[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20210616152946.2709-1-mark-pk.tsai@mediatek.com>
Date: Wed, 16 Jun 2021 23:29:46 +0800
From: Mark-PK Tsai <mark-pk.tsai@...iatek.com>
To: <peterz@...radead.org>
CC: <linux-arm-kernel@...ts.infradead.org>,
<linux-kernel@...r.kernel.org>,
<linux-mediatek@...ts.infradead.org>, <mark-pk.tsai@...iatek.com>,
<matthias.bgg@...il.com>, <mhelsley@...are.com>,
<rostedt@...dmis.org>, <samitolvanen@...gle.com>,
<yj.chiang@...iatek.com>
Subject: Re: [PATCH v3] recordmcount: Correct st_shndx handling
> On Wed, Jun 16, 2021 at 12:23:13AM +0800, Mark-PK Tsai wrote:
> > From: Peter Zijlstra <peterz@...radead.org>
> >
> > One should only use st_shndx when >SHN_UNDEF and <SHN_LORESERVE. When
> > SHN_XINDEX, then use .symtab_shndx. Otherwise use 0.
> >
> > This handles the case: st_shndx >= SHN_LORESERVE && st_shndx != SHN_XINDEX.
> >
> > Reported-by: Mark-PK Tsai <mark-pk.tsai@...iatek.com>
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
> > Tested-by: Mark-PK Tsai <mark-pk.tsai@...iatek.com>
> > [handle endianness of sym->st_shndx]
> > Signed-off-by: Mark-PK Tsai <mark-pk.tsai@...iatek.com>
> > ---
> > scripts/recordmcount.h | 13 +++++++++----
> > 1 file changed, 9 insertions(+), 4 deletions(-)
> >
> > diff --git a/scripts/recordmcount.h b/scripts/recordmcount.h
> > index f9b19524da11..ef9c3425f86b 100644
> > --- a/scripts/recordmcount.h
> > +++ b/scripts/recordmcount.h
> > @@ -194,13 +194,18 @@ static unsigned int get_symindex(Elf_Sym const *sym, Elf32_Word const *symtab,
> > unsigned long offset;
> > int index;
> >
> > + if (w2(sym->st_shndx) > SHN_UNDEF &&
> > + w2(sym->st_shndx) < SHN_LORESERVE)
> > return w2(sym->st_shndx);
> >
> > + if (w2(sym->st_shndx) == SHN_XINDEX) {
> > + offset = (unsigned long)sym - (unsigned long)symtab;
> > + index = offset / sizeof(*sym);
> >
> > + return w(symtab_shndx[index]);
> > + }
> > +
> > + return 0;
> > }
>
> Thanks. However that leads to atrocious codegen because w2 is an
> indirect function, something like the below seems much better.
Oh, I didn't notice that.
I'll update in v4.
Thanks!
>
> 1d00: 41 0f b7 7f 0e movzwl 0xe(%r15),%edi
> 1d05: c0 eb 04 shr $0x4,%bl
> 1d08: ff 15 7a 54 00 00 callq *0x547a(%rip) # 7188 <w2>
> 1d0e: 85 c0 test %eax,%eax
> 1d10: 74 16 je 1d28 <main+0xba8>
> 1d12: 41 0f b7 7f 0e movzwl 0xe(%r15),%edi
> 1d17: ff 15 6b 54 00 00 callq *0x546b(%rip) # 7188 <w2>
> 1d1d: 3d ff fe 00 00 cmp $0xfeff,%eax
> 1d22: 0f 86 00 03 00 00 jbe 2028 <main+0xea8>
> 1d28: 41 0f b7 7f 0e movzwl 0xe(%r15),%edi
> 1d2d: ff 15 55 54 00 00 callq *0x5455(%rip) # 7188 <w2>
>
> vs
>
> 1d0c: 41 0f b7 7f 0e movzwl 0xe(%r15),%edi
> 1d11: ff 15 71 54 00 00 callq *0x5471(%rip) # 7188 <w2>
>
> ---
> diff --git a/scripts/recordmcount.h b/scripts/recordmcount.h
> index f9b19524da11..b3e9d0563c03 100644
> --- a/scripts/recordmcount.h
> +++ b/scripts/recordmcount.h
> @@ -192,15 +192,23 @@ static unsigned int get_symindex(Elf_Sym const *sym, Elf32_Word const *symtab,
> Elf32_Word const *symtab_shndx)
> {
> unsigned long offset;
> + unsigned short shndx;
> int index;
>
> - if (sym->st_shndx != SHN_XINDEX)
> - return w2(sym->st_shndx);
> + shndx = w2(sym->st_shndx);
>
> - offset = (unsigned long)sym - (unsigned long)symtab;
> - index = offset / sizeof(*sym);
> + if (shndx > SHN_UNDEF &&
> + shndx < SHN_LORESERVE)
> + return shndx;
>
> - return w(symtab_shndx[index]);
> + if (shndx == SHN_XINDEX) {
> + offset = (unsigned long)sym - (unsigned long)symtab;
> + index = offset / sizeof(*sym);
> +
> + return w(symtab_shndx[index]);
> + }
> +
> + return 0;
> }
>
> static unsigned int get_shnum(Elf_Ehdr const *ehdr, Elf_Shdr const *shdr0)
Powered by blists - more mailing lists