Re: 5.3-rc3: Frozen graphics with kcompactd migrating i915 pages

From: Chris Wilson
Date: Thu Sep 12 2019 - 08:47:58 EST


Quoting Linus Torvalds (2019-09-12 12:59:25)
> On Thu, Sep 12, 2019 at 12:51 PM Martin Wilck <Martin.Wilck@xxxxxxxx> wrote:
> >
> > Is there an alternative to reverting aa56a292ce62 ("drm/i915/userptr:
> > Acquire the page lock around set_page_dirty()")? And if we do, what
> > would be the consequences? Would other patches need to be reverted,
> > too?
>
> Looking at that commit, and the backtrace of the lockup, I think that
> reverting it is the correct thing to do.
>
> You can't take the page lock in invalidate_range(), since it's called
> from try_to_unmap(), which is called with the page lock already held.
>
> So commit aa56a292ce62 is just fundamentally completely wrong and
> should be reverted.

There's still the dilemma that we get called without the page lock, but
at this moment in time in order to hit 5.3, it needs a revert sent
directly to Linus.
-Chris