Re: next/master bisection: baseline.login on rk3288-rock2-square

[Ard Biesheuvel]

On Thu, 4 Feb 2021 at 12:45, Russell King - ARM Linux admin
<linux@xxxxxxxxxxxxxxx> wrote:
>
> On Thu, Feb 04, 2021 at 11:32:05AM +0000, Guillaume Tucker wrote:
> > Yes it does fix the issue:
> >
> >   https://lava.collabora.co.uk/scheduler/job/3173819
> >
> > with Ard's fix applied to this test branch:
> >
> >   https://gitlab.collabora.com/gtucker/linux/-/commits/next-20210203-ard-fix/
> >
> >
> > +clang +Nick
> >
> > It's worth mentioning that the issue only happens with kernels
> > built with Clang.  As you can see there are several other arm
> > platforms failing with clang-11 builds but booting fine with
> > gcc-8:
>
> My gut feeling is that it isn't Clang specific - it's likely down to
> the exact code/data placement, how things end up during decompression,
> and exactly what state the cache ends up in.
>
> That certainly was the case with the original regression.
>

Agreed.

So given that my queued fix turns this

cache_clean
cache_off

into this

cache_off
cache_clean

for v7 only, and considering that turning this into

cache_clean
cache_off
cache_clean

(as the diff tested by Guillaume does) fixes the reported issue, it
seems like the safest option to me at this point.

Reverting both patches, one of which has been in mainline since v5.7,
seems unwise to me at this point in the cycle.