The BKL is now moved from the video_open function in v4l2-dev.c to the
various drivers. It seems about a third of the drivers already has a
lock of some sort protecting the open(), another third uses
video_exclusive_open (yuck!) and the last third required adding the
BKL in their open function.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
video_device_release() does kfree(), which made the following printk()
doing a use-after-free.
printk() first and release then.
Reported-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
This patch fixes several small issues at vivi register routines:
1) minor and n_devs should be unsigned;
2) n_devs = 0 were not properly handled;
3) if n_devs specify a high number, the driver would just roll back and
won't register any device. The proper behaviour is to keep all succeded
devices registered;
4) both n_devs and minor were using 0 as permissions. Better to have
them with 0444.
With the current patch, if n_devs specify a very large value, it will
register all possible devices. For example, on a machine without any
other V4L drivers loaded, with this patch, we will have something like:
vivi: V4L2 device registered as /dev/video0
vivi: V4L2 device registered as /dev/video1
vivi: V4L2 device registered as /dev/video2
...
vivi: V4L2 device registered as /dev/video31
video_register_device_index: get_index failed
Video Technology Magazine Virtual Video Capture Board ver 0.5.0 successfully loaded.
5) The number of allocated devices on success is now kept at n_devs:
$ cat /sys/module/vivi/parameters/n_devs
32
Thanks to Henne <henne@nachtwindheim.de> for pointing that there were
some issues at vivi.
Cc: Henne <henne@nachtwindheim.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
The type and type2 fields were unused and so could be removed.
Instead add a vfl_type field that contains the type of the video
device.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
All ioctl callbacks are now stored in a new v4l2_ioctl_ops struct. Drivers fill in
a const struct v4l2_ioctl_ops and video_device just contains a const pointer to it.
This ensures a clean separation between the const ops struct and the non-const
video_device struct.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
The functions in a header should not belong to another module. The prio functions
belong to v4l2-common.c, so move them to v4l2-common.h.
The ioctl functions belong to v4l2-ioctl.c, so create a new v4l2-ioctl.h header
and move those functions to it.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
The naming for the callbacks that handle the VIDIOC_ENUM_FMT and
VIDIOC_S/G/TRY_FMT ioctls was very confusing. Renamed it to match
the v4l2_buf_type name.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Move allocation after first check and fix memory leak.
Noticed-by: Daniel Marjamäki <danielm77@spray.se>
Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
some gcc versions complain that fh is used without being defined.
The error report is bogus. However, fixing it is trivial. Better to make
gcc happy.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
__FUNCTION__ is gcc-specific, use __func__
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
vivi previously had a very complex queuing system and held spinlocks while
doing copy_to_user, kmalloc, etc. This caused the driver to easily deadlock
when a multi-threaded application used it and revealed bugs in videobuf too.
This replaces the copy_to_user with memcpy since we were never copying to user
space addresses. And makes the kmalloc atomic.
Signed-off-by: Brandon Philips <bphilips@suse.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
vivi currently doesn't have the infrastructure to handle being opened more than
one time and will crash if it is. So, make it openable only once.
Signed-off-by: Brandon Philips <bphilips@suse.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Manually fixed all pertinent checkpatch.pl errors inside the source code.
Also removed some unused code at the driver and a few minor cleanups.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
mv_count is a counter used to move the vertical bars. Before this patch, it
where a static var. This works fine for just one device. However, when using
multiple devices, every device would increment it.
This patch moves it to its correct place: struct vivi_dev. So, now, each device
has its own data.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Now, it is possible to open multiple vivi devices, by using n_devs parameter.
This makes vivi driver closer to a real one.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
There were a trouble at vivi driver when using non-zero inodes. This where due
to not properly preserving the minor inode after calling video_register. Since
this driver is a reference for newer drivers, and it is possible to have more
than one video device inside the machine, this patch makes vivi to dynamically
allocate video_device struct.
Thanks to Gregor Jasny <jasny@vidsoft.de> for pointing the issue.
Also, this patch removes a very anoying (but useless) message of not having a
proper release call.
CC: Gregor Jasny <jasny@vidsoft.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Drivers were using cookie cutter code for stopping the read/stream. Use the
new videobuf_stop function which is lock safe.
Signed-off-by: Brandon Philips <bphilips@suse.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
struct video_device used to define a .hardware field. While
initialized on severl drivers, this field is never used inside V4L.
However, drivers using it need to include the old V4L1 header.
This seems to cause compilation troubles with some random configs.
Better just to remove it from all drivers.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
The rest of V4L files.
There is one list_for_each+list_entry in cpia_pp.c that
wasn't changed because it expects the loop iterator to remain NULL if
the list is empty.
A bug in vivi is fixed; the 'safe' version needs to be used because the loop
deletes the list entries.
Simplify a second loop in vivi and get rid if an un-used variable in that loop.
Signed-off-by: Trent Piepho <xyzzy@speakeasy.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
The "resource" locking in vivi isn't needed since
streamon/streamoff/read_stream do mutual exclusion using
q->reading/q->streaming.
Plus it is sort of broken:
a) res_locked() use in vivi_read() is racey.
b) res_free() calls mutex_lock twice causing streamoff to break
Signed-off-by: Brandon Philips <bphilips@suse.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Since vivi is using videobuf_read_stream() it can use videobuf_poll_stream()
now.
Signed-off-by: Brandon Philips <bphilips@suse.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
If the client provides V4L2_FIELD_ANY vivi should return a valid field :)
Signed-off-by: Brandon Philips <bphilips@suse.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
While this is not the standard color bar behaviour, having some movement
there allows to check if buffers are being properly handled.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Due to the replace of videobuf_read_one to videobuf_read_stream, poll()
method implementation is wrong. This fixes poll() implementation, making
read of /dev/video? to work again.
With this method, an USB driver can use video-buf, without needing to
request memory from the DMA-safe area.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
This patch removes the usage of videobuf-dma-sg from vivi driver, using
instead videobuf-vmalloc. This way, vivi will be useful for testing the
newer method. Reverting this patch won't hurt vivi, since both methods
work fine.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
The Virtual Video driver uses a semaphore as mutex. Use the mutex API
instead of the (binary) semaphore.
Signed-off-by: Matthias Kaehlcke <matthias.kaehlcke@gmail.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
vivi scatter method were used as a proof of concept. It can be
safelly removed from mainstream, since the current method is
faster and better than the previous solution.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Currently, the freezer treats all tasks as freezable, except for the kernel
threads that explicitly set the PF_NOFREEZE flag for themselves. This
approach is problematic, since it requires every kernel thread to either
set PF_NOFREEZE explicitly, or call try_to_freeze(), even if it doesn't
care for the freezing of tasks at all.
It seems better to only require the kernel threads that want to or need to
be frozen to use some freezer-related code and to remove any
freezer-related code from the other (nonfreezable) kernel threads, which is
done in this patch.
The patch causes all kernel threads to be nonfreezable by default (ie. to
have PF_NOFREEZE set by default) and introduces the set_freezable()
function that should be called by the freezable kernel threads in order to
unset PF_NOFREEZE. It also makes all of the currently freezable kernel
threads call set_freezable(), so it shouldn't cause any (intentional)
change of behaviour to appear. Additionally, it updates documentation to
describe the freezing of tasks more accurately.
[akpm@linux-foundation.org: build fixes]
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Nigel Cunningham <nigel@nigel.suspend2.net>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The result of copy_to_user was not used, so the compiler complained
now a warning will be issued if copy_to_user fails.
Signed-off-by: Marcel Siegert <mws@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Before this patch, vivi were simulating a scatter gather DMA transfer.
While this is academic, showing how stuff really works on a real PCI
device, this means a non-optimized code.
There are only two memory models that vivi implements:
1) kernel alloced memory. This is also used by read() method.
On this case, a vmalloc32 buffer is allocated at kernel;
2) userspace allocated memory. This is used by most userspace apps.
video-buf will store this pointer.
a simple copy_to_user is enough to transfer data.
The third memory model scenario supported by video-buf is overlay mode.
This model is not implemented on vivi and unlikely to be implemented on
newer drivers, since now, most userspace apps do some post-processing
(like de-interlacing).
After this patch, some cleanups may be done at video-buf.c to avoid
allocating pages, when the driver doesn't need a PCI buffer. This is the
case of vivi and usb drivers.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
There are several potential troubles on vivi waitqueue code:
- Watchdog timer should be reset at every received frame;
- Watchdog timer should be reset at the beginning of vivi_thread();
- Checks for errors when creating a newer thread with kernel_thread();
- Wake up vivi_thread() after creating it.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
If for some reason vivi_thread() fails, vivi will suffer an OOPS at
thread stop code, since waitqueue wouldn't be properly initializated.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Many struct file_operations in the kernel can be "const". Marking them const
moves these to the .rodata section, which avoids false sharing with potential
dirty data. In addition it'll catch accidental writes at compile time to
these shared resources.
[akpm@sdl.org: dvb fix]
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
vivi.c uses the KM_BOUNCE_READ with local interrupts enabled.
This means that if a disk interrupt occurs while vivi.c is using this
fixmap slot, the vivi.c driver will, upon return from that interrupt, find
that the fixmap slot now points at a different physical page.
The net result will probably be rare corruption of disk file contents,
because viv.c will now be altering the page which the disk code was
recently using.
Thanks to Andrew Morton for pointing this.
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
The return value of kthread_run() should be checked by IS_ERR().
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Freeing data including list_head in list_for_each() is not safe.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Hans Verkuil
Adrian Bunk
Mauro Carvalho Chehab
Carl Karsten
Marcin Slusarz
Harvey Harrison
Brandon Philips
Trent Piepho
Brandon Philips
Matthias Kaehlcke
Rafael J. Wysocki
Marcel Siegert
Arjan van de Ven
Akinobu Mita