| Message ID | 1349881739-32277-1-git-send-email-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Oct 10, 2012 at 5:08 PM, Chris Wilson <chris@chris-wilson.co.uk> wrote: > If we have hit oom whilst holding our struct_mutex, then currently we > cannot reap our own GPU buffers which likely pin most of memory, making > an outright OOM more likely. So if we are running in direct reclaim and > already hold the mutex, attempt to free buffers knowing that the > original function can not continue until we return. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> I've thought a bit about this, and I fear the implications. It's a very neat trick, but now every memory alloc call could potentially result in unpinned objects getting unbound and in active objects getting retired. Previously we only needed to fear active objects disappearing when calling retire_request (since that could drop the last reference). With this chance we have many more places, and given how often we e.g. fumbled the refcounting in the fence stealing code I'm scared ... /me needs to think more about this Cheers, Daniel > --- > drivers/gpu/drm/i915/i915_gem.c | 24 +++++++++++++++++++++--- > 1 file changed, 21 insertions(+), 3 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > index 1d0cbfb..bed4084 100644 > --- a/drivers/gpu/drm/i915/i915_gem.c > +++ b/drivers/gpu/drm/i915/i915_gem.c > @@ -4589,6 +4589,18 @@ void i915_gem_release(struct drm_device *dev, struct drm_file *file) > spin_unlock(&file_priv->mm.lock); > } > > +static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task) > +{ > + if (!mutex_is_locked(mutex)) > + return false; > + > +#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_SMP) > + return mutex->owner == task; > +#else > + return false; > +#endif > +} > + > static int > i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) > { > @@ -4599,10 +4611,15 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) > struct drm_device *dev = dev_priv->dev; > struct drm_i915_gem_object *obj; > int nr_to_scan = sc->nr_to_scan; > + bool unlock = true; > int cnt; > > - if (!mutex_trylock(&dev->struct_mutex)) > - return 0; > + if (!mutex_trylock(&dev->struct_mutex)) { > + if (mutex_is_locked_by(&dev->struct_mutex, current)) > + unlock = false; > + else > + return 0; > + } > > if (nr_to_scan) { > nr_to_scan -= i915_gem_purge(dev_priv, nr_to_scan); > @@ -4618,6 +4635,7 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) > if (obj->pin_count == 0 && obj->pages_pin_count == 0) > cnt += obj->base.size >> PAGE_SHIFT; > > - mutex_unlock(&dev->struct_mutex); > + if (unlock) > + mutex_unlock(&dev->struct_mutex); > return cnt; > } > -- > 1.7.10.4 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
On Wed, 10 Oct 2012 16:08:59 +0100 Chris Wilson <chris@chris-wilson.co.uk> wrote: > If we have hit oom whilst holding our struct_mutex, then currently we > cannot reap our own GPU buffers which likely pin most of memory, making > an outright OOM more likely. So if we are running in direct reclaim and > already hold the mutex, attempt to free buffers knowing that the > original function can not continue until we return. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> I will try to review this a bit further on the plane ride home. It's quite a scary patch, but it fixes OOM destroying IGT runs, so it's pretty necessary IMO. Meanwhile: Tested-by: Ben Widawsky <ben@bwidawsk.net> > --- > drivers/gpu/drm/i915/i915_gem.c | 24 +++++++++++++++++++++--- > 1 file changed, 21 insertions(+), 3 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > index 1d0cbfb..bed4084 100644 > --- a/drivers/gpu/drm/i915/i915_gem.c > +++ b/drivers/gpu/drm/i915/i915_gem.c > @@ -4589,6 +4589,18 @@ void i915_gem_release(struct drm_device *dev, struct drm_file *file) > spin_unlock(&file_priv->mm.lock); > } > > +static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task) > +{ > + if (!mutex_is_locked(mutex)) > + return false; > + > +#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_SMP) > + return mutex->owner == task; > +#else > + return false; > +#endif > +} > + > static int > i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) > { > @@ -4599,10 +4611,15 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) > struct drm_device *dev = dev_priv->dev; > struct drm_i915_gem_object *obj; > int nr_to_scan = sc->nr_to_scan; > + bool unlock = true; > int cnt; > > - if (!mutex_trylock(&dev->struct_mutex)) > - return 0; > + if (!mutex_trylock(&dev->struct_mutex)) { > + if (mutex_is_locked_by(&dev->struct_mutex, current)) > + unlock = false; > + else > + return 0; > + } > > if (nr_to_scan) { > nr_to_scan -= i915_gem_purge(dev_priv, nr_to_scan); > @@ -4618,6 +4635,7 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) > if (obj->pin_count == 0 && obj->pages_pin_count == 0) > cnt += obj->base.size >> PAGE_SHIFT; > > - mutex_unlock(&dev->struct_mutex); > + if (unlock) > + mutex_unlock(&dev->struct_mutex); > return cnt; > }
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 1d0cbfb..bed4084 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -4589,6 +4589,18 @@ void i915_gem_release(struct drm_device *dev, struct drm_file *file) spin_unlock(&file_priv->mm.lock); } +static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task) +{ + if (!mutex_is_locked(mutex)) + return false; + +#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_SMP) + return mutex->owner == task; +#else + return false; +#endif +} + static int i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) { @@ -4599,10 +4611,15 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) struct drm_device *dev = dev_priv->dev; struct drm_i915_gem_object *obj; int nr_to_scan = sc->nr_to_scan; + bool unlock = true; int cnt; - if (!mutex_trylock(&dev->struct_mutex)) - return 0; + if (!mutex_trylock(&dev->struct_mutex)) { + if (mutex_is_locked_by(&dev->struct_mutex, current)) + unlock = false; + else + return 0; + } if (nr_to_scan) { nr_to_scan -= i915_gem_purge(dev_priv, nr_to_scan); @@ -4618,6 +4635,7 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) if (obj->pin_count == 0 && obj->pages_pin_count == 0) cnt += obj->base.size >> PAGE_SHIFT; - mutex_unlock(&dev->struct_mutex); + if (unlock) + mutex_unlock(&dev->struct_mutex); return cnt; }
If we have hit oom whilst holding our struct_mutex, then currently we cannot reap our own GPU buffers which likely pin most of memory, making an outright OOM more likely. So if we are running in direct reclaim and already hold the mutex, attempt to free buffers knowing that the original function can not continue until we return. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> --- drivers/gpu/drm/i915/i915_gem.c | 24 +++++++++++++++++++++--- 1 file changed, 21 insertions(+), 3 deletions(-)