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>
This patch implements the UFO support in S2io driver. This patch uses the UFO
interface available in linux-2.6.15 kernel.
Signed-off-by: Ananda Raju <ananda.raju@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This patch provides dynamic two buffer-mode and 3 buffer-mode options.
Previously 2 buffer-mode was compilation option. Now with this patch applied
one can load driver in 2 buffer-mode with module-load parameter
ie.
#insmod s2io.ko rx_ring_mode=2
This patch also provides 3 buffer-mode which provides header separation
functionality. In 3 buffer-mode skb->data will have L2/L3/L4 headers and
"skb_shinfo(skb)->frag_list->data" will have have L4 payload.
one can load driver in 3 buffer-mode with same above module-load parameter
ie.
#insmod s2io.ko rx_ring_mode=3
Please review the patch.
Signed-off-by: Ananda Raju <ananda.raju@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
This patch adds support for MSI/MSI-X feature to the driver. It is
a runtime parameter(for now, loadable parameter). Default is INTA.
Patch has been tested on IA64 platform with Xframe II adapter,
both of which support MSI-X feature. An improvement of about 7%
in throughput(both Tx and Rx) was observed and a reduction by 7%
in CPU utilization during Tx test.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This patch contains the following hardware related fixes and other
miscellaneous bug fixes.
1. Updated the definition of single and double-bit ECC errors
2. Earlier we were allocating Transmit descriptors equal to
MAX_SKB_FRAGS. This was causing a boundary condition failure.
Need to allocate MAX_SKB_FRAGS+1 descriptors.
3. On some platforms(like PPC), pci_alloc_consistent() can return
a zero DMA address. Since the NIC cannot handle zero-addresses,
a workaround has been provided. Basically, we don't use such
that page. We reallocate.
4. If list_info allocation failed during driver load, check for
it during driver exit and return instead of trying to dereference
NULL pointer.
5. Increase the debug level of few non-critical debug messages.
6. Reset the card on critical ECC double errors only in case of
XframeI since XframeII can recover from such errors.
7. Print copyright message on driver load.
8. Bumped up the driver version no. to 2.0.8.1
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Damir Perisa <damir.perisa@solnet.ch> reports:
drivers/net/s2io.h:765: error: invalid lvalue in assignment
drivers/net/s2io.h:766: error: invalid lvalue in assignment
That's a gcc4 error. I don't see why the casts are there anyway..
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This is a patch to incorporate comments from earlier 12 patches.
It also fixes a few issues we found during this time.
Following is a list of changes in this patch. Item 1 incorporates
earlier comments. Issues addressed in items 2 to 4 were discovered
recently.
1. wmb() call in s2io_xmit() replaced with mmiowb().
2. The dtx_control register was earlier programmed incorrectly
for Xframe II adapter.
3. As suggested by hardware team, after a reset, in case of
Xframe II adapter, we clear certain spurious errors by
clearing PCI-X ECC status register, "detected parity error"
bit in PCI_STATUS register and PCI_STATUS bit in txpic_int register.
4. On IBM PPC platforms, we found that in the Rx buffer replenish
function, two memory writes(one to the the descriptor length and
another to the ownership) were getting reordered. This was causing
the adapter to see the ownership transfered to it before the length
was updated.
One solution was to add a wmb() but since this would turnout expensive
on some platforms if called for every descriptor, we set the ownership
bit and other fields of '2' to 'N' Rx descriptors followed by a wmb()
and then set the ownership of first descriptor ('1').
Here the value 'N' is configurable by making it a module loadable
parameter (rxsync_frequency).
(NOTE: This parameter is a power of 2).
5. Bumped up the driver version no. to 2.0.2.1
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This is a patch to provide bimodal interrupt moderation support for
Xframe II adapter. Basically, in this moderation scheme, the adapter
raises a traffic interrupt if the no. of packets transmitted and/or
received reaches a programmable threshold.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This patch provides basic support for the Xframe II adapter.
Includes the following changes:
1. New values to program XAUI interface.
2. Print the PCI/PCI-X mode(bus frequency, width).
3. Remove EOI from reset during intialization.
4. Enable all 8 PCCs if Xframe II adapter.
5. Programs the RLDRAM size depending on the device.
(Note: RLDRAM size on XFARME-I is 64Mb whereas on XFRAME-II it's 32 Mb).
6. Enable extended(64-bit) statistics counters.
7. Program timer interrupt duration based on PCI/PCI-X clock speed.
8. Not required to save/restore PCI config space before/after reset.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This patch implements the slow-path handling functions(link
state change, hardware errors) as a timer. It is not
handled in interrupt handler as was done previously.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
This patch relates to mostly performance related changes.
1. Fixed incorrect computation of PANIC level in rx_buffer_level().
2. Removed unnecessary PIOs(read/write of tx_traffic_int and
rx_traffic_int) from interrupt handler and removed read of
general_int_status register from xmit routine.
3. Enable two-buffer mode(for Rx path) automatically for SGI
systems. This improves Rx performance dramatically on
SGI systems.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
Below patch includes fixes for few purely software bugs identified
since last release.
1. Keep track and display(as part of ethtool command output) the no.
of single-bit and double-bit ECC errors.
2. Handle race condition between intr handler and "interface down"
routine.
3. Initial link state setting modified so that the link state displayed
after "interface Up" is correct.
4. Fix for "Incorrect Tx packet count when TSO is enabled".
5. Disable periodic DMA of statistics and schedule one-shot DMA
only when required.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
Below patch addresses few h/w specific issues.
1. Check for additional ownership bit on Rx path before
starting Rx processing.
2. Enable only 4 PCCs(Per Context Controller) for Xframe I
revisions less than 4.
3. Program Rx and Tx round robin registers depending on
no. of rings/FIFOs.
4. Tx continous interrupts is now a loadable parameter.
5. Reset the card if we get double-bit ECC errors.
6. A soft reset of XGXS being done to force a link state change has been
eliminated.
7. After a reset, clear "parity error detected" bit,
PCI-X ECC status register, and PCI_STATUS bit in
tx_pic_int register.
8. The error in the disabling allmulticast implementation has been
rectified.
9. Leave the PCI-X parameters MMRBC, OST etc. at their
BIOS/system defaults.
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Hi,
We are submitting a series of 13 patches to support our Xframe I and
Xframe II line of products. The patches can be categorized as follows:
Patches 1-8 : Changes applicable to both Xframe I and II
Patches 9-11: Xframe II specific features
Patch 12: Addresses issues found during testing cycle.
Patch 13: Incorpoates mostly the review comments from community
and some last moment bug fixes.
Please review the patches and let us know your comments.
Starting with patch 1 below.
This patch involves cosmetic changes(tabs and indentation,
regrouping of transmit and receive data structures, typecasting,
code cleanup).
Signed-off-by: Ravinandan Arakali <ravinandan.arakali@neterion.com>
Signed-off-by: Raghavendra Koushik <raghavendra.koushik@neterion.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Ravinandan Arakali