mbox series

[v19,0/9] Add DELETE_BUF ioctl

Message ID 20240206080219.11951-1-benjamin.gaignard@collabora.com
Headers show
Series Add DELETE_BUF ioctl | expand

Message

Benjamin Gaignard Feb. 6, 2024, 8:02 a.m. UTC
Unlike when resolution change on keyframes, dynamic resolution change
on inter frames doesn't allow to do a stream off/on sequence because
it is need to keep all previous references alive to decode inter frames.
This constraint have two main problems:
- more memory consumption.
- more buffers in use.
To solve these issue this series introduce DELETE_BUFS ioctl and remove
the 32 buffers limit per queue.

VP9 conformance tests using fluster give a score of 210/305.
The 23 of the 24 resize inter tests (vp90-2-21-resize_inter_* files) are ok
but require to use postprocessor.

Kernel branch is available here:
https://gitlab.collabora.com/benjamin.gaignard/for-upstream/-/commits/remove_vb2_queue_limit_v19

GStreamer branch to use DELETE_BUF ioctl and testing dynamic resolution
change is here:
https://gitlab.freedesktop.org/benjamin.gaignard1/gstreamer/-/commits/VP9_drc

changes in version 19:
- Fix typo in commit message.
- Fix ioctl domentation
- Rework q->is_busy patch following Hans's comments
- Change where DELETE_BUFS is enabled

changes in version 18:
- rebased on top of:
  https://patchwork.linuxtv.org/project/linux-media/patch/20240118121452.29151-1-benjamin.gaignard@collabora.com/
  https://patchwork.linuxtv.org/project/linux-media/patch/92975c06-d6e1-4ba6-8d03-b2ef0b199c21@xs4all.nl/
- Add a patch to update vb2_is_busy() logic.
- fix __vb2_queue_alloc() parameters descriptions.
- rework bitmap free range finding loop
- remove per queue capability flag.
- rework v4l_delete_bufs() to check if VIDIOC_CREATE_BUFS is enabled
  and if vidioc_delete_bufs pointer is valid.
- update documentation.
- Direclty use vb2_core_delete_bufs() in v4l2_m2m_ioctl_delete_bufs().
  Remove useless static functions.

changes in version 17:
- rebased on top of:
  https://patchwork.linuxtv.org/project/linux-media/patch/20240118121452.29151-1-benjamin.gaignard@collabora.com/
  https://patchwork.linuxtv.org/project/linux-media/patch/92975c06-d6e1-4ba6-8d03-b2ef0b199c21@xs4all.nl/
- rewrite min_reqbufs_allocation field documentation.
- rewrite vb2_core_create_bufs() first_index parameter documentation.
- rework bitmap allocation usage in __vb2_queue_alloc().
- remove useless i < q->max_num_buffers checks.
- rework DELETE_BUFS documentation.
- change split between patch 7 and patch 8
- v4l2_m2m_delete_bufs() is now a static function.

 
Benjamin Gaignard (9):
  media: videobuf2: Update vb2_is_busy() logic
  videobuf2: Add min_reqbufs_allocation field to vb2_queue structure
  media: test-drivers: Set REQBUFS minimum number of buffers
  media: core: Rework how create_buf index returned value is computed
  media: core: Add bitmap manage bufs array entries
  media: core: Free range of buffers
  media: v4l2: Add DELETE_BUFS ioctl
  media: v4l2: Add mem2mem helpers for DELETE_BUFS ioctl
  media: verisilicon: Support deleting buffers on capture queue

 .../userspace-api/media/v4l/user-func.rst     |   1 +
 .../media/v4l/vidioc-delete-bufs.rst          |  79 +++++++
 .../media/common/videobuf2/videobuf2-core.c   | 223 ++++++++++++------
 .../media/common/videobuf2/videobuf2-v4l2.c   |  26 +-
 .../media/platform/verisilicon/hantro_v4l2.c  |   1 +
 .../media/test-drivers/vicodec/vicodec-core.c |   1 +
 drivers/media/test-drivers/vim2m.c            |   1 +
 .../media/test-drivers/vimc/vimc-capture.c    |   3 +-
 drivers/media/test-drivers/visl/visl-video.c  |   1 +
 drivers/media/test-drivers/vivid/vivid-core.c |   5 +-
 drivers/media/v4l2-core/v4l2-dev.c            |   3 +
 drivers/media/v4l2-core/v4l2-ioctl.c          |  24 ++
 drivers/media/v4l2-core/v4l2-mem2mem.c        |  10 +
 include/media/v4l2-ioctl.h                    |   4 +
 include/media/v4l2-mem2mem.h                  |   2 +
 include/media/videobuf2-core.h                |  52 +++-
 include/media/videobuf2-v4l2.h                |   2 +
 include/uapi/linux/videodev2.h                |  16 ++
 18 files changed, 369 insertions(+), 85 deletions(-)
 create mode 100644 Documentation/userspace-api/media/v4l/vidioc-delete-bufs.rst

Comments

Hans Verkuil Feb. 7, 2024, 10:28 a.m. UTC | #1
On 06/02/2024 09:58, Hans Verkuil wrote:
> On 06/02/2024 09:02, Benjamin Gaignard wrote:
>> Unlike when resolution change on keyframes, dynamic resolution change
>> on inter frames doesn't allow to do a stream off/on sequence because
>> it is need to keep all previous references alive to decode inter frames.
>> This constraint have two main problems:
>> - more memory consumption.
>> - more buffers in use.
>> To solve these issue this series introduce DELETE_BUFS ioctl and remove
>> the 32 buffers limit per queue.
> 
> This v19 looks good. There are three outstanding issues that I need to take a
> look at:
> 
> 1) Can we still signal support for DELETE_BUFS in the V4L2_BUF_CAP_ caps?
>    It would be nice to have, but I'm not sure if and how that can be done.
> 
> 2) I want to take another look at providing a next version of the VIDIOC_CREATE_BUFS
>    ioctl and how that would possibly influence the design of DELETE_BUFS.
>    Just to make sure we're not blocking anything or making life harder.

So I just tried creating a new version of the VIDIOC_CREATE_BUFS ioctl
that is greatly simplified. This just updates the header, no real code yet:

Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
diff --git a/include/uapi/linux/videodev2.h b/include/uapi/linux/videodev2.h
index 03443833aaaa..a7398e4c85e7 100644
--- a/include/uapi/linux/videodev2.h
+++ b/include/uapi/linux/videodev2.h
@@ -2624,6 +2624,39 @@ struct v4l2_create_buffers {
 	__u32			reserved[5];
 };

+/**
+ * struct v4l2_create_buffers_v2 - VIDIOC_CREATE_BUFFERS argument
+ * @type:	enum v4l2_buf_type
+ * @memory:	enum v4l2_memory; buffer memory type
+ * @count:	entry: number of requested buffers,
+ *		return: number of created buffers
+ * @num_planes:	requested number of planes for each buffer
+ * @sizes:	requested plane sizes for each buffer
+ * @start_index:on return, index of the first created buffer
+ * @total_count:on return, the total number of allocated buffers
+ * @capabilities: capabilities of this buffer type.
+ * @flags:	additional buffer management attributes (ignored unless the
+ *		queue has V4L2_BUF_CAP_SUPPORTS_MMAP_CACHE_HINTS capability
+ *		and configured for MMAP streaming I/O).
+ * @max_num_buffers: if V4L2_BUF_CAP_SUPPORTS_MAX_NUM_BUFFERS capability flag is set
+ *		this field indicate the maximum possible number of buffers
+ *		for this queue.
+ * @reserved:	future extensions
+ */
+struct v4l2_create_buffers_v2 {
+	__u32			type;
+	__u32			memory;
+	__u32			count;
+	__u32			num_planes;
+	__u32			size[VIDEO_MAX_PLANES];
+	__u32			start_index;
+	__u32			total_count;
+	__u32			capabilities;
+	__u32			flags;
+	__u32			max_num_buffers;
+	__u32			reserved[7];
+};
+
 /**
  * struct v4l2_delete_buffers - VIDIOC_DELETE_BUFS argument
  * @index:	the first buffer to be deleted
@@ -2738,6 +2771,7 @@ struct v4l2_delete_buffers {

 #define VIDIOC_QUERY_EXT_CTRL	_IOWR('V', 103, struct v4l2_query_ext_ctrl)
 #define VIDIOC_DELETE_BUFS	_IOWR('V', 104, struct v4l2_delete_buffers)
+#define VIDIOC_CREATE_BUFFERS	_IOWR('V', 105, struct v4l2_create_buffers_v2)


 /* Reminder: when adding new ioctls please add support for them to


Sadly, struct v4l2_create_buffers was already used for the existing
VIDIOC_CREATE_BUFS (I wish it was called v4l2_create_bufs...), so I did
have to add a _v2 suffix. Compared to the old struct it just moves the
type, num_planes and sizes from v4l2_format into the new struct and drops
the format field. Note that those fields are the only v4l2_format fields
that VIDIOC_CREATE_BUFS used, so getting rid of the format makes live
much easier. I also renamed 'index' to 'start_index' and added a new 'total_count'
field to report the total number of buffers.

The reason for adding 'total_count' is that VIDIOC_CREATE_BUFS with
count == 0 would report the total number of buffers in the 'index' field,
which was rather odd. Adding a specific field for this purpose is nicer.

One thing that might impact your series is the name of the ioctls.

We have:

VIDIOC_CREATE_BUFS/v4l2_create_buffers
VIDIOC_DELETE_BUFS/v4l2_delete_buffers
VIDIOC_CREATE_BUFFERS/v4l2_create_buffers_v2 (TBD)

I'm wondering if VIDIOC_DELETE_BUFS should be renamed to
VIDIOC_DELETE_BUFFERS, that would make it more consistent with
the proposed VIDIOC_CREATE_BUFFERS.

Alternatively, instead of calling it VIDIOC_CREATE_BUFFERS we
might call it VIDIOC_CREATE_BUFS_V2.

Or perhaps some other naming convention?

Ideas are welcome.

Regards,

	Hans
Hans Verkuil Feb. 7, 2024, 11:20 a.m. UTC | #2
On 07/02/2024 12:15, Benjamin Gaignard wrote:
> 
> Le 07/02/2024 à 10:12, Hans Verkuil a écrit :
>> Hi Benjamin,
>>
>> On 06/02/2024 09:58, Hans Verkuil wrote:
>>> On 06/02/2024 09:02, Benjamin Gaignard wrote:
>>>> Unlike when resolution change on keyframes, dynamic resolution change
>>>> on inter frames doesn't allow to do a stream off/on sequence because
>>>> it is need to keep all previous references alive to decode inter frames.
>>>> This constraint have two main problems:
>>>> - more memory consumption.
>>>> - more buffers in use.
>>>> To solve these issue this series introduce DELETE_BUFS ioctl and remove
>>>> the 32 buffers limit per queue.
>>> This v19 looks good. There are three outstanding issues that I need to take a
>>> look at:
>>>
>>> 1) Can we still signal support for DELETE_BUFS in the V4L2_BUF_CAP_ caps?
>>>     It would be nice to have, but I'm not sure if and how that can be done.
>> So, I came up with the following patch to add back the V4L2_BUF_CAP_SUPPORTS_DELETE_BUFS
>> capability. If the DELETE_BUFS ioctl is valid, then it sets this capability
>> before calling vidioc_reqbufs or vidioc_create_bufs. So right now it will set
>> this for any queue. If we ever want to disable this for a specific queue, then
>> either the driver has to override these two ops and clear the flag, or a new
>> vb2_queue flag (e.g. disable_delete_bufs) is added and vb2_set_flags_and_caps()
>> will clear that capability based on that flag.
>>
>> In any case, for now just set it for both queues by default.
>>
>> If you agree that this is a good way to proceed, then can you incorporate this
>> into a v20? You can add the documentation for this cap from the v17 version.
> 
> Do you want it to be a separate patch or included in the patch introducing DELETE_BUFS ioctl ?

Up to you, whatever makes the most sense.

Regards,

	Hans

> 
>>
>> Regards,
>>
>>     Hans
>>
>> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
>> ---
>> diff --git a/drivers/media/common/videobuf2/videobuf2-v4l2.c b/drivers/media/common/videobuf2/videobuf2-v4l2.c
>> index 8e437104f9c1..64f2d662d068 100644
>> --- a/drivers/media/common/videobuf2/videobuf2-v4l2.c
>> +++ b/drivers/media/common/videobuf2/videobuf2-v4l2.c
>> @@ -685,7 +685,7 @@ static void vb2_set_flags_and_caps(struct vb2_queue *q, u32 memory,
>>           *flags &= V4L2_MEMORY_FLAG_NON_COHERENT;
>>       }
>>
>> -    *caps = V4L2_BUF_CAP_SUPPORTS_ORPHANED_BUFS;
>> +    *caps |= V4L2_BUF_CAP_SUPPORTS_ORPHANED_BUFS;
>>       if (q->io_modes & VB2_MMAP)
>>           *caps |= V4L2_BUF_CAP_SUPPORTS_MMAP;
>>       if (q->io_modes & VB2_USERPTR)
>> diff --git a/drivers/media/v4l2-core/v4l2-ioctl.c b/drivers/media/v4l2-core/v4l2-ioctl.c
>> index a172d33edd19..45bc705e171e 100644
>> --- a/drivers/media/v4l2-core/v4l2-ioctl.c
>> +++ b/drivers/media/v4l2-core/v4l2-ioctl.c
>> @@ -2100,6 +2100,7 @@ static int v4l_overlay(const struct v4l2_ioctl_ops *ops,
>>   static int v4l_reqbufs(const struct v4l2_ioctl_ops *ops,
>>                   struct file *file, void *fh, void *arg)
>>   {
>> +    struct video_device *vfd = video_devdata(file);
>>       struct v4l2_requestbuffers *p = arg;
>>       int ret = check_fmt(file, p->type);
>>
>> @@ -2108,6 +2109,9 @@ static int v4l_reqbufs(const struct v4l2_ioctl_ops *ops,
>>
>>       memset_after(p, 0, flags);
>>
>> +    if (is_valid_ioctl(vfd, VIDIOC_DELETE_BUFS))
>> +        p->capabilities = V4L2_BUF_CAP_SUPPORTS_DELETE_BUFS;
>> +
>>       return ops->vidioc_reqbufs(file, fh, p);
>>   }
>>
>> @@ -2141,6 +2145,7 @@ static int v4l_dqbuf(const struct v4l2_ioctl_ops *ops,
>>   static int v4l_create_bufs(const struct v4l2_ioctl_ops *ops,
>>                   struct file *file, void *fh, void *arg)
>>   {
>> +    struct video_device *vfd = video_devdata(file);
>>       struct v4l2_create_buffers *create = arg;
>>       int ret = check_fmt(file, create->format.type);
>>
>> @@ -2151,6 +2156,9 @@ static int v4l_create_bufs(const struct v4l2_ioctl_ops *ops,
>>
>>       v4l_sanitize_format(&create->format);
>>
>> +    if (is_valid_ioctl(vfd, VIDIOC_DELETE_BUFS))
>> +        create->capabilities = V4L2_BUF_CAP_SUPPORTS_DELETE_BUFS;
>> +
>>       ret = ops->vidioc_create_bufs(file, fh, create);
>>
>>       if (create->format.type == V4L2_BUF_TYPE_VIDEO_CAPTURE ||
>> diff --git a/include/uapi/linux/videodev2.h b/include/uapi/linux/videodev2.h
>> index 03443833aaaa..da307f46f903 100644
>> --- a/include/uapi/linux/videodev2.h
>> +++ b/include/uapi/linux/videodev2.h
>> @@ -1036,6 +1036,7 @@ struct v4l2_requestbuffers {
>>   #define V4L2_BUF_CAP_SUPPORTS_M2M_HOLD_CAPTURE_BUF    (1 << 5)
>>   #define V4L2_BUF_CAP_SUPPORTS_MMAP_CACHE_HINTS        (1 << 6)
>>   #define V4L2_BUF_CAP_SUPPORTS_MAX_NUM_BUFFERS        (1 << 7)
>> +#define V4L2_BUF_CAP_SUPPORTS_DELETE_BUFS        (1 << 8)
>>
>>   /**
>>    * struct v4l2_plane - plane info for multi-planar buffers
>>
>>
>>
Benjamin Gaignard Feb. 7, 2024, 11:25 a.m. UTC | #3
Le 07/02/2024 à 11:28, Hans Verkuil a écrit :
> On 06/02/2024 09:58, Hans Verkuil wrote:
>> On 06/02/2024 09:02, Benjamin Gaignard wrote:
>>> Unlike when resolution change on keyframes, dynamic resolution change
>>> on inter frames doesn't allow to do a stream off/on sequence because
>>> it is need to keep all previous references alive to decode inter frames.
>>> This constraint have two main problems:
>>> - more memory consumption.
>>> - more buffers in use.
>>> To solve these issue this series introduce DELETE_BUFS ioctl and remove
>>> the 32 buffers limit per queue.
>> This v19 looks good. There are three outstanding issues that I need to take a
>> look at:
>>
>> 1) Can we still signal support for DELETE_BUFS in the V4L2_BUF_CAP_ caps?
>>     It would be nice to have, but I'm not sure if and how that can be done.
>>
>> 2) I want to take another look at providing a next version of the VIDIOC_CREATE_BUFS
>>     ioctl and how that would possibly influence the design of DELETE_BUFS.
>>     Just to make sure we're not blocking anything or making life harder.
> So I just tried creating a new version of the VIDIOC_CREATE_BUFS ioctl
> that is greatly simplified. This just updates the header, no real code yet:
>
> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> ---
> diff --git a/include/uapi/linux/videodev2.h b/include/uapi/linux/videodev2.h
> index 03443833aaaa..a7398e4c85e7 100644
> --- a/include/uapi/linux/videodev2.h
> +++ b/include/uapi/linux/videodev2.h
> @@ -2624,6 +2624,39 @@ struct v4l2_create_buffers {
>   	__u32			reserved[5];
>   };
>
> +/**
> + * struct v4l2_create_buffers_v2 - VIDIOC_CREATE_BUFFERS argument
> + * @type:	enum v4l2_buf_type
> + * @memory:	enum v4l2_memory; buffer memory type
> + * @count:	entry: number of requested buffers,
> + *		return: number of created buffers
> + * @num_planes:	requested number of planes for each buffer
> + * @sizes:	requested plane sizes for each buffer
> + * @start_index:on return, index of the first created buffer
> + * @total_count:on return, the total number of allocated buffers
> + * @capabilities: capabilities of this buffer type.
> + * @flags:	additional buffer management attributes (ignored unless the
> + *		queue has V4L2_BUF_CAP_SUPPORTS_MMAP_CACHE_HINTS capability
> + *		and configured for MMAP streaming I/O).
> + * @max_num_buffers: if V4L2_BUF_CAP_SUPPORTS_MAX_NUM_BUFFERS capability flag is set
> + *		this field indicate the maximum possible number of buffers
> + *		for this queue.
> + * @reserved:	future extensions
> + */
> +struct v4l2_create_buffers_v2 {
> +	__u32			type;
> +	__u32			memory;
> +	__u32			count;
> +	__u32			num_planes;
> +	__u32			size[VIDEO_MAX_PLANES];
> +	__u32			start_index;
> +	__u32			total_count;
> +	__u32			capabilities;
> +	__u32			flags;
> +	__u32			max_num_buffers;
> +	__u32			reserved[7];
> +};
> +
>   /**
>    * struct v4l2_delete_buffers - VIDIOC_DELETE_BUFS argument
>    * @index:	the first buffer to be deleted
> @@ -2738,6 +2771,7 @@ struct v4l2_delete_buffers {
>
>   #define VIDIOC_QUERY_EXT_CTRL	_IOWR('V', 103, struct v4l2_query_ext_ctrl)
>   #define VIDIOC_DELETE_BUFS	_IOWR('V', 104, struct v4l2_delete_buffers)
> +#define VIDIOC_CREATE_BUFFERS	_IOWR('V', 105, struct v4l2_create_buffers_v2)
>
>
>   /* Reminder: when adding new ioctls please add support for them to
>
>
> Sadly, struct v4l2_create_buffers was already used for the existing
> VIDIOC_CREATE_BUFS (I wish it was called v4l2_create_bufs...), so I did
> have to add a _v2 suffix. Compared to the old struct it just moves the
> type, num_planes and sizes from v4l2_format into the new struct and drops
> the format field. Note that those fields are the only v4l2_format fields
> that VIDIOC_CREATE_BUFS used, so getting rid of the format makes live
> much easier. I also renamed 'index' to 'start_index' and added a new 'total_count'
> field to report the total number of buffers.
>
> The reason for adding 'total_count' is that VIDIOC_CREATE_BUFS with
> count == 0 would report the total number of buffers in the 'index' field,
> which was rather odd. Adding a specific field for this purpose is nicer.
>
> One thing that might impact your series is the name of the ioctls.
>
> We have:
>
> VIDIOC_CREATE_BUFS/v4l2_create_buffers
> VIDIOC_DELETE_BUFS/v4l2_delete_buffers
> VIDIOC_CREATE_BUFFERS/v4l2_create_buffers_v2 (TBD)
>
> I'm wondering if VIDIOC_DELETE_BUFS should be renamed to
> VIDIOC_DELETE_BUFFERS, that would make it more consistent with
> the proposed VIDIOC_CREATE_BUFFERS.
>
> Alternatively, instead of calling it VIDIOC_CREATE_BUFFERS we
> might call it VIDIOC_CREATE_BUFS_V2.
>
> Or perhaps some other naming convention?

Maybe VIDIOC_NEW_BUFS/v4l2_new_buffers ?

I will wait until naming convention is fixed before send v20.

Regards,
Benjamin

>
> Ideas are welcome.
>
> Regards,
>
> 	Hans
>