp is guaranteed to be valid, and no lock is needed when
writing a u32 variable.
Signed-off-by: Li Zefan <lizf@cn.fujitsuc.com>
Acked-by: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Patch fixes issues with dev->dev_addr changing from array to pointer.
Hopefully there are no others.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds three attribute files in /sys/class/net/$dev/ for tun
devices; allowing userspace to obtain the information which TUNGETIFF
offers, and more, but without having to attach to the device in question
(which may not be possible if it's in use).
It also fixes a bug which has been present in the TUNGETIFF ioctl since
its inception, where it would never set IFF_TUN or IFF_TAP according to
the device type. (Look carefully at the code which I remove from
tun_get_iff() and how the new tun_flags() helper is subtly different).
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ETH_P_SLOW is already defined in include/linux/if_ether.h.
There's no need to define BOND_ETH_P_LACPDU in drivers/net/bonding/bond_3ad.h
Signed-off-by: Richard Genoud <richard.genoud@gmail.com>
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add missed checking of dev_addr_init return value in alloc_netdev_mq.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
net/core/dev.c | 15 ++++++++++++---
1 files changed, 12 insertions(+), 3 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove unnecessary length parameter since it's always 4 bytes.
Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This board doesn't suppot msi-x well due to msi-x table
mapping (hardware) issue.
Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o Fix the order of irq and hardware context teardown.
Also synchronize the interrupt in dev close() before
releasing tx buffers.
o Fix possible msi-x vector leak if available vectors are
less than requested.
o Request multiple msix vectors only if hardware supports
multiple rx queues.
Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Store msi target status register offset in adapter struct.
This avoids contention on msi_tgt_status table from interrupt
hadlers of different pci function.
Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit ead2ceb0ec ("Network Drop Monitor:
Adding kfree_skb_clean for non-drops and modifying end-of-line points
for skbs") established new conventions for identifying dropped packets.
Align skb_kill_datagram() with these conventions so that packets that
get dropped just before the copy to userspace are properly tracked.
Signed-off-by: John Dykstra <john.dykstra1@gmail.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ashish Karkare <akarkare@marvell.com>
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
fix size of remaped iomem, which is 1 byte to small
(e.g. mappes only 0xff bytes instead of 0x100)
Signed-off-by: Matthias Ludwig <mludwig@ultratronik.de>
Acked-by: Steve Glendinning <steve.glendinning@smsc.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
after the recent changes to wired drivers to use only
netif_carrier_off the driver can have outstanding tx work to
complete that will never complete once link is down. Since the
intel hardware will hold this tx work forever, the driver
notices a tx timeout condition internally and might try
to instigate printk and reset of the part with a
netif_stop_queue, which doesn't work because link is down.
Don't bother arming to tx hang detection when link is down.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make sure we don't get any sign-extend issues when we shift a 1
into bit 31.
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We weren't logging the 82598AT fan failure if it occurred before (ixgbe_open)
as we hadn't sent up to catch the interrupt that event caused.
This patch checks for this failure in:
ixgbe_probe - So we can log the failure asap. We check right after we
set up the adapter->flags, which is when we know that we have a fan.
ixgbe_up_complete - To catch failures that may have happened between probe
and when we set up the interrupt that would normally detect the fan failure.
To enable all of this we need to initialize the adapter flag with
IXGBE_FLAG_FAN_FAIL_CAPABLE when the NIC contained a fan.
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This enables L2 header split when packet split is enabled for 82599.
Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current configuration is not setting queue 0 correctly for DCB
configurations. As a result unconfigured queues are being used to setup
the SRRCTL register rx buffer len sizes.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As per the documentation for 82599 in order to support hardware RSC the
header size must be set. This is only currently done for packet split
mode. This patch sets the header buffer length for all modes.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can avoid waking up tasks not interested in receive notifications,
using wake_up_interruptible_poll() instead of wake_up_interruptible()
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Small cleanup patch to reduce line lengths, before a change in
tcp_prequeue().
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Otherwise list_for_each_entry_rcu() et al. aren't visible
and we get build failures in some configurations.
Signed-off-by: David S. Miller <davem@davemloft.net>
By separating the freeing code from the refcounting decrementing.
Probably reducing icache pressure when we still have reference counts to
go.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit ac45f602ee ("net: infrastructure
for hardware time stamping") added two skb initialization actions to
__alloc_skb(), which need to be added to skb_recycle_check() as well.
Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add barrier() to bnx2_get_hw_{tx|rx}_cons() to fix this issue:
http://bugzilla.kernel.org/show_bug.cgi?id=12698
This issue was reported by multiple i386 users. Without barrier(),
the compiled code looks like the following where %eax contains the
address of the tx_cons or rx_cons in the DMA status block. The
status block contents can change between the cmpb and the movzwl
instruction. The driver would crash if the value was not 0xff during
the cmpb instruction, but changed to 0xff during the movzwl
instruction.
6828: 80 38 ff cmpb $0xff,(%eax)
682b: 0f b7 10 movzwl (%eax),%edx
With the added barrier(), the compiled code now looks correct:
683d: 0f b7 10 movzwl (%eax),%edx
6840: 0f b6 c2 movzbl %dl,%eax
6843: 3d ff 00 00 00 cmp $0xff,%eax
Thanks to Pascal de Bruijn <pmjdebruijn@pcode.nl> for reporting the
problem and Holger Noefer <hnoefer@pironet-ndh.com> for patiently
testing test patches for us.
Also updated version to 2.0.1.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When no limit is given, the bfifo uses a default of tx_queue_len * mtu.
Packets handled by qdiscs include the link layer header, so this should
be taken into account, similar to what other qdiscs do.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The setup_rctl call was making a call into the ring structure after it had
been freed. This was causing a panic on shutdown. This call wasn't
necessary since it is possible to get the needed index from
adapter->vfs_allocated_count.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Traditionally Intel based NIC drivers request I/O port even though it
doesn't need that really.
Intel PCIE 10Gb driver (ixgbe) also requests I/O port but it doesn't
need it either.
This is a little inconvenient situation because sometimes we have to
handle those cards on the slots where any I/O space is not attached.
So we made pach which makes ixgbe driver legacy I/O port free.
Signed-off-by: Masayuki Gouji <gouji.masayuki@jp.fujitsu.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
According to the "PCI Error Recovery" document, if after a recovery,
the bus is disabled, the error_detected function should return
PCI_ERS_RESULT_DISCONNECT. Actually ixgbe error_detected function is
always returning PCI_ERS_RESULT_NEED_RESET, even if the bus is in failure.
This patch just check if the bus is disabled and then returns
PCI_ERS_RESULT_DISCONNET.
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
