diff mbox series

[05/10] dma-buf/resv: add dma_resv_prune_fences

Message ID 20190821123147.110736-6-christian.koenig@amd.com (mailing list archive)
State New, archived
Headers show
Series [01/10] dma-buf: make to_dma_fence_array NULL safe | expand

Commit Message

Christian König Aug. 21, 2019, 12:31 p.m. UTC
Add a new dma_resv_prune_fences() function to improve memory management.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
 drivers/dma-buf/dma-resv.c                 | 37 ++++++++++++++++++++++
 drivers/gpu/drm/i915/gem/i915_gem_wait.c   |  3 +-
 drivers/gpu/drm/i915/i915_gem_batch_pool.c |  2 +-
 drivers/gpu/drm/i915/i915_vma.c            |  3 +-
 drivers/gpu/drm/ttm/ttm_bo.c               |  2 +-
 include/linux/dma-resv.h                   |  1 +
 6 files changed, 42 insertions(+), 6 deletions(-)

Comments

Chris Wilson Aug. 21, 2019, 2:55 p.m. UTC | #1
Quoting Christian König (2019-08-21 13:31:42)
> Add a new dma_resv_prune_fences() function to improve memory management.
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
>  drivers/dma-buf/dma-resv.c                 | 37 ++++++++++++++++++++++
>  drivers/gpu/drm/i915/gem/i915_gem_wait.c   |  3 +-
>  drivers/gpu/drm/i915/i915_gem_batch_pool.c |  2 +-
>  drivers/gpu/drm/i915/i915_vma.c            |  3 +-
>  drivers/gpu/drm/ttm/ttm_bo.c               |  2 +-
>  include/linux/dma-resv.h                   |  1 +
>  6 files changed, 42 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c
> index 42a8f3f11681..24adc32d36d4 100644
> --- a/drivers/dma-buf/dma-resv.c
> +++ b/drivers/dma-buf/dma-resv.c
> @@ -301,6 +301,43 @@ void dma_resv_add_excl_fence(struct dma_resv *obj, struct dma_fence *fence)
>  }
>  EXPORT_SYMBOL(dma_resv_add_excl_fence);
>  
> +/**
> + * dma_resv_prune_fences - prune signaled fences from the resv object
> + * @obj: the reservation object to prune
> + *
> + * Prune all signaled fences from the reservation object.
> + */
> +void dma_resv_prune_fences(struct dma_resv *obj)
> +{
> +       struct dma_resv_list *list;
> +       struct dma_fence *fence;
> +       unsigned int i;
> +
> +       dma_resv_assert_held(obj);
> +
> +       fence = dma_resv_get_excl(obj);
> +       if (dma_fence_is_signaled(fence)) {
> +               RCU_INIT_POINTER(obj->fence_excl, NULL);
> +               dma_fence_put(fence);
> +       }
> +
> +       list = dma_resv_get_list(obj);
> +       if (!list)
> +               return;
> +
> +       for (i = 0; i < list->shared_count; ++i) {
> +               fence = rcu_dereference_protected(list->shared[i],
> +                                                 dma_resv_held(obj));
> +
> +               if (!dma_fence_is_signaled(fence))
> +                       continue;
> +
> +               RCU_INIT_POINTER(list->shared[i], dma_fence_get_stub());
> +               dma_fence_put(fence);

Not worth reusing the compaction logic from add_shared_fence?
-Chris
Chris Wilson Aug. 21, 2019, 2:56 p.m. UTC | #2
Quoting Chris Wilson (2019-08-21 15:55:08)
> Quoting Christian König (2019-08-21 13:31:42)
> > Add a new dma_resv_prune_fences() function to improve memory management.
> > 
> > Signed-off-by: Christian König <christian.koenig@amd.com>
> > ---
> >  drivers/dma-buf/dma-resv.c                 | 37 ++++++++++++++++++++++
> >  drivers/gpu/drm/i915/gem/i915_gem_wait.c   |  3 +-
> >  drivers/gpu/drm/i915/i915_gem_batch_pool.c |  2 +-
> >  drivers/gpu/drm/i915/i915_vma.c            |  3 +-
> >  drivers/gpu/drm/ttm/ttm_bo.c               |  2 +-
> >  include/linux/dma-resv.h                   |  1 +
> >  6 files changed, 42 insertions(+), 6 deletions(-)
> > 
> > diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c
> > index 42a8f3f11681..24adc32d36d4 100644
> > --- a/drivers/dma-buf/dma-resv.c
> > +++ b/drivers/dma-buf/dma-resv.c
> > @@ -301,6 +301,43 @@ void dma_resv_add_excl_fence(struct dma_resv *obj, struct dma_fence *fence)
> >  }
> >  EXPORT_SYMBOL(dma_resv_add_excl_fence);
> >  
> > +/**
> > + * dma_resv_prune_fences - prune signaled fences from the resv object
> > + * @obj: the reservation object to prune
> > + *
> > + * Prune all signaled fences from the reservation object.
> > + */
> > +void dma_resv_prune_fences(struct dma_resv *obj)
> > +{
> > +       struct dma_resv_list *list;
> > +       struct dma_fence *fence;
> > +       unsigned int i;
> > +
> > +       dma_resv_assert_held(obj);
> > +
> > +       fence = dma_resv_get_excl(obj);
> > +       if (dma_fence_is_signaled(fence)) {
> > +               RCU_INIT_POINTER(obj->fence_excl, NULL);
> > +               dma_fence_put(fence);
> > +       }
> > +
> > +       list = dma_resv_get_list(obj);
> > +       if (!list)
> > +               return;
> > +
> > +       for (i = 0; i < list->shared_count; ++i) {
> > +               fence = rcu_dereference_protected(list->shared[i],
> > +                                                 dma_resv_held(obj));
> > +
> > +               if (!dma_fence_is_signaled(fence))
> > +                       continue;
> > +
> > +               RCU_INIT_POINTER(list->shared[i], dma_fence_get_stub());
> > +               dma_fence_put(fence);
> 
> Not worth reusing the compaction logic from add_shared_fence?

Scratch that, you're going to rewrite the shared fence container.
-Chris
diff mbox series

Patch

diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c
index 42a8f3f11681..24adc32d36d4 100644
--- a/drivers/dma-buf/dma-resv.c
+++ b/drivers/dma-buf/dma-resv.c
@@ -301,6 +301,43 @@  void dma_resv_add_excl_fence(struct dma_resv *obj, struct dma_fence *fence)
 }
 EXPORT_SYMBOL(dma_resv_add_excl_fence);
 
+/**
+ * dma_resv_prune_fences - prune signaled fences from the resv object
+ * @obj: the reservation object to prune
+ *
+ * Prune all signaled fences from the reservation object.
+ */
+void dma_resv_prune_fences(struct dma_resv *obj)
+{
+	struct dma_resv_list *list;
+	struct dma_fence *fence;
+	unsigned int i;
+
+	dma_resv_assert_held(obj);
+
+	fence = dma_resv_get_excl(obj);
+	if (dma_fence_is_signaled(fence)) {
+		RCU_INIT_POINTER(obj->fence_excl, NULL);
+		dma_fence_put(fence);
+	}
+
+	list = dma_resv_get_list(obj);
+	if (!list)
+		return;
+
+	for (i = 0; i < list->shared_count; ++i) {
+		fence = rcu_dereference_protected(list->shared[i],
+						  dma_resv_held(obj));
+
+		if (!dma_fence_is_signaled(fence))
+			continue;
+
+		RCU_INIT_POINTER(list->shared[i], dma_fence_get_stub());
+		dma_fence_put(fence);
+	}
+}
+EXPORT_SYMBOL(dma_resv_prune_fences);
+
 /**
 * dma_resv_copy_fences - Copy all fences from src to dst.
 * @dst: the destination reservation object
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c b/drivers/gpu/drm/i915/gem/i915_gem_wait.c
index 8af55cd3e690..a76d36f8fb77 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c
@@ -85,8 +85,7 @@  i915_gem_object_wait_reservation(struct dma_resv *resv,
 	 * signaled.
 	 */
 	if (prune_fences && dma_resv_trylock(resv)) {
-		if (dma_resv_test_signaled_rcu(resv, true))
-			dma_resv_add_excl_fence(resv, NULL);
+		dma_resv_prune_fences(resv);
 		dma_resv_unlock(resv);
 	}
 
diff --git a/drivers/gpu/drm/i915/i915_gem_batch_pool.c b/drivers/gpu/drm/i915/i915_gem_batch_pool.c
index 5f82a763e64c..274cf5b19fc9 100644
--- a/drivers/gpu/drm/i915/i915_gem_batch_pool.c
+++ b/drivers/gpu/drm/i915/i915_gem_batch_pool.c
@@ -114,7 +114,7 @@  i915_gem_batch_pool_get(struct i915_gem_batch_pool *pool,
 			 */
 			if (rcu_access_pointer(resv->fence)) {
 				dma_resv_lock(resv, NULL);
-				dma_resv_add_excl_fence(resv, NULL);
+				dma_resv_prune_fences(resv);
 				dma_resv_unlock(resv);
 			}
 		}
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index ebfd03d117cd..fcbe433a968c 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -100,8 +100,7 @@  static void __i915_vma_retire(struct i915_active *ref)
 
 	/* Prune the shared fence arrays iff completely idle (inc. external) */
 	if (dma_resv_trylock(obj->base.resv)) {
-		if (dma_resv_test_signaled_rcu(obj->base.resv, true))
-			dma_resv_add_excl_fence(obj->base.resv, NULL);
+		dma_resv_prune_fences(obj->base.resv);
 		dma_resv_unlock(obj->base.resv);
 	}
 
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 58d1f2b28132..f78f52cc2e6d 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1818,7 +1818,7 @@  int ttm_bo_wait(struct ttm_buffer_object *bo,
 	if (timeout == 0)
 		return -EBUSY;
 
-	dma_resv_add_excl_fence(bo->base.resv, NULL);
+	dma_resv_prune_fences(bo->base.resv);
 	return 0;
 }
 EXPORT_SYMBOL(ttm_bo_wait);
diff --git a/include/linux/dma-resv.h b/include/linux/dma-resv.h
index ee50d10f052b..03b0f95682b0 100644
--- a/include/linux/dma-resv.h
+++ b/include/linux/dma-resv.h
@@ -279,6 +279,7 @@  int dma_resv_reserve_shared(struct dma_resv *obj, unsigned int num_fences);
 void dma_resv_add_shared_fence(struct dma_resv *obj, struct dma_fence *fence);
 
 void dma_resv_add_excl_fence(struct dma_resv *obj, struct dma_fence *fence);
+void dma_resv_prune_fences(struct dma_resv *obj);
 
 int dma_resv_get_fences_rcu(struct dma_resv *obj,
 			    struct dma_fence **pfence_excl,