boot loop since 5.17.6

boot loop since 5.17.6

[Thomas Sattler]

Hi there ...

summary: boot loop after upgrading from 5.17.5 to 5.17.6

I have a Gentoo Desktop where I was successfully running a self
compiled vanilla 5.17.5 kernel. After upgrading to 5.17.12 the
machine was unable to boot: lilo loads the kernel and the intel
microcode 20220510_p20220508 (/boot/intel-uc.img) as an initrd.
Then the monitor turns black and a few moments later the BIOS
logo is shown, followed by lilo.


As there was a recent microcode upgrade I first tried to boot
without it. But that didn't help.

Then I tried to boot other already compiled but not yet booted
kernels 5.17.9 and 5.17.6 without any success. Booting the old
5.17.5 still worked. So I compiled 5.18.1 but that also was
unable to boot.


Then I tried to do a "manual bisecting" with patch-5.17.5-6
and found that the machine is still booting with 197 of 209
diffs applied.

After applying "patch-5.17.5-6.part198.patch" compilation is
broken. Still after applying "patch-5.17.5-6.part199.patch".
After applying "patch-5.17.5-6.part200.patch", compilation
works again but the resulting kernel now fails to boot.


It is an (up to date) Gentoo System, installed ~15 years
ago, running a 64Bit kernel with a 32Bit userland. The
hardware is newer then the installation (i5-7400).

I have to mention that the machine is in a remote location.
With lilo allowing a "default kernel" as well as a kernel
to be booted "just once", I was able to try the above from
remote.

Thomas

Re: boot loop since 5.17.6

[Thomas Sattler]

some more information:

$ cat /proc/version
Linux version 5.17.5 (root@dragon) (x86_64-pc-linux-gnu-gcc (Gentoo 
11.2.1_p20220115 p4) 11.2.1 20220115, GNU ld (Gentoo 2.37_p1 p2) 2.37) 
#130 SMP PREEMPT Thu Apr 28 10:50:24 CEST 2022

$ uname -mi
x86_64 GenuineIntel


I tried to compile 5.17.6 without the three mentioned diffs which
modify the following files:

   tools/objtool/check.c   and
   tools/objtool/elf.c      and
   tools/objtool/include/objtool/elf.h

and was then able to successfully boot 5.17.6.

Re: boot loop since 5.17.6

[Thomas Sattler]

with these three diffs reverted, I was able to boot
all affected 5.17.x kernels (x={6,7,8,9,10,11,12})



Am 02.06.22 um 18:42 schrieb Thomas Sattler:
> some more information:
> 
> $ cat /proc/version
> Linux version 5.17.5 (root@dragon) (x86_64-pc-linux-gnu-gcc (Gentoo
> 11.2.1_p20220115 p4) 11.2.1 20220115, GNU ld (Gentoo 2.37_p1 p2) 2.37)
> #130 SMP PREEMPT Thu Apr 28 10:50:24 CEST 2022
> 
> $ uname -mi
> x86_64 GenuineIntel
> 
> 
> I tried to compile 5.17.6 without the three mentioned diffs which
> modify the following files:
> 
>     tools/objtool/check.c   and
>     tools/objtool/elf.c      and
>     tools/objtool/include/objtool/elf.h
> 
> and was then able to successfully boot 5.17.6.

Re: boot loop since 5.17.6

[Greg KH]

On Thu, Jun 02, 2022 at 06:14:43PM +0200, Thomas Sattler wrote:
> After applying "patch-5.17.5-6.part198.patch" compilation is
> broken. Still after applying "patch-5.17.5-6.part199.patch".
> After applying "patch-5.17.5-6.part200.patch", compilation
> works again but the resulting kernel now fails to boot.

I have no idea what those random patches are, please can you say what
the upstream commit is?

thanks,

greg k-h

Re: boot loop since 5.17.6

[Nathan Chancellor]

On Thu, Jun 02, 2022 at 09:24:18PM +0200, Thomas Sattler wrote:
> with these three diffs reverted, I was able to boot
> all affected 5.17.x kernels (x={6,7,8,9,10,11,12})
> 
> 
> 
> Am 02.06.22 um 18:42 schrieb Thomas Sattler:
> > some more information:
> > 
> > $ cat /proc/version
> > Linux version 5.17.5 (root@dragon) (x86_64-pc-linux-gnu-gcc (Gentoo
> > 11.2.1_p20220115 p4) 11.2.1 20220115, GNU ld (Gentoo 2.37_p1 p2) 2.37)
> > #130 SMP PREEMPT Thu Apr 28 10:50:24 CEST 2022
> > 
> > $ uname -mi
> > x86_64 GenuineIntel
> > 
> > 
> > I tried to compile 5.17.6 without the three mentioned diffs which
> > modify the following files:
> > 
> >     tools/objtool/check.c   and
> >     tools/objtool/elf.c      and
> >     tools/objtool/include/objtool/elf.h
> > 
> > and was then able to successfully boot 5.17.6.

5.17.6 has commit 60d2b0b1018a ("objtool: Fix code relocs vs weak
symbols"), which has a known issue that is fixed with commit
ead165fa1042 ("objtool: Fix symbol creation"). If you apply ead165fa1042
on 5.17.6 or newer, does that resolve your issue?

ead165fa1042 is tagged for stable but I don't think Greg picks up
patches from mainline until they are in a tagged -rc release.

Cheers,
Nathan

Re: boot loop since 5.17.6

[Thomas Sattler]

Am 02.06.22 um 23:32 schrieb Nathan Chancellor:
>> Am 02.06.22 um 18:42 schrieb Thomas Sattler:
>>>
>>> I tried to compile 5.17.6 without the three mentioned diffs which
>>> modify the following files:
>>>
>>>      tools/objtool/check.c   and
>>>      tools/objtool/elf.c      and
>>>      tools/objtool/include/objtool/elf.h
>>>
>>> and was then able to successfully boot 5.17.6.
> 
> 5.17.6 has commit 60d2b0b1018a ("objtool: Fix code relocs vs weak
> symbols"), which has a known issue that is fixed with commit
> ead165fa1042 ("objtool: Fix symbol creation"). If you apply ead165fa1042
> on 5.17.6 or newer, does that resolve your issue?

I applied ead165fa1042 ontop of 5.17.12, but that did not make
my system boot that kernel.

Thomas

Re: boot loop since 5.17.6

[Thomas Sattler]

Am 02.06.22 um 22:08 schrieb Greg KH:
> On Thu, Jun 02, 2022 at 06:14:43PM +0200, Thomas Sattler wrote:
>> After applying "patch-5.17.5-6.part198.patch" compilation is
>> broken. Still after applying "patch-5.17.5-6.part199.patch".
>> After applying "patch-5.17.5-6.part200.patch", compilation
>> works again but the resulting kernel now fails to boot.
> 
> I have no idea what those random patches are, please can you say what
> the upstream commit is?

I took what I reverted from patch-5.17.5-6.xz. In your tree it
matches what Nathan mentioned (60d2b0b1018a) plus d17f64c29512.

Now, knowing that they were two patchsets, I compiled 5.17.12
twice, once without 60d2b0b1018a and once without d17f64c29512.

And it turns out it is 60d2b0b1018a which breaks my system.

Thomas

Re: boot loop since 5.17.6

[Greg KH]

On Fri, Jun 03, 2022 at 01:29:26AM +0200, Thomas Sattler wrote:
> Am 02.06.22 um 22:08 schrieb Greg KH:
> > On Thu, Jun 02, 2022 at 06:14:43PM +0200, Thomas Sattler wrote:
> > > After applying "patch-5.17.5-6.part198.patch" compilation is
> > > broken. Still after applying "patch-5.17.5-6.part199.patch".
> > > After applying "patch-5.17.5-6.part200.patch", compilation
> > > works again but the resulting kernel now fails to boot.
> > 
> > I have no idea what those random patches are, please can you say what
> > the upstream commit is?
> 
> I took what I reverted from patch-5.17.5-6.xz. In your tree it
> matches what Nathan mentioned (60d2b0b1018a) plus d17f64c29512.
> 
> Now, knowing that they were two patchsets, I compiled 5.17.12
> twice, once without 60d2b0b1018a and once without d17f64c29512.
> 
> And it turns out it is 60d2b0b1018a which breaks my system.

Does 5.18.1 also break for you?

thanks,

greg k-h

Re: boot loop since 5.17.6

[Thomas Sattler]

Am 03.06.22 um 14:57 schrieb Greg KH:
> On Fri, Jun 03, 2022 at 01:29:26AM +0200, Thomas Sattler wrote:
>> Am 02.06.22 um 22:08 schrieb Greg KH:
>>> On Thu, Jun 02, 2022 at 06:14:43PM +0200, Thomas Sattler wrote:
>>
>> Now, knowing that they were two patchsets, I compiled 5.17.12
>> twice, once without 60d2b0b1018a and once without d17f64c29512.
>>
>> And it turns out it is 60d2b0b1018a which breaks my system.
> 
> Does 5.18.1 also break for you?
> 

There is no difference between 5.17.6, 5.17.12 and 5.18.1:

   - they do not boot vanilla
   - applying ead165fa1042 is no help
   - reverting 60d2b0b1018a allows booting

Thomas

Re: boot loop since 5.17.6

[Greg KH]

On Fri, Jun 03, 2022 at 03:46:34PM +0200, Thomas Sattler wrote:
> Am 03.06.22 um 14:57 schrieb Greg KH:
> > On Fri, Jun 03, 2022 at 01:29:26AM +0200, Thomas Sattler wrote:
> > > Am 02.06.22 um 22:08 schrieb Greg KH:
> > > > On Thu, Jun 02, 2022 at 06:14:43PM +0200, Thomas Sattler wrote:
> > > 
> > > Now, knowing that they were two patchsets, I compiled 5.17.12
> > > twice, once without 60d2b0b1018a and once without d17f64c29512.
> > > 
> > > And it turns out it is 60d2b0b1018a which breaks my system.
> > 
> > Does 5.18.1 also break for you?
> > 
> 
> There is no difference between 5.17.6, 5.17.12 and 5.18.1:
> 
>   - they do not boot vanilla
>   - applying ead165fa1042 is no help
>   - reverting 60d2b0b1018a allows booting

Great, please work with the developers of that change to track down the
problem and get it fixed.

thanks,

greg k-h

Re: boot loop since 5.17.6

[Thorsten Leemhuis]

Hi, this is your Linux kernel regression tracker. Sorry, I'm behind mail.

On 03.06.22 00:44, Thomas Sattler wrote:
> Am 02.06.22 um 23:32 schrieb Nathan Chancellor:
>>> Am 02.06.22 um 18:42 schrieb Thomas Sattler:
>>>>
>>>> I tried to compile 5.17.6 without the three mentioned diffs which
>>>> modify the following files:
>>>>
>>>>      tools/objtool/check.c   and
>>>>      tools/objtool/elf.c      and
>>>>      tools/objtool/include/objtool/elf.h
>>>>
>>>> and was then able to successfully boot 5.17.6.
>>
>> 5.17.6 has commit 60d2b0b1018a ("objtool: Fix code relocs vs weak
>> symbols"),

FWIW, that is 4abff6d48dbc in mainline

>> which has a known issue that is fixed with commit
>> ead165fa1042 ("objtool: Fix symbol creation"). If you apply ead165fa1042
>> on 5.17.6 or newer, does that resolve your issue?
> 
> I applied ead165fa1042 ontop of 5.17.12, but that did not make
> my system boot that kernel.

Was there any progress to get down to this? Peter, who authored
4abff6d48dbc, is not even CCed yet to this thread yet afaics.

BTW, 5.17 will likely be EOL in a week or two. Thomas, maybe it might be
the best to give 5.19-rc1 a shot and in case the regression is still
there start a new thread about this that focuses on the regression in
mainline. That makes things less confusing and the regression needs to
be fixed in mainline first anyway before it can be fixed in the stable
trees.

Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat)

P.S.: As the Linux kernel's regression tracker I deal with a lot of
reports and sometimes miss something important when writing mails like
this. If that's the case here, don't hesitate to tell me in a public
reply, it's in everyone's interest to set the public record straight.

Re: boot loop since 5.17.6

[Thomas Sattler]

Hi Thorsten,

I just compiled 5.19-rc1 and my issue is solved there.

Thomas

Re: boot loop since 5.17.6

[Greg KH]

On Tue, Jun 07, 2022 at 01:40:03PM +0200, Thomas Sattler wrote:
> Hi Thorsten,
> 
> I just compiled 5.19-rc1 and my issue is solved there.

Any chance you can do 'git bisect' between 5.18 and 5.19-rc1 to find the
commit that fixed the issue?

thanks,

greg k-h

Re: boot loop since 5.17.6

[Thorsten Leemhuis]

On 07.06.22 13:40, Thomas Sattler wrote:
>
> I just compiled 5.19-rc1 and my issue is solved there.

Then in the interest of the greater good it would be good if you could
check 5.18.y as well, as if it doesn't work it might be a good idea to
identify what fixed the problem in mainline and backport the fix to
5.18.y. The same is kinda true for 5.17 as well, but obviously it's your
choice if you want to spend time on this.

Ciao, Thorsten

Re: boot loop since 5.17.6

[Thomas Sattler]

Hi Greg,

I did not yet start a bisect but looked at commits that modify
the same file again. And there was a promising one:

   22682a07acc3 objtool: Fix objtool regression on x32 systems

That one applies cleanly to both, 5.17.12 and 5.18.2, and also
fixes my boot-problem on both.

Thomas



Am 07.06.22 um 13:53 schrieb Greg KH:
> On Tue, Jun 07, 2022 at 01:40:03PM +0200, Thomas Sattler wrote:
>> Hi Thorsten,
>>
>> I just compiled 5.19-rc1 and my issue is solved there.
> 
> Any chance you can do 'git bisect' between 5.18 and 5.19-rc1 to find the
> commit that fixed the issue?
> 
> thanks,
> 
> greg k-h

Re: boot loop since 5.17.6

[Greg KH]

On Tue, Jun 07, 2022 at 04:17:26PM +0200, Thomas Sattler wrote:
> Hi Greg,
> 
> I did not yet start a bisect but looked at commits that modify
> the same file again. And there was a promising one:
> 
>   22682a07acc3 objtool: Fix objtool regression on x32 systems
> 
> That one applies cleanly to both, 5.17.12 and 5.18.2, and also
> fixes my boot-problem on both.

Oh good, that's already queued up for the next round of stable kernels.

thanks,

greg k-h