[Intel-gfx] [PATCH v2 1/2] drm/i915: Do an unlocked wait before set-cache-level ioctl
Chris Wilson
chris at chris-wilson.co.uk
Thu Jan 19 11:43:55 UTC 2017
On Thu, Jan 19, 2017 at 01:36:29PM +0200, Mika Kuoppala wrote:
> Chris Wilson <chris at chris-wilson.co.uk> writes:
>
> > Since a change in cache level is likely to trigger an unbind, avoid
> > waiting under the mutex by preemptively doing an unlocked wait.
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > ---
> > drivers/gpu/drm/i915/i915_gem.c | 29 +++++++++++++++++++----------
> > 1 file changed, 19 insertions(+), 10 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> > index c1c2765bb8d0..0926c291404c 100644
> > --- a/drivers/gpu/drm/i915/i915_gem.c
> > +++ b/drivers/gpu/drm/i915/i915_gem.c
> > @@ -3384,7 +3384,7 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,
> > struct drm_i915_gem_caching *args = data;
> > struct drm_i915_gem_object *obj;
> > enum i915_cache_level level;
> > - int ret;
> > + int ret = 0;
> >
> > switch (args->caching) {
> > case I915_CACHING_NONE:
> > @@ -3409,20 +3409,29 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data,
> > return -EINVAL;
> > }
> >
> > - ret = i915_mutex_lock_interruptible(dev);
> > + obj = i915_gem_object_lookup(file, args->handle);
> > + if (!obj)
> > + return -ENOENT;
> > +
> > + if (obj->cache_level == level)
> > + goto out;
> > +
> > + ret = i915_gem_object_wait(obj,
> > + I915_WAIT_INTERRUPTIBLE,
> > + MAX_SCHEDULE_TIMEOUT,
> > + to_rps_client(file));
> > if (ret)
> > - return ret;
> > + goto out;
> >
> > - obj = i915_gem_object_lookup(file, args->handle);
> > - if (!obj) {
> > - ret = -ENOENT;
> > - goto unlock;
> > - }
>
> Is the intent here that you just make it very likely
> that the wait in set_cache_level will not trigger?
Yes. We don't remove it, since we need it to ensure correctness, but we
prefer to do the waiting without holding any lock.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
More information about the Intel-gfx
mailing list