Re: [PATCH 1/4] locking/ww_mutex: add ww_mutex_is_owned_by function v3

From: Christian KÃnig
Date: Tue Feb 20 2018 - 09:35:40 EST


Am 20.02.2018 um 14:57 schrieb Peter Zijlstra:
On Tue, Feb 20, 2018 at 02:26:55PM +0100, Christian KÃnig wrote:
+static inline bool ww_mutex_is_owned_by(struct ww_mutex *lock,
+ struct ww_acquire_ctx *ctx)
+{
+ if (ctx)
+ return likely(READ_ONCE(lock->ctx) == ctx);
+ else
+ return likely(__mutex_owner(&lock->base) == current);
+}
Much better than the previous version. If you want to bike-shed, you can
leave out the 'else' and unindent the last line.
Thanks for the suggestion, going to do this.
You might also want likely(ctx), since ww_mutex without ctx is
a-typical I would think.

I do worry about potential users of .ctx = NULL, though. It makes it far
too easy to do recursive locking, which is something we should strongly
discourage.
Well, one of the addressed use cases is indeed checking for recursive
locking. But recursive locking is something rather normal for ww_mutex and
we are just exercising an existing code path.
But that would be the ctx case, right? I'm not sure there is a lot of
!ctx use out there, and in that case it really is rather like a normal
mutex.

E.g. the most common use case for the ww_mutex is in the graphics drivers
where usespace sends us a list of buffer objects to work with.

Now when userspace sends us duplicates in that buffer list the expectation
is to get -EALREADY from ww_mutex_lock when we try to lock the same ww_mutex
twice.
Right, I remember that much.. :-)

The intention behind this function is now to a) be able to extend those
checks to make sure user space doesn't sends us potentially harmful nonsense
and b) allow to check for recursion in TTM during buffer object eviction
which uses ww_mutex_trylock instead of ww_mutex_lock.
OK, but neither case would in fact need the !ctx case right? That's just
there for completeness sake?

Unfortunately not. TTM uses trylock to lock BOs which are about to be evicted to make room for all the BOs locked with a ctx.

I need to be able to distinct between the BOs which are trylocked and those which are locked with a ctx.

Writing this I actually noticed the current version is buggy, cause even when we check the mutex owner we still need to make sure that the ctx in the lock is NULL.

Time for v4 of the patch,
Christian.


But yes, I cannot think of a better fallback there either.