lwn.git/drivers/media/video/cx88/cx88-mpeg.c, branch v3.10.3

[media] rename most media/video pci drivers to media/pci

2012-08-15T19:36:22+00:00

Rename all PCI drivers with their own directory under drivers/media/video into drivers/media/pci and update the building system. Signed-off-by: Mauro Carvalho Chehab

[media] cx88: properly maintain decoder config when using MPEG encoder

2011-07-27T20:56:00+00:00

The cx88 driver would force core->input to always be zero when doing the the request_acquire(). While it wasn't actually changing the input register in the hardware, the driver makes decision based on the current input. In particular, it decides whether to do things like enabling the comb filter when on a composite input but disabling it on s-video. So for example, on the HVR-1300, using the s-video input with the MPEG encoder would end up with the video decoder core configured as though the input type were composite. In short, the driver state did not match the hardware state. This patch does two things: 1. It forces the input to zero only if actually switching to DVB mode. This prevents the input from changing when the blackbird driver opens the device. 2. Keep track of what the input was set to when switching to DVB, and reset it back when done. This eliminates a condition where for example the user had the analog side of the board set to capture on the s-video input, then he used DVB for a bit, then the analog input would unexpectedly be set to the tuner input. This work was sponsored by Anevia S.A. Signed-off-by: Devin Heitmueller Cc: Florent Audebert Signed-off-by: Mauro Carvalho Chehab

[media] Stop using linux/version.h on most video drivers

2011-07-27T20:53:12+00:00

All the modified drivers didn't have any version increment since Jan, 1 2011. Several of them didn't have any version increment for a long time, even having new features and important bug fixes happening. As we're now filling the QUERYCAP version with the current Kernel Release, we don't need to maintain a per-driver version control anymore. So, let's just use the default. In order to preserve the Kernel module version history, a KERNEL_VERSION() macro were added to all modified drivers, and the extraver number were incremented. I opted to preserve the per-driver version control to a few pwc, pvrusb2, s2255, s5p-fimc and sh_vou. A few drivers are still using the legacy way to handle ioctl's. So, we can't do such change on them, otherwise, they'll break. Those are: uvc, et61x251 and sn9c102. The rationale is that the per-driver version control seems to be actively maintained on those. Yet, I think that the better for them would be to just use the default version numbering, instead of doing that by themselves. While here, removed a few uneeded include linux/version.h Acked-by: Hans Verkuil Signed-off-by: Mauro Carvalho Chehab

[media] cx88: protect cx8802_devlist with a mutex

2011-05-20T12:29:38+00:00

Add and use a mutex to protect the cx88-mpeg device list. Previously the BKL prevented races. Based on work by Ben Hutchings . Tested-by: Andi Huber Tested-by: Marlon de Boer Signed-off-by: Jonathan Nieder Signed-off-by: Mauro Carvalho Chehab

[media] cx88: hold device lock during sub-driver initialization

2011-05-20T12:29:37+00:00

cx8802_blackbird_probe makes a device node for the mpeg sub-device before it has been added to dev->drvlist. If the device is opened during that time, the open succeeds but request_acquire cannot be called, so the reference count remains zero. Later, when the device is closed, the reference count becomes negative --- uh oh. Close the race by holding core->lock during probe and not releasing until the device is in drvlist and initialization finished. Previously the BKL prevented this race. Reported-by: Andreas Huber Tested-by: Andi Huber Tested-by: Marlon de Boer Cc: stable@kernel.org Signed-off-by: Jonathan Nieder Signed-off-by: Mauro Carvalho Chehab

[media] cx88: fix locking of sub-driver operations

2011-05-20T12:29:36+00:00

The BKL conversion of this driver seems to have gone wrong. Loading the cx88-blackbird driver deadlocks. The cause: mpeg_ops::open in the cx2388x blackbird driver acquires the device lock and calls the sub-driver's request_acquire, which tries to acquire the lock again. Fix it by clarifying the semantics of request_acquire, request_release, advise_acquire, and advise_release: now all will rely on the caller to acquire the device lock. Based on work by Ben Hutchings . Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=31962 Reported-by: Andi Huber Tested-by: Andi Huber Tested-by: Marlon de Boer Cc: stable@kernel.org Signed-off-by: Jonathan Nieder Signed-off-by: Mauro Carvalho Chehab

[media] cx88: protect per-device driver list with device lock

2011-05-20T12:29:34+00:00

The BKL conversion of this driver seems to have gone wrong. Various uses of the sub-device and driver lists appear to be subject to race conditions. In particular, some functions access drvlist without a relevant lock held, which will race against removal of drivers. Let's start with that --- clean up by consistently protecting dev->drvlist with dev->core->lock, noting driver functions that require the device lock to be held or not to be held. After this patch, there are still some races --- e.g., cx8802_blackbird_remove can run between the time the blackbird driver is acquired and the time it is used in mpeg_release, and there's a similar race in cx88_dvb_bus_ctrl. Later patches will address the remaining known races and the deadlock noticed by Andi. This patch just makes the semantics clearer in preparation for those later changes. Based on work by Ben Hutchings . Tested-by: Andi Huber Tested-by: Marlon de Boer Cc: stable@kernel.org Signed-off-by: Jonathan Nieder Signed-off-by: Mauro Carvalho Chehab

[media] use pci_dev->revision

2011-05-20T10:26:16+00:00

pci_setup_device() has saved the PCI revision in the pci_dev struct since Linux 2.6.23. Use it. Cc: Auke Kok Signed-off-by: Bjørn Mork Signed-off-by: Mauro Carvalho Chehab

media/video: explicitly flush request_module work

2010-12-24T15:14:20+00:00

Video drivers request submodules using a work during probe and calls flush_scheduled_work() on exit to make sure the work is complete before being unloaded. This patch makes these drivers flush the work directly instead of using flush_scheduled_work(). While at it, relocate request_submodules() call in saa7134_initdev() right right before successful return as in other drivers to avoid failing after the work is scheduled and returning failure without the work still active. This is in preparation for the deprecation of flush_scheduled_work(). Signed-off-by: Tejun Heo Cc: Mauro Carvalho Chehab

V4L/DVB: drivers/media: Make static data tables and strings const

2010-10-21T03:04:52+00:00

Making static data const avoids allocation of additional r/w memory and reduces initialisation time. It also provides some additional opportunities for compiler optimisations. Signed-off-by: Lawrence Rust Signed-off-by: Mauro Carvalho Chehab