On Wed, Oct 10, 2012 at 5:08 PM, Chris Wilson <chris at 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 at 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 at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
