[FFmpeg-devel] [PATCH V3 2/2] hwaccel_mpeg: check reference pictures are available

[Li, Zhong]

> -----Original Message-----
> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On Behalf
> Of Carl Eugen Hoyos
> Sent: Monday, October 23, 2017 4:53 PM
> To: FFmpeg development discussions and patches
> <ffmpeg-devel at ffmpeg.org>
> Subject: Re: [FFmpeg-devel] [PATCH V3 2/2] hwaccel_mpeg: check reference
> pictures are available
> 
> 2017-10-23 10:49 GMT+02:00 Li, Zhong <zhong.li at intel.com>:
> >> -----Original Message-----
> >> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On
> Behalf
> >> Of Carl Eugen Hoyos
> >> Sent: Friday, October 13, 2017 4:34 AM
> >> To: FFmpeg development discussions and patches
> >> <ffmpeg-devel at ffmpeg.org>
> >> Subject: Re: [FFmpeg-devel] [PATCH V3 2/2] hwaccel_mpeg: check
> >> reference pictures are available
> >>
> >> 2017-10-12 10:20 GMT+02:00 Zhong Li <zhong.li at intel.com>:
> >> > some reference pictures are not existed but valiad surface_ids are
> >> > set, (for example, the second field of the first key frame can be
> >> > P_Picture_Type but there is no preivous frame.) then may cause some
> >> > driver problems
> >>
> >> > (e.g, segment fault on mesa/AMD driver).
> >>
> >> Please also confirm that these driver bugs are fixed.
> >
> > Sorry for late reply. Do you mean this patch should be verified on
> > mesa/AMD environment, or another patch of Meda/AMD driver is needed?
> 
> I believe it is a good idea to add a work-around for a Mesa bug into the
> FFmpeg code-base (not all users will immediately update their graphics driver)
> but only if the Mesa bug is either fixed or at least going to be fixed with
> some certainty.
> 
> The driver should never segfault.

Exactly, a better way should be "return with some error message". 
But I have no AMD GPU on hand, this issue is reported by Mark Thompson when https://patchwork.ffmpeg.org/patch/5538/ applied.
Maybe Mark can give some comments. 
BTW, this patch is not just a work-around, it improves the robustness. Application should make sure the parameters passing to driver are really valid.

> 
> Carl Eugen
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel