* gold --emit-relocs
@ 2015-08-19 5:16 Alan Modra
2015-08-19 7:39 ` Cary Coutant
0 siblings, 1 reply; 3+ messages in thread
From: Alan Modra @ 2015-08-19 5:16 UTC (permalink / raw)
To: binutils; +Cc: Cary Coutant
A symbol value in an ELF final linked binary is absolute, in contrast
to a relocatable object file where the value is section relative. For
--emit-relocs it is therefore incorrect to use the value of a section
symbol as the addend when adjusting relocs against input section
symbols to output section symbols.
I believe it is correct to always subtract os->address() from the
addend. Relocatable linking will set output section addresses to
zero, but if for some reason the addresses changed to be non-zero then
you'd need to subtract os->address() for ld -r too.
Is this OK to commit?
PR gold/18846
* target-reloc.h (relocate_relocs): Move handling of addend for
RELOC_ADJUST_FOR_SECTION_RELA. Subtract os->address() from addend.
* powerpc.cc (relocate_relocs): Likewise.
diff --git a/gold/powerpc.cc b/gold/powerpc.cc
index 540e197..a9d93aa 100644
--- a/gold/powerpc.cc
+++ b/gold/powerpc.cc
@@ -8095,7 +8095,6 @@ Target_powerpc<size, big_endian>::relocate_relocs(
typename elfcpp::Elf_types<size>::Elf_WXword r_info = reloc.get_r_info();
unsigned int r_sym = elfcpp::elf_r_sym<size>(r_info);
unsigned int r_type = elfcpp::elf_r_type<size>(r_info);
- const unsigned int orig_r_sym = r_sym;
typename elfcpp::Elf_types<size>::Elf_Swxword addend
= reloc.get_r_addend();
const Symbol* gsym = NULL;
@@ -8130,6 +8129,8 @@ Target_powerpc<size, big_endian>::relocate_relocs(
// the output section corresponding to input section
// in which this symbol is defined.
gold_assert(r_sym < local_count);
+ const Symbol_value<size>* psymval = object->local_symbol(r_sym);
+ addend = psymval->value(object, addend);
bool is_ordinary;
unsigned int shndx =
object->local_symbol_input_shndx(r_sym, &is_ordinary);
@@ -8138,6 +8139,7 @@ Target_powerpc<size, big_endian>::relocate_relocs(
gold_assert(os != NULL);
gold_assert(os->needs_symtab_index());
r_sym = os->symtab_index();
+ addend -= os->address();
}
break;
@@ -8185,10 +8187,8 @@ Target_powerpc<size, big_endian>::relocate_relocs(
if (strategy == Relocatable_relocs::RELOC_COPY)
;
else if (strategy == Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_RELA)
- {
- const Symbol_value<size>* psymval = object->local_symbol(orig_r_sym);
- addend = psymval->value(object, addend);
- }
+ // Handled above
+ ;
else if (strategy == Relocatable_relocs::RELOC_SPECIAL)
{
if (addend >= 32768)
diff --git a/gold/target-reloc.h b/gold/target-reloc.h
index c135459..3688e4e 100644
--- a/gold/target-reloc.h
+++ b/gold/target-reloc.h
@@ -702,6 +702,18 @@ relocate_relocs(
gold_assert(os != NULL);
gold_assert(os->needs_symtab_index());
new_symndx = os->symtab_index();
+ if (strategy == Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_RELA)
+ {
+ typename elfcpp::Elf_types<size>::Elf_Swxword addend;
+ const Symbol_value<size>* psymval;
+ psymval = object->local_symbol(r_sym);
+ addend = Reloc_types<sh_type, size, big_endian>::
+ get_reloc_addend(&reloc);
+ addend = psymval->value(object, addend);
+ addend -= os->address();
+ Reloc_types<sh_type, size, big_endian>::
+ set_reloc_addend(&reloc_write, addend);
+ }
}
break;
@@ -776,14 +788,7 @@ relocate_relocs(
switch (strategy)
{
case Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_RELA:
- {
- typename elfcpp::Elf_types<size>::Elf_Swxword addend;
- addend = Reloc_types<sh_type, size, big_endian>::
- get_reloc_addend(&reloc);
- addend = psymval->value(object, addend);
- Reloc_types<sh_type, size, big_endian>::
- set_reloc_addend(&reloc_write, addend);
- }
+ // Handled above
break;
case Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_0:
--
Alan Modra
Australia Development Lab, IBM
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: gold --emit-relocs
2015-08-19 5:16 gold --emit-relocs Alan Modra
@ 2015-08-19 7:39 ` Cary Coutant
2015-08-20 2:35 ` Alan Modra
0 siblings, 1 reply; 3+ messages in thread
From: Cary Coutant @ 2015-08-19 7:39 UTC (permalink / raw)
To: Alan Modra; +Cc: Binutils
> A symbol value in an ELF final linked binary is absolute, in contrast
> to a relocatable object file where the value is section relative. For
> --emit-relocs it is therefore incorrect to use the value of a section
> symbol as the addend when adjusting relocs against input section
> symbols to output section symbols.
>
> I believe it is correct to always subtract os->address() from the
> addend. Relocatable linking will set output section addresses to
> zero, but if for some reason the addresses changed to be non-zero then
> you'd need to subtract os->address() for ld -r too.
Yes, I agree with your analysis. Thanks!
> diff --git a/gold/target-reloc.h b/gold/target-reloc.h
> index c135459..3688e4e 100644
> --- a/gold/target-reloc.h
> +++ b/gold/target-reloc.h
> @@ -702,6 +702,18 @@ relocate_relocs(
> gold_assert(os != NULL);
> gold_assert(os->needs_symtab_index());
> new_symndx = os->symtab_index();
> + if (strategy == Relocatable_relocs::RELOC_ADJUST_FOR_SECTION_RELA)
> + {
> + typename elfcpp::Elf_types<size>::Elf_Swxword addend;
> + const Symbol_value<size>* psymval;
> + psymval = object->local_symbol(r_sym);
> + addend = Reloc_types<sh_type, size, big_endian>::
> + get_reloc_addend(&reloc);
> + addend = psymval->value(object, addend);
> + addend -= os->address();
> + Reloc_types<sh_type, size, big_endian>::
> + set_reloc_addend(&reloc_write, addend);
> + }
Rather than move this code to here, I'd prefer to leave it where it
is, but promote "os" to the scope of the loop and adjust the addend
using os->address():
diff --git a/gold/target-reloc.h b/gold/target-reloc.h
index c135459..89906af 100644
--- a/gold/target-reloc.h
+++ b/gold/target-reloc.h
@@ -666,6 +666,7 @@ relocate_relocs(
// Get the new symbol index.
+ Output_section* os = NULL;
unsigned int new_symndx;
if (r_sym < local_count)
{
@@ -698,7 +699,7 @@ relocate_relocs(
unsigned int shndx =
object->local_symbol_input_shndx(r_sym, &is_ordinary);
gold_assert(is_ordinary);
- Output_section* os = object->output_section(shndx);
+ os = object->output_section(shndx);
gold_assert(os != NULL);
gold_assert(os->needs_symtab_index());
new_symndx = os->symtab_index();
@@ -780,7 +781,8 @@ relocate_relocs(
typename elfcpp::Elf_types<size>::Elf_Swxword addend;
addend = Reloc_types<sh_type, size, big_endian>::
get_reloc_addend(&reloc);
- addend = psymval->value(object, addend);
+ gold_assert(os != NULL);
+ addend = psymval->value(object, addend) - os->address();
Reloc_types<sh_type, size, big_endian>::
set_reloc_addend(&reloc_write, addend);
}
... And a similar change in powerpc.cc.
-cary
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: gold --emit-relocs
2015-08-19 7:39 ` Cary Coutant
@ 2015-08-20 2:35 ` Alan Modra
0 siblings, 0 replies; 3+ messages in thread
From: Alan Modra @ 2015-08-20 2:35 UTC (permalink / raw)
To: Cary Coutant; +Cc: Binutils
On Wed, Aug 19, 2015 at 12:39:50AM -0700, Cary Coutant wrote:
> Rather than move this code to here, I'd prefer to leave it where it
> is, but promote "os" to the scope of the loop and adjust the addend
> using os->address():
OK, I probably would have written the patch this way myself if I'd
started with target-reloc.h. Patch revised as you suggest and
committed 9215b98b.
--
Alan Modra
Australia Development Lab, IBM
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2015-08-20 2:35 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-08-19 5:16 gold --emit-relocs Alan Modra
2015-08-19 7:39 ` Cary Coutant
2015-08-20 2:35 ` Alan Modra
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).