[PATCH v3] recordmcount: Correct st_shndx handling

[Peter Zijlstra]

On Wed, Jun 16, 2021 at 12:23:13AM +0800, Mark-PK Tsai wrote:
> From: Peter Zijlstra <peterz at infradead.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 at mediatek.com>
> Signed-off-by: Peter Zijlstra (Intel) <peterz at infradead.org>
> Tested-by: Mark-PK Tsai <mark-pk.tsai at mediatek.com>
> [handle endianness of sym->st_shndx]
> Signed-off-by: Mark-PK Tsai <mark-pk.tsai at mediatek.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.

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)