[v1,2/2] media: staging: tegra-vde: Support V4L stateless video decoder API

Expose Tegra video decoder as a generic V4L M2M stateless video decoder.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/staging/media/tegra-vde/Kconfig       |    7 +
 drivers/staging/media/tegra-vde/Makefile      |    2 +-
 drivers/staging/media/tegra-vde/h264.c        |  345 +++++-
 drivers/staging/media/tegra-vde/h264_reader.c |  264 +++++
 drivers/staging/media/tegra-vde/v4l2.c        | 1013 +++++++++++++++++
 drivers/staging/media/tegra-vde/vde.c         |   74 +-
 drivers/staging/media/tegra-vde/vde.h         |   87 ++
 7 files changed, 1784 insertions(+), 8 deletions(-)
 create mode 100644 drivers/staging/media/tegra-vde/h264_reader.c
 create mode 100644 drivers/staging/media/tegra-vde/v4l2.c

Le mercredi 12 janvier 2022 à 18:39 +0300, Dmitry Osipenko a écrit :
> Expose Tegra video decoder as a generic V4L M2M stateless video decoder.

Thanks for working on this. Note that it would be nice to provide V4L2
compliance test result, and if this is actually possible, provide fluster
conformance results using ffmpeg, gstreamer, chromium or your own decoder,
though if its your own, it would be nice to share a bit more so we can check
that it's not interpreting the uAPI differently from other (we'd like drivers to
work on multiple userland ideally).

As usual I leave to other doing proper review, I added a comment below, pointing
out the presence of a bitstream parser in this driver, and suggested an
amendment to the spec to get rid of this. For me the code looks otherwise quite
straight forward, is there any known issue that would keep this driver in
staging ?

Please see below ...

> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/staging/media/tegra-vde/Kconfig       |    7 +
>  drivers/staging/media/tegra-vde/Makefile      |    2 +-
>  drivers/staging/media/tegra-vde/h264.c        |  345 +++++-
>  drivers/staging/media/tegra-vde/h264_reader.c |  264 +++++
>  drivers/staging/media/tegra-vde/v4l2.c        | 1013 +++++++++++++++++
>  drivers/staging/media/tegra-vde/vde.c         |   74 +-
>  drivers/staging/media/tegra-vde/vde.h         |   87 ++
>  7 files changed, 1784 insertions(+), 8 deletions(-)
>  create mode 100644 drivers/staging/media/tegra-vde/h264_reader.c
>  create mode 100644 drivers/staging/media/tegra-vde/v4l2.c
> 
> diff --git a/drivers/staging/media/tegra-vde/Kconfig b/drivers/staging/media/tegra-vde/Kconfig
> index 0dc78afd09e0..07dbc1f44ca8 100644
> --- a/drivers/staging/media/tegra-vde/Kconfig
> +++ b/drivers/staging/media/tegra-vde/Kconfig
> @@ -2,9 +2,16 @@
>  config TEGRA_VDE
>  	tristate "NVIDIA Tegra Video Decoder Engine driver"
>  	depends on ARCH_TEGRA || COMPILE_TEST
> +	depends on VIDEO_DEV && VIDEO_V4L2
>  	select DMA_SHARED_BUFFER
>  	select IOMMU_IOVA
> +	select MEDIA_CONTROLLER
> +	select MEDIA_CONTROLLER_REQUEST_API
>  	select SRAM
> +	select VIDEOBUF2_DMA_CONTIG
> +	select VIDEOBUF2_DMA_SG
> +	select V4L2_H264
> +	select V4L2_MEM2MEM_DEV
>  	help
>  	    Say Y here to enable support for the NVIDIA Tegra video decoder
>  	    driver.
> diff --git a/drivers/staging/media/tegra-vde/Makefile b/drivers/staging/media/tegra-vde/Makefile
> index 43525b08b3b0..c5b15a822cfa 100644
> --- a/drivers/staging/media/tegra-vde/Makefile
> +++ b/drivers/staging/media/tegra-vde/Makefile
> @@ -1,3 +1,3 @@
>  # SPDX-License-Identifier: GPL-2.0
> -tegra-vde-y := vde.o iommu.o dmabuf-cache.o h264.o
> +tegra-vde-y := vde.o iommu.o dmabuf-cache.o h264.o h264_reader.o v4l2.o
>  obj-$(CONFIG_TEGRA_VDE)	+= tegra-vde.o
> diff --git a/drivers/staging/media/tegra-vde/h264.c b/drivers/staging/media/tegra-vde/h264.c
> index 03faa705bf71..f54722164493 100644
> --- a/drivers/staging/media/tegra-vde/h264.c
> +++ b/drivers/staging/media/tegra-vde/h264.c
> @@ -11,10 +11,18 @@
>  #include <linux/reset.h>
>  #include <linux/slab.h>
>  
> +#include <media/v4l2-h264.h>
> +
>  #include "trace.h"
>  #include "uapi.h"
>  #include "vde.h"
>  
> +struct h264_reflists {
> +	u8 p[V4L2_H264_NUM_DPB_ENTRIES];
> +	u8 b0[V4L2_H264_NUM_DPB_ENTRIES];
> +	u8 b1[V4L2_H264_NUM_DPB_ENTRIES];
> +};
> +
>  static int tegra_vde_wait_mbe(struct tegra_vde *vde)
>  {
>  	u32 tmp;
> @@ -125,8 +133,8 @@ static void tegra_vde_setup_frameid(struct tegra_vde *vde,
>  	u32 y_addr  = frame ? frame->y_addr  : 0x6CDEAD00;
>  	u32 cb_addr = frame ? frame->cb_addr : 0x6CDEAD00;
>  	u32 cr_addr = frame ? frame->cr_addr : 0x6CDEAD00;
> -	u32 value1 = frame ? ((mbs_width << 16) | mbs_height) : 0;
> -	u32 value2 = frame ? ((((mbs_width + 1) >> 1) << 6) | 1) : 0;
> +	u32 value1 = frame ? ((frame->luma_atoms_pitch << 16) | mbs_height) : 0;
> +	u32 value2 = frame ? ((frame->chroma_atoms_pitch << 6) | 1) : 0;
>  
>  	tegra_vde_writel(vde, y_addr  >> 8, vde->frameid, 0x000 + frameid * 4);
>  	tegra_vde_writel(vde, cb_addr >> 8, vde->frameid, 0x100 + frameid * 4);
> @@ -645,3 +653,336 @@ int tegra_vde_decode_h264(struct tegra_vde *vde,
>  
>  	return tegra_vde_decode_end(vde);
>  }
> +
> +static struct vb2_buffer *get_ref_buf(struct tegra_ctx *ctx,
> +				      struct vb2_v4l2_buffer *dst,
> +				      unsigned int dpb_idx)
> +{
> +	const struct v4l2_h264_dpb_entry *dpb = ctx->h264.decode_params->dpb;
> +	struct vb2_queue *cap_q = &ctx->fh.m2m_ctx->cap_q_ctx.q;
> +	int buf_idx = -1;
> +
> +	if (dpb[dpb_idx].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
> +		buf_idx = vb2_find_timestamp(cap_q,
> +					     dpb[dpb_idx].reference_ts, 0);
> +
> +	/*
> +	 * If a DPB entry is unused or invalid, address of current destination
> +	 * buffer is returned.
> +	 */
> +	if (buf_idx < 0)
> +		return &dst->vb2_buf;
> +
> +	return vb2_get_buffer(cap_q, buf_idx);
> +}
> +
> +static int tegra_vde_validate_vb_size(struct tegra_ctx *ctx,
> +				      struct vb2_buffer *vb,
> +				      unsigned int plane_id,
> +				      size_t min_size)
> +{
> +	u64 offset = vb->planes[plane_id].data_offset;
> +	struct device *dev = ctx->vde->dev;
> +
> +	if (offset + min_size > vb2_plane_size(vb, plane_id)) {
> +		dev_err(dev, "Too small plane[%u] size %lu @0x%llX, should be at least %zu\n",
> +			plane_id, vb2_plane_size(vb, plane_id), offset, min_size);
> +		return -EINVAL;
> +	}
> +
> +	return 0;
> +}
> +
> +static int tegra_vde_h264_setup_frame(struct tegra_ctx *ctx,
> +				      struct tegra_vde_h264_decoder_ctx *h264,
> +				      struct v4l2_h264_reflist_builder *b,
> +				      struct vb2_buffer *vb,
> +				      unsigned int ref_id,
> +				      unsigned int id)
> +{
> +	struct v4l2_pix_format_mplane *pixfmt = &ctx->decoded_fmt.fmt.pix_mp;
> +	struct tegra_m2m_buffer *tb = vb_to_tegra_buf(vb);
> +	struct tegra_ctx_h264 *h = &ctx->h264;
> +	struct tegra_vde *vde = ctx->vde;
> +	struct device *dev = vde->dev;
> +	unsigned int cstride, lstride;
> +	unsigned int flags = 0;
> +	size_t lsize, csize;
> +	int err, frame_num;
> +
> +	lsize = h264->pic_width_in_mbs * 16 * h264->pic_height_in_mbs * 16;
> +	csize = h264->pic_width_in_mbs *  8 * h264->pic_height_in_mbs *  8;
> +	lstride = pixfmt->plane_fmt[0].bytesperline;
> +	cstride = pixfmt->plane_fmt[1].bytesperline;
> +
> +	err = tegra_vde_validate_vb_size(ctx, vb, 0, lsize);
> +	if (err)
> +		return err;
> +
> +	err = tegra_vde_validate_vb_size(ctx, vb, 1, csize);
> +	if (err)
> +		return err;
> +
> +	err = tegra_vde_validate_vb_size(ctx, vb, 2, csize);
> +	if (err)
> +		return err;
> +
> +	if (!tb->aux || tb->aux->size < csize) {
> +		dev_err(dev, "Too small aux size %zd, should be at least %zu\n",
> +			tb->aux ? tb->aux->size : -1, csize);
> +		return -EINVAL;
> +	}
> +
> +	if (id == 0) {
> +		frame_num = h->decode_params->frame_num;
> +
> +		if (h->decode_params->nal_ref_idc)
> +			flags |= FLAG_REFERENCE;
> +	} else {
> +		frame_num = b->refs[ref_id].frame_num & 0x7fffff;
> +	}
> +
> +	if (to_vb2_v4l2_buffer(vb)->flags & V4L2_BUF_FLAG_BFRAME)
> +		flags |= FLAG_B_FRAME;
> +
> +	vde->frames[id].flags = flags;
> +	vde->frames[id].y_addr = tb->dma_addr[0];
> +	vde->frames[id].cb_addr = tb->dma_addr[1];
> +	vde->frames[id].cr_addr = tb->dma_addr[2];
> +	vde->frames[id].aux_addr = tb->aux->dma_addr;
> +	vde->frames[id].frame_num = frame_num;
> +	vde->frames[id].luma_atoms_pitch = lstride / VDE_ATOM;
> +	vde->frames[id].chroma_atoms_pitch = cstride / VDE_ATOM;
> +
> +	return 0;
> +}
> +
> +static void tegra_vde_h264_setup_frame_metadata(struct vb2_v4l2_buffer *src,
> +						struct vb2_v4l2_buffer *dst)
> +{
> +	struct vb2_buffer *vb = &src->vb2_buf;
> +	unsigned int bitstream_offset;
> +	unsigned long bitstream_size;
> +	const void *bitstream;
> +	int slice_type;
> +
> +	v4l2_m2m_buf_copy_metadata(src, dst, true);
> +
> +	/*
> +	 * Tegra hardware require information about frame's type, assuming
> +	 * that frame consists of the same type slices. Userspace must tag
> +	 * frame's type appropriately.
> +	 *
> +	 * Decoding of a non-uniform frames isn't supported by hardware and
> +	 * require software preprocessing that we don't implement. Decoding
> +	 * is expected to fail in this case. Such video streams are rare in
> +	 * practice, so not a big deal.
> +	 */
> +	if (dst->flags & (V4L2_BUF_FLAG_KEYFRAME |
> +			  V4L2_BUF_FLAG_PFRAME |
> +			  V4L2_BUF_FLAG_BFRAME))
> +		return;
> +
> +	/*
> +	 * If userspace doesn't tell us frame's type, then we will try to
> +	 * extract it from the bitstream. Otherwise we'll hope for the best
> +	 * and try to decode as-is.
> +	 */
> +	bitstream = vb2_plane_vaddr(vb, 0);
> +	if (!bitstream)
> +		return;
> +
> +	bitstream_offset = vb->planes[0].data_offset;
> +	bitstream_size = vb2_get_plane_payload(vb, 0);
> +
> +	slice_type = tegra_h264_parse_slice_type(bitstream + bitstream_offset,
> +						 bitstream_size);

Oh, this is a bit unfortunate, we didn't expect frame based decoder to ever need
the slice_type (only available to slice based decoders). I've lookahead and
notice a bitstream parsing, with emulation byte handling and Golum code. I
expect to see maintainers concerns with doing this, the goals of the interface
was to avoid parsing in kernel space (security in mind).

If so, I may suggest to drop this fallback, and propose an amendment to the
spec, we can require flagging KEYFRAME/PFRAME/BFRAME on the OUTPUT buffer, this
won't break any drivers/userland on other HW, and will benefit possibly other HW
in the future. I can volunteer to patch GStreamer and LibreELEC ffmpeg if we
agree to this. Not sure how it works for Chromium, or if it actually make sense
to support here.

(expecting feedback from Hans and Ezequiel here)

> +	if (slice_type < 0)
> +		return;
> +
> +	switch (slice_type % 5) {
> +	case V4L2_H264_SLICE_TYPE_I:
> +		dst->flags |= V4L2_BUF_FLAG_KEYFRAME;
> +		break;
> +
> +	case V4L2_H264_SLICE_TYPE_P:
> +		dst->flags |= V4L2_BUF_FLAG_PFRAME;
> +		break;
> +
> +	case V4L2_H264_SLICE_TYPE_B:
> +		dst->flags |= V4L2_BUF_FLAG_BFRAME;
> +		break;
> +
> +	default:
> +		break;
> +	}
> +}
> +
> +static int tegra_vde_h264_setup_frames(struct tegra_ctx *ctx,
> +				       struct tegra_vde_h264_decoder_ctx *h264)
> +{
> +	struct vb2_v4l2_buffer *src = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
> +	struct vb2_v4l2_buffer *dst = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
> +	const struct v4l2_h264_dpb_entry *dpb = ctx->h264.decode_params->dpb;
> +	struct tegra_ctx_h264 *h = &ctx->h264;
> +	struct v4l2_h264_reflist_builder b;
> +	struct h264_reflists reflists;
> +	struct vb2_buffer *ref;
> +	unsigned int i;
> +	u8 *dpb_id;
> +	int err;
> +
> +	tegra_vde_h264_setup_frame_metadata(src, dst);
> +
> +	err = tegra_vde_h264_setup_frame(ctx, h264, NULL, &dst->vb2_buf, 0,
> +					 h264->dpb_frames_nb++);
> +	if (err)
> +		return err;
> +
> +	if (dst->flags & V4L2_BUF_FLAG_KEYFRAME)
> +		return 0;
> +
> +	v4l2_h264_init_reflist_builder(&b, h->decode_params, h->sps, dpb);
> +
> +	if (dst->flags & V4L2_BUF_FLAG_BFRAME) {
> +		v4l2_h264_build_b_ref_lists(&b, reflists.b0, reflists.b1);
> +		dpb_id = reflists.b0;
> +	} else {
> +		v4l2_h264_build_p_ref_list(&b, reflists.p);
> +		dpb_id = reflists.p;
> +	}
> +
> +	for (i = 0; i < b.num_valid; i++) {
> +		ref = get_ref_buf(ctx, dst, dpb_id[i]);
> +
> +		err = tegra_vde_h264_setup_frame(ctx, h264, &b, ref, dpb_id[i],
> +						 h264->dpb_frames_nb++);
> +		if (err)
> +			return err;
> +
> +		if (b.refs[dpb_id[i]].pic_order_count < b.cur_pic_order_count)
> +			h264->dpb_ref_frames_with_earlier_poc_nb++;
> +	}
> +
> +	return 0;
> +}
> +
> +static unsigned int to_tegra_vde_h264_level_idc(unsigned int level_idc)
> +{
> +	switch (level_idc) {
> +	case 11:
> +		return 2;
> +	case 12:
> +		return 3;
> +	case 13:
> +		return 4;
> +	case 20:
> +		return 5;
> +	case 21:
> +		return 6;
> +	case 22:
> +		return 7;
> +	case 30:
> +		return 8;
> +	case 31:
> +		return 9;
> +	case 32:
> +		return 10;
> +	case 40:
> +		return 11;
> +	case 41:
> +		return 12;
> +	case 42:
> +		return 13;
> +	case 50:
> +		return 14;
> +	default:
> +		break;
> +	}
> +
> +	return 15;
> +}
> +
> +static int tegra_vde_h264_setup_context(struct tegra_ctx *ctx,
> +					struct tegra_vde_h264_decoder_ctx *h264)
> +{
> +	struct tegra_ctx_h264 *h = &ctx->h264;
> +	struct tegra_vde *vde = ctx->vde;
> +	struct device *dev = vde->dev;
> +	int err;
> +
> +	memset(h264, 0, sizeof(*h264));
> +	memset(vde->frames, 0, sizeof(vde->frames));
> +
> +	tegra_vde_prepare_control_data(ctx, V4L2_CID_STATELESS_H264_DECODE_PARAMS);
> +	tegra_vde_prepare_control_data(ctx, V4L2_CID_STATELESS_H264_SPS);
> +	tegra_vde_prepare_control_data(ctx, V4L2_CID_STATELESS_H264_PPS);
> +
> +	/* CABAC unsupported by hardware, requires software preprocessing */
> +	if (h->pps->flags & V4L2_H264_PPS_FLAG_ENTROPY_CODING_MODE)
> +		return -EOPNOTSUPP;
> +
> +	if (h->sps->profile_idc == 66)
> +		h264->baseline_profile = 1;
> +
> +	if (h->sps->flags & V4L2_H264_SPS_FLAG_DIRECT_8X8_INFERENCE)
> +		h264->direct_8x8_inference_flag = 1;
> +
> +	if (h->pps->flags & V4L2_H264_PPS_FLAG_CONSTRAINED_INTRA_PRED)
> +		h264->constrained_intra_pred_flag = 1;
> +
> +	if (h->pps->flags & V4L2_H264_PPS_FLAG_DEBLOCKING_FILTER_CONTROL_PRESENT)
> +		h264->deblocking_filter_control_present_flag = 1;
> +
> +	if (h->pps->flags & V4L2_H264_PPS_FLAG_BOTTOM_FIELD_PIC_ORDER_IN_FRAME_PRESENT)
> +		h264->pic_order_present_flag = 1;
> +
> +	h264->level_idc				= to_tegra_vde_h264_level_idc(h->sps->level_idc);
> +	h264->log2_max_pic_order_cnt_lsb	= h->sps->log2_max_pic_order_cnt_lsb_minus4 + 4;
> +	h264->log2_max_frame_num		= h->sps->log2_max_frame_num_minus4 + 4;
> +	h264->pic_order_cnt_type		= h->sps->pic_order_cnt_type;
> +	h264->pic_width_in_mbs			= h->sps->pic_width_in_mbs_minus1 + 1;
> +	h264->pic_height_in_mbs			= h->sps->pic_height_in_map_units_minus1 + 1;
> +
> +	h264->num_ref_idx_l0_active_minus1	= h->pps->num_ref_idx_l0_default_active_minus1;
> +	h264->num_ref_idx_l1_active_minus1	= h->pps->num_ref_idx_l1_default_active_minus1;
> +	h264->chroma_qp_index_offset		= h->pps->chroma_qp_index_offset & 0x1f;
> +	h264->pic_init_qp			= h->pps->pic_init_qp_minus26 + 26;
> +
> +	err = tegra_vde_h264_setup_frames(ctx, h264);
> +	if (err)
> +		return err;
> +
> +	err = tegra_vde_validate_h264_ctx(dev, h264);
> +	if (err)
> +		return err;
> +
> +	return 0;
> +}
> +
> +int tegra_vde_h264_decode_run(struct tegra_ctx *ctx)
> +{
> +	struct vb2_v4l2_buffer *src = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
> +	struct tegra_m2m_buffer *bitstream = vb_to_tegra_buf(&src->vb2_buf);
> +	size_t bitstream_size = vb2_get_plane_payload(&src->vb2_buf, 0);
> +	struct tegra_vde_h264_decoder_ctx h264;
> +	struct tegra_vde *vde = ctx->vde;
> +	int err;
> +
> +	err = tegra_vde_h264_setup_context(ctx, &h264);
> +	if (err)
> +		return err;
> +
> +	err = tegra_vde_decode_begin(vde, &h264, vde->frames,
> +				     bitstream->dma_addr[0],
> +				     bitstream_size);
> +	if (err)
> +		return err;
> +
> +	return 0;
> +}
> +
> +int tegra_vde_h264_decode_wait(struct tegra_ctx *ctx)
> +{
> +	return tegra_vde_decode_end(ctx->vde);
> +}
> diff --git a/drivers/staging/media/tegra-vde/h264_reader.c b/drivers/staging/media/tegra-vde/h264_reader.c
> new file mode 100644
> index 000000000000..37ac4413c2d6
> --- /dev/null
> +++ b/drivers/staging/media/tegra-vde/h264_reader.c
> @@ -0,0 +1,264 @@
> +// SPDX-License-Identifier: GPL-2.0+
> +/*
> + * NVIDIA Tegra Video decoder driver
> + *
> + * Copyright (c) 2016 Dmitry Osipenko <digetx@gmail.com>
> + *
> + */
> +
> +#include "vde.h"
> +
> +struct bitstream_reader {
> +	const u8 *data_ptr;
> +	u32 bitstream_end;
> +	u32 data_offset;
> +	uint bit_shift;
> +	bool rbsp_mode;
> +	bool error;
> +};
> +
> +static inline void bitstream_init(struct bitstream_reader *reader,
> +				  void *data, size_t size)
> +{
> +	reader->bitstream_end = size;
> +	reader->data_ptr = data;
> +	reader->data_offset = 0;
> +	reader->bit_shift = 0;
> +	reader->rbsp_mode = 1;
> +	reader->error = 0;
> +}
> +
> +static inline int check_range(struct bitstream_reader *reader, u32 offset)
> +{
> +	if (reader->data_offset + offset >= reader->bitstream_end)
> +		return -ENOSPC;
> +
> +	return 0;
> +}
> +
> +static inline void bitstream_reader_inc_offset(struct bitstream_reader *reader,
> +					       u32 delta)
> +{
> +	reader->data_offset += delta;
> +}
> +
> +static inline u8 emulation_escape(struct bitstream_reader *reader, u32 offset,
> +				  u8 data, bool inc_offset, bool *escaped)
> +{
> +	u32 seq;
> +
> +	if (data != 0x03 || !reader->rbsp_mode)
> +		return data;
> +
> +	if (offset < 2 || offset == reader->bitstream_end)
> +		return data;
> +
> +	seq = *((u32 *)(reader->data_ptr + offset - 2));
> +	seq = be32_to_cpu(seq);
> +
> +	switch (seq) {
> +	case 0x00000300:
> +	case 0x00000301:
> +	case 0x00000302:
> +	case 0x00000303:
> +		if (inc_offset)
> +			reader->data_offset++;
> +
> +		if (escaped)
> +			*escaped = true;
> +
> +		return seq & 0xFF;
> +	default:
> +		break;
> +	}
> +
> +	return data;
> +}
> +
> +static inline u32 bitstream_read_bits(struct bitstream_reader *reader,
> +				      u8 bits_nb, bool inc_offset)
> +{
> +	u8 rshift, bytes_to_read = (bits_nb + reader->bit_shift - 1) / 8;
> +	u32 data_offset = reader->data_offset;
> +	bool escape_inc_offset = false;
> +	u64 ret = 0;
> +
> +	if (inc_offset && check_range(reader, bytes_to_read))
> +		return 0;
> +
> +	rshift = 8 * (bytes_to_read + 1) - (reader->bit_shift + bits_nb);
> +
> +	do {
> +		u8 byte = *(reader->data_ptr + data_offset);
> +		u8 lshift = bytes_to_read * 8;
> +		bool escaped = false;
> +
> +		byte = emulation_escape(reader, data_offset++, byte,
> +					!escape_inc_offset || inc_offset,
> +					&escaped);
> +		if (escaped)
> +			data_offset++;
> +
> +		escape_inc_offset = true;
> +
> +		ret |= (u64)byte << lshift;
> +	} while (bytes_to_read--);
> +
> +	ret >>= rshift;
> +	ret &= (1llu << bits_nb) - 1;
> +
> +	return ret;
> +}
> +
> +static inline void
> +bitstream_reader_inc_offset_b(struct bitstream_reader *reader, u8 bits_nb)
> +{
> +	u8 bit_shift = reader->bit_shift;
> +
> +	reader->data_offset += (bit_shift + bits_nb) / 8;
> +	reader->bit_shift = (bit_shift + bits_nb) % 8;
> +}
> +
> +static inline u8 bitstream_read_u8_no_inc(struct bitstream_reader *reader)
> +{
> +	u8 ret;
> +
> +	if (reader->error)
> +		return 0;
> +
> +	if (check_range(reader, 0))
> +		return 0;
> +
> +	ret = *(reader->data_ptr + reader->data_offset);
> +
> +	return emulation_escape(reader, reader->data_offset, ret, true, NULL);
> +}
> +
> +static inline u32 bitstream_read_u(struct bitstream_reader *reader, u8 bits_nb)
> +{
> +	u32 ret;
> +
> +	if (reader->bit_shift == 0 && bits_nb == 8) {
> +		ret = bitstream_read_u8_no_inc(reader);
> +		bitstream_reader_inc_offset(reader, 1);
> +	} else {
> +		ret = bitstream_read_bits(reader, bits_nb, true);
> +		bitstream_reader_inc_offset_b(reader, bits_nb);
> +	}
> +
> +	return ret;
> +}
> +
> +static inline unsigned int
> +bitstream_skip_leading_zeros(struct bitstream_reader *reader)
> +{
> +	const u8 bit_shift = reader->bit_shift;
> +	u8 leading_zeros_align = 0;
> +	u8 leading_zeros = 0;
> +
> +	if (bit_shift && !reader->error) {
> +		uint byte = bitstream_read_bits(reader, 8 - bit_shift, false);
> +
> +		if (byte)
> +			leading_zeros_align = __builtin_clz(byte) - 24 - bit_shift;
> +		else
> +			leading_zeros_align = 8 - bit_shift;
> +
> +		if (byte) {
> +			reader->bit_shift += leading_zeros_align;
> +
> +			bitstream_reader_inc_offset_b(reader, 1);
> +
> +			return leading_zeros_align;
> +		}
> +
> +		bitstream_reader_inc_offset_b(reader, leading_zeros_align);
> +	}
> +
> +	while (!reader->error) {
> +		uint byte = bitstream_read_u8_no_inc(reader);
> +
> +		leading_zeros += byte ? __builtin_clz(byte) - 24 : 8;
> +
> +		if (byte) {
> +			reader->bit_shift += leading_zeros % 8;
> +			bitstream_reader_inc_offset_b(reader, 1);
> +			leading_zeros += leading_zeros_align;
> +
> +			return leading_zeros;
> +		}
> +
> +		bitstream_reader_inc_offset(reader, 1);
> +	}
> +
> +	return 0;
> +}
> +
> +static inline u32 exp_golomb_codenum(unsigned int exp, u32 val)
> +{
> +	return (1lu << exp) - 1 + val;
> +}
> +
> +static u32 bitstream_read_ue(struct bitstream_reader *reader)
> +{
> +	unsigned int leading_zeros;
> +	u32 val = 0;
> +
> +	leading_zeros = bitstream_skip_leading_zeros(reader);
> +
> +	if (leading_zeros > 31) {
> +		reader->error = 1;
> +		return 0;
> +	}
> +
> +	if (leading_zeros)
> +		val = bitstream_read_u(reader, leading_zeros);
> +
> +	return exp_golomb_codenum(leading_zeros, val);
> +}
> +
> +static inline int bitstream_start_offset(const char *nal)
> +{
> +	if (nal[0] || nal[1])
> +		return -EINVAL;
> +
> +	if (nal[2] == 1)
> +		return 4;
> +
> +	if (nal[2] == 0 && nal[3] == 1)
> +		return 5;
> +
> +	return -EINVAL;
> +}
> +
> +int tegra_h264_parse_slice_type(const void *bitstream, size_t bitstream_size)
> +{
> +	struct bitstream_reader reader;
> +	unsigned int slice_type;
> +	u8 bitstream_data[8];
> +	int start_offset;
> +
> +	/* assuming that bitstream data is uncached, copy it to CPU cache */
> +	bitstream_size = min(bitstream_size, sizeof(bitstream_data));
> +	memcpy(bitstream_data, bitstream, bitstream_size);
> +
> +	start_offset = bitstream_start_offset(bitstream_data);
> +	if (start_offset < 0)
> +		return start_offset;
> +
> +	if (start_offset >= bitstream_size)
> +		return -EINVAL;
> +
> +	bitstream_init(&reader, bitstream_data, bitstream_size);
> +	bitstream_reader_inc_offset(&reader, start_offset);
> +
> +	bitstream_read_ue(&reader);
> +	if (reader.error)
> +		return -EINVAL;
> +
> +	slice_type = bitstream_read_ue(&reader);
> +	if (reader.error)
> +		return -EINVAL;
> +
> +	return slice_type;
> +}
> diff --git a/drivers/staging/media/tegra-vde/v4l2.c b/drivers/staging/media/tegra-vde/v4l2.c
> new file mode 100644
> index 000000000000..3bd593b96593
> --- /dev/null
> +++ b/drivers/staging/media/tegra-vde/v4l2.c
> @@ -0,0 +1,1013 @@
> +// SPDX-License-Identifier: GPL-2.0+
> +/*
> + * NVIDIA Tegra Video decoder driver
> + *
> + * Copyright (C) 2019-2022 Dmitry Osipenko <digetx@gmail.com>
> + *
> + * Based on Cedrus driver by Bootlin.
> + * Copyright (C) 2016 Florent Revest <florent.revest@free-electrons.com>
> + * Copyright (C) 2018 Paul Kocialkowski <paul.kocialkowski@bootlin.com>
> + *
> + * Based on Rockchip driver by Collabora.
> + * Copyright (C) 2019 Boris Brezillon <boris.brezillon@collabora.com>
> + */
> +
> +#include <linux/err.h>
> +#include <linux/slab.h>
> +
> +#include "vde.h"
> +
> +static const struct v4l2_ctrl_config ctrl_cfgs[] = {
> +	{	.id = V4L2_CID_STATELESS_H264_DECODE_PARAMS,	},
> +	{	.id = V4L2_CID_STATELESS_H264_SPS,		},
> +	{	.id = V4L2_CID_STATELESS_H264_PPS,		},
> +	{
> +		.id = V4L2_CID_STATELESS_H264_DECODE_MODE,
> +		.min = V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED,
> +		.max = V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED,
> +		.def = V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED,
> +	},
> +	{
> +		.id = V4L2_CID_STATELESS_H264_START_CODE,
> +		.min = V4L2_STATELESS_H264_START_CODE_ANNEX_B,
> +		.max = V4L2_STATELESS_H264_START_CODE_ANNEX_B,
> +		.def = V4L2_STATELESS_H264_START_CODE_ANNEX_B,
> +	},
> +	{
> +		.id = V4L2_CID_MPEG_VIDEO_H264_PROFILE,
> +		.min = V4L2_MPEG_VIDEO_H264_PROFILE_BASELINE,
> +		.max = V4L2_MPEG_VIDEO_H264_PROFILE_MAIN,
> +		.def = V4L2_MPEG_VIDEO_H264_PROFILE_MAIN,

No action needed, just be aware that exposing BASELINE is a small lie, since FMO
and ASO feature are not supported in the uAPI.

> +	},
> +	{
> +		.id = V4L2_CID_MPEG_VIDEO_H264_LEVEL,
> +		.min = V4L2_MPEG_VIDEO_H264_LEVEL_1_0,
> +		.max = V4L2_MPEG_VIDEO_H264_LEVEL_5_1,
> +	},
> +};
> +
> +static inline struct tegra_ctx *fh_to_tegra_ctx(struct v4l2_fh *fh)
> +{
> +	return container_of(fh, struct tegra_ctx, fh);
> +}
> +
> +static void tegra_set_control_data(struct tegra_ctx *ctx, void *data, u32 id)
> +{
> +	switch (id) {
> +	case V4L2_CID_STATELESS_H264_DECODE_PARAMS:
> +		ctx->h264.decode_params = data;
> +		break;
> +	case V4L2_CID_STATELESS_H264_SPS:
> +		ctx->h264.sps = data;
> +		break;
> +	case V4L2_CID_STATELESS_H264_PPS:
> +		ctx->h264.pps = data;
> +		break;
> +	}
> +}
> +
> +void tegra_vde_prepare_control_data(struct tegra_ctx *ctx, u32 id)
> +{
> +	unsigned int i;
> +
> +	for (i = 0; i < ARRAY_SIZE(ctrl_cfgs); i++) {
> +		if (ctx->ctrls[i]->id == id) {
> +			tegra_set_control_data(ctx, ctx->ctrls[i]->p_cur.p, id);
> +			return;
> +		}
> +	}
> +
> +	tegra_set_control_data(ctx, NULL, id);
> +}
> +
> +static int tegra_queue_setup(struct vb2_queue *vq,
> +			     unsigned int *nbufs,
> +			     unsigned int *num_planes,
> +			     unsigned int sizes[],
> +			     struct device *alloc_devs[])
> +{
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vq);
> +	struct v4l2_format *f;
> +	unsigned int i;
> +
> +	if (V4L2_TYPE_IS_OUTPUT(vq->type))
> +		f = &ctx->coded_fmt;
> +	else
> +		f = &ctx->decoded_fmt;
> +
> +	if (*num_planes) {
> +		if (*num_planes != f->fmt.pix_mp.num_planes)
> +			return -EINVAL;
> +
> +		for (i = 0; i < f->fmt.pix_mp.num_planes; i++) {
> +			if (sizes[i] < f->fmt.pix_mp.plane_fmt[i].sizeimage)
> +				return -EINVAL;
> +		}
> +	} else {
> +		*num_planes = f->fmt.pix_mp.num_planes;
> +
> +		for (i = 0; i < f->fmt.pix_mp.num_planes; i++)
> +			sizes[i] = f->fmt.pix_mp.plane_fmt[i].sizeimage;
> +	}
> +
> +	return 0;
> +}
> +
> +static int tegra_buf_out_validate(struct vb2_buffer *vb)
> +{
> +	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
> +
> +	vbuf->field = V4L2_FIELD_NONE;
> +	return 0;
> +}
> +
> +static void __tegra_buf_cleanup(struct vb2_buffer *vb, unsigned int i)
> +{
> +	struct vb2_queue *vq = vb->vb2_queue;
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vq);
> +	struct tegra_m2m_buffer *tb = vb_to_tegra_buf(vb);
> +
> +	while (i--) {
> +		if (tb->a[i]) {
> +			tegra_vde_dmabuf_cache_unmap(ctx->vde, tb->a[i], true);
> +			tb->a[i] = NULL;
> +		}
> +
> +		if (tb->iova[i]) {
> +			tegra_vde_iommu_unmap(ctx->vde, tb->iova[i]);
> +			tb->iova[i] = NULL;
> +		}
> +	}
> +
> +	if (tb->aux) {
> +		tegra_vde_free_bo(tb->aux);
> +		tb->aux = NULL;
> +	}
> +}
> +
> +static int tegra_buf_init(struct vb2_buffer *vb)
> +{
> +	struct vb2_queue *vq = vb->vb2_queue;
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vq);
> +	struct tegra_m2m_buffer *tb = vb_to_tegra_buf(vb);
> +	struct tegra_vde *vde = ctx->vde;
> +	enum dma_data_direction dma_dir;
> +	struct sg_table *sgt;
> +	unsigned int i;
> +	int err;
> +
> +	if (V4L2_TYPE_IS_CAPTURE(vq->type) && vb->num_planes > 1) {
> +		/*
> +		 * Tegra decoder writes auxiliary data for I/P frames.
> +		 * This data is needed for decoding of B frames.
> +		 */
> +		err = tegra_vde_alloc_bo(vde, &tb->aux, DMA_FROM_DEVICE,
> +					 vb2_plane_size(vb, 1));
> +		if (err)
> +			return err;
> +	}
> +
> +	if (V4L2_TYPE_IS_OUTPUT(vq->type))
> +		dma_dir = DMA_TO_DEVICE;
> +	else
> +		dma_dir = DMA_FROM_DEVICE;
> +
> +	for (i = 0; i < vb->num_planes; i++) {
> +		if (vq->memory == VB2_MEMORY_DMABUF) {
> +			get_dma_buf(vb->planes[i].dbuf);
> +
> +			err = tegra_vde_dmabuf_cache_map(vde, vb->planes[i].dbuf,
> +							 dma_dir, &tb->a[i],
> +							 &tb->dma_base[i]);
> +			if (err) {
> +				dma_buf_put(vb->planes[i].dbuf);
> +				goto cleanup;
> +			}
> +
> +			continue;
> +		}
> +
> +		if (vde->domain) {
> +			sgt = vb2_dma_sg_plane_desc(vb, i);
> +
> +			err = tegra_vde_iommu_map(vde, sgt, &tb->iova[i],
> +						  vb2_plane_size(vb, i));
> +			if (err)
> +				goto cleanup;
> +
> +			tb->dma_base[i] = iova_dma_addr(&vde->iova, tb->iova[i]);
> +		} else {
> +			tb->dma_base[i] = vb2_dma_contig_plane_dma_addr(vb, i);
> +		}
> +	}
> +
> +	return 0;
> +
> +cleanup:
> +	__tegra_buf_cleanup(vb, i);
> +
> +	return err;
> +}
> +
> +static void tegra_buf_cleanup(struct vb2_buffer *vb)
> +{
> +	__tegra_buf_cleanup(vb, vb->num_planes);
> +}
> +
> +static int tegra_buf_prepare(struct vb2_buffer *vb)
> +{
> +	struct vb2_queue *vq = vb->vb2_queue;
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vq);
> +	struct tegra_m2m_buffer *tb = vb_to_tegra_buf(vb);
> +	size_t hw_align, hw_size, hw_payload, size, offset;
> +	struct v4l2_pix_format_mplane *pixfmt;
> +	unsigned int i;
> +	void *vb_data;
> +
> +	if (V4L2_TYPE_IS_OUTPUT(vq->type)) {
> +		hw_align = BSEV_ALIGN;
> +		pixfmt = &ctx->coded_fmt.fmt.pix_mp;
> +	} else {
> +		hw_align = FRAMEID_ALIGN;
> +		pixfmt = &ctx->decoded_fmt.fmt.pix_mp;
> +	}
> +
> +	for (i = 0; i < vb->num_planes; i++) {
> +		offset = vb->planes[i].data_offset;
> +
> +		if (offset & (hw_align - 1))
> +			return -EINVAL;
> +
> +		if (V4L2_TYPE_IS_CAPTURE(vq->type)) {
> +			size = pixfmt->plane_fmt[i].sizeimage;
> +			hw_payload = ALIGN(size, VDE_ATOM);
> +		} else {
> +			size = vb2_get_plane_payload(vb, i) - offset;
> +			hw_payload = ALIGN(size + VDE_ATOM, SXE_BUFFER);
> +		}
> +
> +		hw_size = offset + hw_payload;
> +
> +		if (vb2_plane_size(vb, i) < hw_size)
> +			return -EINVAL;
> +
> +		vb2_set_plane_payload(vb, i, hw_payload);
> +
> +		if (V4L2_TYPE_IS_OUTPUT(vq->type)) {
> +			vb_data = vb2_plane_vaddr(vb, i);
> +
> +			/*
> +			 * Hardware requires zero-padding of coded data.
> +			 * Otherwise it will fail to parse the trailing
> +			 * data and abort the decoding.
> +			 */
> +			if (vb_data)
> +				memset(vb_data + offset + size, 0,
> +				       hw_size - offset - size);
> +		}
> +
> +		tb->dma_addr[i] = tb->dma_base[i] + offset;
> +	}
> +
> +	switch (pixfmt->pixelformat) {
> +	case V4L2_PIX_FMT_YVU420M:
> +		swap(tb->dma_addr[1], tb->dma_addr[2]);
> +		break;
> +	}
> +
> +	return 0;
> +}
> +
> +static void tegra_buf_queue(struct vb2_buffer *vb)
> +{
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
> +	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
> +
> +	v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
> +}
> +
> +static void tegra_buf_request_complete(struct vb2_buffer *vb)
> +{
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
> +
> +	v4l2_ctrl_request_complete(vb->req_obj.req, &ctx->hdl);
> +}
> +
> +static int tegra_start_streaming(struct vb2_queue *vq, unsigned int count)
> +{
> +	return 0;
> +}
> +
> +static void tegra_stop_streaming(struct vb2_queue *vq)
> +{
> +	struct tegra_ctx *ctx = vb2_get_drv_priv(vq);
> +
> +	while (true) {
> +		struct vb2_v4l2_buffer *vbuf;
> +
> +		if (V4L2_TYPE_IS_OUTPUT(vq->type))
> +			vbuf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
> +		else
> +			vbuf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
> +
> +		if (!vbuf)
> +			break;
> +
> +		v4l2_ctrl_request_complete(vbuf->vb2_buf.req_obj.req, &ctx->hdl);
> +		v4l2_m2m_buf_done(vbuf, VB2_BUF_STATE_ERROR);
> +	}
> +}
> +
> +static const struct vb2_ops tegra_qops = {
> +	.queue_setup = tegra_queue_setup,
> +	.buf_init = tegra_buf_init,
> +	.buf_cleanup = tegra_buf_cleanup,
> +	.buf_prepare = tegra_buf_prepare,
> +	.buf_queue = tegra_buf_queue,
> +	.buf_out_validate = tegra_buf_out_validate,
> +	.buf_request_complete = tegra_buf_request_complete,
> +	.start_streaming = tegra_start_streaming,
> +	.stop_streaming = tegra_stop_streaming,
> +	.wait_prepare = vb2_ops_wait_prepare,
> +	.wait_finish = vb2_ops_wait_finish,
> +};
> +
> +static int tegra_queue_init(void *priv,
> +			    struct vb2_queue *src_vq,
> +			    struct vb2_queue *dst_vq)
> +{
> +	struct tegra_ctx *ctx = priv;
> +	struct tegra_vde *vde = ctx->vde;
> +	const struct vb2_mem_ops *mem_ops;
> +	unsigned long dma_attrs;
> +	int err;
> +
> +	/*
> +	 * TODO: Switch to use of vb2_dma_contig_memops uniformly once we
> +	 * will add IOMMU_DOMAIN support for video decoder to tegra-smmu
> +	 * driver. For now we need to stick with SG ops in order to be able
> +	 * to get SGT table easily. This is suboptimal since SG mappings are
> +	 * wasting CPU cache and we don't need that caching.
> +	 */
> +	if (vde->domain)
> +		mem_ops = &vb2_dma_sg_memops;
> +	else
> +		mem_ops = &vb2_dma_contig_memops;
> +
> +	dma_attrs = DMA_ATTR_WRITE_COMBINE;
> +
> +	src_vq->buf_struct_size = sizeof(struct tegra_m2m_buffer);
> +	src_vq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_COPY;
> +	src_vq->type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
> +	src_vq->io_modes = VB2_DMABUF | VB2_MMAP;
> +	src_vq->supports_requests = true;
> +	src_vq->requires_requests = true;
> +	src_vq->lock = &vde->v4l2_lock;
> +	src_vq->dma_attrs = dma_attrs;
> +	src_vq->mem_ops = mem_ops;
> +	src_vq->ops = &tegra_qops;
> +	src_vq->drv_priv = ctx;
> +	src_vq->dev = vde->dev;
> +
> +	err = vb2_queue_init(src_vq);
> +	if (err) {
> +		v4l2_err(&vde->v4l2_dev,
> +			 "failed to initialize src queue: %d\n", err);
> +		return err;
> +	}
> +
> +	/*
> +	 * We may need to read bitstream in kernel, hence kmap is needed
> +	 * for the coded data. It's not needed for framebuffers.
> +	 */
> +	dma_attrs |= DMA_ATTR_NO_KERNEL_MAPPING;
> +
> +	dst_vq->buf_struct_size = sizeof(struct tegra_m2m_buffer);
> +	dst_vq->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_COPY;
> +	dst_vq->type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
> +	dst_vq->io_modes = VB2_DMABUF | VB2_MMAP;
> +	dst_vq->lock = &vde->v4l2_lock;
> +	dst_vq->dma_attrs = dma_attrs;
> +	dst_vq->mem_ops = mem_ops;
> +	dst_vq->ops = &tegra_qops;
> +	dst_vq->drv_priv = ctx;
> +	dst_vq->dev = vde->dev;
> +
> +	err = vb2_queue_init(dst_vq);
> +	if (err) {
> +		v4l2_err(&vde->v4l2_dev,
> +			 "failed to initialize dst queue: %d\n", err);
> +		return err;
> +	}
> +
> +	return 0;
> +}
> +
> +static void tegra_reset_fmt(struct tegra_ctx *ctx, struct v4l2_format *f,
> +			    u32 fourcc)
> +{
> +	memset(f, 0, sizeof(*f));
> +	f->fmt.pix_mp.pixelformat = fourcc;
> +	f->fmt.pix_mp.field = V4L2_FIELD_NONE;
> +	f->fmt.pix_mp.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> +	f->fmt.pix_mp.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> +	f->fmt.pix_mp.colorspace = V4L2_COLORSPACE_REC709;
> +	f->fmt.pix_mp.quantization = V4L2_QUANTIZATION_DEFAULT;
> +}
> +
> +static void tegra_reset_coded_fmt(struct tegra_ctx *ctx)
> +{
> +	const struct tegra_vde_soc *soc = ctx->vde->soc;
> +	struct v4l2_format *f = &ctx->coded_fmt;
> +
> +	ctx->coded_fmt_desc = &soc->coded_fmts[0];
> +	tegra_reset_fmt(ctx, f, ctx->coded_fmt_desc->fourcc);
> +
> +	f->type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
> +	f->fmt.pix_mp.width = ctx->coded_fmt_desc->frmsize.min_width;
> +	f->fmt.pix_mp.height = ctx->coded_fmt_desc->frmsize.min_height;
> +}
> +
> +static void tegra_fill_pixfmt_mp(struct v4l2_pix_format_mplane *pixfmt,
> +				 u32 pixelformat, u32 width, u32 height)
> +{
> +	const struct v4l2_format_info *info = v4l2_format_info(pixelformat);
> +	struct v4l2_plane_pix_format *plane;
> +	unsigned int i;
> +
> +	switch (pixelformat) {
> +	case V4L2_PIX_FMT_YUV420M:
> +	case V4L2_PIX_FMT_YVU420M:
> +		pixfmt->width = width;
> +		pixfmt->height = height;
> +		pixfmt->pixelformat = pixelformat;
> +		pixfmt->num_planes = info->mem_planes;
> +
> +		for (i = 0; i < pixfmt->num_planes; i++) {
> +			unsigned int hdiv = (i == 0) ? 1 : 2;
> +			unsigned int vdiv = (i == 0) ? 1 : 2;
> +
> +			/*
> +			 * VDE is connected to Graphics Memory using 128bit port,
> +			 * all memory accesses are made using 16B atoms.
> +			 *
> +			 * V4L requires Cb/Cr strides to be exactly half of the
> +			 * Y stride, hence we're aligning Y to 16B x 2.
> +			 */
> +			plane = &pixfmt->plane_fmt[i];
> +			plane->bytesperline = ALIGN(width, VDE_ATOM * 2) / hdiv;
> +			plane->sizeimage = plane->bytesperline * height / vdiv;
> +		}
> +
> +		break;
> +	}
> +}
> +
> +static void tegra_reset_decoded_fmt(struct tegra_ctx *ctx)
> +{
> +	struct v4l2_format *f = &ctx->decoded_fmt;
> +
> +	tegra_reset_fmt(ctx, f, ctx->coded_fmt_desc->decoded_fmts[0]);
> +	f->type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
> +	tegra_fill_pixfmt_mp(&f->fmt.pix_mp,
> +			     ctx->coded_fmt_desc->decoded_fmts[0],
> +			     ctx->coded_fmt.fmt.pix_mp.width,
> +			     ctx->coded_fmt.fmt.pix_mp.height);
> +}
> +
> +static int tegra_init_ctrls(struct tegra_ctx *ctx)
> +{
> +	unsigned int i;
> +	int err;
> +
> +	err = v4l2_ctrl_handler_init(&ctx->hdl, ARRAY_SIZE(ctrl_cfgs));
> +	if (err)
> +		return err;
> +
> +	for (i = 0; i < ARRAY_SIZE(ctrl_cfgs); i++) {
> +		ctx->ctrls[i] = v4l2_ctrl_new_custom(&ctx->hdl, &ctrl_cfgs[i],
> +						     NULL);
> +		if (ctx->hdl.error) {
> +			err = ctx->hdl.error;
> +			goto free_ctrls;
> +		}
> +	}
> +
> +	err = v4l2_ctrl_handler_setup(&ctx->hdl);
> +	if (err)
> +		goto free_ctrls;
> +
> +	ctx->fh.ctrl_handler = &ctx->hdl;
> +
> +	return 0;
> +
> +free_ctrls:
> +	v4l2_ctrl_handler_free(&ctx->hdl);
> +
> +	return err;
> +}
> +
> +static int tegra_init_m2m(struct tegra_ctx *ctx)
> +{
> +	ctx->fh.m2m_ctx = v4l2_m2m_ctx_init(ctx->vde->m2m,
> +					    ctx, tegra_queue_init);
> +	if (IS_ERR(ctx->fh.m2m_ctx))
> +		return PTR_ERR(ctx->fh.m2m_ctx);
> +
> +	return 0;
> +}
> +
> +static void tegra_job_finish(struct tegra_ctx *ctx,
> +			     enum vb2_buffer_state result)
> +{
> +	v4l2_m2m_buf_done_and_job_finish(ctx->vde->m2m, ctx->fh.m2m_ctx,
> +					 result);
> +}
> +
> +static void tegra_decode_complete(struct work_struct *work)
> +{
> +	struct tegra_ctx *ctx = container_of(work, struct tegra_ctx, work);
> +	int err;
> +
> +	err = ctx->coded_fmt_desc->decode_wait(ctx);
> +	if (err)
> +		tegra_job_finish(ctx, VB2_BUF_STATE_ERROR);
> +	else
> +		tegra_job_finish(ctx, VB2_BUF_STATE_DONE);
> +}
> +
> +static int tegra_open(struct file *file)
> +{
> +	struct tegra_vde *vde = video_drvdata(file);
> +	struct tegra_ctx *ctx;
> +	int err;
> +
> +	ctx = kzalloc(offsetof(struct tegra_ctx, ctrls[ARRAY_SIZE(ctrl_cfgs)]),
> +		      GFP_KERNEL);
> +	if (!ctx)
> +		return -ENOMEM;
> +
> +	ctx->vde = vde;
> +	tegra_reset_coded_fmt(ctx);
> +	tegra_reset_decoded_fmt(ctx);
> +	v4l2_fh_init(&ctx->fh, video_devdata(file));
> +	INIT_WORK(&ctx->work, tegra_decode_complete);
> +
> +	err = tegra_init_ctrls(ctx);
> +	if (err) {
> +		v4l2_err(&vde->v4l2_dev, "failed to add controls: %d\n", err);
> +		goto free_ctx;
> +	}
> +
> +	err = tegra_init_m2m(ctx);
> +	if (err) {
> +		v4l2_err(&vde->v4l2_dev, "failed to initialize m2m: %d\n", err);
> +		goto free_ctrls;
> +	}
> +
> +	file->private_data = &ctx->fh;
> +	v4l2_fh_add(&ctx->fh);
> +
> +	return 0;
> +
> +free_ctrls:
> +	v4l2_ctrl_handler_free(&ctx->hdl);
> +free_ctx:
> +	kfree(ctx);
> +
> +	return err;
> +}
> +
> +static int tegra_release(struct file *file)
> +{
> +	struct v4l2_fh *fh = file->private_data;
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(fh);
> +	struct tegra_vde *vde = ctx->vde;
> +
> +	v4l2_fh_del(fh);
> +	v4l2_m2m_ctx_release(fh->m2m_ctx);
> +	v4l2_ctrl_handler_free(&ctx->hdl);
> +	v4l2_fh_exit(fh);
> +	kfree(ctx);
> +
> +	tegra_vde_dmabuf_cache_unmap_sync(vde);
> +
> +	return 0;
> +}
> +
> +static const struct v4l2_file_operations tegra_v4l2_fops = {
> +	.owner = THIS_MODULE,
> +	.open = tegra_open,
> +	.poll = v4l2_m2m_fop_poll,
> +	.mmap = v4l2_m2m_fop_mmap,
> +	.release = tegra_release,
> +	.unlocked_ioctl = video_ioctl2,
> +};
> +
> +static int tegra_querycap(struct file *file, void *priv,
> +			  struct v4l2_capability *cap)
> +{
> +	strscpy(cap->bus_info, "platform:tegra-vde", sizeof(cap->bus_info));
> +	strscpy(cap->driver, "tegra-vde", sizeof(cap->driver));
> +	strscpy(cap->card, "tegra-vde", sizeof(cap->card));
> +
> +	return 0;
> +}
> +
> +static int tegra_enum_decoded_fmt(struct file *file, void *priv,
> +				  struct v4l2_fmtdesc *f)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +
> +	if (WARN_ON(!ctx->coded_fmt_desc))
> +		return -EINVAL;
> +
> +	if (f->index >= ctx->coded_fmt_desc->num_decoded_fmts)
> +		return -EINVAL;
> +
> +	f->pixelformat = ctx->coded_fmt_desc->decoded_fmts[f->index];
> +
> +	return 0;
> +}
> +
> +static int tegra_g_decoded_fmt(struct file *file, void *priv,
> +			       struct v4l2_format *f)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +
> +	*f = ctx->decoded_fmt;
> +	return 0;
> +}
> +
> +static int tegra_try_decoded_fmt(struct file *file, void *priv,
> +				 struct v4l2_format *f)
> +{
> +	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +	const struct tegra_coded_fmt_desc *coded_desc;
> +	unsigned int i;
> +
> +	/*
> +	 * The codec context should point to a coded format desc, if the format
> +	 * on the coded end has not been set yet, it should point to the
> +	 * default value.
> +	 */
> +	coded_desc = ctx->coded_fmt_desc;
> +	if (WARN_ON(!coded_desc))
> +		return -EINVAL;
> +
> +	if (!coded_desc->num_decoded_fmts)
> +		return -EINVAL;
> +
> +	for (i = 0; i < coded_desc->num_decoded_fmts; i++) {
> +		if (coded_desc->decoded_fmts[i] == pix_mp->pixelformat)
> +			break;
> +	}
> +
> +	if (i == coded_desc->num_decoded_fmts)
> +		pix_mp->pixelformat = coded_desc->decoded_fmts[0];
> +
> +	/* always apply the frmsize constraint of the coded end */
> +	v4l2_apply_frmsize_constraints(&pix_mp->width,
> +				       &pix_mp->height,
> +				       &coded_desc->frmsize);
> +
> +	tegra_fill_pixfmt_mp(pix_mp, pix_mp->pixelformat,
> +			     pix_mp->width, pix_mp->height);
> +	pix_mp->field = V4L2_FIELD_NONE;
> +
> +	return 0;
> +}
> +
> +static int tegra_s_decoded_fmt(struct file *file, void *priv,
> +			       struct v4l2_format *f)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +	struct vb2_queue *vq;
> +	int err;
> +
> +	/* change not allowed if queue is busy */
> +	vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx,
> +			     V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE);
> +	if (vb2_is_busy(vq))
> +		return -EBUSY;
> +
> +	err = tegra_try_decoded_fmt(file, priv, f);
> +	if (err)
> +		return err;
> +
> +	ctx->decoded_fmt = *f;
> +
> +	return 0;
> +}
> +
> +static int tegra_enum_coded_fmt(struct file *file, void *priv,
> +				struct v4l2_fmtdesc *f)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +	const struct tegra_vde_soc *soc = ctx->vde->soc;
> +
> +	if (f->index >= soc->num_coded_fmts)
> +		return -EINVAL;
> +
> +	f->pixelformat = soc->coded_fmts[f->index].fourcc;
> +
> +	return 0;
> +}
> +
> +static int tegra_g_coded_fmt(struct file *file, void *priv,
> +			     struct v4l2_format *f)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +
> +	*f = ctx->coded_fmt;
> +	return 0;
> +}
> +
> +static const struct tegra_coded_fmt_desc *
> +tegra_find_coded_fmt_desc(struct tegra_ctx *ctx, u32 fourcc)
> +{
> +	const struct tegra_vde_soc *soc = ctx->vde->soc;
> +	unsigned int i;
> +
> +	for (i = 0; i < soc->num_coded_fmts; i++) {
> +		if (soc->coded_fmts[i].fourcc == fourcc)
> +			return &soc->coded_fmts[i];
> +	}
> +
> +	return NULL;
> +}
> +
> +static int tegra_try_coded_fmt(struct file *file, void *priv,
> +			       struct v4l2_format *f)
> +{
> +	struct v4l2_pix_format_mplane *pix_mp = &f->fmt.pix_mp;
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +	const struct tegra_vde_soc *soc = ctx->vde->soc;
> +	size_t size = pix_mp->plane_fmt[0].sizeimage;
> +	const struct tegra_coded_fmt_desc *desc;
> +
> +	desc = tegra_find_coded_fmt_desc(ctx, pix_mp->pixelformat);
> +	if (!desc) {
> +		pix_mp->pixelformat = soc->coded_fmts[0].fourcc;
> +		desc = &soc->coded_fmts[0];
> +	}
> +
> +	v4l2_apply_frmsize_constraints(&pix_mp->width,
> +				       &pix_mp->height,
> +				       &desc->frmsize);
> +
> +	pix_mp->plane_fmt[0].sizeimage = ALIGN(size + VDE_ATOM, SXE_BUFFER);
> +	pix_mp->field = V4L2_FIELD_NONE;
> +	pix_mp->num_planes = 1;
> +
> +	return 0;
> +}
> +
> +static int tegra_s_coded_fmt(struct file *file, void *priv,
> +			     struct v4l2_format *f)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +	struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx;
> +	const struct tegra_coded_fmt_desc *desc;
> +	struct vb2_queue *peer_vq, *vq;
> +	struct v4l2_format *cap_fmt;
> +	int err;
> +
> +	/*
> +	 * In order to support dynamic resolution change, the decoder admits
> +	 * a resolution change, as long as the pixelformat remains. Can't be
> +	 * done if streaming.
> +	 */
> +	vq = v4l2_m2m_get_vq(m2m_ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE);
> +	if (vb2_is_streaming(vq) ||
> +	    (vb2_is_busy(vq) &&
> +	     f->fmt.pix_mp.pixelformat != ctx->coded_fmt.fmt.pix_mp.pixelformat))
> +		return -EBUSY;
> +
> +	/*
> +	 * Since format change on the OUTPUT queue will reset the CAPTURE
> +	 * queue, we can't allow doing so when the CAPTURE queue has buffers
> +	 * allocated.
> +	 */
> +	peer_vq = v4l2_m2m_get_vq(m2m_ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE);
> +	if (vb2_is_busy(peer_vq))
> +		return -EBUSY;
> +
> +	err = tegra_try_coded_fmt(file, priv, f);
> +	if (err)
> +		return err;
> +
> +	desc = tegra_find_coded_fmt_desc(ctx, f->fmt.pix_mp.pixelformat);
> +	if (!desc)
> +		return -EINVAL;
> +
> +	ctx->coded_fmt_desc = desc;
> +	ctx->coded_fmt = *f;
> +
> +	/*
> +	 * Current decoded format might have become invalid with newly
> +	 * selected codec, so reset it to default just to be safe and
> +	 * keep internal driver state sane. User is mandated to set
> +	 * the decoded format again after we return, so we don't need
> +	 * anything smarter.
> +	 *
> +	 * Note that this will propagates any size changes to the decoded format.
> +	 */
> +	tegra_reset_decoded_fmt(ctx);
> +
> +	/* propagate colorspace information to capture */
> +	cap_fmt = &ctx->decoded_fmt;
> +	cap_fmt->fmt.pix_mp.xfer_func = f->fmt.pix_mp.xfer_func;
> +	cap_fmt->fmt.pix_mp.ycbcr_enc = f->fmt.pix_mp.ycbcr_enc;
> +	cap_fmt->fmt.pix_mp.colorspace = f->fmt.pix_mp.colorspace;
> +	cap_fmt->fmt.pix_mp.quantization = f->fmt.pix_mp.quantization;
> +
> +	return 0;
> +}
> +
> +static int tegra_enum_framesizes(struct file *file, void *priv,
> +				 struct v4l2_frmsizeenum *fsize)
> +{
> +	struct tegra_ctx *ctx = fh_to_tegra_ctx(priv);
> +	const struct tegra_coded_fmt_desc *fmt;
> +
> +	if (fsize->index)
> +		return -EINVAL;
> +
> +	fmt = tegra_find_coded_fmt_desc(ctx, fsize->pixel_format);
> +	if (!fmt)
> +		return -EINVAL;
> +
> +	fsize->type = V4L2_FRMSIZE_TYPE_STEPWISE;
> +	fsize->stepwise = fmt->frmsize;
> +
> +	return 0;
> +}
> +
> +static const struct v4l2_ioctl_ops tegra_v4l2_ioctl_ops = {
> +	.vidioc_querycap = tegra_querycap,
> +	.vidioc_enum_framesizes = tegra_enum_framesizes,
> +
> +	.vidioc_try_fmt_vid_out_mplane = tegra_try_coded_fmt,
> +	.vidioc_g_fmt_vid_out_mplane = tegra_g_coded_fmt,
> +	.vidioc_s_fmt_vid_out_mplane = tegra_s_coded_fmt,
> +	.vidioc_enum_fmt_vid_out = tegra_enum_coded_fmt,
> +
> +	.vidioc_try_fmt_vid_cap_mplane = tegra_try_decoded_fmt,
> +	.vidioc_g_fmt_vid_cap_mplane = tegra_g_decoded_fmt,
> +	.vidioc_s_fmt_vid_cap_mplane = tegra_s_decoded_fmt,
> +	.vidioc_enum_fmt_vid_cap = tegra_enum_decoded_fmt,
> +
> +	.vidioc_reqbufs = v4l2_m2m_ioctl_reqbufs,
> +	.vidioc_querybuf = v4l2_m2m_ioctl_querybuf,
> +	.vidioc_qbuf = v4l2_m2m_ioctl_qbuf,
> +	.vidioc_dqbuf = v4l2_m2m_ioctl_dqbuf,
> +	.vidioc_prepare_buf = v4l2_m2m_ioctl_prepare_buf,
> +	.vidioc_create_bufs = v4l2_m2m_ioctl_create_bufs,
> +	.vidioc_expbuf = v4l2_m2m_ioctl_expbuf,
> +
> +	.vidioc_streamon = v4l2_m2m_ioctl_streamon,
> +	.vidioc_streamoff = v4l2_m2m_ioctl_streamoff,
> +
> +	.vidioc_subscribe_event = v4l2_ctrl_subscribe_event,
> +	.vidioc_unsubscribe_event = v4l2_event_unsubscribe,
> +};
> +
> +static void tegra_device_run(void *priv)
> +{
> +	struct tegra_ctx *ctx = priv;
> +	struct vb2_v4l2_buffer *src = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
> +	struct media_request *src_req = src->vb2_buf.req_obj.req;
> +	int err;
> +
> +	v4l2_ctrl_request_setup(src_req, &ctx->hdl);
> +
> +	err = ctx->coded_fmt_desc->decode_run(ctx);
> +
> +	v4l2_ctrl_request_complete(src_req, &ctx->hdl);
> +
> +	if (err)
> +		tegra_job_finish(ctx, VB2_BUF_STATE_ERROR);
> +	else
> +		queue_work(ctx->vde->wq, &ctx->work);
> +}
> +
> +static const struct v4l2_m2m_ops tegra_v4l2_m2m_ops = {
> +	.device_run = tegra_device_run,
> +};
> +
> +static int tegra_request_validate(struct media_request *req)
> +{
> +	unsigned int count;
> +
> +	count = vb2_request_buffer_cnt(req);
> +	if (!count)
> +		return -ENOENT;
> +	else if (count > 1)
> +		return -EINVAL;
> +
> +	return vb2_request_validate(req);
> +}
> +
> +static const struct media_device_ops tegra_media_device_ops = {
> +	.req_validate = tegra_request_validate,
> +	.req_queue = v4l2_m2m_request_queue,
> +};
> +
> +int tegra_vde_v4l2_init(struct tegra_vde *vde)
> +{
> +	struct device *dev = vde->dev;
> +	int err;
> +
> +	mutex_init(&vde->v4l2_lock);
> +	media_device_init(&vde->mdev);
> +	video_set_drvdata(&vde->vdev, vde);
> +
> +	vde->vdev.lock = &vde->v4l2_lock,
> +	vde->vdev.fops = &tegra_v4l2_fops,
> +	vde->vdev.vfl_dir = VFL_DIR_M2M,
> +	vde->vdev.release = video_device_release_empty,
> +	vde->vdev.v4l2_dev = &vde->v4l2_dev;
> +	vde->vdev.ioctl_ops = &tegra_v4l2_ioctl_ops,
> +	vde->vdev.device_caps = V4L2_CAP_VIDEO_M2M_MPLANE | V4L2_CAP_STREAMING,
> +
> +	vde->v4l2_dev.mdev = &vde->mdev;
> +	vde->mdev.ops = &tegra_media_device_ops;
> +	vde->mdev.dev = dev;
> +
> +	strscpy(vde->mdev.model, "tegra-vde", sizeof(vde->mdev.model));
> +	strscpy(vde->vdev.name,  "tegra-vde", sizeof(vde->vdev.name));
> +	strscpy(vde->mdev.bus_info, "platform:tegra-vde",
> +		sizeof(vde->mdev.bus_info));
> +
> +	vde->wq = create_workqueue("tegra-vde");
> +	if (!vde->wq)
> +		return -ENOMEM;
> +
> +	err = media_device_register(&vde->mdev);
> +	if (err) {
> +		dev_err(dev, "failed to register media device: %d\n", err);
> +		goto clean_up_media_device;
> +	}
> +
> +	err = v4l2_device_register(dev, &vde->v4l2_dev);
> +	if (err) {
> +		dev_err(dev, "failed to register v4l2 device: %d\n", err);
> +		goto unreg_media_device;
> +	}
> +
> +	err = video_register_device(&vde->vdev, VFL_TYPE_VIDEO, -1);
> +	if (err) {
> +		dev_err(dev, "failed to register video device: %d\n", err);
> +		goto unreg_v4l2;
> +	}
> +
> +	vde->m2m = v4l2_m2m_init(&tegra_v4l2_m2m_ops);
> +	err = PTR_ERR_OR_ZERO(vde->m2m);
> +	if (err) {
> +		dev_err(dev, "failed to initialize m2m device: %d\n", err);
> +		goto unreg_video_device;
> +	}
> +
> +	err = v4l2_m2m_register_media_controller(vde->m2m, &vde->vdev,
> +						 MEDIA_ENT_F_PROC_VIDEO_DECODER);
> +	if (err) {
> +		dev_err(dev, "failed to register media controller: %d\n", err);
> +		goto release_m2m;
> +	}
> +
> +	v4l2_info(&vde->v4l2_dev, "v4l2 device registered as /dev/video%d\n",
> +		  vde->vdev.num);
> +
> +	return 0;
> +
> +release_m2m:
> +	v4l2_m2m_release(vde->m2m);
> +unreg_video_device:
> +	video_unregister_device(&vde->vdev);
> +unreg_v4l2:
> +	v4l2_device_unregister(&vde->v4l2_dev);
> +unreg_media_device:
> +	media_device_unregister(&vde->mdev);
> +clean_up_media_device:
> +	media_device_cleanup(&vde->mdev);
> +
> +	destroy_workqueue(vde->wq);
> +
> +	return err;
> +}
> +
> +void tegra_vde_v4l2_deinit(struct tegra_vde *vde)
> +{
> +	v4l2_m2m_unregister_media_controller(vde->m2m);
> +	v4l2_m2m_release(vde->m2m);
> +
> +	video_unregister_device(&vde->vdev);
> +	v4l2_device_unregister(&vde->v4l2_dev);
> +
> +	media_device_unregister(&vde->mdev);
> +	media_device_cleanup(&vde->mdev);
> +
> +	destroy_workqueue(vde->wq);
> +}
> diff --git a/drivers/staging/media/tegra-vde/vde.c b/drivers/staging/media/tegra-vde/vde.c
> index 36f5595c0fd8..c147d58c3bfb 100644
> --- a/drivers/staging/media/tegra-vde/vde.c
> +++ b/drivers/staging/media/tegra-vde/vde.c
> @@ -53,10 +53,10 @@ void tegra_vde_set_bits(struct tegra_vde *vde, u32 mask,
>  	tegra_vde_writel(vde, value | mask, base, offset);
>  }
>  
> -static int tegra_vde_alloc_bo(struct tegra_vde *vde,
> -			      struct tegra_vde_bo **ret_bo,
> -			      enum dma_data_direction dma_dir,
> -			      size_t size)
> +int tegra_vde_alloc_bo(struct tegra_vde *vde,
> +		       struct tegra_vde_bo **ret_bo,
> +		       enum dma_data_direction dma_dir,
> +		       size_t size)
>  {
>  	struct device *dev = vde->miscdev.parent;
>  	struct tegra_vde_bo *bo;
> @@ -126,7 +126,7 @@ static int tegra_vde_alloc_bo(struct tegra_vde *vde,
>  	return err;
>  }
>  
> -static void tegra_vde_free_bo(struct tegra_vde_bo *bo)
> +void tegra_vde_free_bo(struct tegra_vde_bo *bo)
>  {
>  	struct tegra_vde *vde = bo->vde;
>  	struct device *dev = vde->miscdev.parent;
> @@ -332,6 +332,8 @@ static int tegra_vde_ioctl_decode_h264(struct tegra_vde *vde,
>  
>  		dpb_frames[i].flags = frames[i].flags;
>  		dpb_frames[i].frame_num = frames[i].frame_num;
> +		dpb_frames[i].luma_atoms_pitch = ctx.pic_width_in_mbs;
> +		dpb_frames[i].chroma_atoms_pitch = cstride / VDE_ATOM;
>  
>  		dma_dir = (i == 0) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
>  
> @@ -626,8 +628,16 @@ static int tegra_vde_probe(struct platform_device *pdev)
>  		goto err_free_secure_bo;
>  	}
>  
> +	err = tegra_vde_v4l2_init(vde);
> +	if (err) {
> +		dev_err(dev, "Failed to initialize V4L2: %d\n", err);
> +		goto misc_unreg;
> +	}
> +
>  	return 0;
>  
> +misc_unreg:
> +	misc_deregister(&vde->miscdev);
>  err_free_secure_bo:
>  	tegra_vde_free_bo(vde->secure_bo);
>  err_pm_runtime:
> @@ -648,6 +658,7 @@ static int tegra_vde_remove(struct platform_device *pdev)
>  	struct tegra_vde *vde = platform_get_drvdata(pdev);
>  	struct device *dev = &pdev->dev;
>  
> +	tegra_vde_v4l2_deinit(vde);
>  	misc_deregister(&vde->miscdev);
>  
>  	tegra_vde_free_bo(vde->secure_bo);
> @@ -722,20 +733,73 @@ static const struct dev_pm_ops tegra_vde_pm_ops = {
>  				tegra_vde_pm_resume)
>  };
>  
> +static const u32 tegra124_decoded_fmts[] = {
> +	/* TBD: T124 supports only a non-standard Tegra tiled format */
> +};
> +
> +static const struct tegra_coded_fmt_desc tegra124_coded_fmts[] = {
> +	{
> +		.fourcc = V4L2_PIX_FMT_H264_SLICE,
> +		.frmsize = {
> +			.min_width = 16,
> +			.max_width = 1920,
> +			.step_width = 16,
> +			.min_height = 16,
> +			.max_height = 2032,
> +			.step_height = 16,
> +		},
> +		.num_decoded_fmts = ARRAY_SIZE(tegra124_decoded_fmts),
> +		.decoded_fmts = tegra124_decoded_fmts,
> +		.decode_run = tegra_vde_h264_decode_run,
> +		.decode_wait = tegra_vde_h264_decode_wait,
> +	},
> +};
> +
> +static const u32 tegra20_decoded_fmts[] = {
> +	V4L2_PIX_FMT_YUV420M,
> +	V4L2_PIX_FMT_YVU420M,
> +};
> +
> +static const struct tegra_coded_fmt_desc tegra20_coded_fmts[] = {
> +	{
> +		.fourcc = V4L2_PIX_FMT_H264_SLICE,
> +		.frmsize = {
> +			.min_width = 16,
> +			.max_width = 1920,
> +			.step_width = 16,
> +			.min_height = 16,
> +			.max_height = 2032,
> +			.step_height = 16,
> +		},
> +		.num_decoded_fmts = ARRAY_SIZE(tegra20_decoded_fmts),
> +		.decoded_fmts = tegra20_decoded_fmts,
> +		.decode_run = tegra_vde_h264_decode_run,
> +		.decode_wait = tegra_vde_h264_decode_wait,
> +	},
> +};
> +
>  static const struct tegra_vde_soc tegra124_vde_soc = {
>  	.supports_ref_pic_marking = true,
> +	.coded_fmts = tegra124_coded_fmts,
> +	.num_coded_fmts = ARRAY_SIZE(tegra124_coded_fmts),
>  };
>  
>  static const struct tegra_vde_soc tegra114_vde_soc = {
>  	.supports_ref_pic_marking = true,
> +	.coded_fmts = tegra20_coded_fmts,
> +	.num_coded_fmts = ARRAY_SIZE(tegra20_coded_fmts),
>  };
>  
>  static const struct tegra_vde_soc tegra30_vde_soc = {
>  	.supports_ref_pic_marking = false,
> +	.coded_fmts = tegra20_coded_fmts,
> +	.num_coded_fmts = ARRAY_SIZE(tegra20_coded_fmts),
>  };
>  
>  static const struct tegra_vde_soc tegra20_vde_soc = {
>  	.supports_ref_pic_marking = false,
> +	.coded_fmts = tegra20_coded_fmts,
> +	.num_coded_fmts = ARRAY_SIZE(tegra20_coded_fmts),
>  };
>  
>  static const struct of_device_id tegra_vde_of_match[] = {
> diff --git a/drivers/staging/media/tegra-vde/vde.h b/drivers/staging/media/tegra-vde/vde.h
> index 8ba6a71e3e40..c24ef1918261 100644
> --- a/drivers/staging/media/tegra-vde/vde.h
> +++ b/drivers/staging/media/tegra-vde/vde.h
> @@ -15,6 +15,16 @@
>  #include <linux/miscdevice.h>
>  #include <linux/mutex.h>
>  #include <linux/types.h>
> +#include <linux/workqueue.h>
> +
> +#include <media/media-device.h>
> +#include <media/videobuf2-dma-contig.h>
> +#include <media/videobuf2-dma-sg.h>
> +#include <media/v4l2-ctrls.h>
> +#include <media/v4l2-device.h>
> +#include <media/v4l2-event.h>
> +#include <media/v4l2-ioctl.h>
> +#include <media/v4l2-mem2mem.h>
>  
>  #define ICMDQUE_WR		0x00
>  #define CMDQUE_CONTROL		0x08
> @@ -25,9 +35,15 @@
>  #define BSE_ICMDQUE_EMPTY	BIT(3)
>  #define BSE_DMA_BUSY		BIT(23)
>  
> +#define BSEV_ALIGN		SZ_1
> +#define FRAMEID_ALIGN		SZ_256
> +#define SXE_BUFFER		SZ_32K
> +#define VDE_ATOM		SZ_16
> +
>  struct clk;
>  struct dma_buf;
>  struct gen_pool;
> +struct tegra_ctx;
>  struct iommu_group;
>  struct iommu_domain;
>  struct reset_control;
> @@ -46,10 +62,23 @@ struct tegra_video_frame {
>  	dma_addr_t aux_addr;
>  	u32 frame_num;
>  	u32 flags;
> +	u32 luma_atoms_pitch;
> +	u32 chroma_atoms_pitch;
> +};
> +
> +struct tegra_coded_fmt_desc {
> +	u32 fourcc;
> +	struct v4l2_frmsize_stepwise frmsize;
> +	unsigned int num_decoded_fmts;
> +	const u32 *decoded_fmts;
> +	int (*decode_run)(struct tegra_ctx *ctx);
> +	int (*decode_wait)(struct tegra_ctx *ctx);
>  };
>  
>  struct tegra_vde_soc {
>  	bool supports_ref_pic_marking;
> +	const struct tegra_coded_fmt_desc *coded_fmts;
> +	u32 num_coded_fmts;
>  };
>  
>  struct tegra_vde_bo {
> @@ -94,8 +123,59 @@ struct tegra_vde {
>  	dma_addr_t bitstream_data_addr;
>  	dma_addr_t iram_lists_addr;
>  	u32 *iram;
> +	struct v4l2_device v4l2_dev;
> +	struct v4l2_m2m_dev *m2m;
> +	struct media_device mdev;
> +	struct video_device vdev;
> +	struct mutex v4l2_lock;
> +	struct workqueue_struct *wq;
> +	struct tegra_video_frame frames[V4L2_H264_NUM_DPB_ENTRIES + 1];
> +};
> +
> +int tegra_vde_alloc_bo(struct tegra_vde *vde,
> +		       struct tegra_vde_bo **ret_bo,
> +		       enum dma_data_direction dma_dir,
> +		       size_t size);
> +void tegra_vde_free_bo(struct tegra_vde_bo *bo);
> +
> +struct tegra_ctx_h264 {
> +	const struct v4l2_ctrl_h264_decode_params *decode_params;
> +	const struct v4l2_ctrl_h264_sps *sps;
> +	const struct v4l2_ctrl_h264_pps *pps;
> +};
> +
> +struct tegra_ctx {
> +	struct tegra_vde *vde;
> +	struct tegra_ctx_h264 h264;
> +	struct work_struct work;
> +	struct v4l2_fh fh;
> +	struct v4l2_ctrl_handler hdl;
> +	struct v4l2_format coded_fmt;
> +	struct v4l2_format decoded_fmt;
> +	const struct tegra_coded_fmt_desc *coded_fmt_desc;
> +	struct v4l2_ctrl *ctrls[];
>  };
>  
> +struct tegra_m2m_buffer {
> +	struct v4l2_m2m_buffer m2m;
> +	struct dma_buf_attachment *a[VB2_MAX_PLANES];
> +	dma_addr_t dma_base[VB2_MAX_PLANES];
> +	dma_addr_t dma_addr[VB2_MAX_PLANES];
> +	struct iova *iova[VB2_MAX_PLANES];
> +	struct tegra_vde_bo *aux;
> +};
> +
> +static inline struct tegra_m2m_buffer *
> +vb_to_tegra_buf(struct vb2_buffer *vb)
> +{
> +	struct v4l2_m2m_buffer *m2m = container_of(vb, struct v4l2_m2m_buffer,
> +						   vb.vb2_buf);
> +
> +	return container_of(m2m, struct tegra_m2m_buffer, m2m);
> +}
> +
> +void tegra_vde_prepare_control_data(struct tegra_ctx *ctx, u32 id);
> +
>  void tegra_vde_writel(struct tegra_vde *vde, u32 value, void __iomem *base,
>  		      u32 offset);
>  u32 tegra_vde_readl(struct tegra_vde *vde, void __iomem *base, u32 offset);
> @@ -111,6 +191,10 @@ int tegra_vde_decode_h264(struct tegra_vde *vde,
>  			  struct tegra_video_frame *dpb_frames,
>  			  dma_addr_t bitstream_data_addr,
>  			  size_t bitstream_data_size);
> +int tegra_vde_h264_decode_run(struct tegra_ctx *ctx);
> +int tegra_vde_h264_decode_wait(struct tegra_ctx *ctx);
> +
> +int tegra_h264_parse_slice_type(const void *bitstream, size_t bitstream_size);
>  
>  int tegra_vde_iommu_init(struct tegra_vde *vde);
>  void tegra_vde_iommu_deinit(struct tegra_vde *vde);
> @@ -164,4 +248,7 @@ tegra_vde_reg_base_name(struct tegra_vde *vde, void __iomem *base)
>  	return "???";
>  }
>  
> +int tegra_vde_v4l2_init(struct tegra_vde *vde);
> +void tegra_vde_v4l2_deinit(struct tegra_vde *vde);
> +
>  #endif /* TEGRA_VDE_H */

12.01.2022 19:49, Nicolas Dufresne пишет:
> Le mercredi 12 janvier 2022 à 18:39 +0300, Dmitry Osipenko a écrit :
>> Expose Tegra video decoder as a generic V4L M2M stateless video decoder.
> 
> Thanks for working on this. Note that it would be nice to provide V4L2
> compliance test result, and if this is actually possible, provide fluster
> conformance results using ffmpeg, gstreamer, chromium or your own decoder,
> though if its your own, it would be nice to share a bit more so we can check
> that it's not interpreting the uAPI differently from other (we'd like drivers to
> work on multiple userland ideally).

Thank you for taking a look at this patch. Now I recalled that wanted to
run V4L2 compliance test, but forgot to do that.

I'll take a look at fluster, it's new to me.

> As usual I leave to other doing proper review, I added a comment below, pointing
> out the presence of a bitstream parser in this driver, and suggested an
> amendment to the spec to get rid of this. For me the code looks otherwise quite
> straight forward, is there any known issue that would keep this driver in
> staging ?

V4L decoding works on par with the legacy custom UAPI used by this
driver. I wish the hardware spec was made public, so we could support
more complex streams, but since it's not going to happen, I think
nothing keeps this driver in staging. It's working good for what is
supported.

> Please see below ...
> 
...
>> +	slice_type = tegra_h264_parse_slice_type(bitstream + bitstream_offset,
>> +						 bitstream_size);
> 
> Oh, this is a bit unfortunate, we didn't expect frame based decoder to ever need
> the slice_type (only available to slice based decoders). I've lookahead and
> notice a bitstream parsing, with emulation byte handling and Golum code. I
> expect to see maintainers concerns with doing this, the goals of the interface
> was to avoid parsing in kernel space (security in mind).

Initially I patched GStreamer to perform the flagging and it worked
okay. GStreamer even has variable for that, which is unused by the code
[1]. But in the end I decided that a universal solution will be a better
option.

[1]
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/blob/main/subprojects/gst-plugins-bad/gst-libs/gst/codecs/gsth264picture.h#L119

> If so, I may suggest to drop this fallback, and propose an amendment to the
> spec, we can require flagging KEYFRAME/PFRAME/BFRAME on the OUTPUT buffer, this
> won't break any drivers/userland on other HW, and will benefit possibly other HW
> in the future. I can volunteer to patch GStreamer and LibreELEC ffmpeg if we
> agree to this. Not sure how it works for Chromium, or if it actually make sense
> to support here.
> 
> (expecting feedback from Hans and Ezequiel here)

Amending the spec will be great, although it's not clear how to flag
frame that consists of slices having different types.

...
>> +static const struct v4l2_ctrl_config ctrl_cfgs[] = {
>> +	{	.id = V4L2_CID_STATELESS_H264_DECODE_PARAMS,	},
>> +	{	.id = V4L2_CID_STATELESS_H264_SPS,		},
>> +	{	.id = V4L2_CID_STATELESS_H264_PPS,		},
>> +	{
>> +		.id = V4L2_CID_STATELESS_H264_DECODE_MODE,
>> +		.min = V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED,
>> +		.max = V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED,
>> +		.def = V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED,
>> +	},
>> +	{
>> +		.id = V4L2_CID_STATELESS_H264_START_CODE,
>> +		.min = V4L2_STATELESS_H264_START_CODE_ANNEX_B,
>> +		.max = V4L2_STATELESS_H264_START_CODE_ANNEX_B,
>> +		.def = V4L2_STATELESS_H264_START_CODE_ANNEX_B,
>> +	},
>> +	{
>> +		.id = V4L2_CID_MPEG_VIDEO_H264_PROFILE,
>> +		.min = V4L2_MPEG_VIDEO_H264_PROFILE_BASELINE,
>> +		.max = V4L2_MPEG_VIDEO_H264_PROFILE_MAIN,
>> +		.def = V4L2_MPEG_VIDEO_H264_PROFILE_MAIN,
> 
> No action needed, just be aware that exposing BASELINE is a small lie, since FMO
> and ASO feature are not supported in the uAPI.

Okay

Hi Dmitry,

I love your patch! Perhaps something to improve:

[auto build test WARNING on media-tree/master]
[also build test WARNING on next-20220112]
[cannot apply to v5.16]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Dmitry-Osipenko/Add-V4L-stateless-video-decoder-API-support-to-NVIDIA-Tegra-driver/20220112-234115
base:   git://linuxtv.org/media_tree.git master
config: powerpc-randconfig-s031-20220112 (https://download.01.org/0day-ci/archive/20220113/202201130310.BFtsDEXe-lkp@intel.com/config)
compiler: powerpc-linux-gcc (GCC) 11.2.0
reproduce:
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # apt-get install sparse
        # sparse version: v0.6.4-dirty
        # https://github.com/0day-ci/linux/commit/3b11791070fd9de6cd368f28578ebab731386a83
        git remote add linux-review https://github.com/0day-ci/linux
        git fetch --no-tags linux-review Dmitry-Osipenko/Add-V4L-stateless-video-decoder-API-support-to-NVIDIA-Tegra-driver/20220112-234115
        git checkout 3b11791070fd9de6cd368f28578ebab731386a83
        # save the config file to linux build tree
        mkdir build_dir
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' O=build_dir ARCH=powerpc SHELL=/bin/bash drivers/staging/media/tegra-vde/

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>


sparse warnings: (new ones prefixed by >>)
>> drivers/staging/media/tegra-vde/h264_reader.c:57:15: sparse: sparse: cast to restricted __be32
>> drivers/staging/media/tegra-vde/h264_reader.c:57:15: sparse: sparse: cast to restricted __be32
>> drivers/staging/media/tegra-vde/h264_reader.c:57:15: sparse: sparse: cast to restricted __be32
>> drivers/staging/media/tegra-vde/h264_reader.c:57:15: sparse: sparse: cast to restricted __be32

vim +57 drivers/staging/media/tegra-vde/h264_reader.c

    44	
    45	static inline u8 emulation_escape(struct bitstream_reader *reader, u32 offset,
    46					  u8 data, bool inc_offset, bool *escaped)
    47	{
    48		u32 seq;
    49	
    50		if (data != 0x03 || !reader->rbsp_mode)
    51			return data;
    52	
    53		if (offset < 2 || offset == reader->bitstream_end)
    54			return data;
    55	
    56		seq = *((u32 *)(reader->data_ptr + offset - 2));
  > 57		seq = be32_to_cpu(seq);
    58	
    59		switch (seq) {
    60		case 0x00000300:
    61		case 0x00000301:
    62		case 0x00000302:
    63		case 0x00000303:
    64			if (inc_offset)
    65				reader->data_offset++;
    66	
    67			if (escaped)
    68				*escaped = true;
    69	
    70			return seq & 0xFF;
    71		default:
    72			break;
    73		}
    74	
    75		return data;
    76	}
    77	

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org

Le mercredi 12 janvier 2022 à 22:04 +0300, Dmitry Osipenko a écrit :
> > If so, I may suggest to drop this fallback, and propose an amendment to the
> > spec, we can require flagging KEYFRAME/PFRAME/BFRAME on the OUTPUT buffer,
> > this
> > won't break any drivers/userland on other HW, and will benefit possibly
> > other HW
> > in the future. I can volunteer to patch GStreamer and LibreELEC ffmpeg if we
> > agree to this. Not sure how it works for Chromium, or if it actually make
> > sense
> > to support here.
> > 
> > (expecting feedback from Hans and Ezequiel here)
> 
> Amending the spec will be great, although it's not clear how to flag
> frame that consists of slices having different types.

As per spec, all slices of a frame must be of the same type. In short, there is
no problem, adding new flags to the decode_params.flags is fine, and is backward
compatible. I had a second thought that I'd probably prefer this over using the
v4l2_buffer flags, but either way seems backward compatible.

In H264, but also other CODEC, slices are have two types of parameters, some of
the parameters are invariant between slices, but still duplicated so you can
decode some of the frame, even if the very first slice is lost. We tried our
best to place all the slice invariant parameters in decode_params to keep the
slice_params as small as we could.

regards,
Nicolas

18.01.2022 05:43, Nicolas Dufresne пишет:
> Le mercredi 12 janvier 2022 à 22:04 +0300, Dmitry Osipenko a écrit :
>>> If so, I may suggest to drop this fallback, and propose an amendment to the
>>> spec, we can require flagging KEYFRAME/PFRAME/BFRAME on the OUTPUT buffer,
>>> this
>>> won't break any drivers/userland on other HW, and will benefit possibly
>>> other HW
>>> in the future. I can volunteer to patch GStreamer and LibreELEC ffmpeg if we
>>> agree to this. Not sure how it works for Chromium, or if it actually make
>>> sense
>>> to support here.
>>>
>>> (expecting feedback from Hans and Ezequiel here)
>>
>> Amending the spec will be great, although it's not clear how to flag
>> frame that consists of slices having different types.
> 
> As per spec, all slices of a frame must be of the same type. In short, there is
> no problem, adding new flags to the decode_params.flags is fine, and is backward
> compatible. I had a second thought that I'd probably prefer this over using the
> v4l2_buffer flags, but either way seems backward compatible.
> 
> In H264, but also other CODEC, slices are have two types of parameters, some of
> the parameters are invariant between slices, but still duplicated so you can
> decode some of the frame, even if the very first slice is lost. We tried our
> best to place all the slice invariant parameters in decode_params to keep the
> slice_params as small as we could.

Could you please give a direct reference to the spec? (chapter / page or
provide quote)

I'm vaguely recalling that x264 encoder was able to generate such frames.