On 04 May 2021 at 09:47am, Christophe Leroy wrote:

Hi

Le 04/05/2021 à 09:21, Christian Zigotzky a écrit :

quoted

Hi Christophe,

Thanks for your answer but I think I don't know how it works with the 
cherry-pick.

$ git bisect start

As you suspect the problem to be specific to powerpc, I can do

git bisect start -- arch/powerpc

quoted

$ git bisect good 68a32ba14177d4a21c4a9a941cf1d7aea86d436f
$ git bisect bad c70a4be130de333ea079c59da41cc959712bb01c

You said that powerpc-5.13-1 is bad so you can narrow the search I think:

git bisect bad powerpc-5.13-1
git bisect good 887f3ceb51cd3~

I tried it but without any success.

git bisect bad powerpc-5.13-1

Output:
fatal: Needed a single revision
Bad rev input: powerpc-5.13-1

Maybe we should look in the PowerPC updates directly. The CPUs of the 
AmigaOne X5000 and virtual e5500 QEMU machine belong to BookE cpu 
family. The AmigaOne X1000 isn't affected by this issue because the PA6T 
belongs to the Book3S cpu family. [1]

I found this in the PowerPC updates 5.13-1:  - Convert 64-bit BookE to 
do interrupt entry/exit in C.

Maybe we should look more in the modified BookE files:

arch/powerpc/kernel/head_booke.h [2]
arch/powerpc/kernel/head_fsl_booke.S [3]

Please check the BookE commits in the PowerPC updates 5.13-1. You don't 
need an AmigaOne X5000 for testing because the virtual e5500 QEMU 
machine is also affected.

Thanks,
Christian

[1] https://www.kernel.org/doc/Documentation/powerpc/cpu_families.txt
[2] 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/diff/arch/powerpc/kernel/head_booke.h?id=c70a4be130de333ea079c59da41cc959712bb01c
[3] 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/diff/arch/powerpc/kernel/head_fsl_booke.S?id=c70a4be130de333ea079c59da41cc959712bb01c

quoted

Bisecting: 2462 revisions left to test after this (roughly 11 steps)
[47a6959fa331fe892a4fc3b48ca08e92045c6bda] netfilter: allow to turn 
off xtables compat layer

$ git cherry-pick 525642624783
error: could not apply 525642624783... powerpc/signal32: Fix 
erroneous SIGSEGV on RT signal return
hint: after resolving the conflicts, mark the corrected paths
hint: with 'git add <paths>' or 'git rm <paths>'
hint: and commit the result with 'git commit'

How can I fix this error?

This problably means that the step is at a commit which is prior to 
the first bad commit you identified at previous step. If you narrow 
the bisect as explained above, it shouldn't happen unless git decides 
it needs to descend a branch to a merge point. In that case just do 
'git cherry-pick --abort" and go without the fix.

Note that once you have cherry picked the fix and tested the result, 
you have to apply the result to HEAD~ (the commit before the 
cherry-pick).
- git bisect good HEAD~
- git bisect bad HEAD~

Christophe