IPoIB: Fix send lockup due to missed TX completion
authorMike Marciniszyn <mike.marciniszyn@intel.com>
Tue, 26 Feb 2013 15:46:27 +0000 (15:46 +0000)
committerBen Hutchings <ben@decadent.org.uk>
Wed, 10 Apr 2013 02:20:02 +0000 (03:20 +0100)
commit6273c03e47f9192594311b3d06601c695d2c7171
tree85df1bb9c8b6f6d0f246e7e6ede97072604059e8
parent9de8a34b82aea9f6f64460af7e86ab6bbc37014e
IPoIB: Fix send lockup due to missed TX completion

commit 1ee9e2aa7b31427303466776f455d43e5e3c9275 upstream.

Commit f0dc117abdfa ("IPoIB: Fix TX queue lockup with mixed UD/CM
traffic") attempts to solve an issue where unprocessed UD send
completions can deadlock the netdev.

The patch doesn't fully resolve the issue because if more than half
the tx_outstanding's were UD and all of the destinations are RC
reachable, arming the CQ doesn't solve the issue.

This patch uses the IB_CQ_REPORT_MISSED_EVENTS on the
ib_req_notify_cq().  If the rc is above 0, the UD send cq completion
callback is called directly to re-arm the send completion timer.

This issue is seen in very large parallel filesystem deployments
and the patch has been shown to correct the issue.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
drivers/infiniband/ulp/ipoib/ipoib_cm.c