+++ This bug was initially created as a clone of Bug #456686 +++

Description of problem:
There is a missing memory barrier in the current aio_complete in the RHEL5
kernels causing a race between read_events/aio_complete causing the thread in
read_events to sleep indefinitely, hanging the application that is waiting on
I/O completion.

This was reported upstream by Quentin Barnes of Yahoo:

http://lkml.org/lkml/2008/3/12/207

Fix has been merged in 2.6.26:

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=6cb2a21049b8990df4576c5fce4d48d0206c22d5

And was also accepted for 2.6.24.y:

http://git.kernel.org/?p=linux/kernel/git/stable/linux-2.6.24.y.git;a=commit;h=0db49fc729eee503836ea12745b55f7f802d2abb

Version-Release number of selected component (if applicable):


How reproducible:
Unclear. Depends on the AIO application; Quentin reports seeing hangs virtually
100% of the time. Looking for a straightforward reproducer for this now and will
update with details when they are available.

Steps to Reproduce:
< to be filled >
  
Actual results:
Application hangs in read_events

Expected results:
No hang. AIO completes as normal.

Additional info: