Quoting Kent Baxley [EMAIL PROTECTED]:
Subject: Can Mellanox cards handle scatter gather I/O?
Evidently, we're seeing performance problems when a card on one side
is capable of scatter gather IO but the mellanox driver is not able to
handle it. Then a whole lot of time is spent gathering
This email was generated automatically, please do not reply
git_url: git://git.openfabrics.org/ofed_1_2/linux-2.6.git
git_branch: ofed_1_2
Common build parameters: --with-ipoib-mod --with-sdp-mod --with-srp-mod
--with-user_mad-mod --with-user_access-mod --with-mthca-mod --with-core-mod
/ofa_1_2_c_kernel-20070815-0200_linux-2.6.18_powerpc_check/drivers/infiniband/hw/ehca/ehca_main.c:831:
error: invalid type argument of â-â
/home/vlad/tmp/ofa_1_2_c_kernel-20070815-0200_linux-2.6.18_powerpc_check/drivers/infiniband/hw/ehca/ehca_main.c:834:
error: invalid type argument of â-â
/home
- .max_send_sge = 1,
+ .max_send_sge = MAX_SKB_FRAGS + 1,
this makes WQEs much bigger -- do we want to do this for all devices?
Or is it going to hurt on something like Tavor, which doesn't do
checksum offload and hence can't take advantage of
Added valgrind support to the auto-tools configuration file.
Signed-off-by: Dotan Barak [EMAIL PROTECTED]
---
diff --git a/configure.in b/configure.in
index ce15f44..9171319 100644
--- a/configure.in
+++ b/configure.in
@@ -15,6 +15,19 @@ AC_ARG_ENABLE(libcheck, [ --disable-libcheck do not
Added valgrind support to the cma code and mark buffers that were filled
by the kernel level with the macro VALGRIND_MAKE_MEM_DEFINED.
Signed-off-by: Dotan Barak [EMAIL PROTECTED]
---
diff --git a/src/cma.c b/src/cma.c
index 32edc1f..42fa162 100644
--- a/src/cma.c
+++ b/src/cma.c
@@ -57,6
On 8/9/07, Sean Hefty [EMAIL PROTECTED] wrote:
I'd like to propose the following change as a simple solution for handling SA
scalability problems:
Modify the ib_sa module to support an SA LID that's separate from the SM LID.
This concept is supported by the spec through SA redirection;
Roland Dreier wrote:
I am implementing this only for completeness sake and want to do it with
minimal
effort. Given the above, do you still see it necessary to use NAPI?
Is it acceptable that I roll up the previous NOSRQ patches (other than
this one) into
a single patch and can
David Miller wrote:
From: Sean Hefty [EMAIL PROTECTED]
Date: Thu, 09 Aug 2007 14:40:16 -0700
Steve Wise wrote:
Any more comments?
Does anyone have ideas on how to reserve the port space without using a
struct socket?
How about we just remove the RDMA stack altogether? I am not at all
In message [EMAIL PROTECTED],Michael S. Tsirkin writes:
if ((dev-features NETIF_F_SG)
!(dev-features NETIF_F_ALL_CSUM)) {
printk(KERN_NOTICE %s: Dropping NETIF_F_SG since no checksum
feature.\n,
dev-name);
dev-features
Free the device list to prevent memory leaks (and give a good code example).
Signed-off-by: Dotan Barak [EMAIL PROTECTED]
---
diff --git a/examples/devinfo.c b/examples/devinfo.c
index d054999..4e4316a 100644
--- a/examples/devinfo.c
+++ b/examples/devinfo.c
@@ -323,7 +323,7 @@ int main(int
On Wed, 2007-08-15 at 09:05 +0300, Michael S. Tsirkin wrote:
Quoting Kent Baxley [EMAIL PROTECTED]:
Subject: Can Mellanox cards handle scatter gather I/O?
Evidently, we're seeing performance problems when a card on one side
is capable of scatter gather IO but the mellanox driver is
Thank you all for all the comments! - and I know there will be more.
I have reworked the patches so that they pass the checkpatch.pl barrier
(except for one case), split the patches to core, ulp or hw driver and
in some cases referred to your comments on the implementation. Also they
are self
Add high dma support to ipoib
This patch assumes all IB devices support 64 bit DMA.
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_main.c
===
---
Add checksum offload support to the core
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
A device that publishes IB_DEVICE_IP_CSUM actually supports
calculating checksum on transmit and provides indication whether
the checksum is OK on receive.
Index: linux-2.6.23-rc1/include/rdma/ib_verbs.h
Add checksum offload support to mlx4
Signed-off-by: Ali Ayub [EMAIL PROTECTED]
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Index: linux-2.6.23-rc1/include/linux/mlx4/cq.h
===
--- linux-2.6.23-rc1.orig/include/linux/mlx4/cq.h
Add checksum offload support to ipoib
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
Signed-off-by: Ali Ayub [EMAIL PROTECTED]
---
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib.h
===
---
Add LSO support to mlx4
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Index: linux-2.6.23-rc1/drivers/net/mlx4/fw.c
===
--- linux-2.6.23-rc1.orig/drivers/net/mlx4/fw.c 2007-08-15 20:50:30.0
+0300
+++
Add LSO support to ipoib
Using LSO improves performance by allowing the software
to not fragment the payload to mtu sized patckets and also
results in lower rate of interrupts since each such work
request has just one CQE.
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Index:
Configure QP's max message size according to the value queried
by query dev cap.
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Please note the two previous patches were numbered 6. fix typo is 6
and s/g is 7.
Index: linux-2.6.23-rc1/drivers/infiniband/hw/mlx4/qp.c
Fix type - comma instead of semicolon
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
===
--- linux-2.6.23-rc1.orig/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
Add checksum offload support to the core
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
resending this one - I suspect the previous one had formatting problems.
A device that publishes IB_DEVICE_IP_CSUM actually supports
calculating checksum on transmit and provides indication whether
the
the ofa_1_2_c_kernel-20070815-0200.tgz from today.
The changes are applied and we successfully built the package on SLES10 and
rhel-4.5.
All works fine for us.
regards Stefan
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi
Add scatter/gather support to ipoib
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
If the net device does not use scatter/gather then the QP
uses a single send sge.
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib.h
===
Add LSO supprt at the core
This patch provides support at the core level for devices that
support TCP large send offload fragmentation.
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
Index: linux-2.6.23-rc1/include/rdma/ib_verbs.h
Add ethtool support to ipoib
Signed-off-by: Eli Cohen [EMAIL PROTECTED]
---
This one is actually the foundation with no real contecxt. I think
we can add here all the logic of wheather to allow using a certain
feature, e.g. checksum offload, scatter/gather etc. and decide on
all the
On 16:21 Tue 14 Aug , Ira Weiny wrote:
From 6e01f304c1d1cbd8d1163708d907ec0550a0ca31 Mon Sep 17 00:00:00 2001
From: Ira K. Weiny [EMAIL PROTECTED]
Date: Wed, 25 Jul 2007 11:33:08 -0700
Subject: [PATCH] Fix the osm_epi_node_id_t structure name.
The structure osm_epi_node_id_t
On 18:06 Tue 14 Aug , Ira Weiny wrote:
From fc0251b5150afdd5d40d55facdf85c645939b482 Mon Sep 17 00:00:00 2001
From: Ira K. Weiny [EMAIL PROTECTED]
Date: Tue, 14 Aug 2007 18:04:43 -0700
Subject: [PATCH] iblinkinfo.pl: Remove annoying double quote around switch
name.
Signed-off-by:
Roland:
I read the IB specification, it does say posting send WR to QP
in error state is OK, and the WR will be completed in FLUSH error, but
it does not say that for posting receive WR. I want to check with you if
posting receive WR to QP in error state returns SUCCESS, and the WR
will
Fix trivial syntax errors - extra ) and unmatched #endif.
Signed-off-by: Sasha Khapyorsky [EMAIL PROTECTED]
---
opensm/libvendor/osm_vendor_al.c |1 -
opensm/libvendor/osm_vendor_ts.c |2 --
2 files changed, 0 insertions(+), 3 deletions(-)
diff --git a/opensm/libvendor/osm_vendor_al.c
A similar fix to netfilter from Eric Dumazet inspired me to
look around a bit by using some grep/sed stuff as looking for
this kind of bugs seemed easy to automate. This is one of them
I found where it looks like this semicolon is not valid.
Signed-off-by: Ilpo Järvinen [EMAIL PROTECTED]
---
The patches to follow is a roll up of a couple of patches. These patches
have been tested against 2.6.23-rc1 kernel on ppc64 machines.
PATCH 1 -this is a combined patch of V8 and the misc patch along with
some minor changes.
PATCH 2 -this enables a fall back to UD mode when there are not enough
This patch enables IPoIB CM on HCAs that do not support SRQ.
Signed-off-by: Pradeep Satyanarayana [EMAIL PROTECTED]
---
--- a/linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib.h 2007-07-31
13:14:30.0 -0400
+++ b/linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib.h 2007-08-14
+if test x$with_valgrind = x || test x$with_valgrind = xno; then
+want_valgrind=no
+AC_DEFINE([NVALGRIND], 1, [Define to 1 to disable Valgrind annotations.])
+AC_CHECK_HEADER(valgrind/memcheck.h,
+[AC_DEFINE(HAVE_VALGRIND_MEMCHECK_H, 1,
+[Define to 1 if you have the
This patch provides a fall back mechanism to UD mode if enough resources
are not available in CM mode.
Signed-off-by: Pradeep Satyanarayana [EMAIL PROTECTED]
---
--- a/linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_cm.c 2007-08-15
18:16:26.0 -0400
+++
I am trying to develop some code using OFED distribution. I am trying to
create QP but its not able to be created. I am not sure of the reasons
why. The snippet of the code would be something like this. I am not sure
if this what I am doing is correct.
memset (qp_init_attr, 0, sizeof (struct
Steve Wise wrote:
David Miller wrote:
From: Sean Hefty [EMAIL PROTECTED]
Date: Thu, 09 Aug 2007 14:40:16 -0700
Steve Wise wrote:
Any more comments?
Does anyone have ideas on how to reserve the port space without using
a struct socket?
How about we just remove the RDMA stack altogether?
Needing to reach out of the RDMA sandbox and reserve net stack
resources away from itself travels a path we've consistently avoided.
Where did the idea of an RDMA sandbox come from? Obviously no one
disagrees with keeping things clean and maintainable, but the idea
that RDMA is a
I read the IB specification, it does say posting send WR to QP
in error state is OK, and the WR will be completed in FLUSH error, but
it does not say that for posting receive WR. I want to check with you if
posting receive WR to QP in error state returns SUCCESS, and the WR
will be
qp_init_attr.cap.max_recv_sge = IB_MAX_REQ;
qp_init_attr.cap.max_send_sge = IB_MAX_REQ;
IB_MAX_REQ is set 1024.
You didn't say what hardware you are using, but I would suspect that
1024 gather/scatter entries is too many.
Is there any maximum limit to the amount of max wr that I can set
* Roland Dreier ([EMAIL PROTECTED]) wrote:
qp_init_attr.cap.max_recv_sge = IB_MAX_REQ;
qp_init_attr.cap.max_send_sge = IB_MAX_REQ;
IB_MAX_REQ is set 1024.
You didn't say what hardware you are using, but I would suspect that
1024 gather/scatter entries is too many.
Is there any
A similar fix to netfilter from Eric Dumazet inspired me to
look around a bit by using some grep/sed stuff as looking for
this kind of bugs seemed easy to automate. This is one of them
I found where it looks like this semicolon is not valid.
Indeed, this is a bug. It only triggers on the
It's not about being a niche. It's about creating a maintainable
software net stack that has predictable behavior.
Needing to reach out of the RDMA sandbox and reserve net stack resources
away from itself travels a path we've consistently avoided.
We need to ensure that we're also creating a
Since implementor is more accurate that spec. Thanks.
--CQ
-Original Message-
From: Roland Dreier [mailto:[EMAIL PROTECTED]
Sent: Wednesday, August 15, 2007 10:15 PM
To: Tang, Changqing
Cc: general@lists.openfabrics.org; [EMAIL PROTECTED]
Subject: Re: Post receive WR to QP in
Linus, please pull from
master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband.git for-linus
This tree is also available from kernel.org mirrors at:
git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband.git
for-linus
This is basically a resend of the small fixes for
OSM Simulation Regression Summary
[Generated mail - please do NOT reply]
OpenSM rev = Mon_Aug_13_00:33:19_2007 [1094c660253d5749331c30c82817e96a6faf939d]
ibutils rev = Thu_Aug_9_15:39:12_2007 [d1681ad8afb35441ac1741392d9e2e94a9d7e22f]
Total=560 Pass=560 Fail=0
Pass:
42 Stability
46 matches
Mail list logo