[PATCH 0/2] dma-buf/sync_file: rework fences on struct sync_file

[PATCH 0/2] dma-buf/sync_file: rework fences on struct sync_file

[Gustavo Padovan]

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

Hi all, 
  
This is an attempt to improve fence support on Sync File. The basic idea  
is to have only sync_file->fence and store all fences there, either as 
normal fences or fence_arrays. That way we can remove some potential
duplication when using fence_array with sync_file: the duplication of the array 
of fences and the duplication of fence_add_callback() for all fences.  
  
Now when creating a new sync_file during the merge process sync_file_set_fence()
will set sync_file->fence based on the number of fences for that sync_file. If  
there is more than one fence a fence_array is created. One important advantage  
approach is that we only add one fence callback now, no matter how many fences  
there are in a sync_file - the individual callbacks are added by fence_array.
  
Please review! Thanks! 
  
  Gustavo  

Changes since RFC v2 (Comments from Chris Wilson and Christian König):  
 - struct sync_file lost status member in favor of fence_is_signaled()
 - drop use of fence_array_teardown()
 - use sizeof(*fence) to allocate only an array on fence pointers

Changes since RFC v1 (Comments from Chris Wilson and Christian König):  
 - Not using fence_ops anymore.
 - fence_is_array() was created to differentiate fence from fence_array  
 - fence_array_teardown() is now exported and used under fence_is_array()
 - struct sync_file lost num_fences member 

Gustavo Padovan (2):
  dma-buf/fence-array: add fence_is_array()
  dma-buf/sync_file: rework fence storage in struct file

 drivers/dma-buf/sync_file.c          | 163 ++++++++++++++++++++++-------------
 drivers/staging/android/sync_debug.c |  13 ++-
 include/linux/fence-array.h          |  10 +++
 include/linux/sync_file.h            |  17 ++--
 4 files changed, 128 insertions(+), 75 deletions(-)

-- 
2.5.5

[PATCH 0/2] dma-buf/sync_file: rework fences on struct sync_file

[Gustavo Padovan]

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

Hi all, 
  
This is an attempt to improve fence support on Sync File. The basic idea  
is to have only sync_file->fence and store all fences there, either as 
normal fences or fence_arrays. That way we can remove some potential
duplication when using fence_array with sync_file: the duplication of the array 
of fences and the duplication of fence_add_callback() for all fences.  
  
Now when creating a new sync_file during the merge process sync_file_set_fence()
will set sync_file->fence based on the number of fences for that sync_file. If  
there is more than one fence a fence_array is created. One important advantage  
approach is that we only add one fence callback now, no matter how many fences  
there are in a sync_file - the individual callbacks are added by fence_array.
  
Please review! Thanks! 
  
  Gustavo  

Changes since RFC v2 (Comments from Chris Wilson and Christian König):  
 - struct sync_file lost status member in favor of fence_is_signaled()
 - drop use of fence_array_teardown()
 - use sizeof(*fence) to allocate only an array on fence pointers

Changes since RFC v1 (Comments from Chris Wilson and Christian König):  
 - Not using fence_ops anymore.
 - fence_is_array() was created to differentiate fence from fence_array  
 - fence_array_teardown() is now exported and used under fence_is_array()
 - struct sync_file lost num_fences member 

Gustavo Padovan (2):
  dma-buf/fence-array: add fence_is_array()
  dma-buf/sync_file: rework fence storage in struct file

 drivers/dma-buf/sync_file.c          | 163 ++++++++++++++++++++++-------------
 drivers/staging/android/sync_debug.c |  13 ++-
 include/linux/fence-array.h          |  10 +++
 include/linux/sync_file.h            |  17 ++--
 4 files changed, 128 insertions(+), 75 deletions(-)

-- 
2.5.5

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 1/2] dma-buf/fence-array: add fence_is_array()

[Gustavo Padovan]

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

Add helper to check if fence is array.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
---
 include/linux/fence-array.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/linux/fence-array.h b/include/linux/fence-array.h
index 86baaa4..d2e9f40 100644
--- a/include/linux/fence-array.h
+++ b/include/linux/fence-array.h
@@ -52,6 +52,16 @@ struct fence_array {
 extern const struct fence_ops fence_array_ops;
 
 /**
+ * fence_is_array - check if a fence is from the array subsclass
+ *
+ * Return true if it is a fence_array and false otherwise.
+ */
+static inline bool fence_is_array(struct fence *fence)
+{
+	return (fence->ops == &fence_array_ops) ? true : false;
+}
+
+/**
  * to_fence_array - cast a fence to a fence_array
  * @fence: fence to cast to a fence_array
  *
-- 
2.5.5

[PATCH 1/2] dma-buf/fence-array: add fence_is_array()

[Gustavo Padovan]

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

Add helper to check if fence is array.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
---
 include/linux/fence-array.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/linux/fence-array.h b/include/linux/fence-array.h
index 86baaa4..d2e9f40 100644
--- a/include/linux/fence-array.h
+++ b/include/linux/fence-array.h
@@ -52,6 +52,16 @@ struct fence_array {
 extern const struct fence_ops fence_array_ops;
 
 /**
+ * fence_is_array - check if a fence is from the array subsclass
+ *
+ * Return true if it is a fence_array and false otherwise.
+ */
+static inline bool fence_is_array(struct fence *fence)
+{
+	return (fence->ops == &fence_array_ops) ? true : false;
+}
+
+/**
  * to_fence_array - cast a fence to a fence_array
  * @fence: fence to cast to a fence_array
  *
-- 
2.5.5

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 1/2] dma-buf/fence-array: add fence_is_array()

[Chris Wilson]

On Thu, Jun 30, 2016 at 11:09:47AM -0300, Gustavo Padovan wrote:
> From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> 
> Add helper to check if fence is array.
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Christian König <christian.koenig@amd.com>
> Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> ---
>  include/linux/fence-array.h | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/include/linux/fence-array.h b/include/linux/fence-array.h
> index 86baaa4..d2e9f40 100644
> --- a/include/linux/fence-array.h
> +++ b/include/linux/fence-array.h
> @@ -52,6 +52,16 @@ struct fence_array {
>  extern const struct fence_ops fence_array_ops;
>  
>  /**
> + * fence_is_array - check if a fence is from the array subsclass
> + *
> + * Return true if it is a fence_array and false otherwise.
> + */
> +static inline bool fence_is_array(struct fence *fence)
> +{
> +	return (fence->ops == &fence_array_ops) ? true : false;

Just return fence->ops == &fence_array_ops; It's a logical operation, it
is already a boolean. With that,

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: [PATCH 1/2] dma-buf/fence-array: add fence_is_array()

[Chris Wilson]

On Thu, Jun 30, 2016 at 11:09:47AM -0300, Gustavo Padovan wrote:
> From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> 
> Add helper to check if fence is array.
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Christian König <christian.koenig@amd.com>
> Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> ---
>  include/linux/fence-array.h | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/include/linux/fence-array.h b/include/linux/fence-array.h
> index 86baaa4..d2e9f40 100644
> --- a/include/linux/fence-array.h
> +++ b/include/linux/fence-array.h
> @@ -52,6 +52,16 @@ struct fence_array {
>  extern const struct fence_ops fence_array_ops;
>  
>  /**
> + * fence_is_array - check if a fence is from the array subsclass
> + *
> + * Return true if it is a fence_array and false otherwise.
> + */
> +static inline bool fence_is_array(struct fence *fence)
> +{
> +	return (fence->ops == &fence_array_ops) ? true : false;

Just return fence->ops == &fence_array_ops; It's a logical operation, it
is already a boolean. With that,

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 1/2] dma-buf/fence-array: add fence_is_array()

[Chris Wilson]

On Thu, Jun 30, 2016 at 03:23:59PM +0100, Chris Wilson wrote:
> On Thu, Jun 30, 2016 at 11:09:47AM -0300, Gustavo Padovan wrote:
> > From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> > 
> > Add helper to check if fence is array.
> > 
> > Cc: Chris Wilson <chris@chris-wilson.co.uk>
> > Cc: Christian König <christian.koenig@amd.com>
> > Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> > ---
> >  include/linux/fence-array.h | 10 ++++++++++
> >  1 file changed, 10 insertions(+)
> > 
> > diff --git a/include/linux/fence-array.h b/include/linux/fence-array.h
> > index 86baaa4..d2e9f40 100644
> > --- a/include/linux/fence-array.h
> > +++ b/include/linux/fence-array.h
> > @@ -52,6 +52,16 @@ struct fence_array {
> >  extern const struct fence_ops fence_array_ops;
> >  
> >  /**
> > + * fence_is_array - check if a fence is from the array subsclass
> > + *
> > + * Return true if it is a fence_array and false otherwise.
> > + */
> > +static inline bool fence_is_array(struct fence *fence)
> > +{
> > +	return (fence->ops == &fence_array_ops) ? true : false;
> 
> Just return fence->ops == &fence_array_ops; It's a logical operation, it
> is already a boolean. With that,
> 
> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

Oops, this needs an EXPORT_SYMBOL(fence_array_ops);
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: [PATCH 1/2] dma-buf/fence-array: add fence_is_array()

[Chris Wilson]

On Thu, Jun 30, 2016 at 03:23:59PM +0100, Chris Wilson wrote:
> On Thu, Jun 30, 2016 at 11:09:47AM -0300, Gustavo Padovan wrote:
> > From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> > 
> > Add helper to check if fence is array.
> > 
> > Cc: Chris Wilson <chris@chris-wilson.co.uk>
> > Cc: Christian König <christian.koenig@amd.com>
> > Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> > ---
> >  include/linux/fence-array.h | 10 ++++++++++
> >  1 file changed, 10 insertions(+)
> > 
> > diff --git a/include/linux/fence-array.h b/include/linux/fence-array.h
> > index 86baaa4..d2e9f40 100644
> > --- a/include/linux/fence-array.h
> > +++ b/include/linux/fence-array.h
> > @@ -52,6 +52,16 @@ struct fence_array {
> >  extern const struct fence_ops fence_array_ops;
> >  
> >  /**
> > + * fence_is_array - check if a fence is from the array subsclass
> > + *
> > + * Return true if it is a fence_array and false otherwise.
> > + */
> > +static inline bool fence_is_array(struct fence *fence)
> > +{
> > +	return (fence->ops == &fence_array_ops) ? true : false;
> 
> Just return fence->ops == &fence_array_ops; It's a logical operation, it
> is already a boolean. With that,
> 
> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

Oops, this needs an EXPORT_SYMBOL(fence_array_ops);
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 2/2] dma-buf/sync_file: rework fence storage in struct file

[Gustavo Padovan]

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

Create sync_file->fence to abstract the type of fence we are using for
each sync_file. If only one fence is present we use a normal struct fence
but if there is more fences to be added to the sync_file a fence_array
is created.

This change cleans up sync_file a bit. We don't need to have sync_file_cb
array anymore. Instead, as we always have  one fence, only one fence
callback is registered per sync_file.

v3: Comments from Chris Wilson and Christian König
	- struct sync_file lost status member in favor of fence_is_signaled()
	- drop use of fence_array_teardown()
	- use sizeof(*fence) to allocate only an array on fence pointers

v2: Comments from Chris Wilson and Christian König
	- Not using fence_ops anymore
	- fence_is_array() was created to differentiate fence from fence_array
	- fence_array_teardown() is now exported and used under fence_is_array()
	- struct sync_file lost num_fences member

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
---
 drivers/dma-buf/sync_file.c          | 163 ++++++++++++++++++++++-------------
 drivers/staging/android/sync_debug.c |  13 ++-
 include/linux/sync_file.h            |  17 ++--
 3 files changed, 118 insertions(+), 75 deletions(-)

diff --git a/drivers/dma-buf/sync_file.c b/drivers/dma-buf/sync_file.c
index 9aaa608..7a5aaae 100644
--- a/drivers/dma-buf/sync_file.c
+++ b/drivers/dma-buf/sync_file.c
@@ -28,11 +28,11 @@
 
 static const struct file_operations sync_file_fops;
 
-static struct sync_file *sync_file_alloc(int size)
+static struct sync_file *sync_file_alloc(void)
 {
 	struct sync_file *sync_file;
 
-	sync_file = kzalloc(size, GFP_KERNEL);
+	sync_file = kzalloc(sizeof(*sync_file), GFP_KERNEL);
 	if (!sync_file)
 		return NULL;
 
@@ -45,6 +45,8 @@ static struct sync_file *sync_file_alloc(int size)
 
 	init_waitqueue_head(&sync_file->wq);
 
+	INIT_LIST_HEAD(&sync_file->cb.node);
+
 	return sync_file;
 
 err:
@@ -54,14 +56,11 @@ err:
 
 static void fence_check_cb_func(struct fence *f, struct fence_cb *cb)
 {
-	struct sync_file_cb *check;
 	struct sync_file *sync_file;
 
-	check = container_of(cb, struct sync_file_cb, cb);
-	sync_file = check->sync_file;
+	sync_file = container_of(cb, struct sync_file, cb);
 
-	if (atomic_dec_and_test(&sync_file->status))
-		wake_up_all(&sync_file->wq);
+	wake_up_all(&sync_file->wq);
 }
 
 /**
@@ -76,22 +75,18 @@ struct sync_file *sync_file_create(struct fence *fence)
 {
 	struct sync_file *sync_file;
 
-	sync_file = sync_file_alloc(offsetof(struct sync_file, cbs[1]));
+	sync_file = sync_file_alloc();
 	if (!sync_file)
 		return NULL;
 
-	sync_file->num_fences = 1;
-	atomic_set(&sync_file->status, 1);
+	sync_file->fence = fence;
+
 	snprintf(sync_file->name, sizeof(sync_file->name), "%s-%s%llu-%d",
 		 fence->ops->get_driver_name(fence),
 		 fence->ops->get_timeline_name(fence), fence->context,
 		 fence->seqno);
 
-	sync_file->cbs[0].fence = fence;
-	sync_file->cbs[0].sync_file = sync_file;
-	if (fence_add_callback(fence, &sync_file->cbs[0].cb,
-			       fence_check_cb_func))
-		atomic_dec(&sync_file->status);
+	fence_add_callback(fence, &sync_file->cb, fence_check_cb_func);
 
 	return sync_file;
 }
@@ -121,14 +116,48 @@ err:
 	return NULL;
 }
 
-static void sync_file_add_pt(struct sync_file *sync_file, int *i,
-			     struct fence *fence)
+static int sync_file_set_fence(struct sync_file *sync_file,
+			       struct fence **fences, int num_fences)
+{
+	struct fence_array *array;
+
+	/*
+	 * The reference for the fences in the new sync_file and holded
+	 * in add_fence() during the merge procedure, so for num_fences == 1
+	 * we already own a new reference to the fence. For num_fence > 1
+	 * we own the reference of the fence_array creation.
+	 */
+	if (num_fences == 1) {
+		sync_file->fence = fences[0];
+	} else {
+		array = fence_array_create(num_fences, fences,
+					   fence_context_alloc(1), 1, false);
+		if (!array)
+			return -ENOMEM;
+
+		sync_file->fence = (struct fence *)array;
+	}
+
+	return 0;
+}
+
+static struct fence **get_fences(struct sync_file *sync_file, int *num_fences)
+{
+	if (fence_is_array(sync_file->fence)) {
+		struct fence_array *array = to_fence_array(sync_file->fence);
+		*num_fences = array->num_fences;
+		return array->fences;
+	} else {
+		*num_fences = 1;
+		return &sync_file->fence;
+	}
+}
+
+static void add_fence(struct fence **fences, int *i, struct fence *fence)
 {
-	sync_file->cbs[*i].fence = fence;
-	sync_file->cbs[*i].sync_file = sync_file;
+	fences[*i] = fence;
 
-	if (!fence_add_callback(fence, &sync_file->cbs[*i].cb,
-				fence_check_cb_func)) {
+	if (!fence_is_signaled(fence)) {
 		fence_get(fence);
 		(*i)++;
 	}
@@ -147,16 +176,21 @@ static void sync_file_add_pt(struct sync_file *sync_file, int *i,
 static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
 					 struct sync_file *b)
 {
-	int num_fences = a->num_fences + b->num_fences;
 	struct sync_file *sync_file;
-	int i, i_a, i_b;
-	unsigned long size = offsetof(struct sync_file, cbs[num_fences]);
+	struct fence **fences, **a_fences, **b_fences;
+	int i, i_a, i_b, num_fences, a_num_fences, b_num_fences;
 
-	sync_file = sync_file_alloc(size);
+	sync_file = sync_file_alloc();
 	if (!sync_file)
 		return NULL;
 
-	atomic_set(&sync_file->status, num_fences);
+	a_fences = get_fences(a, &a_num_fences);
+	b_fences = get_fences(b, &b_num_fences);
+	num_fences = a_num_fences + b_num_fences;
+
+	fences = kcalloc(num_fences, sizeof(*fences), GFP_KERNEL);
+	if (!fences)
+		goto err;
 
 	/*
 	 * Assume sync_file a and b are both ordered and have no
@@ -165,55 +199,66 @@ static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
 	 * If a sync_file can only be created with sync_file_merge
 	 * and sync_file_create, this is a reasonable assumption.
 	 */
-	for (i = i_a = i_b = 0; i_a < a->num_fences && i_b < b->num_fences; ) {
-		struct fence *pt_a = a->cbs[i_a].fence;
-		struct fence *pt_b = b->cbs[i_b].fence;
+	for (i = i_a = i_b = 0; i_a < a_num_fences && i_b < b_num_fences; ) {
+		struct fence *pt_a = a_fences[i_a];
+		struct fence *pt_b = b_fences[i_b];
 
 		if (pt_a->context < pt_b->context) {
-			sync_file_add_pt(sync_file, &i, pt_a);
+			add_fence(fences, &i, pt_a);
 
 			i_a++;
 		} else if (pt_a->context > pt_b->context) {
-			sync_file_add_pt(sync_file, &i, pt_b);
+			add_fence(fences, &i, pt_b);
 
 			i_b++;
 		} else {
 			if (pt_a->seqno - pt_b->seqno <= INT_MAX)
-				sync_file_add_pt(sync_file, &i, pt_a);
+				add_fence(fences, &i, pt_a);
 			else
-				sync_file_add_pt(sync_file, &i, pt_b);
+				add_fence(fences, &i, pt_b);
 
 			i_a++;
 			i_b++;
 		}
 	}
 
-	for (; i_a < a->num_fences; i_a++)
-		sync_file_add_pt(sync_file, &i, a->cbs[i_a].fence);
+	for (; i_a < a_num_fences; i_a++)
+		add_fence(fences, &i, a_fences[i_a]);
+
+	for (; i_b < b_num_fences; i_b++)
+		add_fence(fences, &i, b_fences[i_b]);
+
+	if (num_fences > i) {
+		fences = krealloc(fences, i * sizeof(**fences),
+				  GFP_KERNEL);
+		if (!fences)
+			goto err;
+	}
 
-	for (; i_b < b->num_fences; i_b++)
-		sync_file_add_pt(sync_file, &i, b->cbs[i_b].fence);
+	if (sync_file_set_fence(sync_file, fences, i) < 0) {
+		kfree(fences);
+		goto err;
+	}
 
-	if (num_fences > i)
-		atomic_sub(num_fences - i, &sync_file->status);
-	sync_file->num_fences = i;
+	fence_add_callback(sync_file->fence, &sync_file->cb,
+			   fence_check_cb_func);
 
 	strlcpy(sync_file->name, name, sizeof(sync_file->name));
 	return sync_file;
+
+err:
+	fput(sync_file->file);
+	return NULL;
+
 }
 
 static void sync_file_free(struct kref *kref)
 {
 	struct sync_file *sync_file = container_of(kref, struct sync_file,
 						     kref);
-	int i;
-
-	for (i = 0; i < sync_file->num_fences; ++i) {
-		fence_remove_callback(sync_file->cbs[i].fence,
-				      &sync_file->cbs[i].cb);
-		fence_put(sync_file->cbs[i].fence);
-	}
 
+	fence_remove_callback(sync_file->fence, &sync_file->cb);
+	fence_put(sync_file->fence);
 	kfree(sync_file);
 }
 
@@ -232,9 +277,9 @@ static unsigned int sync_file_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, &sync_file->wq, wait);
 
-	status = atomic_read(&sync_file->status);
+	status = fence_is_signaled(sync_file->fence);
 
-	if (!status)
+	if (status)
 		return POLLIN;
 	if (status < 0)
 		return POLLERR;
@@ -315,8 +360,9 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 {
 	struct sync_file_info info;
 	struct sync_fence_info *fence_info = NULL;
+	struct fence **fences;
 	__u32 size;
-	int ret, i;
+	int num_fences, ret, i;
 
 	if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
 		return -EFAULT;
@@ -324,6 +370,8 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 	if (info.flags || info.pad)
 		return -EINVAL;
 
+	fences = get_fences(sync_file, &num_fences);
+
 	/*
 	 * Passing num_fences = 0 means that userspace doesn't want to
 	 * retrieve any sync_fence_info. If num_fences = 0 we skip filling
@@ -333,16 +381,16 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 	if (!info.num_fences)
 		goto no_fences;
 
-	if (info.num_fences < sync_file->num_fences)
+	if (info.num_fences < num_fences)
 		return -EINVAL;
 
-	size = sync_file->num_fences * sizeof(*fence_info);
+	size = num_fences * sizeof(*fence_info);
 	fence_info = kzalloc(size, GFP_KERNEL);
 	if (!fence_info)
 		return -ENOMEM;
 
-	for (i = 0; i < sync_file->num_fences; ++i)
-		sync_fill_fence_info(sync_file->cbs[i].fence, &fence_info[i]);
+	for (i = 0; i < num_fences; i++)
+		sync_fill_fence_info(fences[i], &fence_info[i]);
 
 	if (copy_to_user(u64_to_user_ptr(info.sync_fence_info), fence_info,
 			 size)) {
@@ -352,11 +400,8 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 
 no_fences:
 	strlcpy(info.name, sync_file->name, sizeof(info.name));
-	info.status = atomic_read(&sync_file->status);
-	if (info.status >= 0)
-		info.status = !info.status;
-
-	info.num_fences = sync_file->num_fences;
+	info.status = fence_is_signaled(sync_file->fence);
+	info.num_fences = num_fences;
 
 	if (copy_to_user((void __user *)arg, &info, sizeof(info)))
 		ret = -EFAULT;
diff --git a/drivers/staging/android/sync_debug.c b/drivers/staging/android/sync_debug.c
index 5f57499..0638a06 100644
--- a/drivers/staging/android/sync_debug.c
+++ b/drivers/staging/android/sync_debug.c
@@ -159,10 +159,15 @@ static void sync_print_sync_file(struct seq_file *s,
 	int i;
 
 	seq_printf(s, "[%p] %s: %s\n", sync_file, sync_file->name,
-		   sync_status_str(atomic_read(&sync_file->status)));
-
-	for (i = 0; i < sync_file->num_fences; ++i)
-		sync_print_fence(s, sync_file->cbs[i].fence, true);
+		   sync_status_str(!fence_is_signaled(sync_file->fence)));
+
+	if (fence_is_array(sync_file->fence)) {
+		struct fence_array *array = to_fence_array(sync_file->fence);
+		for (i = 0; i < array->num_fences; ++i)
+			sync_print_fence(s, array->fences[i], true);
+	} else {
+		sync_print_fence(s, sync_file->fence, true);
+	}
 }
 
 static int sync_debugfs_show(struct seq_file *s, void *unused)
diff --git a/include/linux/sync_file.h b/include/linux/sync_file.h
index c6ffe8b..2efc5ec 100644
--- a/include/linux/sync_file.h
+++ b/include/linux/sync_file.h
@@ -19,12 +19,7 @@
 #include <linux/list.h>
 #include <linux/spinlock.h>
 #include <linux/fence.h>
-
-struct sync_file_cb {
-	struct fence_cb cb;
-	struct fence *fence;
-	struct sync_file *sync_file;
-};
+#include <linux/fence-array.h>
 
 /**
  * struct sync_file - sync file to export to the userspace
@@ -32,10 +27,9 @@ struct sync_file_cb {
  * @kref:		reference count on fence.
  * @name:		name of sync_file.  Useful for debugging
  * @sync_file_list:	membership in global file list
- * @num_fences:		number of sync_pts in the fence
  * @wq:			wait queue for fence signaling
- * @status:		0: signaled, >0:active, <0: error
- * @cbs:		sync_pts callback information
+ * @fence:		fence with the fences in the sync_file
+ * @cb:			fence callback information
  */
 struct sync_file {
 	struct file		*file;
@@ -44,12 +38,11 @@ struct sync_file {
 #ifdef CONFIG_DEBUG_FS
 	struct list_head	sync_file_list;
 #endif
-	int num_fences;
 
 	wait_queue_head_t	wq;
-	atomic_t		status;
 
-	struct sync_file_cb	cbs[];
+	struct fence		*fence;
+	struct fence_cb cb;
 };
 
 struct sync_file *sync_file_create(struct fence *fence);
-- 
2.5.5

[PATCH 2/2] dma-buf/sync_file: rework fence storage in struct file

[Gustavo Padovan]

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

Create sync_file->fence to abstract the type of fence we are using for
each sync_file. If only one fence is present we use a normal struct fence
but if there is more fences to be added to the sync_file a fence_array
is created.

This change cleans up sync_file a bit. We don't need to have sync_file_cb
array anymore. Instead, as we always have  one fence, only one fence
callback is registered per sync_file.

v3: Comments from Chris Wilson and Christian König
	- struct sync_file lost status member in favor of fence_is_signaled()
	- drop use of fence_array_teardown()
	- use sizeof(*fence) to allocate only an array on fence pointers

v2: Comments from Chris Wilson and Christian König
	- Not using fence_ops anymore
	- fence_is_array() was created to differentiate fence from fence_array
	- fence_array_teardown() is now exported and used under fence_is_array()
	- struct sync_file lost num_fences member

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
---
 drivers/dma-buf/sync_file.c          | 163 ++++++++++++++++++++++-------------
 drivers/staging/android/sync_debug.c |  13 ++-
 include/linux/sync_file.h            |  17 ++--
 3 files changed, 118 insertions(+), 75 deletions(-)

diff --git a/drivers/dma-buf/sync_file.c b/drivers/dma-buf/sync_file.c
index 9aaa608..7a5aaae 100644
--- a/drivers/dma-buf/sync_file.c
+++ b/drivers/dma-buf/sync_file.c
@@ -28,11 +28,11 @@
 
 static const struct file_operations sync_file_fops;
 
-static struct sync_file *sync_file_alloc(int size)
+static struct sync_file *sync_file_alloc(void)
 {
 	struct sync_file *sync_file;
 
-	sync_file = kzalloc(size, GFP_KERNEL);
+	sync_file = kzalloc(sizeof(*sync_file), GFP_KERNEL);
 	if (!sync_file)
 		return NULL;
 
@@ -45,6 +45,8 @@ static struct sync_file *sync_file_alloc(int size)
 
 	init_waitqueue_head(&sync_file->wq);
 
+	INIT_LIST_HEAD(&sync_file->cb.node);
+
 	return sync_file;
 
 err:
@@ -54,14 +56,11 @@ err:
 
 static void fence_check_cb_func(struct fence *f, struct fence_cb *cb)
 {
-	struct sync_file_cb *check;
 	struct sync_file *sync_file;
 
-	check = container_of(cb, struct sync_file_cb, cb);
-	sync_file = check->sync_file;
+	sync_file = container_of(cb, struct sync_file, cb);
 
-	if (atomic_dec_and_test(&sync_file->status))
-		wake_up_all(&sync_file->wq);
+	wake_up_all(&sync_file->wq);
 }
 
 /**
@@ -76,22 +75,18 @@ struct sync_file *sync_file_create(struct fence *fence)
 {
 	struct sync_file *sync_file;
 
-	sync_file = sync_file_alloc(offsetof(struct sync_file, cbs[1]));
+	sync_file = sync_file_alloc();
 	if (!sync_file)
 		return NULL;
 
-	sync_file->num_fences = 1;
-	atomic_set(&sync_file->status, 1);
+	sync_file->fence = fence;
+
 	snprintf(sync_file->name, sizeof(sync_file->name), "%s-%s%llu-%d",
 		 fence->ops->get_driver_name(fence),
 		 fence->ops->get_timeline_name(fence), fence->context,
 		 fence->seqno);
 
-	sync_file->cbs[0].fence = fence;
-	sync_file->cbs[0].sync_file = sync_file;
-	if (fence_add_callback(fence, &sync_file->cbs[0].cb,
-			       fence_check_cb_func))
-		atomic_dec(&sync_file->status);
+	fence_add_callback(fence, &sync_file->cb, fence_check_cb_func);
 
 	return sync_file;
 }
@@ -121,14 +116,48 @@ err:
 	return NULL;
 }
 
-static void sync_file_add_pt(struct sync_file *sync_file, int *i,
-			     struct fence *fence)
+static int sync_file_set_fence(struct sync_file *sync_file,
+			       struct fence **fences, int num_fences)
+{
+	struct fence_array *array;
+
+	/*
+	 * The reference for the fences in the new sync_file and holded
+	 * in add_fence() during the merge procedure, so for num_fences == 1
+	 * we already own a new reference to the fence. For num_fence > 1
+	 * we own the reference of the fence_array creation.
+	 */
+	if (num_fences == 1) {
+		sync_file->fence = fences[0];
+	} else {
+		array = fence_array_create(num_fences, fences,
+					   fence_context_alloc(1), 1, false);
+		if (!array)
+			return -ENOMEM;
+
+		sync_file->fence = (struct fence *)array;
+	}
+
+	return 0;
+}
+
+static struct fence **get_fences(struct sync_file *sync_file, int *num_fences)
+{
+	if (fence_is_array(sync_file->fence)) {
+		struct fence_array *array = to_fence_array(sync_file->fence);
+		*num_fences = array->num_fences;
+		return array->fences;
+	} else {
+		*num_fences = 1;
+		return &sync_file->fence;
+	}
+}
+
+static void add_fence(struct fence **fences, int *i, struct fence *fence)
 {
-	sync_file->cbs[*i].fence = fence;
-	sync_file->cbs[*i].sync_file = sync_file;
+	fences[*i] = fence;
 
-	if (!fence_add_callback(fence, &sync_file->cbs[*i].cb,
-				fence_check_cb_func)) {
+	if (!fence_is_signaled(fence)) {
 		fence_get(fence);
 		(*i)++;
 	}
@@ -147,16 +176,21 @@ static void sync_file_add_pt(struct sync_file *sync_file, int *i,
 static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
 					 struct sync_file *b)
 {
-	int num_fences = a->num_fences + b->num_fences;
 	struct sync_file *sync_file;
-	int i, i_a, i_b;
-	unsigned long size = offsetof(struct sync_file, cbs[num_fences]);
+	struct fence **fences, **a_fences, **b_fences;
+	int i, i_a, i_b, num_fences, a_num_fences, b_num_fences;
 
-	sync_file = sync_file_alloc(size);
+	sync_file = sync_file_alloc();
 	if (!sync_file)
 		return NULL;
 
-	atomic_set(&sync_file->status, num_fences);
+	a_fences = get_fences(a, &a_num_fences);
+	b_fences = get_fences(b, &b_num_fences);
+	num_fences = a_num_fences + b_num_fences;
+
+	fences = kcalloc(num_fences, sizeof(*fences), GFP_KERNEL);
+	if (!fences)
+		goto err;
 
 	/*
 	 * Assume sync_file a and b are both ordered and have no
@@ -165,55 +199,66 @@ static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
 	 * If a sync_file can only be created with sync_file_merge
 	 * and sync_file_create, this is a reasonable assumption.
 	 */
-	for (i = i_a = i_b = 0; i_a < a->num_fences && i_b < b->num_fences; ) {
-		struct fence *pt_a = a->cbs[i_a].fence;
-		struct fence *pt_b = b->cbs[i_b].fence;
+	for (i = i_a = i_b = 0; i_a < a_num_fences && i_b < b_num_fences; ) {
+		struct fence *pt_a = a_fences[i_a];
+		struct fence *pt_b = b_fences[i_b];
 
 		if (pt_a->context < pt_b->context) {
-			sync_file_add_pt(sync_file, &i, pt_a);
+			add_fence(fences, &i, pt_a);
 
 			i_a++;
 		} else if (pt_a->context > pt_b->context) {
-			sync_file_add_pt(sync_file, &i, pt_b);
+			add_fence(fences, &i, pt_b);
 
 			i_b++;
 		} else {
 			if (pt_a->seqno - pt_b->seqno <= INT_MAX)
-				sync_file_add_pt(sync_file, &i, pt_a);
+				add_fence(fences, &i, pt_a);
 			else
-				sync_file_add_pt(sync_file, &i, pt_b);
+				add_fence(fences, &i, pt_b);
 
 			i_a++;
 			i_b++;
 		}
 	}
 
-	for (; i_a < a->num_fences; i_a++)
-		sync_file_add_pt(sync_file, &i, a->cbs[i_a].fence);
+	for (; i_a < a_num_fences; i_a++)
+		add_fence(fences, &i, a_fences[i_a]);
+
+	for (; i_b < b_num_fences; i_b++)
+		add_fence(fences, &i, b_fences[i_b]);
+
+	if (num_fences > i) {
+		fences = krealloc(fences, i * sizeof(**fences),
+				  GFP_KERNEL);
+		if (!fences)
+			goto err;
+	}
 
-	for (; i_b < b->num_fences; i_b++)
-		sync_file_add_pt(sync_file, &i, b->cbs[i_b].fence);
+	if (sync_file_set_fence(sync_file, fences, i) < 0) {
+		kfree(fences);
+		goto err;
+	}
 
-	if (num_fences > i)
-		atomic_sub(num_fences - i, &sync_file->status);
-	sync_file->num_fences = i;
+	fence_add_callback(sync_file->fence, &sync_file->cb,
+			   fence_check_cb_func);
 
 	strlcpy(sync_file->name, name, sizeof(sync_file->name));
 	return sync_file;
+
+err:
+	fput(sync_file->file);
+	return NULL;
+
 }
 
 static void sync_file_free(struct kref *kref)
 {
 	struct sync_file *sync_file = container_of(kref, struct sync_file,
 						     kref);
-	int i;
-
-	for (i = 0; i < sync_file->num_fences; ++i) {
-		fence_remove_callback(sync_file->cbs[i].fence,
-				      &sync_file->cbs[i].cb);
-		fence_put(sync_file->cbs[i].fence);
-	}
 
+	fence_remove_callback(sync_file->fence, &sync_file->cb);
+	fence_put(sync_file->fence);
 	kfree(sync_file);
 }
 
@@ -232,9 +277,9 @@ static unsigned int sync_file_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, &sync_file->wq, wait);
 
-	status = atomic_read(&sync_file->status);
+	status = fence_is_signaled(sync_file->fence);
 
-	if (!status)
+	if (status)
 		return POLLIN;
 	if (status < 0)
 		return POLLERR;
@@ -315,8 +360,9 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 {
 	struct sync_file_info info;
 	struct sync_fence_info *fence_info = NULL;
+	struct fence **fences;
 	__u32 size;
-	int ret, i;
+	int num_fences, ret, i;
 
 	if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
 		return -EFAULT;
@@ -324,6 +370,8 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 	if (info.flags || info.pad)
 		return -EINVAL;
 
+	fences = get_fences(sync_file, &num_fences);
+
 	/*
 	 * Passing num_fences = 0 means that userspace doesn't want to
 	 * retrieve any sync_fence_info. If num_fences = 0 we skip filling
@@ -333,16 +381,16 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 	if (!info.num_fences)
 		goto no_fences;
 
-	if (info.num_fences < sync_file->num_fences)
+	if (info.num_fences < num_fences)
 		return -EINVAL;
 
-	size = sync_file->num_fences * sizeof(*fence_info);
+	size = num_fences * sizeof(*fence_info);
 	fence_info = kzalloc(size, GFP_KERNEL);
 	if (!fence_info)
 		return -ENOMEM;
 
-	for (i = 0; i < sync_file->num_fences; ++i)
-		sync_fill_fence_info(sync_file->cbs[i].fence, &fence_info[i]);
+	for (i = 0; i < num_fences; i++)
+		sync_fill_fence_info(fences[i], &fence_info[i]);
 
 	if (copy_to_user(u64_to_user_ptr(info.sync_fence_info), fence_info,
 			 size)) {
@@ -352,11 +400,8 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file,
 
 no_fences:
 	strlcpy(info.name, sync_file->name, sizeof(info.name));
-	info.status = atomic_read(&sync_file->status);
-	if (info.status >= 0)
-		info.status = !info.status;
-
-	info.num_fences = sync_file->num_fences;
+	info.status = fence_is_signaled(sync_file->fence);
+	info.num_fences = num_fences;
 
 	if (copy_to_user((void __user *)arg, &info, sizeof(info)))
 		ret = -EFAULT;
diff --git a/drivers/staging/android/sync_debug.c b/drivers/staging/android/sync_debug.c
index 5f57499..0638a06 100644
--- a/drivers/staging/android/sync_debug.c
+++ b/drivers/staging/android/sync_debug.c
@@ -159,10 +159,15 @@ static void sync_print_sync_file(struct seq_file *s,
 	int i;
 
 	seq_printf(s, "[%p] %s: %s\n", sync_file, sync_file->name,
-		   sync_status_str(atomic_read(&sync_file->status)));
-
-	for (i = 0; i < sync_file->num_fences; ++i)
-		sync_print_fence(s, sync_file->cbs[i].fence, true);
+		   sync_status_str(!fence_is_signaled(sync_file->fence)));
+
+	if (fence_is_array(sync_file->fence)) {
+		struct fence_array *array = to_fence_array(sync_file->fence);
+		for (i = 0; i < array->num_fences; ++i)
+			sync_print_fence(s, array->fences[i], true);
+	} else {
+		sync_print_fence(s, sync_file->fence, true);
+	}
 }
 
 static int sync_debugfs_show(struct seq_file *s, void *unused)
diff --git a/include/linux/sync_file.h b/include/linux/sync_file.h
index c6ffe8b..2efc5ec 100644
--- a/include/linux/sync_file.h
+++ b/include/linux/sync_file.h
@@ -19,12 +19,7 @@
 #include <linux/list.h>
 #include <linux/spinlock.h>
 #include <linux/fence.h>
-
-struct sync_file_cb {
-	struct fence_cb cb;
-	struct fence *fence;
-	struct sync_file *sync_file;
-};
+#include <linux/fence-array.h>
 
 /**
  * struct sync_file - sync file to export to the userspace
@@ -32,10 +27,9 @@ struct sync_file_cb {
  * @kref:		reference count on fence.
  * @name:		name of sync_file.  Useful for debugging
  * @sync_file_list:	membership in global file list
- * @num_fences:		number of sync_pts in the fence
  * @wq:			wait queue for fence signaling
- * @status:		0: signaled, >0:active, <0: error
- * @cbs:		sync_pts callback information
+ * @fence:		fence with the fences in the sync_file
+ * @cb:			fence callback information
  */
 struct sync_file {
 	struct file		*file;
@@ -44,12 +38,11 @@ struct sync_file {
 #ifdef CONFIG_DEBUG_FS
 	struct list_head	sync_file_list;
 #endif
-	int num_fences;
 
 	wait_queue_head_t	wq;
-	atomic_t		status;
 
-	struct sync_file_cb	cbs[];
+	struct fence		*fence;
+	struct fence_cb cb;
 };
 
 struct sync_file *sync_file_create(struct fence *fence);
-- 
2.5.5

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 2/2] dma-buf/sync_file: rework fence storage in struct file

[Chris Wilson]

On Thu, Jun 30, 2016 at 11:09:48AM -0300, Gustavo Padovan wrote:
> From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> 
> Create sync_file->fence to abstract the type of fence we are using for
> each sync_file. If only one fence is present we use a normal struct fence
> but if there is more fences to be added to the sync_file a fence_array
> is created.
> 
> This change cleans up sync_file a bit. We don't need to have sync_file_cb
> array anymore. Instead, as we always have  one fence, only one fence
> callback is registered per sync_file.
> 
> v3: Comments from Chris Wilson and Christian König
> 	- struct sync_file lost status member in favor of fence_is_signaled()
> 	- drop use of fence_array_teardown()
> 	- use sizeof(*fence) to allocate only an array on fence pointers
> 
> v2: Comments from Chris Wilson and Christian König
> 	- Not using fence_ops anymore
> 	- fence_is_array() was created to differentiate fence from fence_array
> 	- fence_array_teardown() is now exported and used under fence_is_array()
> 	- struct sync_file lost num_fences member
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Christian König <christian.koenig@amd.com>
> Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> ---
>  drivers/dma-buf/sync_file.c          | 163 ++++++++++++++++++++++-------------
>  drivers/staging/android/sync_debug.c |  13 ++-
>  include/linux/sync_file.h            |  17 ++--
>  3 files changed, 118 insertions(+), 75 deletions(-)
> 
> diff --git a/drivers/dma-buf/sync_file.c b/drivers/dma-buf/sync_file.c
> index 9aaa608..7a5aaae 100644
> --- a/drivers/dma-buf/sync_file.c
> +++ b/drivers/dma-buf/sync_file.c
> @@ -28,11 +28,11 @@
>  
>  static const struct file_operations sync_file_fops;
>  
> -static struct sync_file *sync_file_alloc(int size)
> +static struct sync_file *sync_file_alloc(void)
>  {
>  	struct sync_file *sync_file;
>  
> -	sync_file = kzalloc(size, GFP_KERNEL);
> +	sync_file = kzalloc(sizeof(*sync_file), GFP_KERNEL);
>  	if (!sync_file)
>  		return NULL;
>  
> @@ -45,6 +45,8 @@ static struct sync_file *sync_file_alloc(int size)
>  
>  	init_waitqueue_head(&sync_file->wq);
>  
> +	INIT_LIST_HEAD(&sync_file->cb.node);
> +
>  	return sync_file;
>  
>  err:
> @@ -54,14 +56,11 @@ err:
>  
>  static void fence_check_cb_func(struct fence *f, struct fence_cb *cb)
>  {
> -	struct sync_file_cb *check;
>  	struct sync_file *sync_file;
>  
> -	check = container_of(cb, struct sync_file_cb, cb);
> -	sync_file = check->sync_file;
> +	sync_file = container_of(cb, struct sync_file, cb);
>  
> -	if (atomic_dec_and_test(&sync_file->status))
> -		wake_up_all(&sync_file->wq);
> +	wake_up_all(&sync_file->wq);
>  }
>  
>  /**
> @@ -76,22 +75,18 @@ struct sync_file *sync_file_create(struct fence *fence)
>  {
>  	struct sync_file *sync_file;
>  
> -	sync_file = sync_file_alloc(offsetof(struct sync_file, cbs[1]));
> +	sync_file = sync_file_alloc();
>  	if (!sync_file)
>  		return NULL;
>  
> -	sync_file->num_fences = 1;
> -	atomic_set(&sync_file->status, 1);
> +	sync_file->fence = fence;
> +
>  	snprintf(sync_file->name, sizeof(sync_file->name), "%s-%s%llu-%d",
>  		 fence->ops->get_driver_name(fence),
>  		 fence->ops->get_timeline_name(fence), fence->context,
>  		 fence->seqno);
>  
> -	sync_file->cbs[0].fence = fence;
> -	sync_file->cbs[0].sync_file = sync_file;
> -	if (fence_add_callback(fence, &sync_file->cbs[0].cb,
> -			       fence_check_cb_func))
> -		atomic_dec(&sync_file->status);
> +	fence_add_callback(fence, &sync_file->cb, fence_check_cb_func);
>  
>  	return sync_file;
>  }
> @@ -121,14 +116,48 @@ err:
>  	return NULL;
>  }
>  
> -static void sync_file_add_pt(struct sync_file *sync_file, int *i,
> -			     struct fence *fence)
> +static int sync_file_set_fence(struct sync_file *sync_file,
> +			       struct fence **fences, int num_fences)
> +{
> +	struct fence_array *array;
> +
> +	/*
> +	 * The reference for the fences in the new sync_file and holded

s/holded/held/

> +	 * in add_fence() during the merge procedure, so for num_fences == 1
> +	 * we already own a new reference to the fence. For num_fence > 1
> +	 * we own the reference of the fence_array creation.
> +	 */

Thanks, that really does help understanding the reference handling.

> +	if (num_fences == 1) {
> +		sync_file->fence = fences[0];
> +	} else {
> +		array = fence_array_create(num_fences, fences,
> +					   fence_context_alloc(1), 1, false);
> +		if (!array)
> +			return -ENOMEM;
> +
> +		sync_file->fence = (struct fence *)array;

I'd prefer sync_file->fence = &array->base;

> +	}
> +
> +	return 0;
> +}
> +
> +static struct fence **get_fences(struct sync_file *sync_file, int *num_fences)
> +{
> +	if (fence_is_array(sync_file->fence)) {
> +		struct fence_array *array = to_fence_array(sync_file->fence);
> +		*num_fences = array->num_fences;
> +		return array->fences;
> +	} else {
> +		*num_fences = 1;
> +		return &sync_file->fence;
> +	}
> +}
> +
> +static void add_fence(struct fence **fences, int *i, struct fence *fence)
>  {
> -	sync_file->cbs[*i].fence = fence;
> -	sync_file->cbs[*i].sync_file = sync_file;
> +	fences[*i] = fence;
>  
> -	if (!fence_add_callback(fence, &sync_file->cbs[*i].cb,
> -				fence_check_cb_func)) {
> +	if (!fence_is_signaled(fence)) {
>  		fence_get(fence);
>  		(*i)++;
>  	}
> @@ -147,16 +176,21 @@ static void sync_file_add_pt(struct sync_file *sync_file, int *i,
>  static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
>  					 struct sync_file *b)
>  {
> -	int num_fences = a->num_fences + b->num_fences;
>  	struct sync_file *sync_file;
> -	int i, i_a, i_b;
> -	unsigned long size = offsetof(struct sync_file, cbs[num_fences]);
> +	struct fence **fences, **a_fences, **b_fences;
> +	int i, i_a, i_b, num_fences, a_num_fences, b_num_fences;
>  
> -	sync_file = sync_file_alloc(size);
> +	sync_file = sync_file_alloc();
>  	if (!sync_file)
>  		return NULL;
>  
> -	atomic_set(&sync_file->status, num_fences);
> +	a_fences = get_fences(a, &a_num_fences);
> +	b_fences = get_fences(b, &b_num_fences);
> +	num_fences = a_num_fences + b_num_fences;

Do we need overflow paranoia here? Probably, as this is on a user ioctl
path.

> +
> +	fences = kcalloc(num_fences, sizeof(*fences), GFP_KERNEL);
> +	if (!fences)
> +		goto err;
>  
>  	/*
>  	 * Assume sync_file a and b are both ordered and have no
> @@ -165,55 +199,66 @@ static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
>  	 * If a sync_file can only be created with sync_file_merge
>  	 * and sync_file_create, this is a reasonable assumption.
>  	 */
> -	for (i = i_a = i_b = 0; i_a < a->num_fences && i_b < b->num_fences; ) {
> -		struct fence *pt_a = a->cbs[i_a].fence;
> -		struct fence *pt_b = b->cbs[i_b].fence;
> +	for (i = i_a = i_b = 0; i_a < a_num_fences && i_b < b_num_fences; ) {
> +		struct fence *pt_a = a_fences[i_a];
> +		struct fence *pt_b = b_fences[i_b];
>  
>  		if (pt_a->context < pt_b->context) {
> -			sync_file_add_pt(sync_file, &i, pt_a);
> +			add_fence(fences, &i, pt_a);
>  
>  			i_a++;
>  		} else if (pt_a->context > pt_b->context) {
> -			sync_file_add_pt(sync_file, &i, pt_b);
> +			add_fence(fences, &i, pt_b);
>  
>  			i_b++;
>  		} else {
>  			if (pt_a->seqno - pt_b->seqno <= INT_MAX)
> -				sync_file_add_pt(sync_file, &i, pt_a);
> +				add_fence(fences, &i, pt_a);
>  			else
> -				sync_file_add_pt(sync_file, &i, pt_b);
> +				add_fence(fences, &i, pt_b);
>  
>  			i_a++;
>  			i_b++;
>  		}
>  	}
>  
> -	for (; i_a < a->num_fences; i_a++)
> -		sync_file_add_pt(sync_file, &i, a->cbs[i_a].fence);
> +	for (; i_a < a_num_fences; i_a++)
> +		add_fence(fences, &i, a_fences[i_a]);
> +
> +	for (; i_b < b_num_fences; i_b++)
> +		add_fence(fences, &i, b_fences[i_b]);
> +
> +	if (num_fences > i) {
> +		fences = krealloc(fences, i * sizeof(**fences),

Hmm, still too large, sizeof(*fences);

Looking pretty good. Putting fence-array to good use.

With those minor tweaks,
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: [PATCH 2/2] dma-buf/sync_file: rework fence storage in struct file

[Chris Wilson]

On Thu, Jun 30, 2016 at 11:09:48AM -0300, Gustavo Padovan wrote:
> From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> 
> Create sync_file->fence to abstract the type of fence we are using for
> each sync_file. If only one fence is present we use a normal struct fence
> but if there is more fences to be added to the sync_file a fence_array
> is created.
> 
> This change cleans up sync_file a bit. We don't need to have sync_file_cb
> array anymore. Instead, as we always have  one fence, only one fence
> callback is registered per sync_file.
> 
> v3: Comments from Chris Wilson and Christian König
> 	- struct sync_file lost status member in favor of fence_is_signaled()
> 	- drop use of fence_array_teardown()
> 	- use sizeof(*fence) to allocate only an array on fence pointers
> 
> v2: Comments from Chris Wilson and Christian König
> 	- Not using fence_ops anymore
> 	- fence_is_array() was created to differentiate fence from fence_array
> 	- fence_array_teardown() is now exported and used under fence_is_array()
> 	- struct sync_file lost num_fences member
> 
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Christian König <christian.koenig@amd.com>
> Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> ---
>  drivers/dma-buf/sync_file.c          | 163 ++++++++++++++++++++++-------------
>  drivers/staging/android/sync_debug.c |  13 ++-
>  include/linux/sync_file.h            |  17 ++--
>  3 files changed, 118 insertions(+), 75 deletions(-)
> 
> diff --git a/drivers/dma-buf/sync_file.c b/drivers/dma-buf/sync_file.c
> index 9aaa608..7a5aaae 100644
> --- a/drivers/dma-buf/sync_file.c
> +++ b/drivers/dma-buf/sync_file.c
> @@ -28,11 +28,11 @@
>  
>  static const struct file_operations sync_file_fops;
>  
> -static struct sync_file *sync_file_alloc(int size)
> +static struct sync_file *sync_file_alloc(void)
>  {
>  	struct sync_file *sync_file;
>  
> -	sync_file = kzalloc(size, GFP_KERNEL);
> +	sync_file = kzalloc(sizeof(*sync_file), GFP_KERNEL);
>  	if (!sync_file)
>  		return NULL;
>  
> @@ -45,6 +45,8 @@ static struct sync_file *sync_file_alloc(int size)
>  
>  	init_waitqueue_head(&sync_file->wq);
>  
> +	INIT_LIST_HEAD(&sync_file->cb.node);
> +
>  	return sync_file;
>  
>  err:
> @@ -54,14 +56,11 @@ err:
>  
>  static void fence_check_cb_func(struct fence *f, struct fence_cb *cb)
>  {
> -	struct sync_file_cb *check;
>  	struct sync_file *sync_file;
>  
> -	check = container_of(cb, struct sync_file_cb, cb);
> -	sync_file = check->sync_file;
> +	sync_file = container_of(cb, struct sync_file, cb);
>  
> -	if (atomic_dec_and_test(&sync_file->status))
> -		wake_up_all(&sync_file->wq);
> +	wake_up_all(&sync_file->wq);
>  }
>  
>  /**
> @@ -76,22 +75,18 @@ struct sync_file *sync_file_create(struct fence *fence)
>  {
>  	struct sync_file *sync_file;
>  
> -	sync_file = sync_file_alloc(offsetof(struct sync_file, cbs[1]));
> +	sync_file = sync_file_alloc();
>  	if (!sync_file)
>  		return NULL;
>  
> -	sync_file->num_fences = 1;
> -	atomic_set(&sync_file->status, 1);
> +	sync_file->fence = fence;
> +
>  	snprintf(sync_file->name, sizeof(sync_file->name), "%s-%s%llu-%d",
>  		 fence->ops->get_driver_name(fence),
>  		 fence->ops->get_timeline_name(fence), fence->context,
>  		 fence->seqno);
>  
> -	sync_file->cbs[0].fence = fence;
> -	sync_file->cbs[0].sync_file = sync_file;
> -	if (fence_add_callback(fence, &sync_file->cbs[0].cb,
> -			       fence_check_cb_func))
> -		atomic_dec(&sync_file->status);
> +	fence_add_callback(fence, &sync_file->cb, fence_check_cb_func);
>  
>  	return sync_file;
>  }
> @@ -121,14 +116,48 @@ err:
>  	return NULL;
>  }
>  
> -static void sync_file_add_pt(struct sync_file *sync_file, int *i,
> -			     struct fence *fence)
> +static int sync_file_set_fence(struct sync_file *sync_file,
> +			       struct fence **fences, int num_fences)
> +{
> +	struct fence_array *array;
> +
> +	/*
> +	 * The reference for the fences in the new sync_file and holded

s/holded/held/

> +	 * in add_fence() during the merge procedure, so for num_fences == 1
> +	 * we already own a new reference to the fence. For num_fence > 1
> +	 * we own the reference of the fence_array creation.
> +	 */

Thanks, that really does help understanding the reference handling.

> +	if (num_fences == 1) {
> +		sync_file->fence = fences[0];
> +	} else {
> +		array = fence_array_create(num_fences, fences,
> +					   fence_context_alloc(1), 1, false);
> +		if (!array)
> +			return -ENOMEM;
> +
> +		sync_file->fence = (struct fence *)array;

I'd prefer sync_file->fence = &array->base;

> +	}
> +
> +	return 0;
> +}
> +
> +static struct fence **get_fences(struct sync_file *sync_file, int *num_fences)
> +{
> +	if (fence_is_array(sync_file->fence)) {
> +		struct fence_array *array = to_fence_array(sync_file->fence);
> +		*num_fences = array->num_fences;
> +		return array->fences;
> +	} else {
> +		*num_fences = 1;
> +		return &sync_file->fence;
> +	}
> +}
> +
> +static void add_fence(struct fence **fences, int *i, struct fence *fence)
>  {
> -	sync_file->cbs[*i].fence = fence;
> -	sync_file->cbs[*i].sync_file = sync_file;
> +	fences[*i] = fence;
>  
> -	if (!fence_add_callback(fence, &sync_file->cbs[*i].cb,
> -				fence_check_cb_func)) {
> +	if (!fence_is_signaled(fence)) {
>  		fence_get(fence);
>  		(*i)++;
>  	}
> @@ -147,16 +176,21 @@ static void sync_file_add_pt(struct sync_file *sync_file, int *i,
>  static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
>  					 struct sync_file *b)
>  {
> -	int num_fences = a->num_fences + b->num_fences;
>  	struct sync_file *sync_file;
> -	int i, i_a, i_b;
> -	unsigned long size = offsetof(struct sync_file, cbs[num_fences]);
> +	struct fence **fences, **a_fences, **b_fences;
> +	int i, i_a, i_b, num_fences, a_num_fences, b_num_fences;
>  
> -	sync_file = sync_file_alloc(size);
> +	sync_file = sync_file_alloc();
>  	if (!sync_file)
>  		return NULL;
>  
> -	atomic_set(&sync_file->status, num_fences);
> +	a_fences = get_fences(a, &a_num_fences);
> +	b_fences = get_fences(b, &b_num_fences);
> +	num_fences = a_num_fences + b_num_fences;

Do we need overflow paranoia here? Probably, as this is on a user ioctl
path.

> +
> +	fences = kcalloc(num_fences, sizeof(*fences), GFP_KERNEL);
> +	if (!fences)
> +		goto err;
>  
>  	/*
>  	 * Assume sync_file a and b are both ordered and have no
> @@ -165,55 +199,66 @@ static struct sync_file *sync_file_merge(const char *name, struct sync_file *a,
>  	 * If a sync_file can only be created with sync_file_merge
>  	 * and sync_file_create, this is a reasonable assumption.
>  	 */
> -	for (i = i_a = i_b = 0; i_a < a->num_fences && i_b < b->num_fences; ) {
> -		struct fence *pt_a = a->cbs[i_a].fence;
> -		struct fence *pt_b = b->cbs[i_b].fence;
> +	for (i = i_a = i_b = 0; i_a < a_num_fences && i_b < b_num_fences; ) {
> +		struct fence *pt_a = a_fences[i_a];
> +		struct fence *pt_b = b_fences[i_b];
>  
>  		if (pt_a->context < pt_b->context) {
> -			sync_file_add_pt(sync_file, &i, pt_a);
> +			add_fence(fences, &i, pt_a);
>  
>  			i_a++;
>  		} else if (pt_a->context > pt_b->context) {
> -			sync_file_add_pt(sync_file, &i, pt_b);
> +			add_fence(fences, &i, pt_b);
>  
>  			i_b++;
>  		} else {
>  			if (pt_a->seqno - pt_b->seqno <= INT_MAX)
> -				sync_file_add_pt(sync_file, &i, pt_a);
> +				add_fence(fences, &i, pt_a);
>  			else
> -				sync_file_add_pt(sync_file, &i, pt_b);
> +				add_fence(fences, &i, pt_b);
>  
>  			i_a++;
>  			i_b++;
>  		}
>  	}
>  
> -	for (; i_a < a->num_fences; i_a++)
> -		sync_file_add_pt(sync_file, &i, a->cbs[i_a].fence);
> +	for (; i_a < a_num_fences; i_a++)
> +		add_fence(fences, &i, a_fences[i_a]);
> +
> +	for (; i_b < b_num_fences; i_b++)
> +		add_fence(fences, &i, b_fences[i_b]);
> +
> +	if (num_fences > i) {
> +		fences = krealloc(fences, i * sizeof(**fences),

Hmm, still too large, sizeof(*fences);

Looking pretty good. Putting fence-array to good use.

With those minor tweaks,
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel