1 .. SPDX-License-Identifier: GPL-2.0
8 Author: Jonathan Corbet <corbet@lwn.net>
14 The videobuf framework was deprecated in favor of videobuf2. Shouldn't
15 be used on new drivers.
20 The videobuf layer functions as a sort of glue layer between a V4L2 driver
21 and user space. It handles the allocation and management of buffers for
22 the storage of video frames. There is a set of functions which can be used
23 to implement many of the standard POSIX I/O system calls, including read(),
24 poll(), and, happily, mmap(). Another set of functions can be used to
25 implement the bulk of the V4L2 ioctl() calls related to streaming I/O,
26 including buffer allocation, queueing and dequeueing, and streaming
27 control. Using videobuf imposes a few design decisions on the driver
28 author, but the payback comes in the form of reduced code in the driver and
29 a consistent implementation of the V4L2 user-space API.
34 Not all video devices use the same kind of buffers. In fact, there are (at
35 least) three common variations:
37 - Buffers which are scattered in both the physical and (kernel) virtual
38 address spaces. (Almost) all user-space buffers are like this, but it
39 makes great sense to allocate kernel-space buffers this way as well when
40 it is possible. Unfortunately, it is not always possible; working with
41 this kind of buffer normally requires hardware which can do
42 scatter/gather DMA operations.
44 - Buffers which are physically scattered, but which are virtually
45 contiguous; buffers allocated with vmalloc(), in other words. These
46 buffers are just as hard to use for DMA operations, but they can be
47 useful in situations where DMA is not available but virtually-contiguous
48 buffers are convenient.
50 - Buffers which are physically contiguous. Allocation of this kind of
51 buffer can be unreliable on fragmented systems, but simpler DMA
52 controllers cannot deal with anything else.
54 Videobuf can work with all three types of buffers, but the driver author
55 must pick one at the outset and design the driver around that decision.
57 [It's worth noting that there's a fourth kind of buffer: "overlay" buffers
58 which are located within the system's video memory. The overlay
59 functionality is considered to be deprecated for most use, but it still
60 shows up occasionally in system-on-chip drivers where the performance
61 benefits merit the use of this technique. Overlay buffers can be handled
62 as a form of scattered buffer, but there are very few implementations in
63 the kernel and a description of this technique is currently beyond the
64 scope of this document.]
66 Data structures, callbacks, and initialization
67 ----------------------------------------------
69 Depending on which type of buffers are being used, the driver should
70 include one of the following files:
74 <media/videobuf-dma-sg.h> /* Physically scattered */
75 <media/videobuf-vmalloc.h> /* vmalloc() buffers */
76 <media/videobuf-dma-contig.h> /* Physically contiguous */
78 The driver's data structure describing a V4L2 device should include a
79 struct videobuf_queue instance for the management of the buffer queue,
80 along with a list_head for the queue of available buffers. There will also
81 need to be an interrupt-safe spinlock which is used to protect (at least)
84 The next step is to write four simple callbacks to help videobuf deal with
85 the management of buffers:
89 struct videobuf_queue_ops {
90 int (*buf_setup)(struct videobuf_queue *q,
91 unsigned int *count, unsigned int *size);
92 int (*buf_prepare)(struct videobuf_queue *q,
93 struct videobuf_buffer *vb,
94 enum v4l2_field field);
95 void (*buf_queue)(struct videobuf_queue *q,
96 struct videobuf_buffer *vb);
97 void (*buf_release)(struct videobuf_queue *q,
98 struct videobuf_buffer *vb);
101 buf_setup() is called early in the I/O process, when streaming is being
102 initiated; its purpose is to tell videobuf about the I/O stream. The count
103 parameter will be a suggested number of buffers to use; the driver should
104 check it for rationality and adjust it if need be. As a practical rule, a
105 minimum of two buffers are needed for proper streaming, and there is
106 usually a maximum (which cannot exceed 32) which makes sense for each
107 device. The size parameter should be set to the expected (maximum) size
108 for each frame of data.
110 Each buffer (in the form of a struct videobuf_buffer pointer) will be
111 passed to buf_prepare(), which should set the buffer's size, width, height,
112 and field fields properly. If the buffer's state field is
113 VIDEOBUF_NEEDS_INIT, the driver should pass it to:
117 int videobuf_iolock(struct videobuf_queue* q, struct videobuf_buffer *vb,
118 struct v4l2_framebuffer *fbuf);
120 Among other things, this call will usually allocate memory for the buffer.
121 Finally, the buf_prepare() function should set the buffer's state to
124 When a buffer is queued for I/O, it is passed to buf_queue(), which should
125 put it onto the driver's list of available buffers and set its state to
126 VIDEOBUF_QUEUED. Note that this function is called with the queue spinlock
127 held; if it tries to acquire it as well things will come to a screeching
128 halt. Yes, this is the voice of experience. Note also that videobuf may
129 wait on the first buffer in the queue; placing other buffers in front of it
130 could again gum up the works. So use list_add_tail() to enqueue buffers.
132 Finally, buf_release() is called when a buffer is no longer intended to be
133 used. The driver should ensure that there is no I/O active on the buffer,
134 then pass it to the appropriate free routine(s):
138 /* Scatter/gather drivers */
139 int videobuf_dma_unmap(struct videobuf_queue *q,
140 struct videobuf_dmabuf *dma);
141 int videobuf_dma_free(struct videobuf_dmabuf *dma);
143 /* vmalloc drivers */
144 void videobuf_vmalloc_free (struct videobuf_buffer *buf);
146 /* Contiguous drivers */
147 void videobuf_dma_contig_free(struct videobuf_queue *q,
148 struct videobuf_buffer *buf);
150 One way to ensure that a buffer is no longer under I/O is to pass it to:
154 int videobuf_waiton(struct videobuf_buffer *vb, int non_blocking, int intr);
156 Here, vb is the buffer, non_blocking indicates whether non-blocking I/O
157 should be used (it should be zero in the buf_release() case), and intr
158 controls whether an interruptible wait is used.
163 At this point, much of the work is done; much of the rest is slipping
164 videobuf calls into the implementation of the other driver callbacks. The
165 first step is in the open() function, which must initialize the
166 videobuf queue. The function to use depends on the type of buffer used:
170 void videobuf_queue_sg_init(struct videobuf_queue *q,
171 struct videobuf_queue_ops *ops,
174 enum v4l2_buf_type type,
175 enum v4l2_field field,
179 void videobuf_queue_vmalloc_init(struct videobuf_queue *q,
180 struct videobuf_queue_ops *ops,
183 enum v4l2_buf_type type,
184 enum v4l2_field field,
188 void videobuf_queue_dma_contig_init(struct videobuf_queue *q,
189 struct videobuf_queue_ops *ops,
192 enum v4l2_buf_type type,
193 enum v4l2_field field,
197 In each case, the parameters are the same: q is the queue structure for the
198 device, ops is the set of callbacks as described above, dev is the device
199 structure for this video device, irqlock is an interrupt-safe spinlock to
200 protect access to the data structures, type is the buffer type used by the
201 device (cameras will use V4L2_BUF_TYPE_VIDEO_CAPTURE, for example), field
202 describes which field is being captured (often V4L2_FIELD_NONE for
203 progressive devices), msize is the size of any containing structure used
204 around struct videobuf_buffer, and priv is a private data pointer which
205 shows up in the priv_data field of struct videobuf_queue. Note that these
206 are void functions which, evidently, are immune to failure.
208 V4L2 capture drivers can be written to support either of two APIs: the
209 read() system call and the rather more complicated streaming mechanism. As
210 a general rule, it is necessary to support both to ensure that all
211 applications have a chance of working with the device. Videobuf makes it
212 easy to do that with the same code. To implement read(), the driver need
213 only make a call to one of:
217 ssize_t videobuf_read_one(struct videobuf_queue *q,
218 char __user *data, size_t count,
219 loff_t *ppos, int nonblocking);
221 ssize_t videobuf_read_stream(struct videobuf_queue *q,
222 char __user *data, size_t count,
223 loff_t *ppos, int vbihack, int nonblocking);
225 Either one of these functions will read frame data into data, returning the
226 amount actually read; the difference is that videobuf_read_one() will only
227 read a single frame, while videobuf_read_stream() will read multiple frames
228 if they are needed to satisfy the count requested by the application. A
229 typical driver read() implementation will start the capture engine, call
230 one of the above functions, then stop the engine before returning (though a
231 smarter implementation might leave the engine running for a little while in
232 anticipation of another read() call happening in the near future).
234 The poll() function can usually be implemented with a direct call to:
238 unsigned int videobuf_poll_stream(struct file *file,
239 struct videobuf_queue *q,
242 Note that the actual wait queue eventually used will be the one associated
243 with the first available buffer.
245 When streaming I/O is done to kernel-space buffers, the driver must support
246 the mmap() system call to enable user space to access the data. In many
247 V4L2 drivers, the often-complex mmap() implementation simplifies to a
252 int videobuf_mmap_mapper(struct videobuf_queue *q,
253 struct vm_area_struct *vma);
255 Everything else is handled by the videobuf code.
257 The release() function requires two separate videobuf calls:
261 void videobuf_stop(struct videobuf_queue *q);
262 int videobuf_mmap_free(struct videobuf_queue *q);
264 The call to videobuf_stop() terminates any I/O in progress - though it is
265 still up to the driver to stop the capture engine. The call to
266 videobuf_mmap_free() will ensure that all buffers have been unmapped; if
267 so, they will all be passed to the buf_release() callback. If buffers
268 remain mapped, videobuf_mmap_free() returns an error code instead. The
269 purpose is clearly to cause the closing of the file descriptor to fail if
270 buffers are still mapped, but every driver in the 2.6.32 kernel cheerfully
271 ignores its return value.
276 The V4L2 API includes a very long list of driver callbacks to respond to
277 the many ioctl() commands made available to user space. A number of these
278 - those associated with streaming I/O - turn almost directly into videobuf
279 calls. The relevant helper functions are:
283 int videobuf_reqbufs(struct videobuf_queue *q,
284 struct v4l2_requestbuffers *req);
285 int videobuf_querybuf(struct videobuf_queue *q, struct v4l2_buffer *b);
286 int videobuf_qbuf(struct videobuf_queue *q, struct v4l2_buffer *b);
287 int videobuf_dqbuf(struct videobuf_queue *q, struct v4l2_buffer *b,
289 int videobuf_streamon(struct videobuf_queue *q);
290 int videobuf_streamoff(struct videobuf_queue *q);
292 So, for example, a VIDIOC_REQBUFS call turns into a call to the driver's
293 vidioc_reqbufs() callback which, in turn, usually only needs to locate the
294 proper struct videobuf_queue pointer and pass it to videobuf_reqbufs().
295 These support functions can replace a great deal of buffer management
296 boilerplate in a lot of V4L2 drivers.
298 The vidioc_streamon() and vidioc_streamoff() functions will be a bit more
299 complex, of course, since they will also need to deal with starting and
300 stopping the capture engine.
305 Thus far, we have talked about buffers, but have not looked at how they are
306 allocated. The scatter/gather case is the most complex on this front. For
307 allocation, the driver can leave buffer allocation entirely up to the
308 videobuf layer; in this case, buffers will be allocated as anonymous
309 user-space pages and will be very scattered indeed. If the application is
310 using user-space buffers, no allocation is needed; the videobuf layer will
311 take care of calling get_user_pages() and filling in the scatterlist array.
313 If the driver needs to do its own memory allocation, it should be done in
314 the vidioc_reqbufs() function, *after* calling videobuf_reqbufs(). The
315 first step is a call to:
319 struct videobuf_dmabuf *videobuf_to_dma(struct videobuf_buffer *buf);
321 The returned videobuf_dmabuf structure (defined in
322 <media/videobuf-dma-sg.h>) includes a couple of relevant fields:
326 struct scatterlist *sglist;
329 The driver must allocate an appropriately-sized scatterlist array and
330 populate it with pointers to the pieces of the allocated buffer; sglen
331 should be set to the length of the array.
333 Drivers using the vmalloc() method need not (and cannot) concern themselves
334 with buffer allocation at all; videobuf will handle those details. The
335 same is normally true of contiguous-DMA drivers as well; videobuf will
336 allocate the buffers (with dma_alloc_coherent()) when it sees fit. That
337 means that these drivers may be trying to do high-order allocations at any
338 time, an operation which is not always guaranteed to work. Some drivers
339 play tricks by allocating DMA space at system boot time; videobuf does not
340 currently play well with those drivers.
342 As of 2.6.31, contiguous-DMA drivers can work with a user-supplied buffer,
343 as long as that buffer is physically contiguous. Normal user-space
344 allocations will not meet that criterion, but buffers obtained from other
345 kernel drivers, or those contained within huge pages, will work with these
351 The final part of a videobuf implementation has no direct callback - it's
352 the portion of the code which actually puts frame data into the buffers,
353 usually in response to interrupts from the device. For all types of
354 drivers, this process works approximately as follows:
356 - Obtain the next available buffer and make sure that somebody is actually
359 - Get a pointer to the memory and put video data there.
361 - Mark the buffer as done and wake up the process waiting for it.
363 Step (1) above is done by looking at the driver-managed list_head structure
364 - the one which is filled in the buf_queue() callback. Because starting
365 the engine and enqueueing buffers are done in separate steps, it's possible
366 for the engine to be running without any buffers available - in the
367 vmalloc() case especially. So the driver should be prepared for the list
368 to be empty. It is equally possible that nobody is yet interested in the
369 buffer; the driver should not remove it from the list or fill it until a
370 process is waiting on it. That test can be done by examining the buffer's
371 done field (a wait_queue_head_t structure) with waitqueue_active().
373 A buffer's state should be set to VIDEOBUF_ACTIVE before being mapped for
374 DMA; that ensures that the videobuf layer will not try to do anything with
375 it while the device is transferring data.
377 For scatter/gather drivers, the needed memory pointers will be found in the
378 scatterlist structure described above. Drivers using the vmalloc() method
379 can get a memory pointer with:
383 void *videobuf_to_vmalloc(struct videobuf_buffer *buf);
385 For contiguous DMA drivers, the function to use is:
389 dma_addr_t videobuf_to_dma_contig(struct videobuf_buffer *buf);
391 The contiguous DMA API goes out of its way to hide the kernel-space address
392 of the DMA buffer from drivers.
394 The final step is to set the size field of the relevant videobuf_buffer
395 structure to the actual size of the captured image, set state to
396 VIDEOBUF_DONE, then call wake_up() on the done queue. At this point, the
397 buffer is owned by the videobuf layer and the driver should not touch it
400 Developers who are interested in more information can go into the relevant
401 header files; there are a few low-level functions declared there which have
402 not been talked about here. Also worthwhile is the vivi driver
403 (drivers/media/platform/vivi.c), which is maintained as an example of how V4L2
404 drivers should be written. Vivi only uses the vmalloc() API, but it's good
405 enough to get started with. Note also that all of these calls are exported
406 GPL-only, so they will not be available to non-GPL kernel modules.