Re: Linux 5.10-rc4

[Dave Airlie]

On Thu, 19 Nov 2020 at 08:15, Daniel Vetter <daniel.vetter@xxxxxxxx> wrote:
>
> On Wed, Nov 18, 2020 at 11:01 PM David Laight <David.Laight@xxxxxxxxxx> wrote:
> >
> > From: Thomas Zimmermann
> > > Sent: 18 November 2020 19:37
> > >
> > > Hi
> > >
> > > Am 18.11.20 um 19:10 schrieb Linus Torvalds:
> > > > On Wed, Nov 18, 2020 at 4:12 AM David Laight <David.Laight@xxxxxxxxxx> wrote:
> > > >>
> > > >> I've got the 'splat' below during boot.
> > > >> This is an 8-core C2758 Atom cpu using the on-board/cpu graphics.
> > > >> User space is Ubuntu 20.04.
> > > >>
> > > >> Additionally the X display has all the colours and alignment slightly
> > > >> messed up.
> > > >> 5.9.0 was ok.
> > > >> I'm just guessing the two issues are related.
> > > >
> > > > Sounds likely.  But it would be lovely if you could bisect when
> > > > exactly the problem(s) started to both verify that, and just to
> > > > pinpoint the exact change..
> >
> > I don't quite understand what 'git bisect' did.
> > I was bisecting between v5.9 and v5.10-rc1 but it suddenly started
> > generating v5.9.0-rc5+ kernels.
>
> We queue up patches for -rc1 way before the previous kernel is
> released, so this is normal.
>
> > The identified commit was 13a8f46d803 drm/ttm: move ghost object created.
> > (retyped - hope it is right).
> > But the diff to that last 'good' commit is massive.
>
> Yeah that's also normal for non-linear history. If you want to
> double-check, re-test the parent of that commit (which is 2ee476f77ffe
> ("drm/ttm: add a simple assign mem to bo wrapper")), which should
> work, and then the bad commit.
>
> Also is this the first bad commit for both the splat and the screen
> corruption issues?
>
> > So I don't know if that is anywhere near right.
>
> Thomas guessed it could be a ttm change, you hit one, and it looks
> like it could be the culprit. Now I guess it's up to Dave. Also adding
> Christian, in case he has an idea.

I'd be mildly surprised if it's that commit, since it just refactors
what looks to me to be two identical code pieces into one instance
(within the scope of me screwing that up, but reading it I can't see
it).

I'll dig into this today.

Dave.