Re: [PATCH v4 06/10] objtool: Handle unsorted table offset of rodata

From: Josh Poimboeuf
Date: Wed Nov 27 2024 - 20:00:51 EST


On Wed, Nov 27, 2024 at 04:16:29PM -0800, Josh Poimboeuf wrote:
> On Wed, Nov 27, 2024 at 04:10:18PM -0800, Josh Poimboeuf wrote:
> > On Wed, Nov 27, 2024 at 03:01:33PM +0800, Tiezhu Yang wrote:
> > >
> > >
> > > On 11/27/2024 09:20 AM, Josh Poimboeuf wrote:
> > > > On Tue, Nov 26, 2024 at 09:28:19PM +0800, Tiezhu Yang wrote:
> > > > > > + /* Handle the special cases compiled with Clang on LoongArch */
> > > > > > + if (file->elf->ehdr.e_machine == EM_LOONGARCH &&
> > > > > > + reloc->sym->type == STT_SECTION && reloc != table &&
> > > > > > + reloc_offset(reloc) == reloc_offset(table) + rodata_table_size)
> > > > > > + break;
> > > > >
> > > > > I think it can be generic, like this:
> > > > >
> > > > > /* Check for the end of the table: */
> > > > > if (reloc != table && reloc == next_table)
> > > > > break;
> > > > >
> > > > > if (reloc != table &&
> > > > > reloc_offset(reloc) == reloc_offset(table) +
> > > > > rodata_table_size)
> > > > > break;
> > > > >
> > > > > What do you think?
> > > >
> > > > I'm not sure, this patch is hard to review because it uses
> > > > insn->table_size which doesn't get set until the next patch.
> > > >
> > > > Maybe this patch should come after patches 7 & 8, or maybe they should
> > > > be squashed?
> > >
> > > OK, I will squash the changes into patch #7.
> >
> > I remembered Ard already solved a similar problem when he prototyped x86
> > jump table annotation. Can you pull this patch into your series:
> >
> > https://lore.kernel.org/20241011170847.334429-12-ardb+git@xxxxxxxxxx
>
> Actually, I think I'm going to merge patches 2-5 from Ard's series as
> they're a nice cleanup. Let me do that and then you can base your next
> version off tip/objtool/core once it gets updated with Ard's and Peter's
> patches.

Still talking to myself here, I think we'll only merge the above patch,
since we don't know what the generic annotations are going to look like
yet.

--
Josh