On Fri, Apr 09, 2021 at 11:21:55AM +0200, Sumanth Korikkar wrote:
> Hi All,
Hi Sumanth,
> I am currently working on enabling kpatch application on our platform s390x.
> I have made few required changes to kpatch-build and it seems to work with
> the upstream kernel.
>
> My current work on s390x:
> - Marking the profiling functions.
> - Addition of s390x specific compile flags.
> - Handling of s390x specific sections.
> - Analyzing the output of various patches and checking for correctness
Sounds great! Looking forward to seeing the code.
> I checked out the optimization which is being performed in
> kpatch_line_macro_change_only(). To test this:
> - Added few comments in fs/namei.c, which contains various
> WARN_ON()/BUG_ON() - upstream kernel.
> - Currently shows as "no functional changes found" on s390x, without
> implementing kpatch_line_macro_change_only() function, which seems to be
> correct.
> - Then, I also performed some quick test on x86 by commenting out
> kpatch_line_macro_change_only() function and returning 0.
> - This actually provided the same result "no functional changes found"
> found. But as per my understanding, this should have shown various
> section changes due to change in their line number
>
> I was also searching for pattern which was suggested in commit 6b03bc8e
> ("fix WARN*_ONCE detection on newer kernels") on x86 upstream kernel, but
> could not get these patterns. May be I am missing something.
>
> Could you provide some feedback on these?
Hm, I'm not sure about s390, but your x86 results do seem surprising.
It's certainly possible that newer versions of GCC have stopped encoding
the line number (from the __LINE__ macro) directly in the instruction.
I'll try to dig further.
--
Josh
_______________________________________________
kpatch mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/kpatch
