Re: [PATCH 1/9] objtool: Generic annotation infrastructure

[Peter Zijlstra <peterz@infradead.org>]

On Sat, Nov 23, 2024 at 08:16:40PM -0700, Nathan Chancellor wrote:
> On Fri, Nov 22, 2024 at 01:10:17PM +0100, Peter Zijlstra wrote:
> > Avoid endless .discard.foo sections for each annotation, create a
> > single .discard.annotate section that takes an annotation type along
> > with the instruction.
> > 
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ...
> > --- a/tools/objtool/check.c
> > +++ b/tools/objtool/check.c
> > @@ -2373,6 +2373,49 @@ static int read_unwind_hints(struct objt
> >  	return 0;
> >  }
> >  
> > +static int read_annotate(struct objtool_file *file, void (*func)(int type, struct instruction *insn))
> > +{
> > +	struct section *sec;
> > +	struct instruction *insn;
> > +	struct reloc *reloc;
> > +	int type;
> > +
> > +	sec = find_section_by_name(file->elf, ".discard.annotate");
> > +	if (!sec)
> > +		return 0;
> > +
> > +	if (!sec->rsec)
> > +		return 0;
> > +
> > +	if (sec->sh.sh_entsize != 8) {
> > +		static bool warned = false;
> > +		if (!warned) {
> > +			WARN("%s: dodgy linker, sh_entsize != 8", sec->name);
> 
> Thanks to Fangrui, this has been resolved in LLVM main:
> 
> https://github.com/llvm/llvm-project/commit/d4bed617f4378873d7ddf4b53c041e7b39d1a9ca
> https://github.com/ClangBuiltLinux/linux/issues/2057#issuecomment-2495675374
> 
> I have built a version of LLVM from main and verified that this warning
> does not trigger with that version, while it does with LLVM 19.1.4.

Excellent, thanks for getting this sorted.

Since there's a fair number of llvm releases between the minimally
supported version to build a kernel with and this fix, I'll leave the
warning as non fatal.

One question; is there any other means of setting entsize aside from
(ab)using SHF_MERGE ? The (GNU) as documetation for .section only
mentions entsize in combination with SHF_MERGE.