When compiling this for older kernels using the compatibility build
the compiler complains about uninitialized variables:

In file included from include/linux/kernel.h:20:0,
                 from include/linux/cache.h:4,
                 from include/linux/time.h:7,
                 from include/linux/input.h:13,
                 from /home/hans/work/build/media_build/v4l/compat.h:9,
                 from <command-line>:0:
/home/hans/work/build/media_build/v4l/videobuf2-core.c: In function 'vb2_mmap':
include/linux/dynamic_debug.h:60:9: warning: 'plane' may be used uninitialized in this function [-Wmaybe-uninitialized]
   printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__);  \
         ^
/home/hans/work/build/media_build/v4l/videobuf2-core.c:2381:23: note: 'plane' was declared here
  unsigned int buffer, plane;
                       ^
In file included from include/linux/kernel.h:20:0,
                 from include/linux/cache.h:4,
                 from include/linux/time.h:7,
                 from include/linux/input.h:13,
                 from /home/hans/work/build/media_build/v4l/compat.h:9,
                 from <command-line>:0:
include/linux/dynamic_debug.h:60:9: warning: 'buffer' may be used uninitialized in this function [-Wmaybe-uninitialized]
   printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__);  \
         ^
/home/hans/work/build/media_build/v4l/videobuf2-core.c:2381:15: note: 'buffer' was declared here
  unsigned int buffer, plane;
               ^

While these warnings are bogus (the call to __find_plane_by_offset will
set buffer and plane), it doesn't hurt to initialize these variables.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The vb2 core ignores any return code from the stop_streaming op.
And there really isn't anything it can do anyway in case of an error.
So change the return type to void and update any drivers that implement it.

The int return gave drivers the idea that this operation could actually
fail, but that's really not the case.

The pwc amd sdr-msi3101 drivers both had this construction:

        if (mutex_lock_interruptible(&s->v4l2_lock))
                return -ERESTARTSYS;

This has been updated to just call mutex_lock(). The stop_streaming op
expects this to really stop streaming and I very much doubt this will
work reliably if stop_streaming just returns without really stopping the
DMA.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

In order to implement vb2 DVB support you need to be able to start
a kernel thread that queues and dequeues buffers, calling a callback
function for every buffer. This patch adds support for that.

It's based on drivers/media/v4l2-core/videobuf-dvb.c, but with all the DVB
specific stuff stripped out, thus making it much more generic.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The kernel debug messages produced by vb2 started either with a
lower or an upper case character. Switched all to use lower-case
which seemed to be what was used in the majority of the messages.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

It was impossible to read() or write() a frame if the queue type was multiplanar.
Even if the current format is single planar. Change this to just check whether
the number of planes is 1 or more.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Added a vb2_fileio_is_active inline function that returns true if fileio
is in progress. Check for this too in mmap() (you don't want apps mmap()ing
buffers used by fileio) and expbuf() (same reason).

In addition drivers should be able to check for this in queue_setup() to
return an error if an attempt is made to read() or write() with
V4L2_FIELD_ALTERNATE being configured. This is illegal (there is no way
to pass the TOP/BOTTOM information around using file I/O).

However, in order to be able to check for this the init_fileio function
needs to set q->fileio early on, before the buffers are allocated. So switch
to using internal functions (__reqbufs, vb2_internal_qbuf and
vb2_internal_streamon) to skip the fileio check. Well, that's why the internal
functions were created...
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

q->start_streaming_called is always true, so the WARN_ON check against
it being false can be dropped.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

This is not allowed by the spec and does in fact not make any sense.
Return -EINVAL if this is the case.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

When using write() to write data to an output video node the vb2 core
should set timestamps if V4L2_BUF_FLAG_TIMESTAMP_COPY is set. Nobody
else is able to provide this information with the write() operation.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

__qbuf_mmap was sort of hidden in between the much larger __qbuf_userptr
and __qbuf_dmabuf functions. Move it before __qbuf_userptr which is
also conform the usual order these memory models are implemented: first
mmap, then userptr, then dmabuf.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Many dprintk's in vb2 use a hardcoded prefix with the function name. In
many cases that is now outdated. To keep things consistent the dprintk
macro has been changed to print the function name in addition to the "vb2:"
prefix. Superfluous prefixes elsewhere in the code have been removed.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The application should really always fill in bytesused for output
buffers, unfortunately the vb2 framework never checked for that.

So for single planar formats replace a bytesused of 0 by the length
of the buffer, and for multiplanar format do the same if bytesused is
0 for ALL planes.

This seems to be what the user really intended if v4l2_buffer was
just memset to 0.

I'm afraid that just checking for this and returning an error would
break too many applications. Quite a few drivers never check for bytesused
at all and just use the buffer length instead.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The videobuf2-core did not zero the 'planes' array in __qbuf_userptr()
and __qbuf_dmabuf(). That's now memset to 0. Without this the reserved
array in struct v4l2_plane would be non-zero, causing v4l2-compliance
errors.

More serious is the fact that data_offset was not handled correctly:

- for capture devices it was never zeroed, which meant that it was
  uninitialized. Unless the driver sets it it was a completely random
  number. With the memset above this is now fixed.

- __qbuf_dmabuf had a completely incorrect length check that included
  data_offset.

- in __fill_vb2_buffer in the DMABUF case the data_offset field was
  unconditionally copied from v4l2_buffer to v4l2_plane when this
  should only happen in the output case.

- in the single-planar case data_offset was never correctly set to 0.
  The single-planar API doesn't support data_offset, so setting it
  to 0 is the right thing to do. This too is now solved by the memset.

All these issues were found with v4l2-compliance.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Sparse generated a bunch of errors like this:

drivers/media/v4l2-core/videobuf2-core.c:2045:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:136:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:151:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:168:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:183:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:185:9: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:385:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1115:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1268:33: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1270:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1315:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1324:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1396:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1457:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1482:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1484:9: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1523:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1525:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1815:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1828:17: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1914:25: error: incompatible types in conditional expression (different base types)
drivers/media/v4l2-core/videobuf2-core.c:1944:9: error: incompatible types in conditional expression (different base types)

These are caused by the call*op defines which do something like this:

        (ops->op) ? ops->op(args) : 0

which is OK as long as op is not a void function, because in that case one part
of the conditional expression returns void, the other an integer. Hence the sparse
errors.

I've replaced this by introducing three variants of the call_ macros:
call_*op for int returns, call_void_*op for void returns and call_ptr_*op for
pointer returns.

That's the bad news. The good news is that the fail_*op macros could be removed
since the call_*op macros now have enough information to determine if the op
succeeded or not and can increment the op counter only on success. This at least
makes it more robust w.r.t. future changes.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Don't call buf_finish unless we know that the buffer is in a valid state.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

If you request buffers, then queue buffers and then call STREAMOFF
those buffers are not returned to their dequeued state because streamoff
will just return if q->streaming was 0.

This means that afterwards you can never QBUF that same buffer again unless
you do STREAMON, REQBUFS or close the filehandle first.

It is clear that if you do STREAMOFF even if no STREAMON was called before,
you still want to have all buffers returned to their proper dequeued state.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

No need to oops for this, WARN_ON is good enough.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

If __reqbufs was called then existing buffers are freed. However, if that
happens without ever having started STREAMON, but if buffers have been queued,
then the buf_finish op is never called.

Add a call to __vb2_queue_cancel in __reqbufs so that these buffers are
cleaned up there as well.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

In commit 02f142ec support was added to
start_streaming to return -ENOBUFS if insufficient buffers were queued
for the DMA engine to start. The vb2 core would attempt calling
start_streaming again if another buffer would be queued up.

Later analysis uncovered problems with the queue management if start_streaming
would return an error: the buffers are enqueued to the driver before the
start_streaming op is called, so after an error they are never returned to
the vb2 core. The solution for this is to let the driver return them to
the vb2 core in case of an error while starting the DMA engine. However,
in the case of -ENOBUFS that would be weird: it is not a real error, it
just says that more buffers are needed. Requiring start_streaming to give
them back only to have them requeued again the next time the application
calls QBUF is inefficient.

This patch changes this mechanism: it adds a 'min_buffers_needed' field
to vb2_queue that drivers can set with the minimum number of buffers
required to start the DMA engine. The start_streaming op is only called
if enough buffers are queued. The -ENOBUFS handling has been dropped in
favor of this new method.

Drivers are expected to return buffers back to vb2 core with state QUEUED
if start_streaming would return an error. The vb2 core checks for this
and produces a warning if that didn't happen and it will forcefully
reclaim such buffers to ensure that the internal vb2 core state remains
consistent and all buffer-related resources have been correctly freed
and all op calls have been balanced.

__reqbufs() has been updated to check that at least min_buffers_needed
buffers could be allocated. If fewer buffers were allocated then __reqbufs
will free what was allocated and return -ENOMEM. Based on a suggestion from
Pawel Osciak.

__create_bufs() doesn't do that check, since the use of __create_bufs
assumes some advance scenario where the user might want more control.
Instead streamon will check if enough buffers were allocated to prevent
streaming with fewer than the minimum required number of buffers.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

__vb2_queue_free() would init the queued_list at all times, even if
q->num_buffers > 0. This should only happen if num_buffers == 0.

This situation can happen if a CREATE_BUFFERS call couldn't allocate
enough buffers and had to free those it did manage to allocate before
returning an error.

While we're at it: __vb2_queue_alloc() returns the number of buffers
allocated, not an error code. So stick the result in allocated_buffers
instead of ret as that's very confusing.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

'queued_count' is a bit vague since it is not clear to which queue it
refers to: the vb2 internal list of buffers or the driver-owned list
of buffers.

Rename to make it explicit.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Ensure that these ops are properly balanced.

There are two scenarios:

1) for MMAP buf_init is called when the buffers are created and buf_cleanup
   must be called when the queue is finally freed. This scenario was always
   working.

2) for USERPTR and DMABUF it is more complicated. When a buffer is queued
   the code checks if all planes of this buffer have been acquired before.
   If that's the case, then only buf_prepare has to be called. Otherwise
   buf_cleanup needs to be called if the buffer was acquired before, then,
   once all changed planes have been (re)acquired, buf_init has to be
   called followed by buf_prepare. Should buf_prepare fail, then buf_cleanup
   must be called on the newly acquired planes to release them in.

Finally, in __vb2_queue_free we have to check if the buffer was actually
acquired before calling buf_cleanup. While that it always true for MMAP
mode, it is not necessarily true for the other modes. E.g. if you just
call REQBUFS and close the file handle, then buffers were never queued and
so no buf_init was ever called.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

If a queue was canceled, then the buf_finish op was never called for the
pending buffers. So add this call to queue_cancel. Before calling buf_finish
set the buffer state to PREPARED, which is the correct state. That way the
states DONE and ERROR will only be seen in buf_finish if streaming is in
progress.

Since buf_finish can now be called from non-streaming state we need to
adapt the handful of drivers that actually need to know this.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NSakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The buf_finish op should always work, so change the return type to void.
Update the few drivers that use it.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Reviewed-by: NPawel Osciak <pawel@osciak.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

When a vb2_queue is freed check if all the mem_ops and queue ops were balanced.
So the number of calls to e.g. buf_finish has to match the number of calls to
buf_prepare, etc.

This code is only enabled if CONFIG_VIDEO_ADV_DEBUG is set.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NPawel Osciak <pawel@osciak.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Fix an incorrect test in vb2_internal_qbuf() where only DEQUEUED buffers
are allowed. But PREPARED buffers are also OK.

Introduced by commit 4138111a
("vb2: simplify qbuf/prepare_buf by removing callback").
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Commit 88e26870 ("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: NHans Verkuil <hans.verkuil@cisco.com>
Tested-by: NHans Verkuil <hans.verkuil@cisco.com>
Cc: Andy Walls <awalls@md.metrocast.net>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

This patch adds a test preventing streamon() if there is no buffer
ready.

Without this patch, a user could call streamon() before
preparing any buffer. This leads to a situation where if he calls
close() before calling streamoff() the device is kept streaming.
Signed-off-by: NRicardo Ribalda Delgado <ricardo.ribalda@gmail.com>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

For COPY timestamps, buffer timestamp source flags will traverse the queue
untouched.
Signed-off-by: NSakari Ailus <sakari.ailus@iki.fi>
Acked-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Some devices do not produce timestamps that correspond to the end of the
frame. The user space should be informed on the matter. This patch achieves
that by adding buffer flags (and a mask) for timestamp sources since more
possible timestamping points are expected than just two.

A three-bit mask is defined (V4L2_BUF_FLAG_TSTAMP_SRC_MASK) and two of the
eight possible values is are defined V4L2_BUF_FLAG_TSTAMP_SRC_EOF for end of
frame (value zero) V4L2_BUF_FLAG_TSTAMP_SRC_SOE for start of exposure (next
value).
Signed-off-by: NSakari Ailus <sakari.ailus@iki.fi>
Acked-by: NKamil Debski <k.debski@samsung.com>
Acked-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Mask out other bits when comparing timestamp types.
Signed-off-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The timestamp_type field used to contain only the timestamp type. Soon it
will be used for timestamp source flags as well. Rename the field
accordingly.

[m.chehab@samsung.com: do the change also to drivers/staging/media and at s2255]
Signed-off-by: NSakari Ailus <sakari.ailus@iki.fi>
Acked-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

When sending a buffer to a video output device some of the fields need
to be copied so they arrive in the driver. These are the KEY/P/BFRAME
flags and the TIMECODE flag, and, if that flag is set, the timecode field
itself.

There are a number of functions involved in this: the __fill_vb2_buffer()
is called while preparing a buffer. For output buffers the buffer contains
the video data, so any meta data associated with that (KEY/P/BFRAME and
the field information) should be stored at that point.

The timecode, timecode flag and timestamp information is not part of that,
that information will have to be set when vb2_internal_qbuf() is called to
actually queue the buffer to the driver. Usually VIDIOC_QBUF will do the
prepare as well, but you can call PREPARE_BUF first and only later VIDIOC_QBUF.
You most likely will want to set the timestamp and timecode when you actually
queue the buffer, not when you prepare it.

Finally, in buf_prepare() make sure the timestamp and sequence fields are
actually cleared so that when you do a QUERYBUF of a prepared-but-not-yet-queued
buffer you will not see stale timestamp/sequence data.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

This patch adds a test preventing streamon() if there is no buffer
ready.

Without this patch, a user could call streamon() before
preparing any buffer. This leads to a situation where if he calls
close() before calling streamoff() the device is kept streaming.
Signed-off-by: NRicardo Ribalda Delgado <ricardo.ribalda@gmail.com>
Reviewed-by: NMarek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The comment incorrectly explains that the code verifies information
provided by userspace, while verification has been performed earlier in
reality. Fix it.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

videobuf2 file I/O emulation assumed that buffers dequeued from the
driver would return in the order they were enqueued in the driver.

Improve the file I/O emulator's book-keeping to remove this assumption.

Also set the buf->size properly if a write() dequeues a buffer and the
VB2_FILEIO_WRITE_IMMEDIATELY flag is set.

Based on an initial patch by Andy Walls.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: NAndy Walls <awalls@md.metrocast.net>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

Two fixes:

- there is no need to set the index when calling dqbuf: dqbuf will
  overwrite it.
- __vb2_init_fileio already starts streaming for write(), so there is
  no need to do it again in __vb2_perform_fileio. It can never have
  worked anyway: either __vb2_init_fileio succeeds in starting streaming
  or it is never going to happen.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

If start_streaming returns -ENOBUFS, then it will be retried the next time
a buffer is queued. This means applications no longer need to know how many
buffers need to be queued before STREAMON can be called. This is particularly
useful for output stream I/O.

If a DMA engine needs at least X buffers before it can start streaming, then
for applications to get a buffer out as soon as possible they need to know
the minimum number of buffers to queue before STREAMON can be called. You can't
just try STREAMON after every buffer since on failure STREAMON will dequeue
all your buffers. (Is that a bug or a feature? Frankly, I'm not sure).

This patch simplifies applications substantially: they can just call STREAMON
at the beginning and then start queuing buffers and the DMA engine will
kick in automagically once enough buffers are available.

This also fixes using write() to stream video: the fileio implementation
calls streamon without having any queued buffers, which will fail today for
any driver that requires a minimum number of buffers.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NMarek Szyprowski <m.szyprowski@samsung.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

The read/write implementation in vb2 reuses existing vb2 functions, but
it sets q->fileio to NULL before calling them in order to skip the
'q->fileio != NULL' check.

This works today due to the synchronous nature of read/write, but it
1) is ugly, and 2) will fail in an asynchronous use-case such as a
thread queuing and dequeuing buffers. This last example will be necessary
in order to implement vb2 DVB support.

This patch removes the hack by splitting up the dqbuf/qbuf/streamon/streamoff
functions into an external and an internal version. The external version
checks q->fileio and then calls the internal version. The read/write
implementation now just uses the internal version, removing the need to
set q->fileio to NULL.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>

When preparing a buffer the queue lock is released for a short while
if the memory mode is USERPTR (see __buf_prepare for the details), which
would allow a race with a REQBUFS which can free the buffers. Removing the
buffers from underneath __buf_prepare is obviously a bad idea, so we
check if any of the buffers is in the state PREPARING, and if so we
just return -EAGAIN.

If this happens, then the application does something really strange. The
REQBUFS call can be retried safely, since this situation is transient.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <m.chehab@samsung.com>