[media] vb2: fix read/write regression
authorHans Verkuil <hans.verkuil@cisco.com>
Fri, 14 Feb 2014 09:46:50 +0000 (06:46 -0300)
committerMauro Carvalho Chehab <m.chehab@samsung.com>
Tue, 11 Mar 2014 09:56:36 +0000 (06:56 -0300)
Commit 88e268702bfba78448abd20a31129458707383aa ("vb2: Improve file I/O
emulation to handle buffers in any order") broke read/write support if
the size of the buffer being read/written is less than the size of the
image.

When the commit was tested originally I used qv4l2, which calls read()
with exactly the size of the image. But if you try 'cat /dev/video0'
then it will fail and typically hang after reading two buffers.

This patch fixes the behavior by adding a new cur_index field that
contains the index of the field currently being filled/read, or it
is num_buffers in which case a new buffer needs to be dequeued.

The old index field has been renamed to initial_index in order to be
a bit more descriptive.

This has been tested with both read and write.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Tested-by: Hans Verkuil <hans.verkuil@cisco.com>
Cc: Andy Walls <awalls@md.metrocast.net>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
drivers/media/v4l2-core/videobuf2-core.c

index 3c07534e9ba53234c473149d6070c44df432004c..dbc2b8ab8cdb1e50b7aa6752d424dc0db80b3067 100644 (file)
@@ -2310,6 +2310,22 @@ 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.
+ * @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
+ *             queued, instead whenever a buffer is full it is queued up by
+ *             __vb2_perform_fileio(). Only once all available buffers have
+ *             been queued up will __vb2_perform_fileio() start to dequeue
+ *             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
+ *             available buffers have been queued and __vb2_perform_fileio()
+ *             should start the normal dequeue/queue cycle.
+ *
  * vb2 provides a compatibility layer and emulator of file io (read and
  * write) calls on top of streaming API. For proper operation it required
  * this structure to save the driver state between each call of the read
@@ -2319,7 +2335,8 @@ struct vb2_fileio_data {
        struct v4l2_requestbuffers req;
        struct v4l2_buffer b;
        struct vb2_fileio_buf bufs[VIDEO_MAX_FRAME];
-       unsigned int index;
+       unsigned int cur_index;
+       unsigned int initial_index;
        unsigned int q_count;
        unsigned int dq_count;
        unsigned int flags;
@@ -2419,7 +2436,12 @@ static int __vb2_init_fileio(struct vb2_queue *q, int read)
                                goto err_reqbufs;
                        fileio->bufs[i].queued = 1;
                }
-               fileio->index = q->num_buffers;
+               /*
+                * All buffers have been queued, so mark that by setting
+                * initial_index to q->num_buffers
+                */
+               fileio->initial_index = q->num_buffers;
+               fileio->cur_index = q->num_buffers;
        }
 
        /*
@@ -2498,7 +2520,7 @@ static size_t __vb2_perform_fileio(struct vb2_queue *q, char __user *data, size_
        /*
         * Check if we need to dequeue the buffer.
         */
-       index = fileio->index;
+       index = fileio->cur_index;
        if (index >= q->num_buffers) {
                /*
                 * Call vb2_dqbuf to get buffer back.
@@ -2512,7 +2534,7 @@ static size_t __vb2_perform_fileio(struct vb2_queue *q, char __user *data, size_
                        return ret;
                fileio->dq_count += 1;
 
-               index = fileio->b.index;
+               fileio->cur_index = index = fileio->b.index;
                buf = &fileio->bufs[index];
 
                /*
@@ -2588,8 +2610,20 @@ static size_t __vb2_perform_fileio(struct vb2_queue *q, char __user *data, size_
                buf->queued = 1;
                buf->size = vb2_plane_size(q->bufs[index], 0);
                fileio->q_count += 1;
-               if (fileio->index < q->num_buffers)
-                       fileio->index++;
+               /*
+                * If we are queuing up buffers for the first time, then
+                * increase initial_index by one.
+                */
+               if (fileio->initial_index < q->num_buffers)
+                       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
+                * time we need to dequeue a buffer since we've now queued up
+                * all the 'first time' buffers.
+                */
+               fileio->cur_index = fileio->initial_index;
        }
 
        /*