The UVC video driver converts the timestamp from hardware specific unit to one known by the kernel at the time when the buffer is dequeued. This is fine in general, but the streamoff operation consists of the following steps (among other things): 1. uvc_video_clock_cleanup --- the hardware clock sample array is released and the pointer to the array is set to NULL, 2. buffers in active state are returned to the user and 3. buf_finish callback is called on buffers that are prepared. buf_finish includes calling uvc_video_clock_update that accesses the hardware clock sample array. The above is serialised by a queue specific mutex. Address the problem by skipping the clock conversion if the hardware clock sample array is already released. Reported-by: Chiranjeevi Rapolu <chiranjeevi.rapolu@xxxxxxxxx> Tested-by: Chiranjeevi Rapolu <chiranjeevi.rapolu@xxxxxxxxx> Signed-off-by: Sakari Ailus <sakari.ailus@xxxxxxxxxxxxxxx> --- Hi Laurent, This seems like something that's been out there for a while... I'll figure out soon which stable kernels should receive it, if any. drivers/media/usb/uvc/uvc_video.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c index 84525ff047450..a30c5e1893e72 100644 --- a/drivers/media/usb/uvc/uvc_video.c +++ b/drivers/media/usb/uvc/uvc_video.c @@ -676,6 +676,13 @@ void uvc_video_clock_update(struct uvc_streaming *stream, if (!uvc_hw_timestamps_param) return; + /* + * We may get called if there are buffers done but not dequeued by the + * user. Just bail out in that case. + */ + if (!clock->samples) + return; + spin_lock_irqsave(&clock->lock, flags); if (clock->count < clock->size) -- 2.11.0
