mfxSyncOperation returns MFX_ERR_GPU_HANG (-21)

[Bob_B_1]

     Hi,

Made a decoder and almost "everything" seems to be working fine, but when i run multiple services of the same application the decoder using HW returns the MFX_ERR_GPU_HANG. What could the reasons for it be?

CPU: I7-6700 .

[Seunghyuk_P_Intel]

Hi Bob,

GPU Hang happens when there is timeout with GPU command.

It might be a bug in driver side or it might be happened when GPU usage is close to 100%.

Regards,

Peter.

[Bob_B_1]

Hi Peter,

Do you have an idea of a way to fix it, besides resetting the decoder?

Have you ever met this situation?

Regards,

Bob

[Dmitry_E_Intel]

Hi Bob,

Once GPU hang occurs app can only reset decoder to continue processing. GPU hangs is a nasty issue which MSDK dev and validation teams pay a special attention during a release preparation. In case of the decode component GPU hangs usually caused by decode of corrupted content. However GPU hangs can also be triggered by bugs in application, usage of custom environment (e.g. in case of Linux kernel without patches from MSS package, custom BIOS, etc). Which MSDK version do you use? Which decoder? I'd recommend you to attempt to reproduce the issue with sample_multi_transcode at first to whitelist your application. 

Regards,

Dmitry

[Bob_B_1]

Hi Dmitry, I would say the app is fine since with decoder set to software mode everything goes fine without any crashes on the decoder side. Also i am thinking the gpu hang is not that "real" since as i said when running multiple instances one of them gets the gpu-hang error but the other one keeps on using the GPU without a problem. Somehow from time to time mfxSyncOperation returns that error and it blocks the decoder session from coming back unless a reset is done. Regards, Bob

[Dmitry_E_Intel]

Is there any GPU hang related messages in dmesg output after MSDK returns -21?

Regards,

Dmitry

[Bob_B_1]

Im not sure i understand what you mean by dmesg but if it is decoder mesg then no, it only returns -21 and if the decoder is not deleted and reseted mfxSyncOperation keeps returning -21 ( MFX_ERR_GPU_HANG).

Regards,

  Bob

[Bob_B_1]

    Any news? I tried monitoring the GPU with GPA (from intel) but i still got no better understanding of what might cause the HANG.

Regards,

  Bob