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>
hi,
this changes if() BUG(); constructs to BUG_ON() which is
cleaner and can better optimized away
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
hi,
this changes if() BUG(); constructs to BUG_ON() which is
cleaner and can better optimized away
One of the if()s contains a call to de_is_running(),
which seems to be safe to replace, but someone with more
knownledge of the code might want to verify this...
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
hi,
this changes if() BUG(); constructs to BUG_ON() which is
cleaner and can better optimized away
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Add new ->probe_reset operation to ata_port_operations obsoleting
->phy_reset. The main difference from ->phy_reset is that the new
operation is not allowed to manipulate libata internals directly.
It's not allowed to configure or disable the port or devices. It can
only succeed or fail and classify attached devices into passed
@classes.
This change gives more control to higher level and eases sharing reset
methods with EH.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Make ata_dev_try_classify take @r_err to store tf error register value
on completion and return device class instead of directly manipulating
dev->class. This is preparation for new reset mechanism.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Export ata_busy_sleep(), to be used by low level driver reset functions.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Add constants needed to perform SRST. This is preparation for adding
softreset method.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Implement ata_eh_qc_complete/retry() using scsi_eh_finish_cmd() and
scsi_eh_flush_done_q(). This removes all eh scsicmd finish hacks from
low level drivers.
This change was first suggested by Jeff Garzik.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Export two SCSI EH command handling functions. To be used by libata EH.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Add detailed AC_ERR_* flags and use them. Long-term goal is to
describe all errors with err_mask and tf combination (tf for failed
sector information, etc...). After proper error diagnosis is
implemented, sense data should also be generated from err_mask instead
of directly from hardware tf registers as it is currently.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Return AC_ERR_* mask from issue fuctions instead of 0/-1. This
enables things like failing a qc with AC_ERR_HSM when the device
doesn't set DRDY when the qc is about to be issued.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
In ahci_host_intr err_mask is determined from IRQ status but never
used. This patch sets qc->err_mask to the determined err_mask.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
When ata_qc_issue() fails, the qc might have been dma mapped or not.
So, performing only ata_qc_free() results in dma map leak. This patch
makes ata_qc_issue() mark dma map flags correctly on failure and calls
ata_qc_complete() after ata_qc_issue() fails.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
qc used to be freed automatically on command completion. However, as
a qc can carry information about its completion status, it can be
useful to its owner/issuer after command completion. This patch makes
freeing qc responsibility of its owner. This simplifies
ata_exec_internal() and makes command turn-around for atapi request
sensing less hackish.
This change was originally suggested by Jeff Garzik.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
All ata_qc_free() does is calling __ata_qc_complete() which isn't used
anywhere else. Fold __ata_qc_complete() into ata_qc_free().
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
There is a problem with fragmented skb in s2io driver version 2.0.9.4
available in 2.6.16-rc1 kernel. The adapter will fail to transmit if
any scatter-gather skb arrives. This patch provides fix for the above
described problem.
Signed-off-by: Ananda Raju <ananda.raju@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
On my laptop, the b44 device is created and the carrier state defaults
to ON when created by alloc_etherdev. This means tools like NetworkManager
see the carrier as On and try and bring the device up. The correct thing
to do is mark the carrier as Off when device is created.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Since get_settings() returns a signed int and it gets checked
for < 0 to catch an error, res should be a signed int too.
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Make phy 0 actually be read, as it is not being right now as we have:
int mii_status = mdio_read(dev, phy, MII_BMSR);
int phyx = phy & 0x1f;
When we should have instead:
int phyx = phy & 0x1f;
int mii_status = mdio_read(dev, phyx, MII_BMSR);
so that when phy, in the end of the (phy = 1; phy <= 32...) loop gets
to 32 phyx gets to 0, i.e. we were reading at 32, when the intended
read was for 0.
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Unlike their older siblings, ICH6 and 7 use different scheme for MAP
VALUE. This patch makes ata_piix interpret MV properly on ICH6/7.
Pre-ICH6/7
The value of these bits indicate the address range the SATA port
responds to, and whether or not the SATA and IDE functions are
combined.
000 = Non-combined. P0 is primary master. P1 is secondary master.
001 = Non-combined. P0 is secondary master. P1 is primary master.
100 = Combined. P0 is primary master. P1 is primary slave. P-ATA is
2:0 Map Value secondary.
101 = Combined. P0 is primary slave. P1 is primary master. P-ATA is
secondary.
110 = Combined. P-ATA is primary. P0 is secondary master. P1 is
secondary slave.
111 = Combined. P-ATA is primary. P0 is secondary slave. P1 is
secondary master.
ICH6/7
Map Value - R/W. Map Value (MV): The value in the bits below indicate
the address range the SATA ports responds to, and whether or not the
PATA and SATA functions are combined. When in combined mode, the AHCI
memory space is not available and AHCI may not be used.
00 = Non-combined. P0 is primary master, P2 is the primary slave. P1
is secondary master, P3 is the 1:0 secondary slave (desktop
only). P0 is primary master, P2 is the primary slave (mobile
only).
01 = Combined. IDE is primary. P1 is secondary master, P3 is the
secondary slave. (desktop only)
10 = Combined. P0 is primary master. P2 is primary slave. IDE is secondary
11 = Reserved
Signed-off-by: Tejun Heo <htejun@gmail.com>
--
Jeff, without this patch, ata_piix misdetects my ICH7's combined mode,
ending up not applying bridge limits to PX-710SA and configuring IDE
drive on 40-c cable to UDMA/66.
Thanks.
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
In the case where a (broken) BIOS gives us a blank _CRS for
a PCI Interrupt Link Device, the acpi_walk_resources()
will not terminate, but will then give the callback
the resource end tag. Ignore the end tag.
Signed-off-by: Len Brown <len.brown@intel.com>
Handle system controller power down pending events
on SN systems. This allows the system to gracefully shutdown
before the system controller removes power due to
an adverse environmental condition.
Signed-off-by: Aaron Young <ayoung@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Actually I think this is more appropriate so we don't end up with 17
cases that add drivers/sn to the build lib.
Include drivers/sn when CONFIG_IA64_SGI_SN2 or CONFIG_IA64_GENERIC
is enabled.
Acked-by: Dave Jones <davej@redhat.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Convert to use a single mutex instead of two rwsems as this isn't
performance critical.
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Brent Casavant <bcasavan@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
When the scsi_execute_async interface was added it ended up reducing
the flexibility of userspace to send arbitrary scsi commands through
sg using SG_IO. The SG_IO interface allows userspace to specify the
CDB length. This is now ignored in scsi_execute_async and it is
guessed using the COMMAND_SIZE macro, which is not always correct,
particularly for vendor specific commands. This patch adds a cmd_len
parameter to the scsi_execute_async interface to allow the caller
to specify the length of the CDB.
Signed-off-by: Brian King <brking@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
BIOS might change frequency behind our back when BIOS changes allowed
frequencies via _PPC. In this case cpufreq core got out of sync.
Ask driver for current freq and notify governors about a change
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Dave Jones <davej@redhat.com>
Patch from Lucas Correia Villa Real
This patch defines S3C2400 memory map and adds a S3C24XX macro for
common resources between S3C2400, S3C2410 and S3C2440 cpus.
Signed-off-by: Lucas Correia Villa Real <lucasvr@gobolinux.org>
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Some folk want to use RGB555 rather tahn RGB565 with amba-clcd.
Allow amba-clcd to accept either pixel format.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Resetting the adapter causes the ServeRAID driver to exceed
the max time allowed by the softlock watchdog. Resetting the
hardware can easily require 30 or more seconds. To avoid the
"BUG: soft lockup detected on CPU#0!"
result, this patch replaces the mdelay() calls in the
initialization/reset routines with msleep().
Signed-off-by: Jack Hammer <jack_hammer@adaptec.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Internal lun discovery has been removed since fc_transport
integration. Short-circuiting for tape-devices in
qla2x00_update_fcport() could inadvertently result in a
blocked rport timing-out and its targets being reaped.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
A target can LOGO an initiator at any time (i.e. during I/O,
due to a controller hicup, or as a simple authentication
mechanism after an initial CDB command), when this occurs,
the driver attempts to relogin (PLOGI) to the device via the
DPC thread. Add code to make the appropriate upcall to the
FC transport layer (fc_remote_port_add()) upon successful
completion of the PLOGI.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The driver can typically detect port-loss during an
interrupt context (i.e. via interrogation of a status IOCB's
completion status [CS_PORT_LOGGED_OUT]. Due to the calling
requirements of the fc_rport APIs, the driver would defer
removal of the device to the default workqueue. If the
work-item was preceded by an event which caused the port to
obtain visibility (relogin successful, target re-logged into
the topology), deferred removal could inadvertently drop the
rport. The code also no longer defers removal via the
default workqueue, instead opting for use of the driver's
own DPC thread.
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The patch below "fixes" calculation of the virt_addr for the AUTO_REQSENSE
case. I put "fixes" in quotes because the real fix would be to completely
remove it, but that's beyond the scope of this patch.
Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
New versions of the Power5 firmware can send a "re-enable" message to
the virtual scsi adapter. This fix makes us handle the message
correctly. Without it, the driver goes catatonic and the system crashes
unpleasantly.
Signed-off-by: Dave Boutcher <sleddog@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
memset clears once set structure, there is actually no need for memset,
because configure function do it for us. Next, vfree(NULL) is legal, so
avoid useless labels.
Thanks Dave Jones for reporting this.
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
