[Intel-gfx] [PATCH v2] drm/i915: Fix ref->mutex deadlock in i915_active_wait()

[Sultan Alsawaf]

On Mon, Apr 20, 2020 at 12:02:39PM +0300, Joonas Lahtinen wrote:
> I think the the patch should be dropped for now before the issue is
> properly addressed. Either by backporting the mainline fixes or if
> those are too big and there indeed is a smaller alternative patch
> that is properly reviewed. But the above patch is not, at least yet.

Why should a fix for a bona-fide issue be dropped due to political reasons? This
doesn't make sense to me. This just hurts miserable i915 users even more. If my
patch is going to be dropped, it should be replaced by a different fix at the
same time.

Also, the mainline fixes just *happen* to fix this deadlock by removing the
mutex lock from the path in question and creating multiple other bugs in the
process that had to be addressed with "Fixes:" commits. The regression potential
was too high to include those patches for a "stable" kernel, so I made this
patch which fixes the issue in the simplest way possible. We put this patch into
Ubuntu now as well, because praying for a response from i915 maintainers while
the 20.04 release was on the horizon was not an option.

> There is an another similar thread where there's jumping into
> conclusions and doing ad-hoc patches for already fixed issues:
> 
> https://lore.kernel.org/dri-devel/20200414144309.GB2082@sultan-box.localdomain/

Maybe this wouldn't have happened if I had received a proper response for that
issue on gitlab from the get-go... Instead I got the run-around from Chris
claiming that it wasn't an i915 bug:

https://gitlab.freedesktop.org/drm/intel/issues/1599

> I appreciate enthusiasm to provide fixes to i915 but we should
> continue do the regular due diligence to make sure we're properly
> fixing bugs in upstream kernels. And when fixing them, to make
> sure we're not simply papering over them for a single use case.
> 
> It would be preferred to file a bug for the seen issues,
> describing how to reproduce them with vanilla upstream kernels:
> 
> https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs

gitlab.freedesktop.org/drm/intel is where bugs go to be neglected, as noted
above. I really see no reason to send anything there anymore, when the vast
majority of community-sourced bug reports go ignored. 

Sultan