[Intel-gfx] [PATCH] drm: fix deadlock of syncobj v5

[Chris Wilson]

Quoting zhoucm1 (2018-10-23 10:09:01)
> 
> 
> On 2018年10月23日 17:01, Chris Wilson wrote:
> > Quoting Chunming Zhou (2018-10-23 08:57:54)
> >> v2:
> >> add a mutex between sync_cb execution and free.
> >> v3:
> >> clearly separating the roles for pt_lock and cb_mutex (Chris)
> >> v4:
> >> the cb_mutex should be taken outside of the pt_lock around this if() block. (Chris)
> >> v5:
> >> fix a corner case
> >>
> >> Tested by syncobj_basic and syncobj_wait of igt.
> >>
> >> Signed-off-by: Chunming Zhou <david1.zhou at amd.com>
> >> Cc: Daniel Vetter <daniel at ffwll.ch>
> >> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> >> Cc: Christian König <christian.koenig at amd.com>
> >> Cc: intel-gfx at lists.freedesktop.org
> >> Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
> >> ---
> >>   drivers/gpu/drm/drm_syncobj.c | 55 +++++++++++++++++++----------------
> >>   include/drm/drm_syncobj.h     |  8 +++--
> >>   2 files changed, 36 insertions(+), 27 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c
> >> index 57bf6006394d..679a56791e34 100644
> >> --- a/drivers/gpu/drm/drm_syncobj.c
> >> +++ b/drivers/gpu/drm/drm_syncobj.c
> >> @@ -125,23 +125,26 @@ static int drm_syncobj_fence_get_or_add_callback(struct drm_syncobj *syncobj,
> >>          if (!ret)
> >>                  return 1;
> >>   
> >> -       spin_lock(&syncobj->lock);
> >> +       mutex_lock(&syncobj->cb_mutex);
> >>          /* We've already tried once to get a fence and failed.  Now that we
> >>           * have the lock, try one more time just to be sure we don't add a
> >>           * callback when a fence has already been set.
> >>           */
> >> +       spin_lock(&syncobj->pt_lock);
> >>          if (!list_empty(&syncobj->signal_pt_list)) {
> >> -               spin_unlock(&syncobj->lock);
> >> +               spin_unlock(&syncobj->pt_lock);
> >>                  drm_syncobj_search_fence(syncobj, 0, 0, fence);
> > Hmm, just thinking of other ways of tidying this up
> >
> > mutex_lock(cb_lock);
> > spin_lock(pt_lock);
> > *fence = drm_syncobj_find_signal_pt_for_point();
> > spin_unlock(pt_list);
> > if (*!fence)
> >       drm_syncobj_add_callback_locked(syncobj, cb, func);
> > mutex_unlock(cb_lock);
> >
> > i.e. get rid of the early return and we can even drop the int return here
> > as it is unimportant and unused.
> Yes, do you need I send v6? or you make a separate patch as a improvment?

Send it in reply, we still have some time before the shards catch up
with the ml ;)
-Chris