6 Author: Jonathan Corbet <corbet@lwn.net>
12 The videobuf framework was deprecated in favor of videobuf2. Shouldn't
13 be used on new drivers.
18 The videobuf layer functions as a sort of glue layer between a V4L2 driver
19 and user space. It handles the allocation and management of buffers for
20 the storage of video frames. There is a set of functions which can be used
21 to implement many of the standard POSIX I/O system calls, including read(),
22 poll(), and, happily, mmap(). Another set of functions can be used to
23 implement the bulk of the V4L2 ioctl() calls related to streaming I/O,
24 including buffer allocation, queueing and dequeueing, and streaming
25 control. Using videobuf imposes a few design decisions on the driver
26 author, but the payback comes in the form of reduced code in the driver and
27 a consistent implementation of the V4L2 user-space API.
32 Not all video devices use the same kind of buffers. In fact, there are (at
33 least) three common variations:
35 - Buffers which are scattered in both the physical and (kernel) virtual
36 address spaces. (Almost) all user-space buffers are like this, but it
37 makes great sense to allocate kernel-space buffers this way as well when
38 it is possible. Unfortunately, it is not always possible; working with
39 this kind of buffer normally requires hardware which can do
40 scatter/gather DMA operations.
42 - Buffers which are physically scattered, but which are virtually
43 contiguous; buffers allocated with vmalloc(), in other words. These
44 buffers are just as hard to use for DMA operations, but they can be
45 useful in situations where DMA is not available but virtually-contiguous
46 buffers are convenient.
48 - Buffers which are physically contiguous. Allocation of this kind of
49 buffer can be unreliable on fragmented systems, but simpler DMA
50 controllers cannot deal with anything else.
52 Videobuf can work with all three types of buffers, but the driver author
53 must pick one at the outset and design the driver around that decision.
55 [It's worth noting that there's a fourth kind of buffer: "overlay" buffers
56 which are located within the system's video memory. The overlay
57 functionality is considered to be deprecated for most use, but it still
58 shows up occasionally in system-on-chip drivers where the performance
59 benefits merit the use of this technique. Overlay buffers can be handled
60 as a form of scattered buffer, but there are very few implementations in
61 the kernel and a description of this technique is currently beyond the
62 scope of this document.]
64 Data structures, callbacks, and initialization
65 ----------------------------------------------
67 Depending on which type of buffers are being used, the driver should
68 include one of the following files:
72 <media/videobuf-dma-sg.h> /* Physically scattered */
73 <media/videobuf-vmalloc.h> /* vmalloc() buffers */
74 <media/videobuf-dma-contig.h> /* Physically contiguous */
76 The driver's data structure describing a V4L2 device should include a
77 struct videobuf_queue instance for the management of the buffer queue,
78 along with a list_head for the queue of available buffers. There will also
79 need to be an interrupt-safe spinlock which is used to protect (at least)
82 The next step is to write four simple callbacks to help videobuf deal with
83 the management of buffers:
87 struct videobuf_queue_ops {
88 int (*buf_setup)(struct videobuf_queue *q,
89 unsigned int *count, unsigned int *size);
90 int (*buf_prepare)(struct videobuf_queue *q,
91 struct videobuf_buffer *vb,
92 enum v4l2_field field);
93 void (*buf_queue)(struct videobuf_queue *q,
94 struct videobuf_buffer *vb);
95 void (*buf_release)(struct videobuf_queue *q,
96 struct videobuf_buffer *vb);
99 buf_setup() is called early in the I/O process, when streaming is being
100 initiated; its purpose is to tell videobuf about the I/O stream. The count
101 parameter will be a suggested number of buffers to use; the driver should
102 check it for rationality and adjust it if need be. As a practical rule, a
103 minimum of two buffers are needed for proper streaming, and there is
104 usually a maximum (which cannot exceed 32) which makes sense for each
105 device. The size parameter should be set to the expected (maximum) size
106 for each frame of data.
108 Each buffer (in the form of a struct videobuf_buffer pointer) will be
109 passed to buf_prepare(), which should set the buffer's size, width, height,
110 and field fields properly. If the buffer's state field is
111 VIDEOBUF_NEEDS_INIT, the driver should pass it to:
115 int videobuf_iolock(struct videobuf_queue* q, struct videobuf_buffer *vb,
116 struct v4l2_framebuffer *fbuf);
118 Among other things, this call will usually allocate memory for the buffer.
119 Finally, the buf_prepare() function should set the buffer's state to
122 When a buffer is queued for I/O, it is passed to buf_queue(), which should
123 put it onto the driver's list of available buffers and set its state to
124 VIDEOBUF_QUEUED. Note that this function is called with the queue spinlock
125 held; if it tries to acquire it as well things will come to a screeching
126 halt. Yes, this is the voice of experience. Note also that videobuf may
127 wait on the first buffer in the queue; placing other buffers in front of it
128 could again gum up the works. So use list_add_tail() to enqueue buffers.
130 Finally, buf_release() is called when a buffer is no longer intended to be
131 used. The driver should ensure that there is no I/O active on the buffer,
132 then pass it to the appropriate free routine(s):
136 /* Scatter/gather drivers */
137 int videobuf_dma_unmap(struct videobuf_queue *q,
138 struct videobuf_dmabuf *dma);
139 int videobuf_dma_free(struct videobuf_dmabuf *dma);
141 /* vmalloc drivers */
142 void videobuf_vmalloc_free (struct videobuf_buffer *buf);
144 /* Contiguous drivers */
145 void videobuf_dma_contig_free(struct videobuf_queue *q,
146 struct videobuf_buffer *buf);
148 One way to ensure that a buffer is no longer under I/O is to pass it to:
152 int videobuf_waiton(struct videobuf_buffer *vb, int non_blocking, int intr);
154 Here, vb is the buffer, non_blocking indicates whether non-blocking I/O
155 should be used (it should be zero in the buf_release() case), and intr
156 controls whether an interruptible wait is used.
161 At this point, much of the work is done; much of the rest is slipping
162 videobuf calls into the implementation of the other driver callbacks. The
163 first step is in the open() function, which must initialize the
164 videobuf queue. The function to use depends on the type of buffer used:
168 void videobuf_queue_sg_init(struct videobuf_queue *q,
169 struct videobuf_queue_ops *ops,
172 enum v4l2_buf_type type,
173 enum v4l2_field field,
177 void videobuf_queue_vmalloc_init(struct videobuf_queue *q,
178 struct videobuf_queue_ops *ops,
181 enum v4l2_buf_type type,
182 enum v4l2_field field,
186 void videobuf_queue_dma_contig_init(struct videobuf_queue *q,
187 struct videobuf_queue_ops *ops,
190 enum v4l2_buf_type type,
191 enum v4l2_field field,
195 In each case, the parameters are the same: q is the queue structure for the
196 device, ops is the set of callbacks as described above, dev is the device
197 structure for this video device, irqlock is an interrupt-safe spinlock to
198 protect access to the data structures, type is the buffer type used by the
199 device (cameras will use V4L2_BUF_TYPE_VIDEO_CAPTURE, for example), field
200 describes which field is being captured (often V4L2_FIELD_NONE for
201 progressive devices), msize is the size of any containing structure used
202 around struct videobuf_buffer, and priv is a private data pointer which
203 shows up in the priv_data field of struct videobuf_queue. Note that these
204 are void functions which, evidently, are immune to failure.
206 V4L2 capture drivers can be written to support either of two APIs: the
207 read() system call and the rather more complicated streaming mechanism. As
208 a general rule, it is necessary to support both to ensure that all
209 applications have a chance of working with the device. Videobuf makes it
210 easy to do that with the same code. To implement read(), the driver need
211 only make a call to one of:
215 ssize_t videobuf_read_one(struct videobuf_queue *q,
216 char __user *data, size_t count,
217 loff_t *ppos, int nonblocking);
219 ssize_t videobuf_read_stream(struct videobuf_queue *q,
220 char __user *data, size_t count,
221 loff_t *ppos, int vbihack, int nonblocking);
223 Either one of these functions will read frame data into data, returning the
224 amount actually read; the difference is that videobuf_read_one() will only
225 read a single frame, while videobuf_read_stream() will read multiple frames
226 if they are needed to satisfy the count requested by the application. A
227 typical driver read() implementation will start the capture engine, call
228 one of the above functions, then stop the engine before returning (though a
229 smarter implementation might leave the engine running for a little while in
230 anticipation of another read() call happening in the near future).
232 The poll() function can usually be implemented with a direct call to:
236 unsigned int videobuf_poll_stream(struct file *file,
237 struct videobuf_queue *q,
240 Note that the actual wait queue eventually used will be the one associated
241 with the first available buffer.
243 When streaming I/O is done to kernel-space buffers, the driver must support
244 the mmap() system call to enable user space to access the data. In many
245 V4L2 drivers, the often-complex mmap() implementation simplifies to a
250 int videobuf_mmap_mapper(struct videobuf_queue *q,
251 struct vm_area_struct *vma);
253 Everything else is handled by the videobuf code.
255 The release() function requires two separate videobuf calls:
259 void videobuf_stop(struct videobuf_queue *q);
260 int videobuf_mmap_free(struct videobuf_queue *q);
262 The call to videobuf_stop() terminates any I/O in progress - though it is
263 still up to the driver to stop the capture engine. The call to
264 videobuf_mmap_free() will ensure that all buffers have been unmapped; if
265 so, they will all be passed to the buf_release() callback. If buffers
266 remain mapped, videobuf_mmap_free() returns an error code instead. The
267 purpose is clearly to cause the closing of the file descriptor to fail if
268 buffers are still mapped, but every driver in the 2.6.32 kernel cheerfully
269 ignores its return value.
274 The V4L2 API includes a very long list of driver callbacks to respond to
275 the many ioctl() commands made available to user space. A number of these
276 - those associated with streaming I/O - turn almost directly into videobuf
277 calls. The relevant helper functions are:
281 int videobuf_reqbufs(struct videobuf_queue *q,
282 struct v4l2_requestbuffers *req);
283 int videobuf_querybuf(struct videobuf_queue *q, struct v4l2_buffer *b);
284 int videobuf_qbuf(struct videobuf_queue *q, struct v4l2_buffer *b);
285 int videobuf_dqbuf(struct videobuf_queue *q, struct v4l2_buffer *b,
287 int videobuf_streamon(struct videobuf_queue *q);
288 int videobuf_streamoff(struct videobuf_queue *q);
290 So, for example, a VIDIOC_REQBUFS call turns into a call to the driver's
291 vidioc_reqbufs() callback which, in turn, usually only needs to locate the
292 proper struct videobuf_queue pointer and pass it to videobuf_reqbufs().
293 These support functions can replace a great deal of buffer management
294 boilerplate in a lot of V4L2 drivers.
296 The vidioc_streamon() and vidioc_streamoff() functions will be a bit more
297 complex, of course, since they will also need to deal with starting and
298 stopping the capture engine.
303 Thus far, we have talked about buffers, but have not looked at how they are
304 allocated. The scatter/gather case is the most complex on this front. For
305 allocation, the driver can leave buffer allocation entirely up to the
306 videobuf layer; in this case, buffers will be allocated as anonymous
307 user-space pages and will be very scattered indeed. If the application is
308 using user-space buffers, no allocation is needed; the videobuf layer will
309 take care of calling get_user_pages() and filling in the scatterlist array.
311 If the driver needs to do its own memory allocation, it should be done in
312 the vidioc_reqbufs() function, *after* calling videobuf_reqbufs(). The
313 first step is a call to:
317 struct videobuf_dmabuf *videobuf_to_dma(struct videobuf_buffer *buf);
319 The returned videobuf_dmabuf structure (defined in
320 <media/videobuf-dma-sg.h>) includes a couple of relevant fields:
324 struct scatterlist *sglist;
327 The driver must allocate an appropriately-sized scatterlist array and
328 populate it with pointers to the pieces of the allocated buffer; sglen
329 should be set to the length of the array.
331 Drivers using the vmalloc() method need not (and cannot) concern themselves
332 with buffer allocation at all; videobuf will handle those details. The
333 same is normally true of contiguous-DMA drivers as well; videobuf will
334 allocate the buffers (with dma_alloc_coherent()) when it sees fit. That
335 means that these drivers may be trying to do high-order allocations at any
336 time, an operation which is not always guaranteed to work. Some drivers
337 play tricks by allocating DMA space at system boot time; videobuf does not
338 currently play well with those drivers.
340 As of 2.6.31, contiguous-DMA drivers can work with a user-supplied buffer,
341 as long as that buffer is physically contiguous. Normal user-space
342 allocations will not meet that criterion, but buffers obtained from other
343 kernel drivers, or those contained within huge pages, will work with these
349 The final part of a videobuf implementation has no direct callback - it's
350 the portion of the code which actually puts frame data into the buffers,
351 usually in response to interrupts from the device. For all types of
352 drivers, this process works approximately as follows:
354 - Obtain the next available buffer and make sure that somebody is actually
357 - Get a pointer to the memory and put video data there.
359 - Mark the buffer as done and wake up the process waiting for it.
361 Step (1) above is done by looking at the driver-managed list_head structure
362 - the one which is filled in the buf_queue() callback. Because starting
363 the engine and enqueueing buffers are done in separate steps, it's possible
364 for the engine to be running without any buffers available - in the
365 vmalloc() case especially. So the driver should be prepared for the list
366 to be empty. It is equally possible that nobody is yet interested in the
367 buffer; the driver should not remove it from the list or fill it until a
368 process is waiting on it. That test can be done by examining the buffer's
369 done field (a wait_queue_head_t structure) with waitqueue_active().
371 A buffer's state should be set to VIDEOBUF_ACTIVE before being mapped for
372 DMA; that ensures that the videobuf layer will not try to do anything with
373 it while the device is transferring data.
375 For scatter/gather drivers, the needed memory pointers will be found in the
376 scatterlist structure described above. Drivers using the vmalloc() method
377 can get a memory pointer with:
381 void *videobuf_to_vmalloc(struct videobuf_buffer *buf);
383 For contiguous DMA drivers, the function to use is:
387 dma_addr_t videobuf_to_dma_contig(struct videobuf_buffer *buf);
389 The contiguous DMA API goes out of its way to hide the kernel-space address
390 of the DMA buffer from drivers.
392 The final step is to set the size field of the relevant videobuf_buffer
393 structure to the actual size of the captured image, set state to
394 VIDEOBUF_DONE, then call wake_up() on the done queue. At this point, the
395 buffer is owned by the videobuf layer and the driver should not touch it
398 Developers who are interested in more information can go into the relevant
399 header files; there are a few low-level functions declared there which have
400 not been talked about here. Also worthwhile is the vivi driver
401 (drivers/media/platform/vivi.c), which is maintained as an example of how V4L2
402 drivers should be written. Vivi only uses the vmalloc() API, but it's good
403 enough to get started with. Note also that all of these calls are exported
404 GPL-only, so they will not be available to non-GPL kernel modules.