Subject: Re: [PATCH v4] [media] vb2: Add support for capture_dma_bidirectional
 queue flag
To: Thierry Escande <thierry.escande@collabora.com>,
        Mauro Carvalho Chehab <mchehab@kernel.org>,
        Sakari Ailus <sakari.ailus@iki.fi>
Cc: linux-media@vger.kernel.org, linux-kernel@vger.kernel.org,
        Pawel Osciak <pawel@osciak.com>,
        Kyungmin Park <kyungmin.park@samsung.com>,
        Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>,
        Sylwester Nawrocki <s.nawrocki@samsung.com>
From: Marek Szyprowski <m.szyprowski@samsung.com>
Message-id: <b6bfc5f3-df78-0938-73da-b431b91f3335@samsung.com>
Date: Mon, 24 Oct 2016 10:20:04 +0200
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:45.0) Gecko/20100101
        Thunderbird/45.4.0
MIME-version: 1.0
In-reply-to: <1477294221-10912-1-git-send-email-thierry.escande@collabora.com>
Content-type: text/plain; charset=utf-8; format=flowed
Content-transfer-encoding: 7bit
CMS-TYPE: 201P
References: <CGME20161024073048epcas4p39068cc1699f05d7902038d0d4311a60d@epcas4p3.samsung.com>
 <1477294221-10912-1-git-send-email-thierry.escande@collabora.com>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 8545
Lines: 203

Hi Thierry,


On 2016-10-24 09:30, Thierry Escande wrote:
> From: Pawel Osciak <posciak@chromium.org>
>
> When this flag is set for CAPTURE queues by the driver on calling
> vb2_queue_init(), it forces the buffers on the queue to be
> allocated/mapped with DMA_BIDIRECTIONAL direction flag instead of
> DMA_FROM_DEVICE. This allows the device not only to write to the
> buffers, but also read out from them. This may be useful e.g. for codec
> hardware which may be using CAPTURE buffers as reference to decode
> other buffers.
>
> This flag is ignored for OUTPUT queues as we don't want to allow HW to
> be able to write to OUTPUT buffers.
>
> This patch introduces 2 macros:
> VB2_DMA_DIR(q) returns the corresponding dma_dir for the passed queue
> type, tanking care of the capture_dma_birectional flag.
>
> VB2_DMA_DIR_CAPTURE(d) is a test macro returning true if the passed DMA
> direction refers to a capture buffer. This test is used to map virtual
> addresses for writing and to mark pages as dirty.
>
> Signed-off-by: Pawel Osciak <posciak@chromium.org>
> Tested-by: Pawel Osciak <posciak@chromium.org>
> Signed-off-by: Thierry Escande <thierry.escande@collabora.com>

Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>

Thanks for this patch. It reminds me that I have to look again into s5p-mfc
codec driver and also fix it to use this feature. It worked so far only
because the current Exynos IOMMU driver doesn't implement access protection
checks...

> ---
>
> Changes since v1:
> - Renamed use_dma_bidirectional field as capture_dma_bidirectional
> - Added a VB2_DMA_DIR() macro
>
> Changes since v2:
> - Get rid of dma_dir field and therefore squashed the previous patch
>
> Changes since v3:
> - Fixed typos in include/media/videobuf2-core.h
> - Added VB2_DMA_DIR_CAPTURE() test macro
>
>
>   drivers/media/v4l2-core/videobuf2-core.c       |  9 +++------
>   drivers/media/v4l2-core/videobuf2-dma-contig.c |  2 +-
>   drivers/media/v4l2-core/videobuf2-dma-sg.c     |  5 +++--
>   drivers/media/v4l2-core/videobuf2-vmalloc.c    |  4 ++--
>   include/media/videobuf2-core.h                 | 24 ++++++++++++++++++++++++
>   5 files changed, 33 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/media/v4l2-core/videobuf2-core.c b/drivers/media/v4l2-core/videobuf2-core.c
> index 21900202..22d6105 100644
> --- a/drivers/media/v4l2-core/videobuf2-core.c
> +++ b/drivers/media/v4l2-core/videobuf2-core.c
> @@ -194,8 +194,7 @@ static void __enqueue_in_driver(struct vb2_buffer *vb);
>   static int __vb2_buf_mem_alloc(struct vb2_buffer *vb)
>   {
>   	struct vb2_queue *q = vb->vb2_queue;
> -	enum dma_data_direction dma_dir =
> -		q->is_output ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
> +	enum dma_data_direction dma_dir = VB2_DMA_DIR(q);
>   	void *mem_priv;
>   	int plane;
>   	int ret = -ENOMEM;
> @@ -978,8 +977,7 @@ static int __qbuf_userptr(struct vb2_buffer *vb, const void *pb)
>   	void *mem_priv;
>   	unsigned int plane;
>   	int ret = 0;
> -	enum dma_data_direction dma_dir =
> -		q->is_output ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
> +	enum dma_data_direction dma_dir = VB2_DMA_DIR(q);
>   	bool reacquired = vb->planes[0].mem_priv == NULL;
>   
>   	memset(planes, 0, sizeof(planes[0]) * vb->num_planes);
> @@ -1096,8 +1094,7 @@ static int __qbuf_dmabuf(struct vb2_buffer *vb, const void *pb)
>   	void *mem_priv;
>   	unsigned int plane;
>   	int ret = 0;
> -	enum dma_data_direction dma_dir =
> -		q->is_output ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
> +	enum dma_data_direction dma_dir = VB2_DMA_DIR(q);
>   	bool reacquired = vb->planes[0].mem_priv == NULL;
>   
>   	memset(planes, 0, sizeof(planes[0]) * vb->num_planes);
> diff --git a/drivers/media/v4l2-core/videobuf2-dma-contig.c b/drivers/media/v4l2-core/videobuf2-dma-contig.c
> index fb6a177..a44e383 100644
> --- a/drivers/media/v4l2-core/videobuf2-dma-contig.c
> +++ b/drivers/media/v4l2-core/videobuf2-dma-contig.c
> @@ -507,7 +507,7 @@ static void *vb2_dc_get_userptr(struct device *dev, unsigned long vaddr,
>   	buf->dma_dir = dma_dir;
>   
>   	offset = vaddr & ~PAGE_MASK;
> -	vec = vb2_create_framevec(vaddr, size, dma_dir == DMA_FROM_DEVICE);
> +	vec = vb2_create_framevec(vaddr, size, VB2_DMA_DIR_CAPTURE(dma_dir));
>   	if (IS_ERR(vec)) {
>   		ret = PTR_ERR(vec);
>   		goto fail_buf;
> diff --git a/drivers/media/v4l2-core/videobuf2-dma-sg.c b/drivers/media/v4l2-core/videobuf2-dma-sg.c
> index ecff8f49..51c98f6 100644
> --- a/drivers/media/v4l2-core/videobuf2-dma-sg.c
> +++ b/drivers/media/v4l2-core/videobuf2-dma-sg.c
> @@ -238,7 +238,8 @@ static void *vb2_dma_sg_get_userptr(struct device *dev, unsigned long vaddr,
>   	buf->offset = vaddr & ~PAGE_MASK;
>   	buf->size = size;
>   	buf->dma_sgt = &buf->sg_table;
> -	vec = vb2_create_framevec(vaddr, size, buf->dma_dir == DMA_FROM_DEVICE);
> +	vec = vb2_create_framevec(vaddr, size,
> +				  VB2_DMA_DIR_CAPTURE(buf->dma_dir));
>   	if (IS_ERR(vec))
>   		goto userptr_fail_pfnvec;
>   	buf->vec = vec;
> @@ -291,7 +292,7 @@ static void vb2_dma_sg_put_userptr(void *buf_priv)
>   		vm_unmap_ram(buf->vaddr, buf->num_pages);
>   	sg_free_table(buf->dma_sgt);
>   	while (--i >= 0) {
> -		if (buf->dma_dir == DMA_FROM_DEVICE)
> +		if (VB2_DMA_DIR_CAPTURE(buf->dma_dir))
>   			set_page_dirty_lock(buf->pages[i]);
>   	}
>   	vb2_destroy_framevec(buf->vec);
> diff --git a/drivers/media/v4l2-core/videobuf2-vmalloc.c b/drivers/media/v4l2-core/videobuf2-vmalloc.c
> index ab3227b..76649bd 100644
> --- a/drivers/media/v4l2-core/videobuf2-vmalloc.c
> +++ b/drivers/media/v4l2-core/videobuf2-vmalloc.c
> @@ -86,7 +86,7 @@ static void *vb2_vmalloc_get_userptr(struct device *dev, unsigned long vaddr,
>   	buf->dma_dir = dma_dir;
>   	offset = vaddr & ~PAGE_MASK;
>   	buf->size = size;
> -	vec = vb2_create_framevec(vaddr, size, dma_dir == DMA_FROM_DEVICE);
> +	vec = vb2_create_framevec(vaddr, size, VB2_DMA_DIR_CAPTURE(dma_dir));
>   	if (IS_ERR(vec)) {
>   		ret = PTR_ERR(vec);
>   		goto fail_pfnvec_create;
> @@ -136,7 +136,7 @@ static void vb2_vmalloc_put_userptr(void *buf_priv)
>   		pages = frame_vector_pages(buf->vec);
>   		if (vaddr)
>   			vm_unmap_ram((void *)vaddr, n_pages);
> -		if (buf->dma_dir == DMA_FROM_DEVICE)
> +		if (VB2_DMA_DIR_CAPTURE(buf->dma_dir))
>   			for (i = 0; i < n_pages; i++)
>   				set_page_dirty_lock(pages[i]);
>   	} else {
> diff --git a/include/media/videobuf2-core.h b/include/media/videobuf2-core.h
> index ac5898a..98379ba 100644
> --- a/include/media/videobuf2-core.h
> +++ b/include/media/videobuf2-core.h
> @@ -433,6 +433,9 @@ struct vb2_buf_ops {
>    * @quirk_poll_must_check_waiting_for_buffers: Return POLLERR at poll when QBUF
>    *              has not been called. This is a vb1 idiom that has been adopted
>    *              also by vb2.
> + * @capture_dma_bidirectional:	use DMA_BIDIRECTIONAL for CAPTURE buffers; this
> + *				allows HW to read from the CAPTURE buffers in
> + *				addition to writing; ignored for OUTPUT queues.
>    * @lock:	pointer to a mutex that protects the vb2_queue struct. The
>    *		driver can set this to a mutex to let the v4l2 core serialize
>    *		the queuing ioctls. If the driver wants to handle locking
> @@ -499,6 +502,7 @@ struct vb2_queue {
>   	unsigned			fileio_write_immediately:1;
>   	unsigned			allow_zero_bytesused:1;
>   	unsigned		   quirk_poll_must_check_waiting_for_buffers:1;
> +	unsigned			capture_dma_bidirectional:1;
>   
>   	struct mutex			*lock;
>   	void				*owner;
> @@ -554,6 +558,26 @@ struct vb2_queue {
>   #endif
>   };
>   
> +/*
> + * Returns the corresponding DMA direction given the vb2_queue type (capture or
> + * output). Returns DMA_BIDIRECTIONAL for capture buffers if the vb2_queue field
> + * capture_dma_bidirectional is set by the driver.
> + */
> +#define VB2_DMA_DIR(q) (V4L2_TYPE_IS_OUTPUT((q)->type)   \
> +			? DMA_TO_DEVICE                  \
> +			: (q)->capture_dma_bidirectional \
> +			  ? DMA_BIDIRECTIONAL            \
> +			  : DMA_FROM_DEVICE)
> +
> +/*
> + * Returns true if the DMA direction passed as parameter refers to a capture
> + * buffer as capture buffers allow both FROM_DEVICE and BIDIRECTIONAL DMA
> + * direction. This test is used to map virtual addresses for writing and to mark
> + * pages as dirty.
> + */
> +#define VB2_DMA_DIR_CAPTURE(d) \
> +			((d) == DMA_FROM_DEVICE || (d) == DMA_BIDIRECTIONAL)
> +
>   /**
>    * vb2_plane_vaddr() - Return a kernel virtual address of a given plane
>    * @vb:		vb2_buffer to which the plane in question belongs to

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland