On 3/17/21 10:03 AM, Waiman Long wrote:
On 3/17/21 9:31 AM, Peter Zijlstra wrote:
On Wed, Mar 17, 2021 at 02:12:41PM +0100, Peter Zijlstra wrote:
On Wed, Mar 17, 2021 at 12:38:21PM -0000, tip-bot2 for Waiman Long
wrote:
+ /*
+ * Treat as trylock for ww_mutex.
+ */
+ mutex_acquire_nest(&lock->dep_map, subclass, !!ww_ctx,
nest_lock, ip);
I'm confused... why isn't nest_lock working here?
For ww_mutex, we're supposed to have ctx->dep_map as a nest_lock, and
all lock acquisitions under a nest lock should be fine. Afaict the
above
is just plain wrong.
To clarify:
mutex_lock(&A); ww_mutex_lock(&B, ctx);
ww_mutex_lock(&B, ctx); mutex_lock(&A);
should still very much be a deadlock, but your 'fix' makes it not report
that.
Only order within the ww_ctx can be ignored, and that's exactly what
nest_lock should be doing.
I will take a deeper look into why that is the case.
From reading the source code, nest_lock check is done in
check_deadlock() so that it won't complain. However, nest_lock isn't
considered in check_noncircular() which causes the splat to come out.
Maybe we should add a check for nest_lock there. I will fiddle with the
code to see if it can address the issue.
Cheers,
Longman