Re: [PATCH v2 04/10] objtool: check: Ignore empty alternative groups

[Julien Thierry]

On 4/1/20 1:53 PM, Miroslav Benes wrote:
> On Fri, 27 Mar 2020, Julien Thierry wrote:
>
> > Atlernative section can contain entries for alternatives with no
> > instructions. Objtool will currently crash when handling such an entry.
> >
> > Just skip that entry, but still give a warning to discourage useless
> > entries.
> >
> > Signed-off-by: Julien Thierry <jthierry@xxxxxxxxxx>
> > ---
> >   tools/objtool/check.c | 6 ++++++
> >   1 file changed, 6 insertions(+)
> >
> > diff --git a/tools/objtool/check.c b/tools/objtool/check.c
> > index 74353b2c39ce..5c03460f1f07 100644
> > --- a/tools/objtool/check.c
> > +++ b/tools/objtool/check.c
> > @@ -904,6 +904,12 @@ static int add_special_section_alts(struct objtool_file *file)
> >   		}
> >   
> >   		if (special_alt->group) {
> > +			if (!special_alt->orig_len) {
> > +				WARN_FUNC("empty alternative entry",
> > +					  orig_insn->sec, orig_insn->offset);
> > +				continue;
> > +			}
> > +
> >   			ret = handle_group_alt(file, special_alt, orig_insn,
> >   					       &new_insn);
> >   			if (ret)
>
> Probably the first time I am looking at alternatives handling in objtool,
> so I must be missing something, but is this even possible now? I mean
> get_alt_entry() in special.c sets alt->orig_len when alt->group is true
> (which means .alternatives section) to something which cannot be zero.

What I see is:

	if (alt->group) {
		alt->orig_len = *(unsigned char *)(sec->data->d_buf + offset +
						   entry->orig_len);
		alt->new_len = *(unsigned char *)(sec->data->d_buf + offset +
                                                  entry->new_len);
	}


And as far as I can tell, "alt->orig_len" can be 0 if the entry in the 
.altinstructions section of the .o file has the length set to 0.

I don't know how the alternative section generation works on x86, but on 
arm64 it's just a computed assembly offset which can be 0.

> Is this a preparatory patch for arm64, where this could happen? If yes, it
> would be better to mention it in the changelog.

It used to happen on arm64, but the fix [1] was picked.

I can add that link to the commit if necessary.

[1] https://lkml.org/lkml/2020/1/9/708

Cheers,

--
Julien Thierry