On 13/09/2012 10:35, Jack Morgenstein wrote:
I seem to recall that there were problems with IPoIB when partial membership
pkeys are used.
There are some issues in the overall solution, since ARPs sent over the
broadcast group reach
also nodes with partial membership their HCA generated pkey v
On 11/09/2012 19:52, Doug Ledford wrote:
On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
>Enhance the cached and non-cached pkey table lookups to enable limited and full
>members of the same pkey to co-exist in the pkey table.
>
>This is necessary for SRIOV to allow for a scheme where some guests w
On 13/09/2012 18:53, Or Gerlitz wrote:
The physical PKey table can contain both full and partial memberships
of the same Pkey.
This is needed to serve 2 VFs that are granted access to the same
PKey, albeit with different membership types.
Example use case -- RDMA or IPoIB network storage
On Thu, Aug 30, 2012 at 12:09 PM, Or Gerlitz wrote:
> CMA multicast joins done under the ipoib port state must be carried
> out using the same component mask used by ipoib. Otherwise, its possible
> for the CMA to create a group to which a join made by ipoib will fail,
> or vise-v
Hi Sean,
We have a case here where an app which uses librdmacm wants to preserve its
QP over HA events such IB link down/up, specifically the sequence of
operations
done by the app is the following:
1. rdma_create_id using the IPoIB port space
2. rdma_bind _addr
3. rdma_create_qp using UD QP
On 19/09/2012 18:48, Hefty, Sean wrote:
Can this flushing be somehow done with the current librdmacm/libibverbs APIs
or we need some enhancement?
You can call verbs directly to transition the QP state. That leaves the CM
state unchanged, which doesn't really matter for UD QPs anyway.
Alex,
helpers to deal with new InfiniBand link speeds
Fix resource leaks in the pingpong examples present in the failure/error
flows.
Or Gerlitz (1):
Add raw packet QP type
Makefile.am|6 +++-
examples/rc_pingpong.c | 43 ++---
examples
with the NET_RAW capability are allowed to create raw packet QPs (the
name "raw packet QP" is supposed to suggest an analogy to AF_PACKET /
SOL_RAW sockets).
Signed-off-by: Or Gerlitz
---
include/infiniband/verbs.h |3 ++-
man/ibv_create_qp.3|2 +-
man/ibv_modify_qp.3
From: Dotan Barak
Introduce support for the following extended speeds:
FDR:IBA extended speed 14.0625 Gbps.
EDR:IBA extended speed 25.78125 Gbps.
Signed-off-by: Dotan Barak
Reviewed-by: Hal Rosenstock
Signed-off-by: Or Gerlitz
---
Makefile.am|6 +++-
include
From: Dotan Barak
Signed-off-by: Dotan Barak
Signed-off-by: Or Gerlitz
---
examples/rc_pingpong.c | 43 ---
examples/srq_pingpong.c | 51 ++
examples/uc_pingpong.c | 43
-logs and cleaned
some checkpatch comments.
Or.
Dotan Barak (5):
Replace sscanf() to strtol()
Allow to use the whole BF buffer
Use BlueFlame for RDMA_WRITE/WITH_IMM without data
Change enumeration names for masked atomic opcodes
When calling ibv_modify_qp() return right value
Or Gerlitz
Implement raw packet QPs for Ethernet ports.
Signed-off-by: Or Gerlitz
---
src/qp.c |4
1 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/src/qp.c b/src/qp.c
index 40a6689..90c4e80 100644
--- a/src/qp.c
+++ b/src/qp.c
@@ -286,6 +286,10 @@ int mlx4_post_send(struct ibv_qp
From: Sagi Grimberg
Use the limits reported in ib_query_device(). Make sure that the limits
returned to the caller following qp creation also lie within the
reported device limits.
Signed-off-by: Sagi Grimberg
Signed-off-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
src/mlx4.h | 14
From: Dotan Barak
Change the enumeration names of the masked atomic opcodes to be
consistent with the ones used by the mlx4 kernel driver.
Signed-off-by: Dotan Barak
Signed-off-by: Or Gerlitz
---
src/mlx4.h |4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/mlx4
From: Dotan Barak
Use Blue-Flame for RDMA Write and RDMA Write with immediate without any
data (no s/g). This improves latency for those messages.
Signed-off-by: Dotan Barak
Reviewed-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
src/qp.c |2 ++
1 files changed, 2 insertions(+), 0
From: Dotan Barak
Increase the maximum size of messages (from 192 to 208) that will use the blue
flame buffer.
Signed-off-by: Dotan Barak
Reviewed-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
src/qp.c |2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/src/qp.c b
From: Dotan Barak
When converting a string to a numeric value, strtol() is more safe to use.
Signed-off-by: Dotan Barak
Signed-off-by: Or Gerlitz
---
src/mlx4.c |4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/mlx4.c b/src/mlx4.c
index 4989c46..0a9139f 100644
From: Dotan Barak
When the ibv_query_port() call made by mlx4_modify_qp() fails, the return
value from the latter should indicate the error status of the former and
not simply -1.
Signed-off-by: Dotan Barak
Signed-off-by: Or Gerlitz
---
src/verbs.c |6 --
1 files changed, 4
On Tue, Sep 11, 2012 at 8:10 PM, Doug Ledford wrote:
> On 8/3/2012 4:40 AM, Jack Morgenstein wrote:
> > struct mlx4_ib_sriov{} is created by the master only.
> > It is a container for the following:
> > 1. All the info required by the PPF to multiplex and de-multiplex MADs
> >(including those
On Fri, Sep 21, 2012 at 3:51 AM, Luick, Dean wrote:
>> @@ -286,6 +286,10 @@ int mlx4_post_send(struct ibv_qp *ibqp, struct
>> ibv_send_wr *wr,
>> size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
>> break;
>>
>> + case IBV_QPT_RAW_PACKET:
>
Implement raw packet QPs for Ethernet ports.
Signed-off-by: Or Gerlitz
---
changes from previous version:
- addressed reviewer comment to add break on the post send
flow for the new QP type
src/qp.c |6 ++
1 files changed, 6 insertions(+), 0 deletions(-)
diff --git a/src/qp.c b
ce this is made, the RX flow
processing of IO responses will now be distributed across multiple CPUs.
QPs (--> iser sessions) are assigned to CQs in round robin manner
using the current CQ with minimal number of sessions attached to it.
Signed-off-by: Or Gerlitz
Signed-off-by: Alex Tabachnik
On Wed, Sep 26, 2012 at 7:14 PM, Roland Dreier wrote:
> So I have SR-IOV enabled on a ConnectX-3 adapter, and I loaded the
> driver with num_vfs=1 probe_vf=1, so in the host I see:
>
> # The master device
> $ ibv_devinfo -d mlx4_1
> hca_id: mlx4_1
> transport: InfiniBa
On Wed, Sep 26, 2012 at 10:22 PM, Or Gerlitz wrote:
> On Wed, Sep 26, 2012 at 7:14 PM, Roland Dreier wrote:
>> What do I need for the slave VF's port to become active?
>> I'm running opensm 3.3.13 on a different box, is that new enough?
>> (does SR-IOV require a
On Wed, Sep 26, 2012 at 6:42 AM, Roland Dreier wrote:
> From: Roland Dreier
>
> On an SR-IOV master device, __mlx4_init_one() calls mlx4_init_hca()
> before mlx4_multi_func_init(). However, for unlucky configurations,
> mlx4_init_hca() might call mlx4_SENSE_PORT() (via mlx4_dev_cap()), and
> tha
On 27/09/2012 08:47, Roland Dreier wrote:
On Wed, Sep 26, 2012 at 2:30 PM, Or Gerlitz wrote:
Roland, did this help? do you have IB link for the VF? IPoIB working on it?
Sorry, replied to Hal only by accident. Yes, latest opensm makes things
work fine for me.
Good, so you have basic IB
On 27/09/2012 08:46, Roland Dreier wrote:
On Wed, Sep 26, 2012 at 2:51 PM, Or Gerlitz wrote:
What exactly did you mean by saying "for unlucky configurations"
above? what value did you use for mlx4_core's port_array_type module
param?
I didn't set the parameter at a
On 27/09/2012 10:17, Roland Dreier wrote:
I think I had it cabled up directly to another HCA, and that HCA was in a
system that was either off or at least didn't have the driver loaded. So the
port was in the physically DOWN state...
However, I just tried it and even with that other HCA enabled
On 27/09/2012 08:46, Roland Dreier wrote:
On Wed, Sep 26, 2012 at 2:51 PM, Or Gerlitz wrote:
What exactly did you mean by saying "for unlucky configurations"
above? what value did you use for mlx4_core's port_array_type module
param?
I didn't set the parameter at a
command for multi-function devices.
makes sense, nice doing!
Acked-by: Or Gerlitz for patches 1-3
Roland, I see that these three patches are queued @ your for-next and
also the initial
patch which in a way is more lengthy and heavy. I wonder whether
wouldn't it be fare to allow for Jack to r
On Sun, Sep 23, 2012 at 5:17 PM, Or Gerlitz wrote:
> From: Alex Tabachnik
>
> RX/TX CQs will now be selected from a per HCA pool, for the RX flow
> this has the effect of using different interrupt vectors over low level
> drivers (such as mlx4) who map the "vector" par
On Wed, Oct 3, 2012 at 6:33 AM, Roland Dreier wrote:
>
> From: Roland Dreier
>
> I'll be sending the following to Linus shortly:
>
> [PATCH] IPoIB: Fix build with CONFIG_INFINIBAND_IPOIB_CM=n
>
> With the new netlink support in commit 862096a8bbf8 ("IB/ipoib: Add more
> rtnl_link_ops callbacks")
On 04/10/2012 03:47, Steve Wise wrote:
Not used by iwarp drivers...
Which one, the retry counter or the RNR retry counter?
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.ke
On 29/09/2012 00:53, Hefty, Sean wrote:
[...] Infrastructure to support extended verbs capabilities in a
forward/backward
manner. [...]
Sean,
Just two nits re the patch set...
1. it would be easier to eventually pick these patches if the library
name would be part of the text in brackets, e
file at ~ogerlitz/tmp-patches/0001-NET-IB-mlx4-64-byte-CQE-EQE-support.patch
Jack, I'd like you to review the part in this patch which
relates to SRIOV, I've tested it now, applied on Roland's for-next,
and it works OK on a system with a VF probed on the host and
doing ipoib ping. Both VF and PF n
On 04/10/2012 15:05, Or Gerlitz wrote:
I'd like to try and push this for 3.7
Indeed, but this is WIP so please ignore (actually if you have something
send comments also now...)
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a me
Hi Roland,
We got a report that on a system with multiple (say two) ConnectX HCAs,
its possible
for the order of device probing to be different across simple reboots,
that is sometimes
the device with PCI BDF X is probed 1st and gets to be IB device mlx4_0
and some
other-timesthe device with B
On Thu, Oct 11, 2012 at 10:44 PM, Roland Dreier wrote:
> No one has really ever tried to deal with the issue of userspace RDMA on
> a cache-incoherent architecture. Basically if you try the current stack, the
> in-kernel users (IPoIB etc) should be OK but libibverbs etc. will be
> completely br
From: Eli Cohen
A client re-register event invokes cleanup of all MCGs.
This is required to protect against misbehaved guests leading to
corruption of join/leave database. However, since cleaning up the MCGs
is a heavy operation, it is pushed to a work queue for further processing.
Client re-regi
From: Jack Morgenstein
In the MAD paravirtualization code, one of the checks performed when forwarding
QP1 (GSI) packets from wire to slave was a pkey check: The pkey received in the
MAD must be present in the guest's paravirtualized pkey table, and at least one
of the (packet pkey, guest pkey) m
read from the device
uverbs
sysfs entry, and uses it as the key to realize the CQE size if/as advertized by
the kernel mlx4_ib driver. Older kernel mlx4_ib ABI versions are still
supported.
Signed-off-by: Eli Cohen
Signed-off-by: Or Gerlitz
---
src/cq.c | 39
key processing in the Primary Physical Function
(PPF)
Or Gerlitz (1):
{NET,IB}/mlx4: 64 byte CQE/EQE support
drivers/infiniband/hw/mlx4/cq.c| 34 +++--
drivers/infiniband/hw/mlx4/mad.c | 89 +++-
drivers/infiniband/hw/mlx4/main.c
river does use 64B CQEs or future device capabilities which
must be in sync by user space. This practice allows to work with unmodified
libmlx4 on older devices (e.g A0, B0) which don't support 64 byte CQEs.
Signed-off-by: Eli Cohen
Signed-off-by: Or Gerlitz
---
drivers/infiniband/hw
On Thu, Oct 18, 2012 at 7:33 PM, Roland Dreier wrote:
> thanks, applied
thanks, any insight/s on patches 3/4?
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordom
On Thu, Oct 18, 2012 at 4:58 PM, Roland Dreier wrote:
>
> On Wed, Oct 17, 2012 at 9:42 AM, Or Gerlitz wrote:
> >
> > Also, a respin of the 64B CQE/EQE patches (kernel and user-space), over
> > the V1 posting
> > you were asking for max flexibility - e.g expose/forc
On Fri, Oct 19, 2012 at 1:58 AM, Roland Dreier wrote:
> [...] So I think we need some flag passed to the mlx4_core (that drives the
> PPF) that lets the user opt into 64B CQEs. I would suggest that we
> start with the default value be "disabled" and then flip that after a few
> kernel versions (a
rt mgmt event propagation" make it
pretty hard to actually use the mlx4_core debug messages when running in
SRIOV/IB
mode -- for example, the module load sequence of a device with one VF yielded
631 debug prints, with 408 of them being from this set. Let's just remove them.
Signed-o
From: Dotan Barak
Fixed the resource cleanup to act correctly and prevent a kernel oops when
mlx4_QUERY_ADAPTER() fails.
Signed-off-by: Dotan Barak
Reviewed-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
drivers/net/ethernet/mellanox/mlx4/main.c |5 -
1 files changed, 4
to be off by default and a knob to
set it on, as was requested during the V2 review.
Or
Dotan Barak (1):
net/mlx4: Perform correct resource cleanup if mlx4_QUERY_ADAPTER() fails
Or Gerlitz (2):
net/mlx4_core: Remove more annoying debug messages from the SRIOV flow
{NET,IB}/mlx4:
s function capabilities change towards VFs and ABI change
towards libmlx4 -- a knob was left in the driver under which the new
capabilities will take effect only under specific admin directive, of setting
the enable_64b_cqe_eqe module param, whose fault value is false.
Signed-off-by: Or Gerlitz
--
On 23/10/2012 16:32, Klaus Wacker wrote:
we are implementing a Linux/RDMA solution based on Mellanox/RoCE. Our memory registration is done
via "ib_get_dma_mr()". During a problem follow-up someone asked us about the
"no-snoop" flag and how it is set during memory registration in Linux.
Can yo
Hi Sean,
FYI -- the below warnings seen while building librdmacm 1.0.16 with gcc
4.4.6 through rpmbuild
Or.
make[1]: Entering directory `/root/rpmbuild/BUILD/librdmacm-1.0.16'
CC src_librdmacm_la-cma.lo
CC src_librdmacm_la-addrinfo.lo
CC src_librdmacm_la-acm.lo
CC src
On Thu, Sep 20, 2012 at 10:30 PM, Or Gerlitz wrote:
> Roland,
> This batch of libmlx4 patch contains the patch to support raw
> packet QP, two patches from Sagi that relate to resource
> limitations, and few simple fixes/cleanups from Dotan.
> The first three were submitted pretty
On 19/10/2012 23:58, Kleber Sacilotto de Souza wrote:
During PCI error recovery, the calls to wait_for_completion() in the
infiniband core path can hang waiting for some tasks that will never
complete, since the hardware is nonfunctional.
INFO: task eehd:16029 blocked for more than 120 seconds.
On Mon, Nov 5, 2012 at 7:21 PM, Hefty, Sean wrote:
>
> > > drivers/infiniband/core/ucm.c |2 +-
> > > drivers/infiniband/core/ucma.c|2 +-
> >
> > On these files, as far as I understand this code from quick looking, I'm
> > not sure on what
> > exactly the completion obje
> On Mon, Nov 5, 2012 at 9:54 PM, Kleber Sacilotto de Souza
> wrote:
>>
>>
>> The driver is not returning the completions because during EEH (Extended
>> Error Handling) recovery on powerpc systems the PCI slot is frozen, and
>> we are not going to receive any interrupt from the adapter until we
>
On Tue, Nov 6, 2012 at 11:58 AM, Kleber Sacilotto de Souza
wrote:
> During my tests I've seen the wait_for_completion() call hanging on
> different parts of the code, but not on ucm/ucma. So would it be OK to
> change the other calls and leave the ucm/ucma as it is?
ucm/ucma you should remove in
On Tue, Nov 6, 2012 at 6:44 PM, Or Gerlitz wrote:
> the other parts are OK to me.
I wanted to say that you stepped on real problem and provided a
solution, Roland's wondering is ofcourse
correct, how do we avoid use after free in very slow non pci hotunplug cases?!
--
To unsubscribe f
Bart Van Assche wrote:
> This patch series makes the ib_srp driver better suited for use in a H.A.
> setup because:
> - multipathd is notified faster about transport layer failures.
> - Transport layer failures reliably result in SCSI host removal.
> - Switchover can be triggered explicitly by de
Bart Van Assche wrote:
> This patch series makes the ib_srp driver better suited for use in a H.A.
> setup because:
> - multipathd is notified faster about transport layer failures.
> - Transport layer failures reliably result in SCSI host removal.
> - Switchover can be triggered explicitly by de
On Wed, May 30, 2012 at 7:06 AM, David Dillow wrote:
> On Tue, 2012-05-29 at 17:07 -0400, Karandeep Chahal wrote:
>> Subject: [PATCH] Infiniband srp fast failover patch.
>
> This conflicts with Bart's patches to improve failover; it will be much
> better to use his approach to block the target rat
Bart Van Assche wrote:
> On 11/12/12 23:40, Or Gerlitz wrote:
> This patch is not an essential part of this patch series. All it does
> is to trigger failover more quickly if a port down event has been
> received. Without this patch, if an IB cable has been disconnected long
> eno
Bart Van Assche wrote:
> On 11/12/12 23:36, Or Gerlitz wrote:
> This patch series reduces path failover time significantly. Instead of
> having to wait until the SCSI error handler has finished recovery,
When a SCSI device is selected by mpath and used as a path, aren't failed
com
On Tue, Nov 13, 2012 at 1:46 AM, Karandeep Chahal wrote:
> In my experience with ib-srp, I am not sure if there is such a thing as link
> going down for a short time. When the link goes down, no matter how short
> the duration (< 1s), IO fails over, this failover takes very long time
> currently.
On Tue, Nov 13, 2012 at 11:20 PM, Bart Van Assche wrote:
> On 11/13/12 21:54, Or Gerlitz wrote:
>>
>> Bart Van Assche wrote:
>> [ ... ]
>>
>> this gets way too much complicated, and just for patch which you said
>> "is not an essential part of t
On Tue, Nov 13, 2012 at 11:30 PM, Bart Van Assche wrote:
> These attributes have been added to the SRP transport layer. Since the
> ib_srp driver registers itself with the SRP transport layer the SRP
> transport layer creates these two attributes for the ib_srp driver. This is
> similar to how the
On Sun, Oct 21, 2012 at 4:59 PM, Or Gerlitz wrote:
[...]
Hi Roland,
You basically accepted V3 but said its going to be for 3.8, which is
coming soon, I see that for-next is freshly rebased to 3.7-rc6,
picking this?
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdm
On Mon, Nov 26, 2012 at 6:44 AM, David Dillow wrote:
> One may also pull this series from github:
> git pull git://github.com/dillow/srp-initiator.git ha-merge-v1
Hi Dave,
The kernel maintainers file specifies the following tree
git://git.kernel.org/pub/scm/linux/kernel/git/dad/srp-initi
On Fri, Nov 23, 2012 at 2:10 PM, Bart Van Assche wrote:
> Apparently unloading the ib_ipoib kernel module triggers a circular locking
> dependency complaint. Has anyone already been looking into this ?
Yes, I see that this happens here e.g when doing hot-unplug to the
underlying HW driver, seems
Hi Roland,
I see that you picked this, good! not sure how it happened but Eli's
signature was missing from V3 which you took, can you somehow add this
there?!
Signed-off-by: Eli Cohen
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@v
On 27/11/2012 06:04, David Dillow wrote:
We can push it through James's tree if need be, but Bart's code is
pretty self-contained, and going through the SCSI tree will introduce
merge dependencies. It'd be much easier to push it all through the
RDMA tree
Yep, this makes sense to me even witho
On 27/11/2012 02:47, Stephen Rothwell wrote:
Hi all,
Today's linux-next merge of the net-next tree got a conflict in
drivers/net/ethernet/mellanox/mlx4/en_rx.c between commit 08ff32352d6f
("mlx4: 64-byte CQE/EQE support") from the infiniband tree and commit
f1d29a3fa68b ("mlx4_en: Remove remnant
oiding
the warning and the depencency.
Signed-off-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
drivers/infiniband/hw/mlx4/cm.c |4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
index 80079e5..d
From: Jack Morgenstein
The slave_state_lock spinlock is used both in the interrupt context and
in a process context, hence irq locking must be used. Found by lockdep.
Signed-off-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
drivers/net/ethernet/mellanox/mlx4/cmd.c |9
Hi Roland,
Pretty late indeed, but here are two SRIOV locking related fixes for issues
spotted by lockdep. Basically, as we're late in the cycle, I guess there are
two options here, either push to 3.8 and later to -stable once 3.7 is released,
or push to 3.7 this week. This way or another, would
On Tue, Nov 27, 2012 at 6:34 PM, Bart Van Assche wrote:
> Thanks Dave for doing all this work. A reworked and retested patch series
> that should address all comments that have been posted so far can be found
> here: http://github.com/bvanassche/linux/srp-ha. I can repost the entire
> patch series
On 28/11/2012 11:46, Roland Dreier wrote:
On Tue, Nov 27, 2012 at 12:35 PM, Christoph Lameter wrote:
Roland, we need the raw eth qp patches in the git trees for the libraries.
Could you please bring the trees up to date so that the userspace raw eth
support is in sync with the kernel?
Do you
On Fri, Aug 24, 2012 at 8:35 PM, Hefty, Sean wrote:
>> I don't think those mmap()s should be an issue with fork they are
>> mapping adapter PCI space into userspace, but it should work across fork.
> makes sense
> Do you have any ideas on ways to identify what in the initialization paths
> m
rsocket.c wasn't passing compilation as of missing definition for the
container_of macro, fix it. Reported-by: Eyal Salamon
Signed-off-by: Or Gerlitz
---
src/cma.c |5 -
src/cma.h |6 ++
src/rsocket.c |1 +
3 files changed, 7 insertions(+), 5 deletions(-)
On 29/11/2012 18:27, Hefty, Sean wrote:
Fork support in rsockets is available now, and netperf should work.
Sean,
Trying the latest librdamcm git (+ the patch I sent to have it build
OK), we weren't able to run netperf
here's the server output
starting netserver with host '192.168.20.126'
On 29/11/2012 18:27, Hefty, Sean wrote:
Fork support in rsockets is available now, and netperf should work.
Sean,
Trying the latest librdamcm git (+ the patch I sent to have it build
OK), we weren't able to run netperf
here's the server output
starting netserver with host '192.168.20.126'
On Mon, Dec 3, 2012 at 9:35 PM, Hefty, Sean wrote:
> I'm using a locally modified version of verbs, so I don't end up seeing this
> error.
could this be related to different results while running netperf .w
fork in our setup vs. yours?
Or.
--
To unsubscribe from this list: send the line "unsubs
On Mon, Dec 3, 2012 at 9:48 PM, Hefty, Sean wrote:
>
> According to my notes, I ran netperf with fork support using this:
yep, works just fine with this directive - we got 46Gbs, impressive!
Or.
> server
> --
> export RDMAV_FORK_SAFE=1
> LD_PRELOAD=/usr/local/lib/rsocket/librspreload.so ne
On Fri, Nov 30, 2012 at 4:21 AM, David Dillow wrote:
[...]
> Modulo a few style issues (braces around one line if branches, etc.) and
> having three state variables vs one, I can live with everything up to
> aabfa852acd27962 at git://github.com/bvanassche/linux.git#srp-ha. Those
> two are small th
On Wed, Dec 5, 2012 at 8:50 PM, Bart Van Assche wrote:
[...]
> The only way to make I/O work reliably if a failure can occur at the
> transport layer is to use multipathd on top of ib_srp. If a connection fails
> for some reason, then the SRP SCSI host will be removed after the SCSI error
> handle
On 06/12/2012 16:10, Bart Van Assche wrote:
On 12/05/12 22:32, Or Gerlitz wrote:
On Wed, Dec 5, 2012 at 8:50 PM, Bart Van Assche
wrote:
[...]
The only way to make I/O work reliably if a failure can occur at the
transport layer is to use multipathd on top of ib_srp. If a
connection fails
for
On 06/12/2012 17:04, Bart Van Assche wrote:
On 12/06/12 15:27, Or Gerlitz wrote:
The core problem here seems to be that scsi_remove_host simply never
ends.
Hello Or,
The later patches in the srp-ha patch series avoided such behavior by
checking whether the connection between SRP initiator
river will use
the value of log_num_mgm_entry_size and B0 steering.
Signed-off-by: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
drivers/net/ethernet/mellanox/mlx4/main.c | 52
drivers/net/ethernet/mellanox/mlx4/mcg.c |7 +---
drivers/net/ethernet/mellanox/mlx4/m
From: Hadar Hen Zion
The error flow of the flow steering wrapper had a typo which caused the
wrong firmware command to be called, fix it.
Signed-off-by: Hadar Hen Zion
Signed-off-by: Or Gerlitz
---
.../net/ethernet/mellanox/mlx4/resource_tracker.c |2 +-
1 files changed, 1 insertions
ation from
the resource tracker point of view.
Signed-off-by: Hadar Hen Zion
Signed-off-by: Or Gerlitz
---
.../net/ethernet/mellanox/mlx4/resource_tracker.c | 26 +++
1 files changed, 20 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/mellanox
Hi Roland,
The patch set fixes few issues we've identified and fixed in the
mlx4_core driver around SRIOV and device managed flow steering.
This includes adding QPN enforcement in the resource tracker for
flow-steering rules set by VFs, and a fix to wrong error flow in
the very same area in pa
: Jack Morgenstein
Signed-off-by: Or Gerlitz
---
drivers/net/ethernet/mellanox/mlx4/fw.c | 15 +-
drivers/net/ethernet/mellanox/mlx4/fw.h |1 +
drivers/net/ethernet/mellanox/mlx4/main.c | 71 -
3 files changed, 62 insertions(+), 25 deletions(-)
diff --git a
On Thu, Dec 6, 2012 at 4:27 PM, Or Gerlitz wrote:
[...]
> looking on the current locks in the system, we see that this kworker task
> holds four locks, but none of them seems to be mutually held by another task,
That was ofcourse a wrong assertion, as a lock can't be mutually held
b
On Thu, Dec 6, 2012 at 7:11 PM, Or Gerlitz wrote:
>> The patch set fixes few issues we've identified and fixed in the
>> mlx4_core driver around SRIOV and device managed flow steering.
>> This includes adding QPN enforcement in the resource tracker for
>> flow-ste
On Thu, Dec 6, 2012 at 7:11 PM, Or Gerlitz wrote:
> From: Jack Morgenstein
>
> Separate flow steering capability detection from the decision to activate.
>
> For the master (and for native), detect flow steering capability in
> mlx4_dev_cap, but activate the appropriate stee
On Mon, Dec 10, 2012 at 3:14 PM, Klaus Wacker wrote:
> 2. The switch (BNT virt. fabric 10GB) has configured all ports as "vlan
> tagging=enabled" (trunk-mode)
[...]
> Failed status transport retry counter exceeded (12) for wr_id 2
[...]
> Now my questions are: Any ideas what causes the "ibv_rc_pin
On Tue, Dec 11, 2012 at 7:59 AM, Roland Dreier wrote:
> First batch of InfiniBand/RDMA changes for the 3.8 merge window:
> - A good chunk of Bart Van Assche's SRP fixes
> - UAPI disintegration from David Howells
> - mlx4 support for "64-byte CQE" hardware feature
On 19/12/2012 16:20, Bart Van Assche wrote:
This patch series avoids that SCSI error handling triggers an endless
loop and also restores reporting of QP errors in the kernel log.
Changes between v3 and v2:
- As proposed by Dave, added a patch that prevents sending of a task
management functio
On 20/12/2012 15:19, Bart Van Assche wrote:
On 12/20/12 13:38, Or Gerlitz wrote:
I think few days ago you had a patch on your tree named "Save and
restore host_scribble during error handling", is it possible we need
this here for happy removal of the scsi host?
No. Host removal
On 20/12/2012 17:10, David Dillow wrote:
Works for me, ship it!
Dave, did you gave the patches a try?
Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info
501 - 600 of 1832 matches
Mail list logo