Skip to contentSkip to conversation

drm/vc4: Fulfill user BO creation requests from the kernel BO cache.

PATCH RESEND In review Eric Anholt 2 participants 4 messages Mar 1, 2017 – Mar 2, 2017

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/2 drm/vc4: Fulfill user BO creation requests from the kernel BO cache.

The from_cache flag was actually "the BO is invisible to userspace", so we can repurpose it to just zero out a cached BO and return it to userspace.

Improves wall time for a loop of 5 glsl-algebraic-add-add-1 by -1.44989% +/- 0.862891% (n=28, 1 outlier removed from each that appeared to be other system noise)

Note that there's an intel-gpu-tools test to check for the proper zeroing behavior here, which we continue to pass.

---
 drivers/gpu/drm/vc4/vc4_bo.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
drivers/gpu/drm/vc4/vc4_bo.c +7 -6
index 7abcd9c5dbe2..e5c7aa935b4b 100644
@@ -211,21 +211,22 @@ struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size)
 }
 
 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t unaligned_size,
-			     bool from_cache)
+			     bool allow_unzeroed)
 {
 	size_t size = roundup(unaligned_size, PAGE_SIZE);
 	struct vc4_dev *vc4 = to_vc4_dev(dev);
 	struct drm_gem_cma_object *cma_obj;
+	struct vc4_bo *bo;
 
 	if (size == 0)
 		return ERR_PTR(-EINVAL);
 
 	/* First, try to get a vc4_bo from the kernel BO cache. */
-	if (from_cache) {
-		struct vc4_bo *bo = vc4_bo_get_from_cache(dev, size);
-
-		if (bo)
-			return bo;
+	bo = vc4_bo_get_from_cache(dev, size);
+	if (bo) {
+		if (!allow_unzeroed)
+			memset(bo->base.vaddr, 0, bo->base.base.size);
+		return bo;
 	}
 
 	cma_obj = drm_gem_cma_create(dev, size);
Signature · 6 lines
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH RESEND 2/2 drm/vc4: Fix OOPSes from trying to cache a partially constructed BO.

If a CMA allocation failed, the partially constructed BO would be unreferenced through the normal path, and we might choose to put it in the BO cache. If we then reused it before it expired from the cache, the kernel would OOPS.

Fixes c826a6e10644 ("drm/vc4: Add a BO cache.")
---
 drivers/gpu/drm/vc4/vc4_bo.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/gpu/drm/vc4/vc4_bo.c +8 -0
index e5c7aa935b4b..af29432a6471 100644
@@ -317,6 +317,14 @@ void vc4_free_object(struct drm_gem_object *gem_bo)
 		goto out;
 	}
 
+	/* If this object was partially constructed but CMA allocation
+	 * had failed, just free it.
+	 */
+	if (!bo->base.vaddr) {
+		vc4_bo_destroy(bo);
+		goto out;
+	}
+
 	cache_list = vc4_get_cache_list_for_size(dev, gem_bo->size);
 	if (!cache_list) {
 		vc4_bo_destroy(bo);
Signature · 6 lines
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Reviewed-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2 RESEND] drm/vc4: Fix OOPSes from trying to cache a partially constructed BO.

Quoted 7 lines · On Wed, 1 Mar 2017 10:56:02 -0800 Eric Anholt <[email protected]> wrote:

If a CMA allocation failed, the partially constructed BO would be unreferenced through the normal path, and we might choose to put it in the BO cache. If we then reused it before it expired from the cache, the kernel would OOPS.

Fixes c826a6e10644 ("drm/vc4: Add a BO cache.")
Quoted 13 lines
---
 drivers/gpu/drm/vc4/vc4_bo.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/gpu/drm/vc4/vc4_bo.c +1 -0
index e5c7aa935b4b..af29432a6471 100644
@@ -317,6 +317,14 @@ void vc4_free_object(struct drm_gem_object *gem_bo)
 		goto out;
 	}
 
+	/* If this object was partially constructed but CMA allocation

drivers/gpu/drm/vc4/vc4_bo.c void vc4_free_object(struct drm_gem_object *gem_bo) · line 330

Quoted 10 lines
drivers/gpu/drm/vc4/vc4_bo.c +7 -0
+	 * had failed, just free it.
+	 */
+	if (!bo->base.vaddr) {
+		vc4_bo_destroy(bo);
+		goto out;
+	}
+
 	cache_list = vc4_get_cache_list_for_size(dev, gem_bo->size);
 	if (!cache_list) {
 		vc4_bo_destroy(bo);

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Reviewed-by on v1 1/2
#3 lore raw
In reply to Eric Anholt
Quoted 4 lines · On Wed, 1 Mar 2017 10:56:01 -0800 Eric Anholt <[email protected]> wrote:

The from_cache flag was actually "the BO is invisible to userspace", so we can repurpose it to just zero out a cached BO and return it to userspace.

Quoted 8 lines

Improves wall time for a loop of 5 glsl-algebraic-add-add-1 by -1.44989% +/- 0.862891% (n=28, 1 outlier removed from each that appeared to be other system noise)

Note that there's an intel-gpu-tools test to check for the proper zeroing behavior here, which we continue to pass.

Quoted 27 lines
---
 drivers/gpu/drm/vc4/vc4_bo.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
drivers/gpu/drm/vc4/vc4_bo.c +2 -4
index 7abcd9c5dbe2..e5c7aa935b4b 100644
@@ -211,21 +211,22 @@ struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size)
 }
 
 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t unaligned_size,
-			     bool from_cache)
+			     bool allow_unzeroed)
 {
 	size_t size = roundup(unaligned_size, PAGE_SIZE);
 	struct vc4_dev *vc4 = to_vc4_dev(dev);
 	struct drm_gem_cma_object *cma_obj;
+	struct vc4_bo *bo;
 
 	if (size == 0)
 		return ERR_PTR(-EINVAL);
 
 	/* First, try to get a vc4_bo from the kernel BO cache. */
-	if (from_cache) {
-		struct vc4_bo *bo = vc4_bo_get_from_cache(dev, size);
-

drivers/gpu/drm/vc4/vc4_bo.c struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size) · line 232

Quoted 10 lines
drivers/gpu/drm/vc4/vc4_bo.c +5 -2
-		if (bo)
-			return bo;
+	bo = vc4_bo_get_from_cache(dev, size);
+	if (bo) {
+		if (!allow_unzeroed)
+			memset(bo->base.vaddr, 0, bo->base.base.size);
+		return bo;
 	}
 
 	cma_obj = drm_gem_cma_create(dev, size);

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel