On Wed, Apr 08, 2020 at 02:14:38PM -0700, Linus Torvalds wrote: > (a) what the i915 driver does is at a minimum questionable, and quite > likely actively buggy By the way, this doesn't even seem like the only place in i915 where locking and object lifetime semantics are... special. I found a combined deadlock and use-after-free last week, inside of interwoven recursive call chains, and posted a patch here [1]. Unfortunately, it's been ignored thus far. But the interesting thing is how the bizarre object lifetime semantics there required even more weird things to enact a proper fix. That patch really should be merged (Chris?), but moreover, it'd be nice to see a whole release cycle devoted to just cleaning some of this stuff up. Showstopper bugs and hangs in i915 have really multiplied in recent times. Sultan [1] https://lore.kernel.org/lkml/20200407064007.7599-1-sultan@xxxxxxxxxxxxxxx
