Re: rmk/for-next bisection: baseline.login on bcm2836-rpi-2-b

From: Ard Biesheuvel
Date: Fri Nov 13 2020 - 11:15:47 EST


On Fri, 13 Nov 2020 at 16:58, Russell King - ARM Linux admin
<linux@xxxxxxxxxxxxxxx> wrote:
>
> On Fri, Nov 13, 2020 at 03:43:27PM +0000, Guillaume Tucker wrote:
> > On 13/11/2020 10:35, Ard Biesheuvel wrote:
> > > On Fri, 13 Nov 2020 at 11:31, Guillaume Tucker
> > > <guillaume.tucker@xxxxxxxxxxxxx> wrote:
> > >>
> > >> Hi Ard,
> > >>
> > >> Please see the bisection report below about a boot failure on
> > >> RPi-2b.
> > >>
> > >> Reports aren't automatically sent to the public while we're
> > >> trialing new bisection features on kernelci.org but this one
> > >> looks valid.
> > >>
> > >> There's nothing in the serial console log, probably because it's
> > >> crashing too early during boot. I'm not sure if other platforms
> > >> on kernelci.org were hit by this in the same way, but there
> > >> doesn't seem to be any.
> > >>
> > >> The same regression can be see on rmk's for-next branch as well
> > >> as in linux-next. It happens with both bcm2835_defconfig and
> > >> multi_v7_defconfig.
> > >>
> > >> Some more details can be found here:
> > >>
> > >> https://kernelci.org/test/case/id/5fae44823818ee918adb8864/
> > >>
> > >> If this looks like a real issue but you don't have a platform at
> > >> hand to reproduce it, please let us know if you would like the
> > >> KernelCI test to be re-run with earlyprintk or some debug config
> > >> turned on, or if you have a fix to try.
> > >>
> > >> Best wishes,
> > >> Guillaume
> > >>
> > >
> > > Hello Guillaume,
> > >
> > > That patch did have an issue, but it was already fixed by
> > >
> > > https://www.armlinux.org.uk/developer/patches/viewpatch.php?id=9020/1
> > > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=fc2933c133744305236793025b00c2f7d258b687
> > >
> > > Could you please double check whether cherry-picking that on top of
> > > the first bad commit fixes the problem?
> >
> > Sadly this doesn't appear to be fixing the issue. I've
> > cherry-picked your patch on top of the commit found by the
> > bisection but it still didn't boot, here's the git log
> >
> > cbb9656e83ca ARM: 9020/1: mm: use correct section size macro to describe the FDT virtual address
> > 7a1be318f579 ARM: 9012/1: move device tree mapping out of linear region
> > e9a2f8b599d0 ARM: 9011/1: centralize phys-to-virt conversion of DT/ATAGS address
> > 3650b228f83a Linux 5.10-rc1
> >
> > Test log: https://people.collabora.com/~gtucker/lava/boot/rpi-2-b/v5.10-rc1-3-gcbb9656e83ca/
> >
> > There's no output so it's hard to tell what is going on, but
> > reverting the bad commmit does make the board to boot (that's
> > what "revert: PASS" means in the bisect report). So it's
> > unlikely that there is another issue causing the boot failure.
>
> These silent boot failures are precisely what the DEBUG_LL stuff (and
> early_printk) is supposed to help with - getting the kernel messages
> out when there is an oops before the serial console is initialised.
>

If this is indeed related to the FDT mapping, I would assume
earlycon=... to be usable here.

I will try to reproduce this on a RPi3 but I don't have a RPi2 at
hand, unfortunately.

Would you mind having a quick try whether you can reproduce this on
QEMU, using the raspi2 machine model? If so, that would be a *lot*
easier to diagnose.