Intel Graphics Engine-Reset Functionality Driver Work Revived

[phoronix]

Phoronix: Intel Graphics Engine-Reset Functionality Driver Work Revived

In early 2017 was when there was initial work underway for the Intel Linux graphics driver on a new engine reset capability for Broadwell "Gen 8" hardware and newer. This capability allows for per-engine resets rather than resorting to a full GPU reset in the case of hangs. The code at the time didn't end up being merged to the Linux kernel but there is now a revised implementation...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-Engine-Reset-Revived

[cybertraveler]

Nice. This should improve the user experience in case of partial failures.

It's nice to think that the chip has been designed in a modular enough fashion that this is even possible.

[milkylainen]

Are engine resets due to:
Bad drivers?
Bad hardware?
Bad user code pushing through the GPU?

All of the above?

[SofS]

Are engine resets due to:
Bad drivers?
Bad hardware?
Bad user code pushing through the GPU?

All of the above?

If both the hardware and the driver are good then any sort of bad user code should be handled gracefully. In practice this is wishful thinking however.

[starshipeleven]

Are engine resets due to:
Bad drivers?
Bad hardware?
Bad user code pushing through the GPU?

All of the above?

Does it matter? If the GPU hangs it hangs.

[milkylainen]

Does it matter? If the GPU hangs it hangs.

Sure does not matter for the end user. Probably can't make a reasonable distinction between them.
But the blame is an entirely different matter than end user perspective.
Broken GPU/drivers or broken applications?

I'm inclined to think the former pair rather than the latter.

[starshipeleven]

But the blame is an entirely different matter than end user perspective.
Broken GPU/drivers or broken applications?

I'm inclined to think the former pair rather than the latter.

Blame for what?
If the issue causes some hang **RARELY** it's better dealt with by a reset than actually having someone try to reproduce, debug and fix it.

I don't see Intel iGPUs crashing left and right so I'm not blaming them if they want to paper over some exotic stuff they will never have time to look into.