[FFmpeg-devel] [PATCH] avcodec/dxva2: d3d11va: don't try to get surface description for nullptr

Carl Eugen Hoyos ceffmpeg at gmail.com
Fri Jan 11 23:03:18 EET 2019


2019-01-11 8:07 GMT+01:00, Anton Fedchin <Anton.Fedchin at syncplicity.com>:
>>2019-01-05 11:44 GMT+01:00, Anton Fedchin <afedchin at weezlabs.com>:
>>> From: Anton Fedchin <afedchin at ruswizards.com>
>>>
>>> after 153b36f there is a possibility to crash when trying to get index of
>>> a surface which points to nirvana. it may occurs when a stream starts
>>> with
>>> non i-frame.
>>> ---
>>>  libavcodec/dxva2.c | 10 ++++++----
>>>  1 file changed, 6 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/libavcodec/dxva2.c b/libavcodec/dxva2.c
>>> index 32416112bf..dfae500444 100644
>>> --- a/libavcodec/dxva2.c
>>> +++ b/libavcodec/dxva2.c
>>> @@ -771,16 +771,18 @@ unsigned ff_dxva2_get_surface_index(const
>>> AVCodecContext *avctx,
>>>  #if CONFIG_D3D11VA
>>>      if (avctx->pix_fmt == AV_PIX_FMT_D3D11)
>>>          return (intptr_t)frame->data[1];
>>> -    if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD) {
>>> +    if (avctx->pix_fmt == AV_PIX_FMT_D3D11VA_VLD && surface) {
>>>          D3D11_VIDEO_DECODER_OUTPUT_VIEW_DESC viewDesc;
>>>
>>> ID3D11VideoDecoderOutputView_GetDesc((ID3D11VideoDecoderOutputView*)
>>> surface, &viewDesc);
>>>          return viewDesc.Texture2D.ArraySlice;
>>>      }
>>>  #endif
>>
>>>  #if CONFIG_DXVA2
>>> -    for (i = 0; i < DXVA_CONTEXT_COUNT(avctx, ctx); i++) {
>>> -        if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD &&
>>> ctx->dxva2.surface[i]
>>> == surface)
>>> -            return i;
>>> +    if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
>>> +        for (i = 0; i < DXVA_CONTEXT_COUNT(avctx, ctx); i++) {
>>> +            if (ctx->dxva2.surface[i] == surface)
>>> +                return i;
>>> +        }
>>
>>How is this change related?
>>
>>Carl Eugen
>
> Hi Carl,
>
> If previous condition is failed (i.e. pix_fmt is AV_PIX_FMT_D3D11VA_VLD  and
> surface is null) it will iterate through all surfaces again with no sense.

> Yes, this is not related directly to the fix, but the fix changes method's
> behavior if surface is null and I found it necessary to change this block
> also

Please make it a separate patch.

Carl Eugen


More information about the ffmpeg-devel mailing list