[PATCH v2 6/8] [Media] vcodec: mediatek: Add Mediatek V4L2 Video Encoder Driver

tiffany lin tiffany.lin at mediatek.com
Wed Dec 16 21:52:27 PST 2015


On Wed, 2015-12-16 at 14:47 +0100, Hans Verkuil wrote:
> On 12/16/15 14:17, tiffany lin wrote:
> > Hi Hans,
> > 
> > 
> > On Tue, 2015-12-15 at 15:17 +0100, Hans Verkuil wrote:
> >>
> >> On 12/15/15 14:51, tiffany lin wrote:
> >>> We are not familiar with v4l2-compliance utility, we will check how to
> >>> use it.
> >>
> >> It's part of v4l-utils.git (http://git.linuxtv.org/v4l-utils.git/). There is a
> >> fairly decent man page. It does exhaustive compliance tests for V4L2 devices.
> >>
> >> That said, the support for memory-to-memory codec devices is not great, so I wouldn't
> >> trust any failures it reports when using the streaming tests (i.e. the --stream*
> >> options). By default just run 'v4l2-compliance -d /dev/videoX' to do the compliance
> >> test.
> >>
> >> Note: before I accept this driver I do want to see that compliance test output!
> >>
> > Got it. We will provide it in next version.
> > Now our driver is developed and run base on kernel v3.18.
> > V4L2 and vb2 have some difference between Linux 4.4-rc1 and 3.18 kernel.
> > Is it ok we provided test output base on v3.18 or we need to base on
> > 4.4-rc1?
> 
> I'm actually not sure if the latest v4l2-compliance test suite will work with a 3.18
> kernel. so either you have to go back to an older version of v4l2-compliance that
> works with 3.18 (go back to commit 4a57509a8334aca6ca8e81cd3beb08d5be397dac, that
> might do the trick) or (and that's what I recommend) go with the latest kernel.
> 
> For the media tree that is http://git.linuxtv.org/media_tree.git/log/.
> 
> The final version of the patch has to be against that kernel anyway.
> 
Got it. Thanks for your help, we will try it. ^_^

> >>>>> +}
> >>>>> +
> >>>>> +int m2mctx_venc_queue_init(void *priv, struct vb2_queue *src_vq,
> >>>>> +			   struct vb2_queue *dst_vq)
> >>>>> +{
> >>>>> +	struct mtk_vcodec_ctx *ctx = priv;
> >>>>> +	int ret;
> >>>>> +
> >>>>> +	src_vq->type		= V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
> >>>>> +	src_vq->io_modes	= VB2_DMABUF | VB2_MMAP | VB2_USERPTR;
> >>>>
> >>>> You're using videobuf2-dma-contig, so VB2_USERPTR is generally useless in that
> >>>> case. I would drop it.
> >>>>
> >>> Sorry, I don't get it. 
> >>> We are using videobuf2-dma-contig, but we also using VB2_USERPTR.
> >>
> >> ???? In that case the user pointer you pass in must point to physically contiguous
> >> memory. Which means you got it through some magic. Typically what should be used
> >> are dmabuf handles to pass buffers around between different subsystems.
> >>
> >> The use of VB2_USERPTR for that purpose is deprecated.
> >>
> >> Or am I misunderstanding you as well?
> >>
> > Our encoder support all three modes.
> > In case that A driver + Encode driver flow, OUTPUT buffer will be
> > VB2_DMABUF from A driver.
> > In case that read YCbCr frame data from file and encode them to bit
> > stream flow, we use VB2_USERPTR and VB2_MMAP.
> > In VB2_USERPTR case, videobuf2-dma-contig will help us get continuous
> > dma address.
> > Our chip has IOMMU and M4U that help us get continuous phy address for
> > encode HW.
> > 
> > http://lists.infradead.org/pipermail/linux-mediatek/2015-October/002525.html
> 
> Ah, OK. Have you tested this with malloc()ed buffers? Just asking :-)
> 
Yes. Actually we tested this with new()ed buffers. I think it default
call malloc.

> Regards,
> 
> 	Hans

best regards,
Tiffany




More information about the linux-arm-kernel mailing list