kernel-packages team mailing list archive

[bugproxy <bugproxy@xxxxxxxxxx>]

------- Comment From gpiccoli@xxxxxxxxxx 2016-01-21 10:10 EDT-------
The commit's author is not too "friendly" to the idea of reverting it. He keeps asking for more evidence that it crashes the system. We can easily reproduce it on our side, and by reverting the commit in Ubuntu source, we can build a kernel that does not crash anymore. The commit's author didn't provide any fix for this, things are in the same page as before.

So, what do you think about reverting this in Ubuntu even if in mainline
it's not reverted yet? The iSCSI maintainer is friendly to the idea of
reverting too.

In IBM's internal distro we already reverted. My recommendation is to
revert in Ubuntu, if you agree.

Thanks,

Guilherme

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1517142

Title:
  ubuntu guest with 10G n/w and Texan iSCSI crashes during FIO

Status in linux package in Ubuntu:
  Triaged

Bug description:
  Issues were found in iSCSI tests with hardware remote targets.
  Specifically, kernel crash happens due to dereferencing a null pointer
  (sc->device->lun at libiscsi.c:369, with sc==NULL). During the crash,
  lots of messages regarding lists invalid accesses are showed in kernel
  log.

  The commit 659743b02c41 ("[SCSI] libiscsi: Reduce locking contention
  in fast path") appears to be the cause.

  Reverting the commit solve the issue, at least until we can discuss and
  find the exact problem (and its solution) in the commit
  659743b02c41 ("[SCSI] libiscsi: Reduce locking contention in fast path").

  A test kernel was patched to revert the offend commit - Prashantha is
  running tests to check if the problem is solved.

  With the patched kernel, I am unable to recreate the crash. The patch appears
  to be working.

  A discussion is ongoing in linux-scsi mailing list, to revert the
  patch upstream (look the following link).

  http://marc.info/?l=linux-scsi&m=144730474819919

  Another quick discussion, started by me, it's on open-iscsi mailing
  list, on Google Groups:

  https://groups.google.com/forum/#!topic/open-iscsi/0S5fEM_Aafk

  The iscsi maintainer wants to revert, but patch co-author wants more
  study before reverting. Prashantha is performing some performance
  analysis to check the impact of the patch on iscsi performance.

  Mirroring to Launchpad for Canonical's awareness. Once the discussion
  settles on the final solution, a patch or link to the upstream commit
  will be provided for Canonical to review for acceptance in the 14.04
  LTS kernel and SRU.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1517142/+subscriptions