[ofa-general] Re: [PATCH/RFC] IB/mad: Fix possible deadlock (cancel_delayed_work inside spinlock)

[Bart Van Assche]

On Sun, Aug 23, 2009 at 5:21 PM, Roland Dreier <rdreier at cisco.com> wrote:
>
>  > I'm now using the SRP initiator from a kernel compiled from the
>  > http://git.kernel.org/?p=linux/kernel/git/roland/infiniband.git
>  > repository, and the lockdep complaints also occur on this system. The
>  > system even deadlocks during boot about one out of two times. Do you
>  > already know when you will have the time to commit the
>  > locking-inversion fixes to the infiniband.git repository ?
>
> Everything I know of is already in my tree now.  And I just checked
> http://git.kernel.org/?p=linux/kernel/git/roland/infiniband.git;a=shortlog;h=for-next
> and I see both "IB/mad: Fix possible lock-lock-timer deadlock" and
> "IPoIB: Drop priv->lock before calling ipoib_send()" there.  Those are
> all the lockdep-related things I know of.
>
> I have a hard time imagining that either of those issues could cause a
> deadlock on half of boots either.  Are you sure the deadlock you see is
> related to one of those fixes?

After having switched from the master branch to the for-next branch  I
do now also see the patches mentioned above.

And apparently the phenomenon I observed during boot was not a
deadlock but some other strange phenomenon. See also
http://bugzilla.kernel.org/show_bug.cgi?id=14043 for the details.

Bart.