On 06/16/2016 07:02 PM, Steve Longerbeam wrote: > On 06/16/2016 02:49 AM, Jack Mitchell wrote: >> >> On 16/06/16 02:37, Steve Longerbeam wrote: >>> Hi Jack, >>> >>> On 06/15/2016 03:43 AM, Jack Mitchell wrote: >>>> <snip> >>>> Trying to use a user pointer rather than mmap also fails and causes a kernel splat. >>>> >>> >>> Hmm, I've tested userptr with the mem2mem driver, but maybe never >>> with video capture. I tried "v4l2-ctl -d/dev/video0 --stream-user=8" but >>> that returns "VIDIOC_QBUF: failed: Invalid argument", haven't tracked >>> down why (could be a bug in v4l2-ctl). Can you share the splat? >>> >> >> On re-checking the splat was the same v4l_cropcap that was mentioned before so I don't think it's related. The error I get back is: >> >> VIDIOC_QBUF error 22, Invalid argument >> >> I'm using the example program the the v4l2 docs [1]. > > I found the cause at least in my case. After enabling dynamic debug in > videobuf2-dma-contig.c, "v4l2-ctl -d/dev/video0 --stream-user=8" gives > me > > [ 468.826046] user data must be aligned to 64 bytes > > > > But even getting past that alignment issue, I've only tested userptr (in mem2mem > driver) by giving the driver a user address of a mmap'ed kernel contiguous > buffer. A true discontiguous user buffer may not work, the IPU DMA does not > support scatter-gather. I don't think VB2_USERPTR should be enabled in this case due to the DMA limitations. It won't allow you to use malloc()ed memory and the hack that allows you to pass contiguous memory is superseded by the DMABUF mode. Regards, Hans -- 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
