[1/2] LoongArch: bfd: Remove elf_seg_map condition in loongarch_elf_relax_section
Checks
Commit Message
When I tried "./ld-new sth." the relax does not work because the
elf_seg_map (info->output_bfd) was NULL in loongarch_elf_relax_section.
There is no need for the segment info, we can get the address through
the section info. Thus, remove this condition.
bfd/ChangeLog:
* elfnn-loongarch.c (loongarch_elf_relax_section): Remove
elf_seg_map condition.
ld/ChangeLog:
* testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
* testsuite/ld-loongarch-elf/relax.dd: New test.
---
bfd/elfnn-loongarch.c | 1 -
ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
ld/testsuite/ld-loongarch-elf/relax.exp | 35 +++++++------------------
3 files changed, 16 insertions(+), 26 deletions(-)
create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
Comments
Hi,
Without this condition, the value of 'symbol - pc' may be differences
between
loongarch_elf_relax_section and loongarch_elf_relocate_section.
And it will cause some spec2006 test errors.
在 2023/7/11 下午4:49, Jinyang He 写道:
> When I tried "./ld-new sth." the relax does not work because the
> elf_seg_map (info->output_bfd) was NULL in loongarch_elf_relax_section.
> There is no need for the segment info, we can get the address through
> the section info. Thus, remove this condition.
>
> bfd/ChangeLog:
>
> * elfnn-loongarch.c (loongarch_elf_relax_section): Remove
> elf_seg_map condition.
>
> ld/ChangeLog:
>
> * testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
> * testsuite/ld-loongarch-elf/relax.dd: New test.
> ---
> bfd/elfnn-loongarch.c | 1 -
> ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
> ld/testsuite/ld-loongarch-elf/relax.exp | 35 +++++++------------------
> 3 files changed, 16 insertions(+), 26 deletions(-)
> create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
>
> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> index d3d8419d8..01f349a24 100644
> --- a/bfd/elfnn-loongarch.c
> +++ b/bfd/elfnn-loongarch.c
> @@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd, asection *sec,
> || sec->sec_flg0
> || (sec->flags & SEC_RELOC) == 0
> || sec->reloc_count == 0
> - || elf_seg_map (info->output_bfd) == NULL
> || (info->disable_target_specific_optimizations
> && info->relax_pass == 0)
> /* The exp_seg_relro_adjust is enum phase_enum (0x4),
> diff --git a/ld/testsuite/ld-loongarch-elf/relax.dd b/ld/testsuite/ld-loongarch-elf/relax.dd
> new file mode 100644
> index 000000000..e266beb8f
> --- /dev/null
> +++ b/ld/testsuite/ld-loongarch-elf/relax.dd
> @@ -0,0 +1,6 @@
> +#...
> +.*pcaddi.*
> +.*ld.w.*
> +.*pcaddi.*
> +.*ld.w.*
> +#pass
> diff --git a/ld/testsuite/ld-loongarch-elf/relax.exp b/ld/testsuite/ld-loongarch-elf/relax.exp
> index 7ff876d79..d2bb967c6 100644
> --- a/ld/testsuite/ld-loongarch-elf/relax.exp
> +++ b/ld/testsuite/ld-loongarch-elf/relax.exp
> @@ -21,31 +21,6 @@
>
> if [istarget loongarch64-*-*] {
>
> - if [isbuild loongarch64-*-*] {
> - set testname "loongarch relax build"
> - set pre_builds [list \
> - [list \
> - "$testname" \
> - "" \
> - "" \
> - {relax.s} \
> - {} \
> - "relax" \
> - ] \
> - ]
> -
> - run_cc_link_tests $pre_builds
> -
> - if [file exist "tmpdir/relax"] {
> - set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
> - if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
> - pass "loongarch relax"
> - } {
> - fail "loongarch relax"
> - }
> - }
> - }
> -
> run_ld_link_tests \
> [list \
> [list \
> @@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
> ] \
> "relax-align" \
> ] \
> + [list \
> + "relax" \
> + "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
> + "" \
> + {relax.s} \
> + [list \
> + [list objdump -d relax.dd] \
> + ] \
> + "relax" \
> + ] \
> ]
>
> set objdump_flags "-s -j .data"
Hi,
I'm not fimiliar with the process of output binary. I just noted that it
calls lang_relax_sections() before calling
_bfd_elf_map_sections_to_segments() in ldelf_map_segments(). Thus I
think the condition, elf_seg_map (info->output_bfd) == NULL, always be
true when first time it calls loongarch_elf_relax_section(). Actually we
can get right symbol value when relaxation, but ...
You reminder me that the symbol value will be reduced due to relaxation,
which like the instruction pc is reduced. Then we cannot get right
symbol value if the symbol in relaxable section. And I write a test,
$ cat c.s
.section ".another.text", "x"
nop
la.local $a0, sym1
sym2:
nop
.text
sym1:
nop
nop
nop
la.local $a0, sym2
$ ./gas/as-new -mabi=lp64d -o c.o c.s
$ ./ld/ld-new c.o -o c -e 0 --section-start=.another.text=0x120000000
-Ttext 0x120200000
Disassembly of section .another.text:
0000000120000000 <sym2-0x8>:
120000000: 03400000 andi $zero, $zero, 0x0
120000004: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
0000000120000008 <sym2>:
120000008: 03400000 andi $zero, $zero, 0x0
Disassembly of section .text:
0000000120200000 <sym1>:
120200000: 03400000 andi $zero, $zero, 0x0
120200004: 03400000 andi $zero, $zero, 0x0
120200008: 03400000 andi $zero, $zero, 0x0
12020000c: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
Due to the bug repaired by Patch2 and the bug reported by this test. And
I also did not see that condition in other archs. I'm not sure that the
relaxation spec2006 error is caused by that condition or those bugs.
On 2023-07-12 14:35, mengqinggang wrote:
> Hi,
>
> Without this condition, the value of 'symbol - pc' may be differences
> between
> loongarch_elf_relax_section and loongarch_elf_relocate_section.
> And it will cause some spec2006 test errors.
>
> 在 2023/7/11 下午4:49, Jinyang He 写道:
>> When I tried "./ld-new sth." the relax does not work because the
>> elf_seg_map (info->output_bfd) was NULL in loongarch_elf_relax_section.
>> There is no need for the segment info, we can get the address through
>> the section info. Thus, remove this condition.
>>
>> bfd/ChangeLog:
>>
>> * elfnn-loongarch.c (loongarch_elf_relax_section): Remove
>> elf_seg_map condition.
>>
>> ld/ChangeLog:
>>
>> * testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
>> * testsuite/ld-loongarch-elf/relax.dd: New test.
>> ---
>> bfd/elfnn-loongarch.c | 1 -
>> ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
>> ld/testsuite/ld-loongarch-elf/relax.exp | 35 +++++++------------------
>> 3 files changed, 16 insertions(+), 26 deletions(-)
>> create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
>>
>> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
>> index d3d8419d8..01f349a24 100644
>> --- a/bfd/elfnn-loongarch.c
>> +++ b/bfd/elfnn-loongarch.c
>> @@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd,
>> asection *sec,
>> || sec->sec_flg0
>> || (sec->flags & SEC_RELOC) == 0
>> || sec->reloc_count == 0
>> - || elf_seg_map (info->output_bfd) == NULL
>> || (info->disable_target_specific_optimizations
>> && info->relax_pass == 0)
>> /* The exp_seg_relro_adjust is enum phase_enum (0x4),
>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.dd
>> b/ld/testsuite/ld-loongarch-elf/relax.dd
>> new file mode 100644
>> index 000000000..e266beb8f
>> --- /dev/null
>> +++ b/ld/testsuite/ld-loongarch-elf/relax.dd
>> @@ -0,0 +1,6 @@
>> +#...
>> +.*pcaddi.*
>> +.*ld.w.*
>> +.*pcaddi.*
>> +.*ld.w.*
>> +#pass
>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.exp
>> b/ld/testsuite/ld-loongarch-elf/relax.exp
>> index 7ff876d79..d2bb967c6 100644
>> --- a/ld/testsuite/ld-loongarch-elf/relax.exp
>> +++ b/ld/testsuite/ld-loongarch-elf/relax.exp
>> @@ -21,31 +21,6 @@
>> if [istarget loongarch64-*-*] {
>> - if [isbuild loongarch64-*-*] {
>> - set testname "loongarch relax build"
>> - set pre_builds [list \
>> - [list \
>> - "$testname" \
>> - "" \
>> - "" \
>> - {relax.s} \
>> - {} \
>> - "relax" \
>> - ] \
>> - ]
>> -
>> - run_cc_link_tests $pre_builds
>> -
>> - if [file exist "tmpdir/relax"] {
>> - set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
>> - if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
>> - pass "loongarch relax"
>> - } {
>> - fail "loongarch relax"
>> - }
>> - }
>> - }
>> -
>> run_ld_link_tests \
>> [list \
>> [list \
>> @@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
>> ] \
>> "relax-align" \
>> ] \
>> + [list \
>> + "relax" \
>> + "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
>> + "" \
>> + {relax.s} \
>> + [list \
>> + [list objdump -d relax.dd] \
>> + ] \
>> + "relax" \
>> + ] \
>> ]
>> set objdump_flags "-s -j .data"
Hi,
I will apply this two patches to see if there are any errors in specc2006.
在 2023/7/12 下午4:08, Jinyang He 写道:
> Hi,
>
>
> I'm not fimiliar with the process of output binary. I just noted that
> it calls lang_relax_sections() before calling
> _bfd_elf_map_sections_to_segments() in ldelf_map_segments(). Thus I
> think the condition, elf_seg_map (info->output_bfd) == NULL, always be
> true when first time it calls loongarch_elf_relax_section(). Actually
> we can get right symbol value when relaxation, but ...
>
> You reminder me that the symbol value will be reduced due to
> relaxation, which like the instruction pc is reduced. Then we cannot
> get right symbol value if the symbol in relaxable section. And I write
> a test,
>
> $ cat c.s
> .section ".another.text", "x"
> nop
> la.local $a0, sym1
> sym2:
> nop
>
> .text
> sym1:
> nop
> nop
> nop
> la.local $a0, sym2
>
> $ ./gas/as-new -mabi=lp64d -o c.o c.s
> $ ./ld/ld-new c.o -o c -e 0 --section-start=.another.text=0x120000000
> -Ttext 0x120200000
>
> Disassembly of section .another.text:
>
> 0000000120000000 <sym2-0x8>:
> 120000000: 03400000 andi $zero, $zero, 0x0
> 120000004: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
>
> 0000000120000008 <sym2>:
> 120000008: 03400000 andi $zero, $zero, 0x0
>
> Disassembly of section .text:
>
> 0000000120200000 <sym1>:
> 120200000: 03400000 andi $zero, $zero, 0x0
> 120200004: 03400000 andi $zero, $zero, 0x0
> 120200008: 03400000 andi $zero, $zero, 0x0
> 12020000c: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
>
> Due to the bug repaired by Patch2 and the bug reported by this test.
> And I also did not see that condition in other archs. I'm not sure
> that the relaxation spec2006 error is caused by that condition or
> those bugs.
>
>
> On 2023-07-12 14:35, mengqinggang wrote:
>> Hi,
>>
>> Without this condition, the value of 'symbol - pc' may be differences
>> between
>> loongarch_elf_relax_section and loongarch_elf_relocate_section.
>> And it will cause some spec2006 test errors.
>>
>> 在 2023/7/11 下午4:49, Jinyang He 写道:
>>> When I tried "./ld-new sth." the relax does not work because the
>>> elf_seg_map (info->output_bfd) was NULL in loongarch_elf_relax_section.
>>> There is no need for the segment info, we can get the address through
>>> the section info. Thus, remove this condition.
>>>
>>> bfd/ChangeLog:
>>>
>>> * elfnn-loongarch.c (loongarch_elf_relax_section): Remove
>>> elf_seg_map condition.
>>>
>>> ld/ChangeLog:
>>>
>>> * testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
>>> * testsuite/ld-loongarch-elf/relax.dd: New test.
>>> ---
>>> bfd/elfnn-loongarch.c | 1 -
>>> ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
>>> ld/testsuite/ld-loongarch-elf/relax.exp | 35
>>> +++++++------------------
>>> 3 files changed, 16 insertions(+), 26 deletions(-)
>>> create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
>>>
>>> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
>>> index d3d8419d8..01f349a24 100644
>>> --- a/bfd/elfnn-loongarch.c
>>> +++ b/bfd/elfnn-loongarch.c
>>> @@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd,
>>> asection *sec,
>>> || sec->sec_flg0
>>> || (sec->flags & SEC_RELOC) == 0
>>> || sec->reloc_count == 0
>>> - || elf_seg_map (info->output_bfd) == NULL
>>> || (info->disable_target_specific_optimizations
>>> && info->relax_pass == 0)
>>> /* The exp_seg_relro_adjust is enum phase_enum (0x4),
>>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.dd
>>> b/ld/testsuite/ld-loongarch-elf/relax.dd
>>> new file mode 100644
>>> index 000000000..e266beb8f
>>> --- /dev/null
>>> +++ b/ld/testsuite/ld-loongarch-elf/relax.dd
>>> @@ -0,0 +1,6 @@
>>> +#...
>>> +.*pcaddi.*
>>> +.*ld.w.*
>>> +.*pcaddi.*
>>> +.*ld.w.*
>>> +#pass
>>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.exp
>>> b/ld/testsuite/ld-loongarch-elf/relax.exp
>>> index 7ff876d79..d2bb967c6 100644
>>> --- a/ld/testsuite/ld-loongarch-elf/relax.exp
>>> +++ b/ld/testsuite/ld-loongarch-elf/relax.exp
>>> @@ -21,31 +21,6 @@
>>> if [istarget loongarch64-*-*] {
>>> - if [isbuild loongarch64-*-*] {
>>> - set testname "loongarch relax build"
>>> - set pre_builds [list \
>>> - [list \
>>> - "$testname" \
>>> - "" \
>>> - "" \
>>> - {relax.s} \
>>> - {} \
>>> - "relax" \
>>> - ] \
>>> - ]
>>> -
>>> - run_cc_link_tests $pre_builds
>>> -
>>> - if [file exist "tmpdir/relax"] {
>>> - set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
>>> - if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
>>> - pass "loongarch relax"
>>> - } {
>>> - fail "loongarch relax"
>>> - }
>>> - }
>>> - }
>>> -
>>> run_ld_link_tests \
>>> [list \
>>> [list \
>>> @@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
>>> ] \
>>> "relax-align" \
>>> ] \
>>> + [list \
>>> + "relax" \
>>> + "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
>>> + "" \
>>> + {relax.s} \
>>> + [list \
>>> + [list objdump -d relax.dd] \
>>> + ] \
>>> + "relax" \
>>> + ] \
>>> ]
>>> set objdump_flags "-s -j .data"
Hi,
We should pay attention to them, the bug in relaxation. Relaxation may
not work correctly because of the bug1 reported by the second patch and
the bug2 reported in the my previous email, although it works well in
most of the time. Those 2 bugs are just what we've found, but there may
be hidden problems.
I also tested the bug2 on RISCV and although it couldn't be resolved, it
reported an error.
Test is,
$ cat c.s
.section ".another.text", "x"
.fill 1, 4, 0
call fn2
call fn2
fn1:
nop
.text
fn2:
.fill 4, 4, 0
call fn1
$ ./gas/as-new -march=rv32ic_zicsr -mno-arch-attr -o c.o c.s
$ ./ld/ld-new c.o -o c -e 0 --section-start=.another.text=0x1000000
-Ttext 0x1100000
c.o: in function `fn2':
(.text+0x10): relocation truncated to fit: R_RISCV_JAL against `fn1'
So, I'll find the reason why reloc_bits_pcrel20_s2() not reports the
overfollow error, and add error report which I think it is more
important than these two patches so that we can get feedback when we link.
Thanks!
On 2023-07-13 11:13, mengqinggang wrote:
> Hi,
>
> I will apply this two patches to see if there are any errors in
> specc2006.
>
> 在 2023/7/12 下午4:08, Jinyang He 写道:
>> Hi,
>>
>>
>> I'm not fimiliar with the process of output binary. I just noted that
>> it calls lang_relax_sections() before calling
>> _bfd_elf_map_sections_to_segments() in ldelf_map_segments(). Thus I
>> think the condition, elf_seg_map (info->output_bfd) == NULL, always
>> be true when first time it calls loongarch_elf_relax_section().
>> Actually we can get right symbol value when relaxation, but ...
>>
>> You reminder me that the symbol value will be reduced due to
>> relaxation, which like the instruction pc is reduced. Then we cannot
>> get right symbol value if the symbol in relaxable section. And I
>> write a test,
>>
>> $ cat c.s
>> .section ".another.text", "x"
>> nop
>> la.local $a0, sym1
>> sym2:
>> nop
>>
>> .text
>> sym1:
>> nop
>> nop
>> nop
>> la.local $a0, sym2
>>
>> $ ./gas/as-new -mabi=lp64d -o c.o c.s
>> $ ./ld/ld-new c.o -o c -e 0 --section-start=.another.text=0x120000000
>> -Ttext 0x120200000
>>
>> Disassembly of section .another.text:
>>
>> 0000000120000000 <sym2-0x8>:
>> 120000000: 03400000 andi $zero, $zero, 0x0
>> 120000004: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
>>
>> 0000000120000008 <sym2>:
>> 120000008: 03400000 andi $zero, $zero, 0x0
>>
>> Disassembly of section .text:
>>
>> 0000000120200000 <sym1>:
>> 120200000: 03400000 andi $zero, $zero, 0x0
>> 120200004: 03400000 andi $zero, $zero, 0x0
>> 120200008: 03400000 andi $zero, $zero, 0x0
>> 12020000c: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
>>
>> Due to the bug repaired by Patch2 and the bug reported by this test.
>> And I also did not see that condition in other archs. I'm not sure
>> that the relaxation spec2006 error is caused by that condition or
>> those bugs.
>>
>>
>> On 2023-07-12 14:35, mengqinggang wrote:
>>> Hi,
>>>
>>> Without this condition, the value of 'symbol - pc' may be
>>> differences between
>>> loongarch_elf_relax_section and loongarch_elf_relocate_section.
>>> And it will cause some spec2006 test errors.
>>>
>>> 在 2023/7/11 下午4:49, Jinyang He 写道:
>>>> When I tried "./ld-new sth." the relax does not work because the
>>>> elf_seg_map (info->output_bfd) was NULL in
>>>> loongarch_elf_relax_section.
>>>> There is no need for the segment info, we can get the address through
>>>> the section info. Thus, remove this condition.
>>>>
>>>> bfd/ChangeLog:
>>>>
>>>> * elfnn-loongarch.c (loongarch_elf_relax_section): Remove
>>>> elf_seg_map condition.
>>>>
>>>> ld/ChangeLog:
>>>>
>>>> * testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
>>>> * testsuite/ld-loongarch-elf/relax.dd: New test.
>>>> ---
>>>> bfd/elfnn-loongarch.c | 1 -
>>>> ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
>>>> ld/testsuite/ld-loongarch-elf/relax.exp | 35
>>>> +++++++------------------
>>>> 3 files changed, 16 insertions(+), 26 deletions(-)
>>>> create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
>>>>
>>>> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
>>>> index d3d8419d8..01f349a24 100644
>>>> --- a/bfd/elfnn-loongarch.c
>>>> +++ b/bfd/elfnn-loongarch.c
>>>> @@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd,
>>>> asection *sec,
>>>> || sec->sec_flg0
>>>> || (sec->flags & SEC_RELOC) == 0
>>>> || sec->reloc_count == 0
>>>> - || elf_seg_map (info->output_bfd) == NULL
>>>> || (info->disable_target_specific_optimizations
>>>> && info->relax_pass == 0)
>>>> /* The exp_seg_relro_adjust is enum phase_enum (0x4),
>>>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.dd
>>>> b/ld/testsuite/ld-loongarch-elf/relax.dd
>>>> new file mode 100644
>>>> index 000000000..e266beb8f
>>>> --- /dev/null
>>>> +++ b/ld/testsuite/ld-loongarch-elf/relax.dd
>>>> @@ -0,0 +1,6 @@
>>>> +#...
>>>> +.*pcaddi.*
>>>> +.*ld.w.*
>>>> +.*pcaddi.*
>>>> +.*ld.w.*
>>>> +#pass
>>>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.exp
>>>> b/ld/testsuite/ld-loongarch-elf/relax.exp
>>>> index 7ff876d79..d2bb967c6 100644
>>>> --- a/ld/testsuite/ld-loongarch-elf/relax.exp
>>>> +++ b/ld/testsuite/ld-loongarch-elf/relax.exp
>>>> @@ -21,31 +21,6 @@
>>>> if [istarget loongarch64-*-*] {
>>>> - if [isbuild loongarch64-*-*] {
>>>> - set testname "loongarch relax build"
>>>> - set pre_builds [list \
>>>> - [list \
>>>> - "$testname" \
>>>> - "" \
>>>> - "" \
>>>> - {relax.s} \
>>>> - {} \
>>>> - "relax" \
>>>> - ] \
>>>> - ]
>>>> -
>>>> - run_cc_link_tests $pre_builds
>>>> -
>>>> - if [file exist "tmpdir/relax"] {
>>>> - set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
>>>> - if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
>>>> - pass "loongarch relax"
>>>> - } {
>>>> - fail "loongarch relax"
>>>> - }
>>>> - }
>>>> - }
>>>> -
>>>> run_ld_link_tests \
>>>> [list \
>>>> [list \
>>>> @@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
>>>> ] \
>>>> "relax-align" \
>>>> ] \
>>>> + [list \
>>>> + "relax" \
>>>> + "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
>>>> + "" \
>>>> + {relax.s} \
>>>> + [list \
>>>> + [list objdump -d relax.dd] \
>>>> + ] \
>>>> + "relax" \
>>>> + ] \
>>>> ]
>>>> set objdump_flags "-s -j .data"
On 2023-07-13 15:19, Jinyang He wrote:
> Hi,
>
>
> We should pay attention to them, the bug in relaxation. Relaxation may
> not work correctly because of the bug1 reported by the second patch
> and the bug2 reported in the my previous email, although it works well
> in most of the time. Those 2 bugs are just what we've found, but there
> may be hidden problems.
> I also tested the bug2 on RISCV and although it couldn't be resolved,
> it reported an error.
>
> Test is,
> $ cat c.s
> .section ".another.text", "x"
> .fill 1, 4, 0
> call fn2
> call fn2
> fn1:
> nop
>
> .text
> fn2:
> .fill 4, 4, 0
> call fn1
>
> $ ./gas/as-new -march=rv32ic_zicsr -mno-arch-attr -o c.o c.s
> $ ./ld/ld-new c.o -o c -e 0 --section-start=.another.text=0x1000000
> -Ttext 0x1100000
> c.o: in function `fn2':
> (.text+0x10): relocation truncated to fit: R_RISCV_JAL against `fn1'
>
> So, I'll find the reason why reloc_bits_pcrel20_s2() not reports the
> overfollow error, and add error report which I think it is more
> important than these two patches so that we can get feedback when we
> link.
>
Now, I find that the sign-overflow-check has bug, include
reloc_bits_pcrel20_s2(), reloc_bits_b26() and (others, I did not test
them). And I think the "mask" should be
bfd_signed_vma mask = ((bfd_signed_vma)0x1 << (howto->bitsize -
(sign ? 1 : 0)) - 1;
The test about reloc_bits_b26() is,
$ cat d.s
.section ".another.text", "x"
b sym1
.text
nop
sym1:
nop
$ ./gas/as-new -mabi=lp64d -o d.o d.s
$ ./ld/ld-new d.o -o d -e 0 --section-start=.another.text=0x130000000
-Ttext 0x120000000
I'll do fix and send v2 after you test spec2006. Or other responses?
Thanks!
>
> On 2023-07-13 11:13, mengqinggang wrote:
>> Hi,
>>
>> I will apply this two patches to see if there are any errors in
>> specc2006.
>>
>> 在 2023/7/12 下午4:08, Jinyang He 写道:
>>> Hi,
>>>
>>>
>>> I'm not fimiliar with the process of output binary. I just noted
>>> that it calls lang_relax_sections() before calling
>>> _bfd_elf_map_sections_to_segments() in ldelf_map_segments(). Thus I
>>> think the condition, elf_seg_map (info->output_bfd) == NULL, always
>>> be true when first time it calls loongarch_elf_relax_section().
>>> Actually we can get right symbol value when relaxation, but ...
>>>
>>> You reminder me that the symbol value will be reduced due to
>>> relaxation, which like the instruction pc is reduced. Then we cannot
>>> get right symbol value if the symbol in relaxable section. And I
>>> write a test,
>>>
>>> $ cat c.s
>>> .section ".another.text", "x"
>>> nop
>>> la.local $a0, sym1
>>> sym2:
>>> nop
>>>
>>> .text
>>> sym1:
>>> nop
>>> nop
>>> nop
>>> la.local $a0, sym2
>>>
>>> $ ./gas/as-new -mabi=lp64d -o c.o c.s
>>> $ ./ld/ld-new c.o -o c -e 0
>>> --section-start=.another.text=0x120000000 -Ttext 0x120200000
>>>
>>> Disassembly of section .another.text:
>>>
>>> 0000000120000000 <sym2-0x8>:
>>> 120000000: 03400000 andi $zero, $zero, 0x0
>>> 120000004: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
>>>
>>> 0000000120000008 <sym2>:
>>> 120000008: 03400000 andi $zero, $zero, 0x0
>>>
>>> Disassembly of section .text:
>>>
>>> 0000000120200000 <sym1>:
>>> 120200000: 03400000 andi $zero, $zero, 0x0
>>> 120200004: 03400000 andi $zero, $zero, 0x0
>>> 120200008: 03400000 andi $zero, $zero, 0x0
>>> 12020000c: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
>>>
>>> Due to the bug repaired by Patch2 and the bug reported by this test.
>>> And I also did not see that condition in other archs. I'm not sure
>>> that the relaxation spec2006 error is caused by that condition or
>>> those bugs.
>>>
>>>
>>> On 2023-07-12 14:35, mengqinggang wrote:
>>>> Hi,
>>>>
>>>> Without this condition, the value of 'symbol - pc' may be
>>>> differences between
>>>> loongarch_elf_relax_section and loongarch_elf_relocate_section.
>>>> And it will cause some spec2006 test errors.
>>>>
>>>> 在 2023/7/11 下午4:49, Jinyang He 写道:
>>>>> When I tried "./ld-new sth." the relax does not work because the
>>>>> elf_seg_map (info->output_bfd) was NULL in
>>>>> loongarch_elf_relax_section.
>>>>> There is no need for the segment info, we can get the address through
>>>>> the section info. Thus, remove this condition.
>>>>>
>>>>> bfd/ChangeLog:
>>>>>
>>>>> * elfnn-loongarch.c (loongarch_elf_relax_section): Remove
>>>>> elf_seg_map condition.
>>>>>
>>>>> ld/ChangeLog:
>>>>>
>>>>> * testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
>>>>> * testsuite/ld-loongarch-elf/relax.dd: New test.
>>>>> ---
>>>>> bfd/elfnn-loongarch.c | 1 -
>>>>> ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
>>>>> ld/testsuite/ld-loongarch-elf/relax.exp | 35
>>>>> +++++++------------------
>>>>> 3 files changed, 16 insertions(+), 26 deletions(-)
>>>>> create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
>>>>>
>>>>> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
>>>>> index d3d8419d8..01f349a24 100644
>>>>> --- a/bfd/elfnn-loongarch.c
>>>>> +++ b/bfd/elfnn-loongarch.c
>>>>> @@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd,
>>>>> asection *sec,
>>>>> || sec->sec_flg0
>>>>> || (sec->flags & SEC_RELOC) == 0
>>>>> || sec->reloc_count == 0
>>>>> - || elf_seg_map (info->output_bfd) == NULL
>>>>> || (info->disable_target_specific_optimizations
>>>>> && info->relax_pass == 0)
>>>>> /* The exp_seg_relro_adjust is enum phase_enum (0x4),
>>>>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.dd
>>>>> b/ld/testsuite/ld-loongarch-elf/relax.dd
>>>>> new file mode 100644
>>>>> index 000000000..e266beb8f
>>>>> --- /dev/null
>>>>> +++ b/ld/testsuite/ld-loongarch-elf/relax.dd
>>>>> @@ -0,0 +1,6 @@
>>>>> +#...
>>>>> +.*pcaddi.*
>>>>> +.*ld.w.*
>>>>> +.*pcaddi.*
>>>>> +.*ld.w.*
>>>>> +#pass
>>>>> diff --git a/ld/testsuite/ld-loongarch-elf/relax.exp
>>>>> b/ld/testsuite/ld-loongarch-elf/relax.exp
>>>>> index 7ff876d79..d2bb967c6 100644
>>>>> --- a/ld/testsuite/ld-loongarch-elf/relax.exp
>>>>> +++ b/ld/testsuite/ld-loongarch-elf/relax.exp
>>>>> @@ -21,31 +21,6 @@
>>>>> if [istarget loongarch64-*-*] {
>>>>> - if [isbuild loongarch64-*-*] {
>>>>> - set testname "loongarch relax build"
>>>>> - set pre_builds [list \
>>>>> - [list \
>>>>> - "$testname" \
>>>>> - "" \
>>>>> - "" \
>>>>> - {relax.s} \
>>>>> - {} \
>>>>> - "relax" \
>>>>> - ] \
>>>>> - ]
>>>>> -
>>>>> - run_cc_link_tests $pre_builds
>>>>> -
>>>>> - if [file exist "tmpdir/relax"] {
>>>>> - set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
>>>>> - if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
>>>>> - pass "loongarch relax"
>>>>> - } {
>>>>> - fail "loongarch relax"
>>>>> - }
>>>>> - }
>>>>> - }
>>>>> -
>>>>> run_ld_link_tests \
>>>>> [list \
>>>>> [list \
>>>>> @@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
>>>>> ] \
>>>>> "relax-align" \
>>>>> ] \
>>>>> + [list \
>>>>> + "relax" \
>>>>> + "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
>>>>> + "" \
>>>>> + {relax.s} \
>>>>> + [list \
>>>>> + [list objdump -d relax.dd] \
>>>>> + ] \
>>>>> + "relax" \
>>>>> + ] \
>>>>> ]
>>>>> set objdump_flags "-s -j .data"
Hi Jinyang,
Any progress on this? During my attempt trying to balance relaxation
and scheduling better in GCC I found the elf_seg_map condition is
preventing relaxation on *every* shared library (when we are linking a
shared library the elf_seg_map of .text is NULL).
On a modern system most code paths are in shared libraries, so it's
really bad not to perform the relaxation on them. Esp. now we are
disabling explicit relocs for relaxation, so if we don't relax shared
libraries we are likely regressing the overall performance of the
system.
Sorry for disturbing you during the national holiday, feel free to defer
the reply until the holiday ends!
On Thu, 2023-07-13 at 16:18 +0800, Jinyang He wrote:
> On 2023-07-13 15:19, Jinyang He wrote:
>
> > Hi,
> >
> >
> > We should pay attention to them, the bug in relaxation. Relaxation may
> > not work correctly because of the bug1 reported by the second patch
> > and the bug2 reported in the my previous email, although it works well
> > in most of the time. Those 2 bugs are just what we've found, but there
> > may be hidden problems.
> > I also tested the bug2 on RISCV and although it couldn't be resolved,
> > it reported an error.
> >
> > Test is,
> > $ cat c.s
> > .section ".another.text", "x"
> > .fill 1, 4, 0
> > call fn2
> > call fn2
> > fn1:
> > nop
> >
> > .text
> > fn2:
> > .fill 4, 4, 0
> > call fn1
> >
> > $ ./gas/as-new -march=rv32ic_zicsr -mno-arch-attr -o c.o c.s
> > $ ./ld/ld-new c.o -o c -e 0 --section-start=.another.text=0x1000000
> > -Ttext 0x1100000
> > c.o: in function `fn2':
> > (.text+0x10): relocation truncated to fit: R_RISCV_JAL against `fn1'
> >
> > So, I'll find the reason why reloc_bits_pcrel20_s2() not reports the
> > overfollow error, and add error report which I think it is more
> > important than these two patches so that we can get feedback when we
> > link.
> >
>
> Now, I find that the sign-overflow-check has bug, include
> reloc_bits_pcrel20_s2(), reloc_bits_b26() and (others, I did not test
> them). And I think the "mask" should be
> bfd_signed_vma mask = ((bfd_signed_vma)0x1 << (howto->bitsize -
> (sign ? 1 : 0)) - 1;
>
> The test about reloc_bits_b26() is,
>
> $ cat d.s
> .section ".another.text", "x"
> b sym1
>
> .text
> nop
> sym1:
> nop
>
> $ ./gas/as-new -mabi=lp64d -o d.o d.s
> $ ./ld/ld-new d.o -o d -e 0 --section-start=.another.text=0x130000000
> -Ttext 0x120000000
>
> I'll do fix and send v2 after you test spec2006. Or other responses?
>
>
> Thanks!
>
>
> >
> > On 2023-07-13 11:13, mengqinggang wrote:
> > > Hi,
> > >
> > > I will apply this two patches to see if there are any errors in
> > > specc2006.
> > >
> > > 在 2023/7/12 下午4:08, Jinyang He 写道:
> > > > Hi,
> > > >
> > > >
> > > > I'm not fimiliar with the process of output binary. I just noted
> > > > that it calls lang_relax_sections() before calling
> > > > _bfd_elf_map_sections_to_segments() in ldelf_map_segments(). Thus I
> > > > think the condition, elf_seg_map (info->output_bfd) == NULL, always
> > > > be true when first time it calls loongarch_elf_relax_section().
> > > > Actually we can get right symbol value when relaxation, but ...
> > > >
> > > > You reminder me that the symbol value will be reduced due to
> > > > relaxation, which like the instruction pc is reduced. Then we cannot
> > > > get right symbol value if the symbol in relaxable section. And I
> > > > write a test,
> > > >
> > > > $ cat c.s
> > > > .section ".another.text", "x"
> > > > nop
> > > > la.local $a0, sym1
> > > > sym2:
> > > > nop
> > > >
> > > > .text
> > > > sym1:
> > > > nop
> > > > nop
> > > > nop
> > > > la.local $a0, sym2
> > > >
> > > > $ ./gas/as-new -mabi=lp64d -o c.o c.s
> > > > $ ./ld/ld-new c.o -o c -e 0
> > > > --section-start=.another.text=0x120000000 -Ttext 0x120200000
> > > >
> > > > Disassembly of section .another.text:
> > > >
> > > > 0000000120000000 <sym2-0x8>:
> > > > 120000000: 03400000 andi $zero, $zero, 0x0
> > > > 120000004: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
> > > >
> > > > 0000000120000008 <sym2>:
> > > > 120000008: 03400000 andi $zero, $zero, 0x0
> > > >
> > > > Disassembly of section .text:
> > > >
> > > > 0000000120200000 <sym1>:
> > > > 120200000: 03400000 andi $zero, $zero, 0x0
> > > > 120200004: 03400000 andi $zero, $zero, 0x0
> > > > 120200008: 03400000 andi $zero, $zero, 0x0
> > > > 12020000c: 18ffffe4 pcaddi $a0, 524287(0x7ffff)
> > > >
> > > > Due to the bug repaired by Patch2 and the bug reported by this test.
> > > > And I also did not see that condition in other archs. I'm not sure
> > > > that the relaxation spec2006 error is caused by that condition or
> > > > those bugs.
> > > >
> > > >
> > > > On 2023-07-12 14:35, mengqinggang wrote:
> > > > > Hi,
> > > > >
> > > > > Without this condition, the value of 'symbol - pc' may be
> > > > > differences between
> > > > > loongarch_elf_relax_section and loongarch_elf_relocate_section.
> > > > > And it will cause some spec2006 test errors.
> > > > >
> > > > > 在 2023/7/11 下午4:49, Jinyang He 写道:
> > > > > > When I tried "./ld-new sth." the relax does not work because the
> > > > > > elf_seg_map (info->output_bfd) was NULL in
> > > > > > loongarch_elf_relax_section.
> > > > > > There is no need for the segment info, we can get the address through
> > > > > > the section info. Thus, remove this condition.
> > > > > >
> > > > > > bfd/ChangeLog:
> > > > > >
> > > > > > * elfnn-loongarch.c (loongarch_elf_relax_section): Remove
> > > > > > elf_seg_map condition.
> > > > > >
> > > > > > ld/ChangeLog:
> > > > > >
> > > > > > * testsuite/ld-loongarch-elf/relax.exp: Update trigger method.
> > > > > > * testsuite/ld-loongarch-elf/relax.dd: New test.
> > > > > > ---
> > > > > > bfd/elfnn-loongarch.c | 1 -
> > > > > > ld/testsuite/ld-loongarch-elf/relax.dd | 6 +++++
> > > > > > ld/testsuite/ld-loongarch-elf/relax.exp | 35
> > > > > > +++++++------------------
> > > > > > 3 files changed, 16 insertions(+), 26 deletions(-)
> > > > > > create mode 100644 ld/testsuite/ld-loongarch-elf/relax.dd
> > > > > >
> > > > > > diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> > > > > > index d3d8419d8..01f349a24 100644
> > > > > > --- a/bfd/elfnn-loongarch.c
> > > > > > +++ b/bfd/elfnn-loongarch.c
> > > > > > @@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd,
> > > > > > asection *sec,
> > > > > > || sec->sec_flg0
> > > > > > || (sec->flags & SEC_RELOC) == 0
> > > > > > || sec->reloc_count == 0
> > > > > > - || elf_seg_map (info->output_bfd) == NULL
> > > > > > || (info->disable_target_specific_optimizations
> > > > > > && info->relax_pass == 0)
> > > > > > /* The exp_seg_relro_adjust is enum phase_enum (0x4),
> > > > > > diff --git a/ld/testsuite/ld-loongarch-elf/relax.dd
> > > > > > b/ld/testsuite/ld-loongarch-elf/relax.dd
> > > > > > new file mode 100644
> > > > > > index 000000000..e266beb8f
> > > > > > --- /dev/null
> > > > > > +++ b/ld/testsuite/ld-loongarch-elf/relax.dd
> > > > > > @@ -0,0 +1,6 @@
> > > > > > +#...
> > > > > > +.*pcaddi.*
> > > > > > +.*ld.w.*
> > > > > > +.*pcaddi.*
> > > > > > +.*ld.w.*
> > > > > > +#pass
> > > > > > diff --git a/ld/testsuite/ld-loongarch-elf/relax.exp
> > > > > > b/ld/testsuite/ld-loongarch-elf/relax.exp
> > > > > > index 7ff876d79..d2bb967c6 100644
> > > > > > --- a/ld/testsuite/ld-loongarch-elf/relax.exp
> > > > > > +++ b/ld/testsuite/ld-loongarch-elf/relax.exp
> > > > > > @@ -21,31 +21,6 @@
> > > > > > if [istarget loongarch64-*-*] {
> > > > > > - if [isbuild loongarch64-*-*] {
> > > > > > - set testname "loongarch relax build"
> > > > > > - set pre_builds [list \
> > > > > > - [list \
> > > > > > - "$testname" \
> > > > > > - "" \
> > > > > > - "" \
> > > > > > - {relax.s} \
> > > > > > - {} \
> > > > > > - "relax" \
> > > > > > - ] \
> > > > > > - ]
> > > > > > -
> > > > > > - run_cc_link_tests $pre_builds
> > > > > > -
> > > > > > - if [file exist "tmpdir/relax"] {
> > > > > > - set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
> > > > > > - if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
> > > > > > - pass "loongarch relax"
> > > > > > - } {
> > > > > > - fail "loongarch relax"
> > > > > > - }
> > > > > > - }
> > > > > > - }
> > > > > > -
> > > > > > run_ld_link_tests \
> > > > > > [list \
> > > > > > [list \
> > > > > > @@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
> > > > > > ] \
> > > > > > "relax-align" \
> > > > > > ] \
> > > > > > + [list \
> > > > > > + "relax" \
> > > > > > + "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
> > > > > > + "" \
> > > > > > + {relax.s} \
> > > > > > + [list \
> > > > > > + [list objdump -d relax.dd] \
> > > > > > + ] \
> > > > > > + "relax" \
> > > > > > + ] \
> > > > > > ]
> > > > > > set objdump_flags "-s -j .data"
>
On Thu, 2023-10-05 at 18:09 +0800, Xi Ruoyao via Binutils wrote:
> Hi Jinyang,
>
> Any progress on this? During my attempt trying to balance relaxation
> and scheduling better in GCC I found the elf_seg_map condition is
> preventing relaxation on *every* shared library (when we are linking a
> shared library the elf_seg_map of .text is NULL).
>
> On a modern system most code paths are in shared libraries, so it's
> really bad not to perform the relaxation on them. Esp. now we are
> disabling explicit relocs for relaxation, so if we don't relax shared
> libraries we are likely regressing the overall performance of the
> system.
Phew. It's not only a performance issue. It's actually a *correctness*
issue because the condition also causes R_LARCH_ALIGN skipped, so some
programs depending on code alignment (for e.g. duff-device-like code
using pcaddi for eg) will be broken.
Generally skipping the entire relaxation pass for any section containing
R_LARCH_ALIGN is wrong (as we've discussed in
https://github.com/llvm/llvm-project/pull/67424). Even if we don't
really relax a thing we still *must* process R_LARCH_ALIGN.
> Sorry for disturbing you during the national holiday, feel free to defer
> the reply until the holiday ends!
Hi,
I will send a patch in the next few days to solve this issue and other
issues
related to relaxation.
在 2023/10/5 下午7:19, Xi Ruoyao 写道:
> On Thu, 2023-10-05 at 18:09 +0800, Xi Ruoyao via Binutils wrote:
>> Hi Jinyang,
>>
>> Any progress on this? During my attempt trying to balance relaxation
>> and scheduling better in GCC I found the elf_seg_map condition is
>> preventing relaxation on *every* shared library (when we are linking a
>> shared library the elf_seg_map of .text is NULL).
>>
>> On a modern system most code paths are in shared libraries, so it's
>> really bad not to perform the relaxation on them. Esp. now we are
>> disabling explicit relocs for relaxation, so if we don't relax shared
>> libraries we are likely regressing the overall performance of the
>> system.
> Phew. It's not only a performance issue. It's actually a *correctness*
> issue because the condition also causes R_LARCH_ALIGN skipped, so some
> programs depending on code alignment (for e.g. duff-device-like code
> using pcaddi for eg) will be broken.
>
> Generally skipping the entire relaxation pass for any section containing
> R_LARCH_ALIGN is wrong (as we've discussed in
> https://github.com/llvm/llvm-project/pull/67424). Even if we don't
> really relax a thing we still *must* process R_LARCH_ALIGN.
>
>> Sorry for disturbing you during the national holiday, feel free to defer
>> the reply until the holiday ends!
On Sat, 2023-10-07 at 09:23 +0800, mengqinggang wrote:
>
> Hi,
>
> I will send a patch in the next few days to solve this issue and other
> issues
> related to relaxation.
You can refer to a related bug report,
https://sourceware.org/bugzilla/show_bug.cgi?id=30944, in the commit
message.
Thanks!
@@ -3845,7 +3845,6 @@ loongarch_elf_relax_section (bfd *abfd, asection *sec,
|| sec->sec_flg0
|| (sec->flags & SEC_RELOC) == 0
|| sec->reloc_count == 0
- || elf_seg_map (info->output_bfd) == NULL
|| (info->disable_target_specific_optimizations
&& info->relax_pass == 0)
/* The exp_seg_relro_adjust is enum phase_enum (0x4),
new file mode 100644
@@ -0,0 +1,6 @@
+#...
+.*pcaddi.*
+.*ld.w.*
+.*pcaddi.*
+.*ld.w.*
+#pass
@@ -21,31 +21,6 @@
if [istarget loongarch64-*-*] {
- if [isbuild loongarch64-*-*] {
- set testname "loongarch relax build"
- set pre_builds [list \
- [list \
- "$testname" \
- "" \
- "" \
- {relax.s} \
- {} \
- "relax" \
- ] \
- ]
-
- run_cc_link_tests $pre_builds
-
- if [file exist "tmpdir/relax"] {
- set objdump_output [run_host_cmd "objdump" "-d tmpdir/relax"]
- if { [ regexp ".*pcaddi.*pcaddi.*" $objdump_output] } {
- pass "loongarch relax"
- } {
- fail "loongarch relax"
- }
- }
- }
-
run_ld_link_tests \
[list \
[list \
@@ -58,6 +33,16 @@ if [istarget loongarch64-*-*] {
] \
"relax-align" \
] \
+ [list \
+ "relax" \
+ "-e 0 -Ttext 0x10000 -Tdata 0x20000" "" \
+ "" \
+ {relax.s} \
+ [list \
+ [list objdump -d relax.dd] \
+ ] \
+ "relax" \
+ ] \
]
set objdump_flags "-s -j .data"