On 1/16/19 4:25 PM, Dafna Hirschfeld wrote:
> If the the queues are not streaming then the first resolution
> change is handled in the buf_queue callback.
> The following resolution change events are handled in job_ready.
>
> Signed-off-by: Dafna Hirschfeld <dafna3@xxxxxxxxx>
> ---
> drivers/media/platform/vicodec/vicodec-core.c | 381 ++++++++++++++----
> 1 file changed, 304 insertions(+), 77 deletions(-)
>
> diff --git a/drivers/media/platform/vicodec/vicodec-core.c b/drivers/media/platform/vicodec/vicodec-core.c
> index c97300344fbd..44f2ed59f4a3 100644
> --- a/drivers/media/platform/vicodec/vicodec-core.c
> +++ b/drivers/media/platform/vicodec/vicodec-core.c
> @@ -129,6 +129,8 @@ struct vicodec_ctx {
> u32 comp_frame_size;
> bool comp_has_frame;
> bool comp_has_next_frame;
> + bool first_source_change_sent;
> + bool source_changed;
> };
>
> static inline struct vicodec_ctx *file2ctx(struct file *file)
> @@ -322,6 +324,107 @@ static void job_remove_src_buf(struct vicodec_ctx *ctx, u32 state)
> spin_unlock(ctx->lock);
> }
>
> +static const struct v4l2_fwht_pixfmt_info *info_from_header(struct fwht_cframe_hdr p_hdr)
> +{
> + unsigned int flags = ntohl(p_hdr.flags);
> + unsigned int width_div = (flags & FWHT_FL_CHROMA_FULL_WIDTH) ? 1 : 2;
> + unsigned int height_div = (flags & FWHT_FL_CHROMA_FULL_HEIGHT) ? 1 : 2;
> + unsigned int components_num = 3;
> + unsigned int pixenc = 0;
> + unsigned int version = ntohl(p_hdr.version);
> +
> + if (version == FWHT_VERSION) {
> + components_num = 1 + ((flags & FWHT_FL_COMPONENTS_NUM_MSK) >>
> + FWHT_FL_COMPONENTS_NUM_OFFSET);
> + pixenc = (flags & FWHT_FL_PIXENC_MSK);
> + }
> + return v4l2_fwht_default_fmt(width_div, height_div, version,
> + components_num, pixenc, 0);
> +}
> +
> +static bool is_header_valid(struct fwht_cframe_hdr p_hdr)
> +{
> + const struct v4l2_fwht_pixfmt_info *info;
> + unsigned int w = ntohl(p_hdr.width);
> + unsigned int h = ntohl(p_hdr.height);
> + unsigned int version = ntohl(p_hdr.version);
> + unsigned int flags = ntohl(p_hdr.flags);
> +
> + if (p_hdr.magic1 != FWHT_MAGIC1 || p_hdr.magic2 != FWHT_MAGIC2)
> + return false;
I don't think this check is needed since you can never get here unless
the magic sequence was found.
> +
> + if (!version || version > FWHT_VERSION)
> + return false;
> +
> + if (w < MIN_WIDTH || w > MAX_WIDTH || h < MIN_HEIGHT || h > MAX_HEIGHT)
> + return false;
> +
> + if (version == FWHT_VERSION) {
> + unsigned int components_num = 1 +
> + ((flags & FWHT_FL_COMPONENTS_NUM_MSK) >>
> + FWHT_FL_COMPONENTS_NUM_OFFSET);
> + unsigned int pixenc = flags & FWHT_FL_PIXENC_MSK;
> +
> + if (components_num == 0 || components_num > 4 || pixenc == BIT(19))
> + return false;
> + }
> +
> + info = info_from_header(p_hdr);
> + if (!info)
> + return false;
> + return true;
> +}
> +
> +static void update_capture_data_from_header(struct vicodec_ctx *ctx)
> +{
> + struct vicodec_q_data *q_dst = get_q_data(ctx,
> + V4L2_BUF_TYPE_VIDEO_CAPTURE);
> + struct fwht_cframe_hdr p_hdr = ctx->state.header;
> + const struct v4l2_fwht_pixfmt_info *info = info_from_header(p_hdr);
> + unsigned int flags = ntohl(p_hdr.flags);
> + unsigned int hdr_width_div = (flags & FWHT_FL_CHROMA_FULL_WIDTH) ? 1 : 2;
> + unsigned int hdr_height_div = (flags & FWHT_FL_CHROMA_FULL_HEIGHT) ? 1 : 2;
> +
> + q_dst->info = info;
> + q_dst->visible_width = ntohl(p_hdr.width);
> + q_dst->visible_height = ntohl(p_hdr.height);
> + q_dst->coded_width = vic_round_dim(q_dst->visible_width, hdr_width_div);
> + q_dst->coded_height = vic_round_dim(q_dst->visible_height,
> + hdr_height_div);
> +
> + q_dst->sizeimage = q_dst->coded_width * q_dst->coded_height *
> + q_dst->info->sizeimage_mult / q_dst->info->sizeimage_div;
> + ctx->state.colorspace = ntohl(p_hdr.colorspace);
> +
> + ctx->state.xfer_func = ntohl(p_hdr.xfer_func);
> + ctx->state.ycbcr_enc = ntohl(p_hdr.ycbcr_enc);
> + ctx->state.quantization = ntohl(p_hdr.quantization);
> +}
> +
> +static void set_last_buffer(struct vb2_v4l2_buffer *dst_buf,
> + struct vb2_v4l2_buffer *src_buf,
src_buf should be made const.
> + struct vicodec_ctx *ctx)
> +{
> + struct vicodec_q_data *q_dst = get_q_data(ctx,
> + V4L2_BUF_TYPE_VIDEO_CAPTURE);
> +
> + vb2_set_plane_payload(&dst_buf->vb2_buf, 0, 0);
> + dst_buf->sequence = q_dst->sequence++;
> + dst_buf->vb2_buf.timestamp = src_buf->vb2_buf.timestamp;
> +
> + if (src_buf->flags & V4L2_BUF_FLAG_TIMECODE)
> + dst_buf->timecode = src_buf->timecode;
> + dst_buf->field = src_buf->field;
> + dst_buf->flags |= src_buf->flags &
> + (V4L2_BUF_FLAG_TIMECODE |
> + V4L2_BUF_FLAG_KEYFRAME |
> + V4L2_BUF_FLAG_PFRAME |
> + V4L2_BUF_FLAG_BFRAME |
> + V4L2_BUF_FLAG_TSTAMP_SRC_MASK);
Use v4l2_m2m_buf_copy_data(src_buf, dst_buf, true)
> + dst_buf->flags |= V4L2_BUF_FLAG_LAST;
> + v4l2_m2m_buf_done(dst_buf, VB2_BUF_STATE_DONE);
> +}
> +
> static int job_ready(void *priv)
> {
> static const u8 magic[] = {
> @@ -333,7 +436,15 @@ static int job_ready(void *priv)
> u8 *p;
> u32 sz;
> u32 state;
> -
> + struct vicodec_q_data *q_dst = get_q_data(ctx,
> + V4L2_BUF_TYPE_VIDEO_CAPTURE);
> + unsigned int flags;
> + unsigned int hdr_width_div;
> + unsigned int hdr_height_div;
> + unsigned int max_to_copy;
> +
> + if (ctx->source_changed)
> + return 0;
> if (ctx->is_enc || ctx->comp_has_frame)
> return 1;
>
> @@ -358,11 +469,17 @@ static int job_ready(void *priv)
>
> ctx->comp_frame_size = ntohl(ctx->state.header.size);
>
> - if (ctx->comp_frame_size > ctx->comp_max_size)
> - ctx->comp_frame_size = ctx->comp_max_size;
> + /*
> + * The current scanned frame might be the first frame of a new
> + * resolution so its size might be larger than ctx->comp_max_size.
> + * In that case it is copied up to the current buffer capacity and
> + * the copy will continue after allocating new large enough buffer
> + * when restreaming
> + */
> + max_to_copy = min(ctx->comp_frame_size, ctx->comp_max_size);
>
> - if (ctx->comp_size < ctx->comp_frame_size) {
> - u32 copy = ctx->comp_frame_size - ctx->comp_size;
> + if (ctx->comp_size < max_to_copy) {
> + u32 copy = max_to_copy - ctx->comp_size;
>
> if (copy > p_src + sz - p)
> copy = p_src + sz - p;
> @@ -371,15 +488,16 @@ static int job_ready(void *priv)
> p, copy);
> p += copy;
> ctx->comp_size += copy;
> - if (ctx->comp_size < ctx->comp_frame_size) {
> + if (ctx->comp_size < max_to_copy) {
> job_remove_src_buf(ctx, state);
> goto restart;
> }
> }
> ctx->cur_buf_offset = p - p_src;
> - ctx->comp_has_frame = true;
> + if (ctx->comp_size == ctx->comp_frame_size)
> + ctx->comp_has_frame = true;
> ctx->comp_has_next_frame = false;
> - if (sz - ctx->cur_buf_offset >= sizeof(struct fwht_cframe_hdr)) {
> + if (ctx->comp_has_frame && sz - ctx->cur_buf_offset >= sizeof(struct fwht_cframe_hdr)) {
> struct fwht_cframe_hdr *p_hdr = (struct fwht_cframe_hdr *)p;
> u32 frame_size = ntohl(p_hdr->size);
> u32 remaining = sz - ctx->cur_buf_offset - sizeof(*p_hdr);
> @@ -387,6 +505,36 @@ static int job_ready(void *priv)
> if (!memcmp(p, magic, sizeof(magic)))
> ctx->comp_has_next_frame = remaining >= frame_size;
> }
> + /*
> + * if the header is invalid the device_run will just drop the frame
> + * with an error
> + */
> + if (!is_header_valid(ctx->state.header) && ctx->comp_has_frame)
> + return 1;
> + flags = ntohl(ctx->state.header.flags);
> + hdr_width_div = (flags & FWHT_FL_CHROMA_FULL_WIDTH) ? 1 : 2;
> + hdr_height_div = (flags & FWHT_FL_CHROMA_FULL_HEIGHT) ? 1 : 2;
> +
> + if (ntohl(ctx->state.header.width) != q_dst->visible_width ||
> + ntohl(ctx->state.header.height) != q_dst->visible_height ||
> + !q_dst->info ||
> + hdr_width_div != q_dst->info->width_div ||
> + hdr_height_div != q_dst->info->height_div) {
> + static const struct v4l2_event rs_event = {
> + .type = V4L2_EVENT_SOURCE_CHANGE,
> + .u.src_change.changes = V4L2_EVENT_SRC_CH_RESOLUTION,
> + };
> +
> + struct vb2_v4l2_buffer *dst_buf =
> + v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
> +
> + update_capture_data_from_header(ctx);
> + ctx->first_source_change_sent = true;
> + v4l2_event_queue_fh(&ctx->fh, &rs_event);
> + set_last_buffer(dst_buf, src_buf, ctx);
> + ctx->source_changed = true;
> + return 0;
> + }
> return 1;
> }
>
> @@ -426,7 +574,7 @@ static int enum_fmt(struct v4l2_fmtdesc *f, struct vicodec_ctx *ctx, bool is_out
> if (is_uncomp) {
> const struct v4l2_fwht_pixfmt_info *info = get_q_data(ctx, f->type)->info;
>
> - if (ctx->is_enc)
> + if (!info || ctx->is_enc)
> info = v4l2_fwht_get_pixfmt(f->index);
> else
> info = v4l2_fwht_default_fmt(info->width_div,
> @@ -477,6 +625,9 @@ static int vidioc_g_fmt(struct vicodec_ctx *ctx, struct v4l2_format *f)
> q_data = get_q_data(ctx, f->type);
> info = q_data->info;
>
> + if (!info)
> + info = v4l2_fwht_get_pixfmt(0);
> +
> switch (f->type) {
> case V4L2_BUF_TYPE_VIDEO_CAPTURE:
> case V4L2_BUF_TYPE_VIDEO_OUTPUT:
> @@ -950,6 +1101,7 @@ static int vicodec_subscribe_event(struct v4l2_fh *fh,
> {
> switch (sub->type) {
> case V4L2_EVENT_EOS:
> + case V4L2_EVENT_SOURCE_CHANGE:
> return v4l2_event_subscribe(fh, sub, 0, NULL);
> default:
> return v4l2_ctrl_subscribe_event(fh, sub);
> @@ -1058,7 +1210,64 @@ static void vicodec_buf_queue(struct vb2_buffer *vb)
> {
> struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
> struct vicodec_ctx *ctx = vb2_get_drv_priv(vb->vb2_queue);
> + unsigned int sz = vb2_get_plane_payload(&vbuf->vb2_buf, 0);
> + u8 *p_src = vb2_plane_vaddr(&vbuf->vb2_buf, 0);
> + u8 *p = p_src;
> + struct vb2_queue *vq_out = v4l2_m2m_get_vq(ctx->fh.m2m_ctx,
> + V4L2_BUF_TYPE_VIDEO_OUTPUT);
> + struct vb2_queue *vq_cap = v4l2_m2m_get_vq(ctx->fh.m2m_ctx,
> + V4L2_BUF_TYPE_VIDEO_CAPTURE);
> +
> + /* buf_queue handles only the first source change event */
> + if (ctx->first_source_change_sent) {
> + v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
> + return;
> + }
> + /*
> + * if both queues are streaming, the source change event is
> + * handled in job_ready
> + */
> + if (vb2_is_streaming(vq_cap) && vb2_is_streaming(vq_out)) {
> + v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
> + return;
> + }
> +
> + if (!ctx->is_enc && V4L2_TYPE_IS_OUTPUT(vb->vb2_queue->type)) {
Do this instead:
if (ctx->is_enc || !V4L2_TYPE_IS_OUTPUT(vb->vb2_queue->type)) {
v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
return;
}
Now you can shift the remainder of the code one tab to the left.
> + bool header_valid = false;
> + static const struct v4l2_event rs_event = {
> + .type = V4L2_EVENT_SOURCE_CHANGE,
> + .u.src_change.changes = V4L2_EVENT_SRC_CH_RESOLUTION,
> + };
> +
> + do {
> + enum vb2_buffer_state state = get_next_header(ctx, p_src, sz,
> + (u8 *)&ctx->state.header,
> + &p);
> +
> + if (ctx->header_size < sizeof(struct fwht_cframe_hdr)) {
> + v4l2_m2m_buf_done(vbuf, state);
> + return;
> + }
> + header_valid = is_header_valid(ctx->state.header);
> + /*
> + * p points right after the end of the header in the
> + * buffer. If the header is invalid we set p to point
> + * to the next byte after the start of the header
> + */
> + if (!header_valid) {
> + p = p - sizeof(struct fwht_cframe_hdr) + 1;
> + ctx->header_size = 0;
> + ctx->comp_magic_cnt = 0;
> + }
> +
> + } while (!header_valid);
> + if (p < p_src + sz)
> + ctx->cur_buf_offset = p - p_src;
>
> + update_capture_data_from_header(ctx);
> + ctx->first_source_change_sent = true;
> + v4l2_event_queue_fh(&ctx->fh, &rs_event);
> + }
> v4l2_m2m_buf_queue(ctx->fh.m2m_ctx, vbuf);
> }
>
> @@ -1087,73 +1296,80 @@ static int vicodec_start_streaming(struct vb2_queue *q,
> struct vicodec_q_data *q_data = get_q_data(ctx, q->type);
> struct v4l2_fwht_state *state = &ctx->state;
> const struct v4l2_fwht_pixfmt_info *info = q_data->info;
> - unsigned int size = q_data->coded_width * q_data->coded_height;
> - unsigned int chroma_div = info->width_div * info->height_div;
> - unsigned int total_planes_size;
>
> - /*
> - * we don't know ahead how many components are in the encoding type
> - * V4L2_PIX_FMT_FWHT, so we will allocate space for 4 planes.
> - */
> - if (info->id == V4L2_PIX_FMT_FWHT || info->components_num == 4)
> - total_planes_size = 2 * size + 2 * (size / chroma_div);
> - else if (info->components_num == 3)
> - total_planes_size = size + 2 * (size / chroma_div);
> - else
> - total_planes_size = size;
> + if (!info)
> + return -EINVAL;
>
> q_data->sequence = 0;
>
> - if (!V4L2_TYPE_IS_OUTPUT(q->type)) {
> - if (!ctx->is_enc) {
> - state->visible_width = q_data->visible_width;
> - state->visible_height = q_data->visible_height;
> - state->coded_width = q_data->coded_width;
> - state->coded_height = q_data->coded_height;
> - state->stride = q_data->coded_width * info->bytesperline_mult;
> + ctx->last_src_buf = NULL;
> + ctx->last_dst_buf = NULL;
> + state->gop_cnt = 0;
> +
> + if ((!V4L2_TYPE_IS_OUTPUT(q->type) && !ctx->is_enc) ||
> + (V4L2_TYPE_IS_OUTPUT(q->type) && ctx->is_enc)) {
> + unsigned int size = q_data->coded_width * q_data->coded_height;
> + unsigned int chroma_div = info->width_div * info->height_div;
> + unsigned int total_planes_size;
> + u8 *new_comp_frame;
> +
> + if (!info || info->id == V4L2_PIX_FMT_FWHT) {
> + vicodec_return_bufs(q, VB2_BUF_STATE_QUEUED);
> + return -EINVAL;
> }
> - return 0;
> - }
> + if (info->components_num == 4)
> + total_planes_size = 2 * size + 2 * (size / chroma_div);
> + else if (info->components_num == 3)
> + total_planes_size = size + 2 * (size / chroma_div);
> + else
> + total_planes_size = size;
>
> - if (ctx->is_enc) {
> state->visible_width = q_data->visible_width;
> state->visible_height = q_data->visible_height;
> state->coded_width = q_data->coded_width;
> state->coded_height = q_data->coded_height;
> state->stride = q_data->coded_width * info->bytesperline_mult;
> - }
> - state->ref_frame.luma = kvmalloc(total_planes_size, GFP_KERNEL);
> - ctx->comp_max_size = total_planes_size;
> - state->compressed_frame = kvmalloc(ctx->comp_max_size, GFP_KERNEL);
> - if (!state->ref_frame.luma || !state->compressed_frame) {
> - kvfree(state->ref_frame.luma);
> - kvfree(state->compressed_frame);
> - vicodec_return_bufs(q, VB2_BUF_STATE_QUEUED);
> - return -ENOMEM;
> - }
> - if (info->id == V4L2_PIX_FMT_FWHT || info->components_num >= 3) {
> - state->ref_frame.cb = state->ref_frame.luma + size;
> - state->ref_frame.cr = state->ref_frame.cb + size / chroma_div;
> - } else {
> - state->ref_frame.cb = NULL;
> - state->ref_frame.cr = NULL;
> - }
>
> - if (info->id == V4L2_PIX_FMT_FWHT || info->components_num == 4)
> - state->ref_frame.alpha =
> - state->ref_frame.cr + size / chroma_div;
> - else
> - state->ref_frame.alpha = NULL;
> + state->ref_frame.luma = kvmalloc(total_planes_size, GFP_KERNEL);
> + ctx->comp_max_size = total_planes_size;
> + new_comp_frame = kvmalloc(ctx->comp_max_size, GFP_KERNEL);
>
> - ctx->last_src_buf = NULL;
> - ctx->last_dst_buf = NULL;
> - state->gop_cnt = 0;
> - ctx->cur_buf_offset = 0;
> - ctx->comp_size = 0;
> - ctx->header_size = 0;
> - ctx->comp_magic_cnt = 0;
> - ctx->comp_has_frame = false;
> + if (!state->ref_frame.luma || !new_comp_frame) {
> + kvfree(state->ref_frame.luma);
> + kvfree(new_comp_frame);
> + vicodec_return_bufs(q, VB2_BUF_STATE_QUEUED);
> + return -ENOMEM;
> + }
> + /*
> + * if state->compressed_frame was already allocated then
> + * it contain data of the first frame of the new resolution
> + */
> + if (state->compressed_frame) {
> + if (ctx->comp_size > ctx->comp_max_size) {
> + ctx->comp_size = ctx->comp_max_size;
> + ctx->comp_frame_size = ctx->comp_max_size;
> + }
> + memcpy(new_comp_frame,
> + state->compressed_frame, ctx->comp_size);
> + }
> +
> + kvfree(state->compressed_frame);
> + state->compressed_frame = new_comp_frame;
> +
> + if (info->components_num >= 3) {
> + state->ref_frame.cb = state->ref_frame.luma + size;
> + state->ref_frame.cr = state->ref_frame.cb + size / chroma_div;
> + } else {
> + state->ref_frame.cb = NULL;
> + state->ref_frame.cr = NULL;
> + }
>
> + if (info->components_num == 4)
> + state->ref_frame.alpha =
> + state->ref_frame.cr + size / chroma_div;
> + else
> + state->ref_frame.alpha = NULL;
> + }
> return 0;
> }
>
> @@ -1163,11 +1379,21 @@ static void vicodec_stop_streaming(struct vb2_queue *q)
>
> vicodec_return_bufs(q, VB2_BUF_STATE_ERROR);
>
> - if (!V4L2_TYPE_IS_OUTPUT(q->type))
> - return;
> -
> - kvfree(ctx->state.ref_frame.luma);
> - kvfree(ctx->state.compressed_frame);
> + if ((!V4L2_TYPE_IS_OUTPUT(q->type) && !ctx->is_enc) ||
> + (V4L2_TYPE_IS_OUTPUT(q->type) && ctx->is_enc)) {
> + kvfree(ctx->state.ref_frame.luma);
> + ctx->source_changed = false;
> + }
> + if (V4L2_TYPE_IS_OUTPUT(q->type) && !ctx->is_enc) {
> + ctx->cur_buf_offset = 0;
> + ctx->comp_max_size = 0;
> + ctx->comp_size = 0;
> + ctx->header_size = 0;
> + ctx->comp_magic_cnt = 0;
> + ctx->comp_frame_size = 0;
> + ctx->comp_has_frame = 0;
> + ctx->comp_has_next_frame = 0;
> + }
> }
>
> static const struct vb2_ops vicodec_qops = {
> @@ -1319,16 +1545,17 @@ static int vicodec_open(struct file *file)
> else
> ctx->q_data[V4L2_M2M_SRC].sizeimage =
> size + sizeof(struct fwht_cframe_hdr);
> - ctx->q_data[V4L2_M2M_DST] = ctx->q_data[V4L2_M2M_SRC];
> - ctx->q_data[V4L2_M2M_DST].info =
> - ctx->is_enc ? &pixfmt_fwht : v4l2_fwht_get_pixfmt(0);
> - size = 1280 * 720 * ctx->q_data[V4L2_M2M_DST].info->sizeimage_mult /
> - ctx->q_data[V4L2_M2M_DST].info->sizeimage_div;
> - if (ctx->is_enc)
> - ctx->q_data[V4L2_M2M_DST].sizeimage =
> - size + sizeof(struct fwht_cframe_hdr);
> - else
> - ctx->q_data[V4L2_M2M_DST].sizeimage = size;
> + if (ctx->is_enc) {
> + ctx->q_data[V4L2_M2M_DST] = ctx->q_data[V4L2_M2M_SRC];
> + ctx->q_data[V4L2_M2M_DST].info = &pixfmt_fwht;
> + ctx->q_data[V4L2_M2M_DST].sizeimage = 1280 * 720 *
> + ctx->q_data[V4L2_M2M_DST].info->sizeimage_mult /
> + ctx->q_data[V4L2_M2M_DST].info->sizeimage_div +
> + sizeof(struct fwht_cframe_hdr);
> + } else {
> + ctx->q_data[V4L2_M2M_DST].info = NULL;
> + }
> +
> ctx->state.colorspace = V4L2_COLORSPACE_REC709;
>
> if (ctx->is_enc) {
>
To be honest, this patch is hard to follow. Not your fault, it simply is difficult
code. I think that I will merge a v3 (with all the comments fixed) and then see if
I can refactor things to make the code easier to understand.
And while I do that, you can start work on the stateless codec, which is after all
the goal of your project :-)
Regards,
Hans
