On Tue, 2020-10-06 at 07:44 -0300, Ezequiel Garcia wrote:
> It's possible that the VPU was initialized using just one buffer,
> containing only codec headers.
>
> In this case, right after the initialization and after updating
> the FIFO read pointer, we need to iterate through all the coda_buffer_meta
> and release any metas that have been already used by the VPU.
>
> This issue is affecting indirectly the bitstream buffer fill
> threshold, which depends on the meta end position of the first
> queued meta, which is passed to coda_bitstream_can_fetch_past().
>
> Without this fix, it's possible that for certain videos, the
> bitstream buffer level is not filled properly, resulting in a PIC_RUN
> timeout.
>
> Reported-by: Benjamin Bara <benjamin.bara@xxxxxxxxxxx>
> Signed-off-by: Ezequiel Garcia <ezequiel@xxxxxxxxxxxxx>
> ---
> drivers/media/platform/coda/coda-bit.c | 42 +++++++++++++++++++++++---
> drivers/media/platform/coda/coda.h | 1 +
> 2 files changed, 39 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/media/platform/coda/coda-bit.c b/drivers/media/platform/coda/coda-bit.c
> index 659fcf77b0ed..928a640b0056 100644
> --- a/drivers/media/platform/coda/coda-bit.c
> +++ b/drivers/media/platform/coda/coda-bit.c
> @@ -1836,6 +1836,29 @@ static bool coda_reorder_enable(struct coda_ctx *ctx)
> return profile > V4L2_MPEG_VIDEO_H264_PROFILE_BASELINE;
> }
>
> +static void __coda_decoder_drop_used_metas(struct coda_ctx *ctx)
Let's just call this coda_decoder_drop_used_metas.
Otherwise this looks fine,
Reviewed-by: Philipp Zabel <p.zabel@xxxxxxxxxxxxxx>
> +{
> + struct coda_buffer_meta *meta, *tmp;
> +
> + /*
> + * All metas that end at or before the RD pointer (fifo out),
> + * are now consumed by the VPU and should be released.
> + */
> + spin_lock(&ctx->buffer_meta_lock);
> + list_for_each_entry_safe(meta, tmp, &ctx->buffer_meta_list, list) {
> + if (ctx->bitstream_fifo.kfifo.out >= meta->end) {
> + coda_dbg(2, ctx, "releasing meta: seq=%d start=%d end=%d\n",
> + meta->sequence, meta->start, meta->end);
> +
> + list_del(&meta->list);
> + ctx->num_metas--;
> + ctx->first_frame_sequence++;
> + kfree(meta);
> + }
Here it should also be possible to
else {
break;
}
as meta->end ever increases.
regards
Philipp
