|[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Hi Hans, On 06/24/2012 01:25 PM, Hans Verkuil wrote:
Hi all, In the 3.5 kernel a change was made regarding core locking (i.e. how locking is done if the 'lock' field is set in struct video_device). Before kernel 3.5 all file operations would be locked in that case. But this turned out to be problematic, adding unnecessary latencies in some cases (poll) and potentially introducing a deadlock in mmap: the kernel takes the mm semaphore before calling the driver's file operation, where the core lock is taken, whereas other file ops take the core lock first, and in case of a page fault the mm semaphore will be taken. This scenario is very unlikely, but the lockdep checker will complain about it. So in kernel 3.5 we decided not to take the core lock anymore for file operations other than unlocked_ioctl. Instead the driver will have to do the locking. As a temporary measure the flag V4L2_FL_LOCK_ALL_FOPS was introduced that, if set, would force the core to take the lock anyway for all file operations. In other words, drivers that were not converted yet to do their own locking for non-unlocked_ioctl file operations would just set this flag and keep the old behavior. This patch series goes through all those drivers and pushed the locking down into the driver and removes this legacy flag. These patches just push the locking down and do not do anything smart (except for some additional dead code removal in saa7146, or if I was 100% certain no locking was needed for a particular file operation). In particular for mmap it will still take the core lock, it just does it in the driver now, making it easier to change in a future patch. I have already tested ivtv, saa7146, cpia2, usbvision, em28xx, tm6000, mem2mem_testdev and cx231xx. I hope to test vpif_capture/display tomorrow. The others need to be tested and/or reviewed by others (i.e. you!). I'd really want to get rid of this flag as soon as possible as it makes the v4l2 core lock handling unnecessarily complex, and it also complicates the core and vb2 enhancements patch series I am working on: http://www.spinics.net/lists/linux-media/msg49299.html So please take a look, test if you can, and let me know if it is OK.
I prepared patches removing the V4L2_FL_LOCK_ALL_FOPS flag in s5p-fimc driver a few weeks ago (and also for fimc-lite), just didn't post it yet. I'm going to do it in coming week, when I'm back in the office. Please feel free to drop patches for s5p-fimc in this series and mark off this driver on your list. -- Kind regards, Sylwester -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html