[Intel-gfx] [PATCH v3 03/37] drm/i915/region: support basic eviction

[Chris Wilson]

Quoting Matthew Auld (2019-08-09 23:26:09)
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 6ff01a404346..8735dea74809 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1105,6 +1105,23 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data,
>             !i915_gem_object_has_pages(obj))
>                 i915_gem_object_truncate(obj);
>  
> +       if (obj->mm.region) {
> +               mutex_lock(&obj->mm.region->obj_lock);
> +
> +               switch (obj->mm.madv) {
> +               case I915_MADV_WILLNEED:
> +                       list_move(&obj->mm.region_link,
> +                                 &obj->mm.region->objects);
> +                       break;
> +               default:
> +                       list_move(&obj->mm.region_link,
> +                                 &obj->mm.region->purgeable);
> +                       break;
> +               }
> +
> +               mutex_unlock(&obj->mm.region->obj_lock);
> +       }
> +
>         args->retained = obj->mm.madv != __I915_MADV_PURGED;

Little bit of an impedance mismatch, I hope this turns out fine when
everything is a memory region.

>         mutex_unlock(&obj->mm.lock);
>  
> diff --git a/drivers/gpu/drm/i915/intel_memory_region.c b/drivers/gpu/drm/i915/intel_memory_region.c
> index ef12e462acb8..3a3caaadea1f 100644
> --- a/drivers/gpu/drm/i915/intel_memory_region.c
> +++ b/drivers/gpu/drm/i915/intel_memory_region.c
> @@ -12,6 +12,51 @@ const u32 intel_region_map[] = {
>         [INTEL_MEMORY_STOLEN] = BIT(INTEL_STOLEN + INTEL_MEMORY_TYPE_SHIFT) | BIT(0),
>  };
>  
> +static int
> +intel_memory_region_evict(struct intel_memory_region *mem,
> +                         resource_size_t target,
> +                         unsigned int flags)
> +{
> +       struct drm_i915_gem_object *obj;
> +       resource_size_t found;
> +       int err;
> +
> +       err = 0;
> +       found = 0;
> +
> +       mutex_lock(&mem->obj_lock);
> +       list_for_each_entry(obj, &mem->purgeable, mm.region_link) {
> +               if (!i915_gem_object_has_pages(obj))
> +                       continue;
> +
> +               if (READ_ONCE(obj->pin_global))
> +                       continue;
> +
> +               if (atomic_read(&obj->bind_count))
> +                       continue;
> +
> +               mutex_unlock(&mem->obj_lock);
> +
> +               __i915_gem_object_put_pages(obj, I915_MM_SHRINKER);

So we don't really care about the object being bound then? As all we
care about is the page's pin_count.

So instead of obj->pin_global, obj->bind_bound, you just want

if (atomic_read(&obj->pages.pin_count))
	continue;

as the quick check to see if it is worth preceding.

> +               mutex_lock_nested(&obj->mm.lock, I915_MM_SHRINKER);
> +               if (!i915_gem_object_has_pages(obj)) {
> +                       obj->mm.madv = __I915_MADV_PURGED;
> +                       found += obj->base.size;
> +               }
> +               mutex_unlock(&obj->mm.lock);

The locking here accomplishes what? You just want a boolean from
put_pages().

> +
> +               if (found >= target)
> +                       return 0;
> +
> +               mutex_lock(&mem->obj_lock);
> +       }
> +
> +       err = -ENOSPC;
> +       mutex_unlock(&mem->obj_lock);
> +       return err;
> +}