Hi all. I propose the following plan to shutdown the general list:
1) unsubscribe all current subscribers
2) set the list to discard any incoming messages with an auto-discard
message that points you to linux-r...@vger.kernel.org
Sounds like a perfect plan to me... sorry for not
It's probably just me but I'm not ready yet. I haven't been able to
post a patch to linux-rdma yet :-(
What is going wrong when you try?
- R.
___
general mailing list
general@lists.openfabrics.org
If an SRP target processes SRP I/O slow enough, the SRP initiator locks up.
INFO: task fio:6389 blocked for more than 120 seconds.
echo 0 /proc/sys/kernel/hung_task_timeout_secs disables this message.
fio D 0 6389 6388 0x
880071dc5bd8
thanks, applied.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Thanks, applied.
I almost missed this one because you sent it to general@ and not
linux-r...@vger.kernel.org -- and I was using the fancy
patchwork.kernel.org patch tracking stuff to see what I had to apply.
So sending things to linux-rdma@ helps you too!
Thanks,
Roland
thanks, applied.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
: Don't turn on carrier for a non-active port
Roland Dreier (2):
IB/mad: Fix lock-lock-timer deadlock in RMPP code
Merge branches 'ipoib', 'mad', 'mlx4', 'mthca' and 'nes' into for-linus
drivers/infiniband/core/mad_rmpp.c | 17 +
drivers/infiniband/hw/mthca
Reviewed-by: Sean Hefty sean.he...@intel.com
Thanks, I applied this.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
+#if defined(cache_line_size)
Why the #if here? Do we just need to include linux/cache.h explicitly
to make sure we get the define?
+*((u8 *) mailbox-buf + INIT_HCA_CACHELINE_SZ_OFFSET) =
+order_base_2(cache_line_size() / 16) 5;
Trivial but I think it's safe to assume a
The locking is needed to protect against items being removed from rmpp_list
in
recv_timeout_handler() and recv_cleanup_handler(). No new items should be
added
to the rmpp_list when ib_cancel_rmpp_recvs() is running (or there's a
separate
bug).
OK so how about something like
I agree with you on both comments. Would you like me to resend or will
you make the necessary changes?
please resend, thanks.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To
I may miss this but I don't see how ipoib_ib_dev_down() is called with rtnl
held.
It's called from ipoib_stop(), and .ndo_stop is called with rtnl held.
Anyway, the new patch doesn't use delayed work.
great
___
general mailing list
Anton Blanchard suggested a while back that this might be integrated
with perf-counters, since perf-counters already does mmap() tracking and
also provides events through an mmap()'ed buffer.
Has anybody looked into this?
I didn't see the original suggestion. Certainly hooking in to
So getting those events in the kernel is no problem -- we have the MMU
notifier hooks that tell us exactly what we need to know. The issue is
purely the way userspace registers interest in address ranges, and how
to kernel returns the events.
For perf counters it seems that one
+if (ib_query_port(priv-ca, priv-port, attr) ||
+attr.state != IB_PORT_ACTIVE) {
+ipoib_dbg(priv, wait with carrier until IB port is active\n);
+if (test_bit(IPOIB_FLAG_OPER_UP, priv-flags))
+queue_delayed_work(ipoib_workqueue,
Hmm, or are you saying you can only get 1 event per registered range and
allocate the thing on registration? That'd need some registration limit
to avoid DoS scenarios.
Yes, that's what I do. You're right, I should add a limit... although
their are lots of ways for userspace to consume
Hmm, or are you saying you can only get 1 event per registered range and
allocate the thing on registration? That'd need some registration limit
to avoid DoS scenarios.
Yes, that's what I do. You're right, I should add a limit... although
their are lots of ways for
And by the way, this current patch has a deadlock I think:
@@ -724,6 +724,8 @@ int ipoib_ib_dev_down(struct net_device *dev, int flush)
ipoib_dbg(priv, downing ib_dev\n);
clear_bit(IPOIB_FLAG_OPER_UP, priv-flags);
+cancel_delayed_work(priv-carrier_on_task);
I noticed that there are seven SRP patches (bug fixes) present in OFED
1.4.1 that are not present in mainstream Linux kernels up to and
including version 2.6.30. Do you know whether it is documented
anywhere which process is followed for merging such patches in the
mainstream Linux
several review iterations on lkml and was in
-mm and -next for quite a few weeks. Andrew is OK with merging it (I
think -- Andrew please correct me if I misunderstood you).
Roland Dreier (1):
ummunotify: Userspace support for MMU notifications
Documentation/Makefile
- I guess you have your MPI implementaion w/ ummunotify, right?
Yes, Jeff Squyres (cc'ed) has an Open MPI prototype (mercurial tree at
http://bitbucket.org/jsquyres/ummunot/).
- I guess you have test sevaral pattern, right?
if so, can we see your test result?
Open MPI has a pretty
I don't remember seeing discussion of this on lkml. Yes it is in
-next...
eg http://lkml.org/lkml/2009/7/31/197 and followups, or search for v2
and earlier patches.
Basically it allows app to 'trace itself'? ...with interesting mmap()
interface, exporting int to userspace, hoping it
Does ipoib map IPv6 multicast addresses to MGIDs directly?
No, the IPv6 network stack is responsible for doing the mapping and
passing the HW (ie IPoIB) address in... cf ipv6_ib_mc_map() in
include/net/if_inet6.h
___
general mailing list
Case m is missing a handy break statement. Setting mtu also sets rx_depth.
Can workaround by following -m by -r which resets rx_depth.
Yes, looks that way. Care to send a patch?
- R.
___
general mailing list
general@lists.openfabrics.org
OK, let's try this out and see how it goes... applied, thanks.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
thanks, applied both this and mlx4 version. seems that libcxgb3,
libipathverbs and libnes would want similar treatment?
- R.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To
your mailer seems to have mangled whitespace but I applied it by hand, thanks.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
Can I this version already solved fork() + COW issue? if so, could you
please explain what happen at fork. Obviously RDMA point to either parent
or child page, not both. but Corrent COW rule is, first touch process
get copyed page and other process still own original page. I think it's
My understanding of the code is that fork will end-up calling
copy_page_range() on all VMA, and copy_page_range() calls
mmu_notifier_invalidate_range_start() if is_cow_mapping() is true,
which should be the case here. So you should get some invalidate events
on fork.
Yes, I agree
So.. What is the problem with fork? The semantics of what should
happen seem natural enough to me, the PD doesn't get copied to the
child, so the MR stays with the parent. COW events on the pinned
region must be resolved so that the physical page stays with the
process that has pinned
Anybody has seen lro doesn't work for port2 but works for port1 before?
We found this issue on RHEL5.
I haven't seen this. But I haven't really tried.
Are you talking about mlx4_en or ipoib with mlx4_ib? What do you mean
by lro not working -- the interface doesn't work with lro
):
IB: Use printk_once() for driver versions
Roel Kluin (2):
IB/ipath: strncpy() doesn't always NUL-terminate
RDMA/amso1100: Check kmalloc() result in c2_register_device()
Roland Dreier (15):
IPoIB: Remove unused rdma/ib_cache.h includes
IPoIB: Drop priv-lock before calling
review iterations on lkml and was in
-mm and -next for quite a few weeks. Andrew is OK with merging it (I
think -- Andrew please correct me if I misunderstood you).
Roland Dreier (1):
ummunotify: Userspace support for MMU notifications
Documentation/Makefile |3
thanks, applied both patches.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Holding agent-lock across cancel_delayed_work() (which does
del_timer_sync()) in ib_cancel_rmpp_recvs() leads to lockdep reports of
possible lock-timer deadlocks if a consumer ever does something that
connects agent-lock to a lock taken in IRQ context (cf
The locking is needed to protect against items being removed from rmpp_list
in
recv_timeout_handler() and recv_cleanup_handler(). No new items should be
added
to the rmpp_list when ib_cancel_rmpp_recvs() is running (or there's a
separate
bug).
Ah, I see.
That's trickier I
The above issue does not occur with the for-next branch of the
infiniband git tree, but does occur with 2.6.31-rc9 + aforementioned
patches.
As far as I can see commit 721d67cdca5b7642b380ca0584de8dceecf6102f
Update: patch 721d67cdca5b7642b380ca0584de8dceecf6102f does not apply
cleanly to 2.6.31-rc9, so I have been using a slightly modified
version of this patch
(http://bugzilla.kernel.org/attachment.cgi?id=22624).
I have retested the 2.6.31-rc9 kernel with the following patches applied
.
Finally, I'll plan to merge the following for 2.6.32:
MAINTAINERS: InfiniBand/RDMA mailing list transition to vger
InfiniBand/RDMA development discussion is moving from
general@lists.openfabrics.org to linux-r...@vger.kernel.org.
Signed-off-by: Roland Dreier rola...@cisco.com
---
MAINTAINERS
With 2.6.31-rc9 + patch 4e49627b9bc29a14b393c480e8c979e3bc922ef7 + the
patch you posted at the start of this thread the following lockdep
complaint was triggered on the SRP initiator system during SRP login:
==
[ INFO: HARDIRQ-safe
What about RDMAoE?
Patches were sent few weeks ago and it seems you ignore them.
Sorry, I should have mentioned that.
Yes, I have been ignoring the patches -- I want to get through XRC
first, and also I would like to see a real spec for IBoE (that resolves
issues like multicast interaction
thanks for updating, applied.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
applied -- would be nice to have a way to do this automatically instead
of yet another tunable to sysadmins to worry about, but oh well.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
I understand that ops.poll_cq is actually ibv_cmd_poll_cq(), right ?
No, not for most devices. Look at libmthca, etc to see what the poll_cq
method is set to.
Do you mean during ibv_poll_cq() call, there is no system call involved
?
Right, for most devices poll_cq can be done
Slusarz (1):
IB: Use printk_once() for driver versions
Roel Kluin (2):
IB/ipath: strncpy() doesn't always NUL-terminate
RDMA/amso1100: Check kmalloc() result in c2_register_device()
Roland Dreier (13):
IPoIB: Remove unused rdma/ib_cache.h includes
IB: Use DEFINE_SPINLOCK
The idea seems sound but checkpatch.pl gives 6 errors for this small
patch! Also:
+static int check_mcast(const u8 *addr,unsigned int addrlen,
+ const u8 *broadcast)
name of the function could make it clearer what the expected return
value is ... eg mcast_addr_is_valid()
applied, thanks
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Illegal multicast address can be handed for IPoIB from userspace. For example
the command ip maddr add 33:33:00:00:00:01 dev ib0 injects an illegal
muticast
address to IPoIB that will start a join task for this address. However,
whenever
an illegal multicast address is passed to
n linux-r...@vger.kernel.org
It's there, ready and waiting, should you choose to use it :-)
Thanks again... how do we get archive links added -- is it manually?
Right now we have http://www.spinics.net/lists/linux-rdma/
http://www.mail-archive.com/linux-r...@vger.kernel.org/ should be up
I tend to prefer patches that compile :) --
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c
b/drivers/infiniband/hw/amso1100/c2_provider.c
index 3f2ee64..ad723bd 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -874,7
Applied, thanks for redoing this.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -851,6 +851,10 @@ int c2_register_device(struct c2_dev *dev)
dev-ibdev.post_recv = c2_post_receive;
dev-ibdev.iwcm = kmalloc(sizeof(*dev-ibdev.iwcm), GFP_KERNEL);
+
this seems reasonable to me, applied, thanks.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
I would like know the QDR Infinibad cards will support to back to
back connectivity or not ie with out IB swicth to enable the IB
communication between the two machines .
Yes, any IB port should be able to connect to any other IB port. You do
need a subnet manager (SM) on every IB
Given that you seem to like the rest of the code and Jason hasn't spoken
up yet, I think we can have Roland merge this patch. Roland, what do you
think?
I don't see any problem with the idea and this does sound like a step
forward, so I am planning on merging this (pending review).
It seems we only had positive responses to moving from general@ to a new
linux-r...@vger.kernel.org list, so I'll work on a transition plan.
For now, please continue to use gene...@lists.openfabrics.org. However,
you may want to subscribe to the vger list to be ready for the
transition; for
checkpatch output:
WARNING: suspect code indent for conditional statements (8, 12)
#88: FILE: drivers/infiniband/hw/mlx4/main.c:345:
+ if (!dev-ib_active)
+ return ERR_PTR(-EAGAIN);
ERROR: code indent should use tabs where possible
#107: FILE:
thanks, applied (and thanks for the detailed changelog, that really
makes things easier)
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
Thanks, at long last I applied both the mthca and mlx4 versions of these
patches (with some cleanups).
- R.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please
The one question I would have would be do the kernel.org people want to
see all of the traffic that we currently have on the open fabrics
general list for all of the user-space components?
I don't believe there's any problem with that. vger already hosts quite
a few lists that are
linux-r...@vger.kernel.org
It's there, ready and waiting, should you choose to use it :-)
Thanks!
- R.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please
+#define DT_Mdep_yield pthread_yield
Be aware that on Linux I believe this turns into sched_yield(), which
basically means put me at the end of the thread list ie wait for
everyone else to get a turn ie possibly huge latency...
___
general mailing
Is sleep(0) a preferred way to go?
I think the best solution is not coding spin-loops. Not sure what
sleep(0) ends up turning into, but if you can tell the system I'm
waiting for this object, wake me up when it's available then that
should produce the best behavior.
- R.
thanks, applied.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
this with possible_cpus=32 and it still works for
me -- if you get a chance on your Dell systems that would be helpful too)
commit 58cafda0c3010fc2cdb0fc9be3fbd6d09640dd6f
Author: Roland Dreier rola...@cisco.com
Date: Thu Aug 20 14:26:21 2009 -0700
mlx4_core: Allocate and map sufficient ICM memory for EQ
Lately, I've had a few emails that I thought would have been of interest
to both lkml and also to gene...@lists.openfabrics.org. I've held back
on cross-posting them because I know that general@ is subscribers-only,
and the bounce messages are quite annoying to replies coming from lkml.
The
The attached modification to ibv_rc_pingpong makes it never complete.
Forking seems to do something bad. I noticed that forking right after
ibv_post_recv() cancels the posted WR (as if it was never issued):
sender keeps retrying.
Is it expected behavior or a bug?
Yes, forking is
I am getting this error on a verb send operation and I can't figure
out what could be the cause; I searched for all instances of this
error in the IB code and while I found 4, none was illuminating.
IBV_WC_REM_OP_ERR corresponds to Remote Operation Error, which the IB
spec describes as:
In an attempt to get unexpected DISCONNECT notifications during
ib communication, I'm trying to use 'select()' on the cm channel's file
descriptor, testing it for readability. I've found that this works some of
the time, but not all of the time.
What happens when it doesn't work?
How about this approach? Basically it just open-codes delayed work by
splitting the timer and the work struct, and switches to mod_timer()
instead of del_timer() + add_timer(). It passes very light testing here
(basically I started ipoib and nothing blew up).
---
drivers/infiniband/core/mad.c
Changed module paramater permissions to 0644
Does it really work if someone changes the module parameter at runtime
after the module is loaded?
- R.
___
general mailing list
general@lists.openfabrics.org
this is a continuation of thread:
http://lists.openfabrics.org/pipermail/general/2009-July/060668.html
Thanks for the pointer... it lets me reload my context. I see you
didn't answer the question about mthca -- does it suffer from this
problem as well?
- R.
thanks, applied.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Even if it is really unlikely that this lock cycle would cause a
deadlock, it would be great if this lock cycle could be removed. I'm
not the only developer of kernel modules who runs tests with lockdep
enabled, and it is unpractical to analyze long logfiles full of known
lock cycles to
Has anyone ever encountered a message like the one below ? This message was
generated while booting a 2.6.30.4 kernel with CONFIG_DMA_API_DEBUG=y and
before any out-of-tree kernel modules were loaded.
[ cut here ]
WARNING: at lib/dma-debug.c:635
The lockdep report I obtained this morning with a 2.6.30.4 kernel and
the two patches applied has been attached to the kernel bugzilla
entry. This lockdep report was generated while testing the SRPT target
software. I have double checked that the SRPT target implementation
does not hold
drivers/infiniband/hw/cxgb3/iwch.c |4 +---
drivers/infiniband/hw/mlx4/main.c |6 +-
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -540,15 +540,11 @@ static struct device_attribute
*mlx4_class_attributes[] = {
static void
Thanks, applied with a few cleanups:
ilog2(roundup_pow_of_two()) - order_base_2()
xxx * (1 yy) - xxx yy
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please
thanks for respinning, got em both.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
A fix like the one below ?
I think this gets us part of the way, but not quite.
--- linux-2.6.30.4/drivers/infiniband/ulp/srp/ib_srp-orig.c 2009-08-03
12:13:11.0 +0200
+++ linux-2.6.30.4/drivers/infiniband/ulp/srp/ib_srp.c 2009-08-07
10:23:27.0 +0200
@@
After having enabled CONFIG_DEBUG_KOBJECT=y in the kernel config I
noticed that messages appeared in the kernel log about the IB modules
missing a kobject release() method. This happens both with a vanilla
2.6.30.4 kernel and with a 2.6.27.29 kernel + OFED 1.4.1. Has anyone
noticed
What about multicast though? Switches are going to have trouble with
group membership lists for non IP packets.. Even just sending a ICMPv6
packet (with an IPv6 ethertype) isn't guaranteed to fix it.
In this patch set, all multicast packets use the broadcast mac. We
will address
Specifically scmnd-host_scribble can just be Zero.
I see at last, thanks!
The issue is that SRP is using host_scribble to hold an index, and index
0 is valid for us.
I guess the fix is a bit complex, but basically we should use
host_scribble to point to the request, and if we don't find a
Next I started (/etc/init.d/openibd start; /etc/init.d/opensmd start)
and then stopped (/etc/init.d/opensmd stop; /etc/init.d/openibd stop)
the IB subsystem. The broken message was logged during module unload
only, not during module load.
Oh I see... yes I get it on unload too, for any
Are you sure that this indicates a shortcoming in the kobject
debugging code ? The most recent messages related to the message does
not have a release() function, it is broken and must be fixed I could
find on the LKML date from July 16, 2009
(http://lkml.org/lkml/2009/7/16/306 and
thanks, applied all 10 pending patches.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Now I'm confused about this patch for another reason:
@@ -1429,6 +1431,8 @@ static int srp_reset_device(struct scsi_
return FAILED;
if (req-tsk_status)
return FAILED;
+if (!req-scmnd-device)
+return FAILED;
The NULL pointer dereference happens when srp_reset_device() calls
srp_send_tsk_mgmt(target, req, SRP_TSK_LUN_RESET) with
req-scmnd-device == NULL. When the sg_reset command issues an
SG_SCSI_RESET ioctl, scsi_reset_provider() is invoked and allocates an
scmnd structure and sets
- Endpoint flags now need to be set via atomic bitops because they can
be set on both the iw_cxgb3 workqueue thread and user disconnect threads.
+if (!test_bit(ABORT_REQ_IN_PROGRESS, ep-com.flags)) {
+set_bit(ABORT_REQ_IN_PROGRESS, ep-com.flags);
for atomicity, should
Reported-by: Bart Van Assche bart.vanass...@gmail.com
Signed-off-by: Roland Dreier rola...@cisco.com
---
drivers/infiniband/ulp/ipoib/ipoib_main.c |7 ++-
drivers/infiniband/ulp/ipoib/ipoib_multicast.c |2 ++
2 files changed, 8 insertions(+), 1 deletions(-)
diff --git a/drivers
thanks, applied
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
this one won't apply without 2/5, so I'll wait for you to resend both patches...
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit
thanks, applied this and 5/5.
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
thanks, applied
___
general mailing list
general@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
--- a/drivers/infiniband/hw/ipath/ipath_mad.c
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c
@@ -60,7 +60,7 @@ static int recv_subn_get_nodedescription(struct ib_smp
*smp,
if (smp-attr_mod)
smp-status |= IB_SMP_INVALID_FIELD;
-strncpy(smp-data,
rdma_destroy_id and rdma_leave_multicast call ib_sa_free_multicast. This
call
will block until the join callback completes or is canceled. Can you
describe
the race with cma_ib_mc_handler in more detail?
Also, cma_leave_mc_groups is only called from rdma_destroy_id. Locking
The reason why I need such small rates is because I interface the Infiniband
HCA to an FPGA via an Infiniband physical link. Imagine the FPGA as a
simple repeater that simply forwards the infiniband signals to the Target
HCA. The FPGA cannot handle such a high data rate and neither do I
Maybe it's just a loose connection but yet, it seems to me that
operations on id_priv-mc_list should be protected. Should I send a
different patch?
seems ... should be is very weak justification for locking. What
should they be protected from?
- R.
An update: apparently it is possible to trigger scmnd-device == NULL even
without triggering a prior IB CM disconnect. The following shell commands
are sufficient to trigger the WARN_ON statement in the patch below:
rmmod ib_srp
modprobe ib_srp
ibsrpdm -c | while read target_info; do
By the way, Vladislav Bolkhovitin was so kind to inform me that this
issue is not specific to the SRP initiator. For more information, see
also http://thread.gmane.org/gmane.linux.scsi/26166.
I'm not sure I follow this exactly -- the idea is that sg_reset
generates SCSI commands that are
1 - 100 of 2272 matches
Mail list logo