[PATCH] recordmcount: avoid using ABS symbol as reference

[Ard Biesheuvel]

On Mon, 7 Jun 2021 at 08:59, Mark-PK Tsai <mark-pk.tsai at mediatek.com> wrote:
>
> > > On Mon, 7 Jun 2021 at 04:42, Mark-PK Tsai <mark-pk.tsai at mediatek.com> wrote:
> > >
> > > Avoid using ABS symbol, which won't be relocate, as reference.
> > >
> > > On arm64 platform, if there's shndx equals SHN_ABS(0xfff1).
> > >
> > > Section Headers:
> > > [Nr]    Name                         Type      Address          Off      Size   ES  Flg Lk     Inf    Al
> > > [65521] .text.n_tty_receive_buf      PROGBITS  0000000000000000 3cdab520 000054 00  AX  0      0      4
> > > [65522] .rela.text.n_tty_receive_buf RELA      0000000000000000 3cdab578 000030 18  I   152076 65521  8
> > >
> >
> > A RELA section's r_info field points to the section to which it
> > applies. This is why in the example above section #65522 points to
> > section #65521. This has nothing to do with the numerical value of
> > SHN_ABS.
>
> If the r_info of RELA section is 65521(0xfff1),
> find_secsym_ndx() will use it to find the base symbol.
>

But what does that have to do with the sh_info field of the RELA
section's Elf_Shdr struct? IOW, what is the relevance of section
#65521 here?

> And in the symbol search loop in find_secsym_ndx(), get_symindex will
> return 0xfff1 if the symbol is in ABS section.
>
> In this case, find_secsym_ndx() will return a absolute symbol as
> base, which won't be relocate, if an ABS symbol is found before the
> real symbol in section 65521.
>

I see your point here.

> >
> > > find_secsym_ndx, which use r_info in rela section to find the reference
> > > symbol, may take ABS symbol as base.
> > >
> > > Symbol table '.symtab' contains 453285 entries:
> > >    Num:    Value          Size Type    Bind   Vis       Ndx Name
> > >      6: 0000000000000002     0 NOTYPE  LOCAL  DEFAULT   ABS section_count
> > >
> > > Which cause an invalid address in __mcount_loc.
> > >
> >
> > Could you give a better account of the error you are trying to address?
> >
> > Also, arm64 no longer defines a section_count symbol (since v5.11), so
> > please make sure that the diagnostics of the issue you are addressing
> > are accurate for mainline.
> >
>
> My kernel version is 5.4.61.
> But as I explained, I suppose mainline also have this issue.
>

Mainline is what we work on. So please base your changes (and your
commit log) on mainline.


> >
> > > Signed-off-by: Mark-PK Tsai <mark-pk.tsai at mediatek.com>
> > > ---
> > >  scripts/recordmcount.h | 4 ++++
> > >  1 file changed, 4 insertions(+)
> > >
> > > diff --git a/scripts/recordmcount.h b/scripts/recordmcount.h
> > > index f9b19524da11..9b69167fb7ff 100644
> > > --- a/scripts/recordmcount.h
> > > +++ b/scripts/recordmcount.h
> > > @@ -526,6 +526,10 @@ static int find_secsym_ndx(unsigned const txtndx,
> > >         for (symp = sym0, t = nsym; t; --t, ++symp) {
> > >                 unsigned int const st_bind = ELF_ST_BIND(symp->st_info);
> > >
> > > +               /* avoid absolute symbols */
> > > +               if (symp->st_shndx == SHN_ABS)
> > > +                       continue;
> > > +
> > >                 if (txtndx == get_symindex(symp, symtab, symtab_shndx)
> > >                         /* avoid STB_WEAK */
> > >                     && (STB_LOCAL == st_bind || STB_GLOBAL == st_bind)) {