Re: [PATCH] media: vimc: Add support for contiguous DMA buffers

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Nicolas,

On Fri, Jul 30, 2021 at 11:59:27AM -0400, Nicolas Dufresne wrote:
> Le vendredi 30 juillet 2021 à 17:15 +0300, Laurent Pinchart a écrit :
> > On Fri, Jul 30, 2021 at 04:08:11PM +0200, Dafna Hirschfeld wrote:
> > > On 30.07.21 15:11, Laurent Pinchart wrote:
> > > > On Fri, Jul 30, 2021 at 02:35:20PM +0200, Dafna Hirschfeld wrote:
> > > > > On 30.07.21 02:19, Laurent Pinchart wrote:
> > > > > > The vimc driver is used for testing purpose, and some test use cases
> > > > > > involve sharing buffers with a consumer device. Consumers often require
> > > > > > DMA contiguous memory, which vimc doesn't currently support. This leads
> > > > > > in the best case to usage of bounce buffers, which is very slow, and in
> > > > > > the worst case in a complete failure.
> > > > > > 
> > > > > > Add support for the dma-contig allocator in vimc to support those use
> > > > > > cases properly. The allocator is selected through a new "allocator"
> > > > > > module parameter, which defaults to vmalloc.
> > > > > > 
> > > > > > Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
> > > > > > ---
> > > > > >    drivers/media/test-drivers/vimc/vimc-capture.c |  9 +++++++--
> > > > > >    drivers/media/test-drivers/vimc/vimc-common.h  |  2 ++
> > > > > >    drivers/media/test-drivers/vimc/vimc-core.c    | 10 ++++++++++
> > > > > >    3 files changed, 19 insertions(+), 2 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/media/test-drivers/vimc/vimc-capture.c b/drivers/media/test-drivers/vimc/vimc-capture.c
> > > > > > index 5e9fd902cd37..92b69a6529fb 100644
> > > > > > --- a/drivers/media/test-drivers/vimc/vimc-capture.c
> > > > > > +++ b/drivers/media/test-drivers/vimc/vimc-capture.c
> > > > > > @@ -7,6 +7,7 @@
> > > > > >    
> > > > > >    #include <media/v4l2-ioctl.h>
> > > > > >    #include <media/videobuf2-core.h>
> > > > > > +#include <media/videobuf2-dma-contig.h>
> > > > > >    #include <media/videobuf2-vmalloc.h>
> > > > > >    
> > > > > >    #include "vimc-common.h"
> > > > > > @@ -423,14 +424,18 @@ static struct vimc_ent_device *vimc_cap_add(struct vimc_device *vimc,
> > > > > >    	/* Initialize the vb2 queue */
> > > > > >    	q = &vcap->queue;
> > > > > >    	q->type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
> > > > > > -	q->io_modes = VB2_MMAP | VB2_DMABUF | VB2_USERPTR;
> > > > > > +	q->io_modes = VB2_MMAP | VB2_DMABUF;
> > > > > 
> > > > > maybe to be on the safe side VB2_DMABUF should be set only if vimc_allocator==1 ?
> > > > 
> > > > Why so ? vb2-vmalloc can import dma-bufs.
> > > 
> > > oh, I meant that exporting should not be supported.
> > > I see that vimc set ".vidioc_expbuf = vb2_ioctl_expbuf", maybe remove that if allocator is vmalloc?
> > 
> > If the importer support non-contiguous buffers, vb2-vmalloc can be used
> > as an exporter. I've successfully used this to test sharing buffers
> > between vimc in vmalloc mode and the R-Car H3 display driver with an
> > IOMMU.
> 
> Having an IOMMU is not sufficient, as this is shown with Intel DRM. The DRM
> driver needs to support CPU cache. Note that in GStreamer this is used a lot
> from UVC camera to GL (but it breaks, corrupted images, with kmssink).

That depends on the platform. When the sink device isn't cache-coherent,
it needs to perform a cache sync operation. V4L2 handles this
automatically for codecs (technically speaking for cameras as well, but
cameras are rarely sinks :-)). In DRM/KMS, there's still quite a bit of
work to do. It's not an issue in vimc in any case.

> > > > > > +	if (vimc_allocator != 1)
> > > > > 
> > > > > maybe define a macro instead of `1` ?
> > > > 
> > > > Good idea.
> > > > 
> > > > > > +		q->io_modes |= VB2_USERPTR;
> > > > > >    	q->drv_priv = vcap;
> > > > > >    	q->buf_struct_size = sizeof(struct vimc_cap_buffer);
> > > > > >    	q->ops = &vimc_cap_qops;
> > > > > > -	q->mem_ops = &vb2_vmalloc_memops;
> > > > > > +	q->mem_ops = vimc_allocator == 1
> > > > > > +		   ? &vb2_dma_contig_memops : &vb2_vmalloc_memops;
> > > > > >    	q->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC;
> > > > > >    	q->min_buffers_needed = 2;
> > > > > >    	q->lock = &vcap->lock;
> > > > > > +	q->dev = v4l2_dev->dev;
> > > > > >    
> > > > > >    	ret = vb2_queue_init(q);
> > > > > >    	if (ret) {
> > > > > > diff --git a/drivers/media/test-drivers/vimc/vimc-common.h b/drivers/media/test-drivers/vimc/vimc-common.h
> > > > > > index a289434e75ba..b77939123501 100644
> > > > > > --- a/drivers/media/test-drivers/vimc/vimc-common.h
> > > > > > +++ b/drivers/media/test-drivers/vimc/vimc-common.h
> > > > > > @@ -35,6 +35,8 @@
> > > > > >    
> > > > > >    #define VIMC_PIX_FMT_MAX_CODES 8
> > > > > >    
> > > > > > +extern unsigned int vimc_allocator;
> > > > > > +
> > > > > >    /**
> > > > > >     * vimc_colorimetry_clamp - Adjust colorimetry parameters
> > > > > >     *
> > > > > > diff --git a/drivers/media/test-drivers/vimc/vimc-core.c b/drivers/media/test-drivers/vimc/vimc-core.c
> > > > > > index 4b0ae6f51d76..7badcecb7aed 100644
> > > > > > --- a/drivers/media/test-drivers/vimc/vimc-core.c
> > > > > > +++ b/drivers/media/test-drivers/vimc/vimc-core.c
> > > > > > @@ -5,6 +5,7 @@
> > > > > >     * Copyright (C) 2015-2017 Helen Koike <helen.fornazier@gmail.com>
> > > > > >     */
> > > > > >    
> > > > > > +#include <linux/dma-mapping.h>
> > > > > >    #include <linux/font.h>
> > > > > >    #include <linux/init.h>
> > > > > >    #include <linux/module.h>
> > > > > > @@ -15,6 +16,12 @@
> > > > > >    
> > > > > >    #include "vimc-common.h"
> > > > > >    
> > > > > > +unsigned int vimc_allocator;
> > > > > > +module_param_named(allocator, vimc_allocator, uint, 0444);
> > > > > > +MODULE_PARM_DESC(allocator, " memory allocator selection, default is 0.\n"
> > > > > > +			     "\t\t    0 == vmalloc\n"
> > > > > > +			     "\t\t    1 == dma-contig");
> > > > > > +
> > > > > 
> > > > > There is a section 'Module options' in vimc.rst. So a doc should be added there.
> > > > 
> > > > OK, I'll update that.
> > > > 
> > > > > >    #define VIMC_MDEV_MODEL_NAME "VIMC MDEV"
> > > > > >    
> > > > > >    #define VIMC_ENT_LINK(src, srcpad, sink, sinkpad, link_flags) {	\
> > > > > > @@ -278,6 +285,9 @@ static int vimc_probe(struct platform_device *pdev)
> > > > > >    
> > > > > >    	tpg_set_font(font->data);
> > > > > >    
> > > > > > +	if (vimc_allocator == 1)
> > > > > > +		dma_coerce_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
> > > > > > +
> > > > > >    	vimc = kzalloc(sizeof(*vimc), GFP_KERNEL);
> > > > > >    	if (!vimc)
> > > > > >    		return -ENOMEM;
> > > > > > 

-- 
Regards,

Laurent Pinchart