Re: [PATCH v3 0/6] livepatch: klp-convert tool - Minimal version

[Joe Lawrence]

On Tue, Aug 27, 2024 at 02:30:45PM +0200, Lukas Hruska wrote:
> Summary
> -------
> 
> This is a significantly simplified version of the original klp-convert tool.
> The klp-convert code has never got a proper review and also clean ups
> were not easy. The last version was v7, see
> https://lore.kernel.org/r/20230306140824.3858543-1-joe.lawrence@xxxxxxxxxx
> 
> The main change is that the tool does not longer search for the
> symbols which would need the livepatch specific relocation entry.
> Also klp.symbols file is not longer needed.
> 
> Instead, the needed information is appended to the symbol declaration
> via a new macro KLP_RELOC_SYMBOL(). It creates symbol with all needed
> metadata. For example:
> 
>   extern char *saved_command_line \
>                  KLP_RELOC_SYMBOL(vmlinux, vmlinux, saved_command_line, 0);
> 
> would create symbol
> 
> $>readelf -r -W <compiled livepatch module>:
> Relocation section '.rela.text' at offset 0x32e60 contains 10 entries:
>     Offset             Info             Type               Symbol's Value  Symbol's Name + Addend
> [...]
> 0000000000000068  0000003c00000002 R_X86_64_PC32          0000000000000000 .klp.sym.rela.vmlinux.vmlinux.saved_command_line,0 - 4
> [...]
> 
> 
> The simplified klp-convert tool just transforms symbols
> created by KLP_RELOC_SYMBOL() to object specific rela sections
> and rela entries which would later be proceed when the livepatch
> or the livepatched object is loaded.
> 
> For example, klp-convert would replace the above symbols with:
> 
> $> readelf -r -W <livepatch_module_proceed_by_klp_convert>
> Relocation section '.klp.rela.vmlinux.text' at offset 0x5cb60 contains 1 entry:
>     Offset             Info             Type               Symbol's Value  Symbol's Name + Addend
> 0000000000000068  0000003c00000002 R_X86_64_PC32          0000000000000000 .klp.sym.vmlinux.saved_command_line,0 - 4
> 
> 
> Note that similar macro was needed also in the original version
> to handle more symbols of the same name (sympos).
> 
> Given the above, add klp-convert tool; integrate klp-convert tool into
> kbuild; add data-structure and macros to enable users to annotate
> livepatch source code; make modpost stage compatible with livepatches;
> update livepatch-sample and update documentation.
> 
> 
> Testing
> -------
> 
> The patchset selftests build and execute on x86_64, s390x, and ppc64le
> for both default config (with added livepatch dependencies) and a larger
> SLE-15-ish config.
> 
> 
> Summary of changes in this minimal version v3
> ------------------------
> 
> - klp-convert: symbol format changes (suggested by jlawrence)
> - samples: fixed name of added sample in Makefile (suggested by pmladek)
> - selftests: added ibt test case as an example (DON'T MERGE)
> - fixed all suggested small changes in v2
> 
> Previous versions
> -----------------
> 
> RFC:
>   https://lore.kernel.org/r/cover.1477578530.git.jpoimboe@xxxxxxxxxx/
> v2:
>   https://lore.kernel.org/r/f52d29f7-7d1b-ad3d-050b-a9fa8878faf2@xxxxxxxxxx/
> v3:
>   https://lore.kernel.org/r/20190410155058.9437-1-joe.lawrence@xxxxxxxxxx/
> v4:
>   https://lore.kernel.org/r/20190509143859.9050-1-joe.lawrence@xxxxxxxxxx/
> v5:
>   (not posted)
>   https://github.com/joe-lawrence/klp-convert-tree/tree/klp-convert-v5-devel
> v6:
>   https://lore.kernel.org/r/20220216163940.228309-1-joe.lawrence@xxxxxxxxxx/
> v7:
>   https://lore.kernel.org/r/20230306140824.3858543-1-joe.lawrence@xxxxxxxxxx/
> v1 minimal:
>   https://lore.kernel.org/r/20231106162513.17556-1-lhruska@xxxxxxx/
> v2 minimal:
>   https://lore.kernel.org/r/20240516133009.20224-1-lhruska@xxxxxxx/
> 

Hi Lukas,

Thanks again for posting the patchset and trying a simpler approach.

I tested with latest kpatch-build tree with no ill effects --
essentially klp-convert is safe to run against .ko files that already
contain klp-relocations.

I would prefer more extensive selftests for various klp-relocation types
(as well as symbol position), however I believe wasn't the point of the
minimal version of this patchset.  We can add more tests later.

Anyway, now we have two RFC / patchsets supporting in-tree creation of
klp-relocations (klp-convert and Josh's objtool patchset).  I think we
need to figure out whether one precludes the other, can they co-exist,
or does that even make sense.

Since LPC is right around the corner, does it make sense for folks to
sync up at some point and talk pros/cons to various approaches?  We
don't have a microconference this year, but perhaps over lunch or beers?

--
Joe