From: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
When instantiating a tiled object on an L-shaped memory machine, we mark
the object as unshrinkable to prevent the shrinker from trying to swap
out the pages. We have to do this as we do not know the swizzling on the
individual pages, and so the data will be scrambled across swap out/in.
Not only do we need to move the object off the shrinker list, we need to
mark the object with shrink_pin so that the counter is consistent across
calls to madvise.
Fixes: 0175969e489a ("drm/i915/gem: Use shrinkable status for unknown swizzle quirks")
References: https://gitlab.freedesktop.org/drm/intel/-/issues/3293
Reported-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx>
---
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index aed8a37ccdc9..7361971c177d 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -63,6 +63,8 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
i915->quirks & QUIRK_PIN_SWIZZLED_PAGES) {
GEM_BUG_ON(i915_gem_object_has_tiling_quirk(obj));
i915_gem_object_set_tiling_quirk(obj);
+ GEM_BUG_ON(!list_empty(&obj->mm.link));
+ atomic_inc(&obj->mm.shrink_pin);
shrinkable = false;
}
--
2.26.3
