OSDN Git Service

drm/i915: Remove leftover vma->obj->pages_pin_count on insert/remove
authorChris Wilson <chris@chris-wilson.co.uk>
Tue, 15 Oct 2019 10:01:55 +0000 (11:01 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Tue, 15 Oct 2019 10:46:52 +0000 (11:46 +0100)
We now do the page pin count upfront in vma_get_pages/vma_put_pages, so
that we do the allocations before we enter the vm->mutex. Our vma
page references we are tracked in vma->pages_count and the extra
obj->pages_pin_count being performed later in i915_vma_insert and
i915_vma_remove is redundant, and worse throws off the shrinker's logic
on when it can free an object by unbinding it.

Reported-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reported-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191015100155.10376-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_vma.c

index 7d93625..e90c4d0 100644 (file)
@@ -703,7 +703,6 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
        list_add_tail(&vma->vm_link, &vma->vm->bound_list);
 
        if (vma->obj) {
-               atomic_inc(&vma->obj->mm.pages_pin_count);
                atomic_inc(&vma->obj->bind_count);
                assert_bind_count(vma->obj);
        }
@@ -726,14 +725,12 @@ i915_vma_remove(struct i915_vma *vma)
        if (vma->obj) {
                struct drm_i915_gem_object *obj = vma->obj;
 
-               atomic_dec(&obj->bind_count);
-
                /*
                 * And finally now the object is completely decoupled from this
                 * vma, we can drop its hold on the backing storage and allow
                 * it to be reaped by the shrinker.
                 */
-               i915_gem_object_unpin_pages(obj);
+               atomic_dec(&obj->bind_count);
                assert_bind_count(obj);
        }