Re: [PATCH v14 08/56] media: videobuf2: Use vb2_get_num_buffers() helper

From: Benjamin Gaignard
Date: Wed Nov 08 2023 - 08:22:33 EST



Le 08/11/2023 à 10:42, Tomasz Figa a écrit :
On Tue, Oct 31, 2023 at 05:30:16PM +0100, Benjamin Gaignard wrote:
Stop using queue num_buffers field directly, instead use
vb2_get_num_buffers().
This prepares for the future 'delete buffers' feature where there are
holes in the buffer indices.

Signed-off-by: Benjamin Gaignard <benjamin.gaignard@xxxxxxxxxxxxx>
---
.../media/common/videobuf2/videobuf2-core.c | 92 +++++++++++--------
.../media/common/videobuf2/videobuf2-v4l2.c | 4 +-
2 files changed, 54 insertions(+), 42 deletions(-)

diff --git a/drivers/media/common/videobuf2/videobuf2-core.c b/drivers/media/common/videobuf2/videobuf2-core.c
index b406a30a9b35..c5c5ae4d213d 100644
--- a/drivers/media/common/videobuf2/videobuf2-core.c
+++ b/drivers/media/common/videobuf2/videobuf2-core.c
@@ -444,13 +444,14 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum vb2_memory memory,
unsigned int num_buffers, unsigned int num_planes,
const unsigned plane_sizes[VB2_MAX_PLANES])
{
+ unsigned int q_num_buffers = vb2_get_num_buffers(q);
unsigned int buffer, plane;
struct vb2_buffer *vb;
int ret;
/* Ensure that q->num_buffers+num_buffers is below VB2_MAX_FRAME */
num_buffers = min_t(unsigned int, num_buffers,
- VB2_MAX_FRAME - q->num_buffers);
+ VB2_MAX_FRAME - q_num_buffers);
I guess it's safe in this specific situation, but was there any reason
behind not just calling vb2_get_num_buffers() directly here?

for (buffer = 0; buffer < num_buffers; ++buffer) {
/* Allocate vb2 buffer structures */
@@ -470,7 +471,7 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum vb2_memory memory,
vb->planes[plane].min_length = plane_sizes[plane];
}
- vb2_queue_add_buffer(q, vb, q->num_buffers + buffer);
+ vb2_queue_add_buffer(q, vb, q_num_buffers + buffer);
In this case it should also be fine, but actually now this is a loop and if
somone doesn't know what the other code in the loop does, one could be
concerned that the num buffers actually could have changed, but we still
use the cached one that we got at the beginning of the function.

(Ideally I'd imagine vb2_queue_add_buffer() to append the buffer
at the end of the queue and increment the num_buffers internally, but it
doesn't have to happen now, as this series is already quite complex...)

That will be the case later in the series when I replace num_buffers field
by a bitmap. Until that I prefer to limit the changes in this loop.


call_void_bufop(q, init_buffer, vb);
/* Allocate video buffer memory for the MMAP type */
[snip]
@@ -2513,7 +2519,8 @@ void vb2_core_queue_release(struct vb2_queue *q)
__vb2_cleanup_fileio(q);
__vb2_queue_cancel(q);
mutex_lock(&q->mmap_lock);
- __vb2_queue_free(q, q->num_buffers);
+ __vb2_queue_free(q, vb2_get_num_buffers(q));
+ q->num_buffers = 0;
Unrelated change?

No because I found a case where q->num_buffers wasn't correctly reset while testing.


mutex_unlock(&q->mmap_lock);
}
EXPORT_SYMBOL_GPL(vb2_core_queue_release);
@@ -2542,7 +2549,7 @@ __poll_t vb2_core_poll(struct vb2_queue *q, struct file *file,
/*
* Start file I/O emulator only if streaming API has not been used yet.
*/
- if (q->num_buffers == 0 && !vb2_fileio_is_active(q)) {
+ if (vb2_get_num_buffers(q) == 0 && !vb2_fileio_is_active(q)) {
if (!q->is_output && (q->io_modes & VB2_READ) &&
(req_events & (EPOLLIN | EPOLLRDNORM))) {
if (__vb2_init_fileio(q, 1))
@@ -2580,7 +2587,7 @@ __poll_t vb2_core_poll(struct vb2_queue *q, struct file *file,
* For output streams you can call write() as long as there are fewer
* buffers queued than there are buffers available.
*/
- if (q->is_output && q->fileio && q->queued_count < q->num_buffers)
+ if (q->is_output && q->fileio && q->queued_count < vb2_get_num_buffers(q))
return EPOLLOUT | EPOLLWRNORM;
if (list_empty(&q->done_list)) {
@@ -2629,8 +2636,8 @@ struct vb2_fileio_buf {
* struct vb2_fileio_data - queue context used by file io emulator
*
* @cur_index: the index of the buffer currently being read from or
- * written to. If equal to q->num_buffers then a new buffer
- * must be dequeued.
+ * written to. If equal to number of already queued buffers
+ * then a new buffer must be dequeued.
Hmm, that's a significant meaning change compared to the original text. Is
it indended?

Does "If equal to number of buffers in the vb2_queue then a new buffer must be dequeued."
sound better for you ?


* @initial_index: in the read() case all buffers are queued up immediately
* in __vb2_init_fileio() and __vb2_perform_fileio() just cycles
* buffers. However, in the write() case no buffers are initially
@@ -2640,7 +2647,7 @@ struct vb2_fileio_buf {
* buffers. This means that initially __vb2_perform_fileio()
* needs to know what buffer index to use when it is queuing up
* the buffers for the first time. That initial index is stored
- * in this field. Once it is equal to q->num_buffers all
+ * in this field. Once it is equal to num_buffers all
It's not clear what num_buffers means here. Would it make sense to instead
say "number of buffers in the vb2_queue"?

Yes I will change that


* available buffers have been queued and __vb2_perform_fileio()
* should start the normal dequeue/queue cycle.
*
@@ -2690,7 +2697,7 @@ static int __vb2_init_fileio(struct vb2_queue *q, int read)
/*
* Check if streaming api has not been already activated.
*/
- if (q->streaming || q->num_buffers > 0)
+ if (q->streaming || vb2_get_num_buffers(q) > 0)
return -EBUSY;
/*
@@ -2740,7 +2747,7 @@ static int __vb2_init_fileio(struct vb2_queue *q, int read)
/*
* Get kernel address of each buffer.
*/
- for (i = 0; i < q->num_buffers; i++) {
+ for (i = 0; i < vb2_get_num_buffers(q); i++) {
/* vb can never be NULL when using fileio. */
vb = vb2_get_buffer(q, i);
@@ -2759,18 +2766,23 @@ static int __vb2_init_fileio(struct vb2_queue *q, int read)
/*
* Queue all buffers.
*/
- for (i = 0; i < q->num_buffers; i++) {
- ret = vb2_core_qbuf(q, q->bufs[i], NULL, NULL);
+ for (i = 0; i < vb2_get_num_buffers(q); i++) {
+ struct vb2_buffer *vb2 = vb2_get_buffer(q, i);
+
+ if (!vb2)
+ continue;
+
+ ret = vb2_core_qbuf(q, vb2, NULL, NULL);
if (ret)
goto err_reqbufs;
fileio->bufs[i].queued = 1;
}
Doesn't this part belong to the previous patch that changes q->bufs[x] to
vb2_get_buffer()?

Yes I will change that too.


/*
* All buffers have been queued, so mark that by setting
- * initial_index to q->num_buffers
+ * initial_index to num_buffers
What num_buffers?

I will use your wording: "the number of buffers in the vb2_queue"


*/
- fileio->initial_index = q->num_buffers;
- fileio->cur_index = q->num_buffers;
+ fileio->initial_index = vb2_get_num_buffers(q);
+ fileio->cur_index = fileio->initial_index;
}
/*
@@ -2964,12 +2976,12 @@ static size_t __vb2_perform_fileio(struct vb2_queue *q, char __user *data, size_
* If we are queuing up buffers for the first time, then
* increase initial_index by one.
*/
- if (fileio->initial_index < q->num_buffers)
+ if (fileio->initial_index < vb2_get_num_buffers(q))
fileio->initial_index++;
/*
* The next buffer to use is either a buffer that's going to be
- * queued for the first time (initial_index < q->num_buffers)
- * or it is equal to q->num_buffers, meaning that the next
+ * queued for the first time (initial_index < num_buffers)
+ * or it is equal to num_buffers, meaning that the next
What num_buffers?

Same here


Best regards,
Tomasz
_______________________________________________
Kernel mailing list -- kernel@xxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to kernel-leave@xxxxxxxxxxxxxxxxxxxxx