* [PATCH V2] GAS fix alignment for aarch64-pe
@ 2022-11-14 17:24 Zac Walker
2022-11-15 10:50 ` Nick Clifton
0 siblings, 1 reply; 9+ messages in thread
From: Zac Walker @ 2022-11-14 17:24 UTC (permalink / raw)
To: binutils
[-- Attachment #1: Type: text/plain, Size: 933 bytes --]
For the aarch64-pe target various values of '.align' causes writing of COFF files to fail.
This version of the patch differs from the first by using the write.c implementation of SUB_SEGMENT_ALIGN. This means there was no need to modify md_section_align or aarch64_handle_align. The change is specific to the aarch64-pe target.
In my testing generated aligned code was correctly padded with nop.
Thanks,
Zac
---
gas/config/tc-aarch64.h | 3 +++
1 file changed, 3 insertions(+)
diff --git a/gas/config/tc-aarch64.h b/gas/config/tc-aarch64.h
index 2d514ff610..0ea73021f2 100644
--- a/gas/config/tc-aarch64.h
+++ b/gas/config/tc-aarch64.h
@@ -191,7 +191,10 @@ struct aarch64_frag_type
goto LABEL; \
}
+/* COFF sub section alignment calculated using the write.c implementation. */
+#ifndef OBJ_COFF
#define SUB_SEGMENT_ALIGN(SEG, FRCHAIN) 0
+#endif
#define DWARF2_LINE_MIN_INSN_LENGTH 4
--
2.25.1
[-- Attachment #2: 0001-GAS-fix-alignment-for-aarch64-pe.patch --]
[-- Type: text/x-diff, Size: 719 bytes --]
From 392e919f35ad790b21185ab1ce387bb0694cddd9 Mon Sep 17 00:00:00 2001
From: Zac Walker <zac.walker@linaro.org>
Date: Mon, 14 Nov 2022 17:48:58 +0100
Subject: [PATCH] GAS fix alignment for aarch64-pe
---
gas/config/tc-aarch64.h | 3 +++
1 file changed, 3 insertions(+)
diff --git a/gas/config/tc-aarch64.h b/gas/config/tc-aarch64.h
index 2d514ff610..0ea73021f2 100644
--- a/gas/config/tc-aarch64.h
+++ b/gas/config/tc-aarch64.h
@@ -191,7 +191,10 @@ struct aarch64_frag_type
goto LABEL; \
}
+/* COFF sub section alignment calculated using the write.c implementation. */
+#ifndef OBJ_COFF
#define SUB_SEGMENT_ALIGN(SEG, FRCHAIN) 0
+#endif
#define DWARF2_LINE_MIN_INSN_LENGTH 4
--
2.25.1
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH V2] GAS fix alignment for aarch64-pe
2022-11-14 17:24 [PATCH V2] GAS fix alignment for aarch64-pe Zac Walker
@ 2022-11-15 10:50 ` Nick Clifton
2022-11-15 11:23 ` Zac Walker
0 siblings, 1 reply; 9+ messages in thread
From: Nick Clifton @ 2022-11-15 10:50 UTC (permalink / raw)
To: Zac Walker, binutils
Hi Zac,
> For the aarch64-pe target various values of '.align' causes writing of COFF files to fail.
>
> This version of the patch differs from the first by using the write.c implementation of SUB_SEGMENT_ALIGN. This means there was no need to modify md_section_align or
> aarch64_handle_align. The change is specific to the aarch64-pe target.
>
> In my testing generated aligned code was correctly padded with nop.
Patch approved - please apply.
Cheers
Nick
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH V2] GAS fix alignment for aarch64-pe
2022-11-15 10:50 ` Nick Clifton
@ 2022-11-15 11:23 ` Zac Walker
2022-11-15 21:53 ` aarch64-pe can't fill 16 bytes in section .text Alan Modra
0 siblings, 1 reply; 9+ messages in thread
From: Zac Walker @ 2022-11-15 11:23 UTC (permalink / raw)
To: Nick Clifton, binutils
Thanks Nick.
On 11/15/22 11:50, Nick Clifton wrote:
> Hi Zac,
>
>> For the aarch64-pe target various values of '.align' causes writing of COFF files to fail.
>>
>> This version of the patch differs from the first by using the write.c implementation of SUB_SEGMENT_ALIGN. This means there was no need to modify md_section_align or aarch64_handle_align. The change is specific to the aarch64-pe target.
>>
>> In my testing generated aligned code was correctly padded with nop.
>
> Patch approved - please apply.
>
> Cheers
> Nick
>
^ permalink raw reply [flat|nested] 9+ messages in thread
* aarch64-pe can't fill 16 bytes in section .text
2022-11-15 11:23 ` Zac Walker
@ 2022-11-15 21:53 ` Alan Modra
2022-11-16 17:03 ` Zac Walker
0 siblings, 1 reply; 9+ messages in thread
From: Alan Modra @ 2022-11-15 21:53 UTC (permalink / raw)
To: Zac Walker; +Cc: Nick Clifton, binutils
I'm going to fix the underlying coff_frob_section problem too.
Without commit b66e671854, this:
.p2align 4
nop
.p2align 3
nop
results in an error when coff_frob_section attempts to pad out the
section to a 16-byte boundary. Due to miscalculating the pad pattern
repeat count, write.c:write_contents attempts to shove 16 bytes of
padding into the remaining 4 bytes of the .text section.
* config/obj-coff.c (coff_frob_section): Correct fill count.
Don't pad after errors.
diff --git a/gas/config/obj-coff.c b/gas/config/obj-coff.c
index 98c39e43907..9be697fb62e 100644
--- a/gas/config/obj-coff.c
+++ b/gas/config/obj-coff.c
@@ -1725,7 +1725,8 @@ coff_frob_section (segT sec)
bfd_vma align_power = (bfd_vma) sec->alignment_power + OCTETS_PER_BYTE_POWER;
bfd_vma mask = ((bfd_vma) 1 << align_power) - 1;
- if (size & mask)
+ if (!do_not_pad_sections_to_alignment
+ && (size & mask) != 0)
{
bfd_vma new_size;
fragS *last;
@@ -1740,7 +1741,10 @@ coff_frob_section (segT sec)
while (fragp->fr_next != last)
fragp = fragp->fr_next;
last->fr_address = size;
- fragp->fr_offset += new_size - size;
+ if ((new_size - size) % fragp->fr_var == 0)
+ fragp->fr_offset += (new_size - size) / fragp->fr_var;
+ else
+ abort ();
}
#endif
--
Alan Modra
Australia Development Lab, IBM
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: aarch64-pe can't fill 16 bytes in section .text
2022-11-15 21:53 ` aarch64-pe can't fill 16 bytes in section .text Alan Modra
@ 2022-11-16 17:03 ` Zac Walker
2022-11-17 3:01 ` Alan Modra
0 siblings, 1 reply; 9+ messages in thread
From: Zac Walker @ 2022-11-16 17:03 UTC (permalink / raw)
To: Alan Modra; +Cc: Nick Clifton, binutils
Thanks Alan,
Do you think my SUB_SEGMENT_ALIGN patch is still needed. I didn't merge it yet because I wanted to do more testing with your fix.
Zac
---
gas/config/tc-aarch64.h | 3 +++
1 file changed, 3 insertions(+)
diff --git a/gas/config/tc-aarch64.h b/gas/config/tc-aarch64.h
index 2d514ff610..0ea73021f2 100644
--- a/gas/config/tc-aarch64.h
+++ b/gas/config/tc-aarch64.h
@@ -191,7 +191,10 @@ struct aarch64_frag_type
goto LABEL; \
}
+/* COFF sub section alignment calculated using the write.c implementation. */
+#ifndef OBJ_COFF
#define SUB_SEGMENT_ALIGN(SEG, FRCHAIN) 0
+#endif
#define DWARF2_LINE_MIN_INSN_LENGTH 4
--
2.25.1
On 11/15/22 22:53, Alan Modra wrote:
> I'm going to fix the underlying coff_frob_section problem too.
> Without commit b66e671854, this:
> .p2align 4
> nop
> .p2align 3
> nop
> results in an error when coff_frob_section attempts to pad out the
> section to a 16-byte boundary. Due to miscalculating the pad pattern
> repeat count, write.c:write_contents attempts to shove 16 bytes of
> padding into the remaining 4 bytes of the .text section.
>
> * config/obj-coff.c (coff_frob_section): Correct fill count.
> Don't pad after errors.
>
> diff --git a/gas/config/obj-coff.c b/gas/config/obj-coff.c
> index 98c39e43907..9be697fb62e 100644
> --- a/gas/config/obj-coff.c
> +++ b/gas/config/obj-coff.c
> @@ -1725,7 +1725,8 @@ coff_frob_section (segT sec)
> bfd_vma align_power = (bfd_vma) sec->alignment_power + OCTETS_PER_BYTE_POWER;
> bfd_vma mask = ((bfd_vma) 1 << align_power) - 1;
>
> - if (size & mask)
> + if (!do_not_pad_sections_to_alignment
> + && (size & mask) != 0)
> {
> bfd_vma new_size;
> fragS *last;
> @@ -1740,7 +1741,10 @@ coff_frob_section (segT sec)
> while (fragp->fr_next != last)
> fragp = fragp->fr_next;
> last->fr_address = size;
> - fragp->fr_offset += new_size - size;
> + if ((new_size - size) % fragp->fr_var == 0)
> + fragp->fr_offset += (new_size - size) / fragp->fr_var;
> + else
> + abort ();
> }
> #endif
>
>
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: aarch64-pe can't fill 16 bytes in section .text
2022-11-16 17:03 ` Zac Walker
@ 2022-11-17 3:01 ` Alan Modra
2022-11-18 8:52 ` zac.walker
0 siblings, 1 reply; 9+ messages in thread
From: Alan Modra @ 2022-11-17 3:01 UTC (permalink / raw)
To: Zac Walker; +Cc: Nick Clifton, binutils
On Wed, Nov 16, 2022 at 06:03:56PM +0100, Zac Walker wrote:
> Thanks Alan,
>
> Do you think my SUB_SEGMENT_ALIGN patch is still needed. I didn't merge it yet because I wanted to do more testing with your fix.
Huh, I thought it had already gone in, but it was just in my local
tree.
Your patch isn't needed with my fix to obj-coff.c, but it wouldn't be
a bad idea to apply it anyway. We have a few too many places doing
section padding. It would be nice to get rid of the alignment code
in coff_frob_section.
--
Alan Modra
Australia Development Lab, IBM
^ permalink raw reply [flat|nested] 9+ messages in thread
* RE: aarch64-pe can't fill 16 bytes in section .text
2022-11-17 3:01 ` Alan Modra
@ 2022-11-18 8:52 ` zac.walker
2022-11-18 9:08 ` Jan Beulich
0 siblings, 1 reply; 9+ messages in thread
From: zac.walker @ 2022-11-18 8:52 UTC (permalink / raw)
To: 'Alan Modra'; +Cc: 'Nick Clifton', binutils
Patch merged just now. (Was a holiday here yesterday)
I am still investigating another alignment problem so might be another follow up patch. This code from OpenBLAS:
data_ar:
.word 0x3e44fae6
data_ai:
.word 0x3d320fa2
start:
ldr s20, data_ar
ldr s21, data_ai
Produces:
zscal-min.s:2: Warning: value 0x3e44fae6 truncated to 0xfae6
zscal-min.s:4: Warning: value 0x3d320fa2 truncated to 0xfa2
zscal-min.s:7: Error: pc-relative load offset not word aligned
Works ok with clang. I am not sure if .word data should be aligned by default in gas or this is just a difference in behaviour.
Thanks,
Zac
-----Original Message-----
From: Alan Modra <amodra@gmail.com>
Sent: 17 November 2022 04:02
To: Zac Walker <zac.walker@linaro.org>
Cc: Nick Clifton <nickc@redhat.com>; binutils@sourceware.org
Subject: Re: aarch64-pe can't fill 16 bytes in section .text
On Wed, Nov 16, 2022 at 06:03:56PM +0100, Zac Walker wrote:
> Thanks Alan,
>
> Do you think my SUB_SEGMENT_ALIGN patch is still needed. I didn't merge it yet because I wanted to do more testing with your fix.
Huh, I thought it had already gone in, but it was just in my local tree.
Your patch isn't needed with my fix to obj-coff.c, but it wouldn't be a bad idea to apply it anyway. We have a few too many places
doing section padding. It would be nice to get rid of the alignment code in coff_frob_section.
--
Alan Modra
Australia Development Lab, IBM
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: aarch64-pe can't fill 16 bytes in section .text
2022-11-18 8:52 ` zac.walker
@ 2022-11-18 9:08 ` Jan Beulich
2022-11-18 9:35 ` zac.walker
0 siblings, 1 reply; 9+ messages in thread
From: Jan Beulich @ 2022-11-18 9:08 UTC (permalink / raw)
To: zac.walker; +Cc: 'Nick Clifton', binutils, 'Alan Modra'
On 18.11.2022 09:52, Zac Walker via Binutils wrote:
> Patch merged just now. (Was a holiday here yesterday)
>
> I am still investigating another alignment problem so might be another follow up patch. This code from OpenBLAS:
>
> data_ar:
> .word 0x3e44fae6
> data_ai:
> .word 0x3d320fa2
> start:
> ldr s20, data_ar
> ldr s21, data_ai
>
> Produces:
>
> zscal-min.s:2: Warning: value 0x3e44fae6 truncated to 0xfae6
> zscal-min.s:4: Warning: value 0x3d320fa2 truncated to 0xfa2
> zscal-min.s:7: Error: pc-relative load offset not word aligned
>
> Works ok with clang. I am not sure if .word data should be aligned by default in gas or this is just a difference in behaviour.
.word is overridden by tc-aarch64.c only for ELF. So in your COFF (aiui)
case .word emits just two bytes instead of the expected four. Hence also
the warnings, not just the error.
Jan
^ permalink raw reply [flat|nested] 9+ messages in thread
* RE: aarch64-pe can't fill 16 bytes in section .text
2022-11-18 9:08 ` Jan Beulich
@ 2022-11-18 9:35 ` zac.walker
0 siblings, 0 replies; 9+ messages in thread
From: zac.walker @ 2022-11-18 9:35 UTC (permalink / raw)
To: 'Jan Beulich'
Cc: 'Nick Clifton', binutils, 'Alan Modra'
Thanks for the help Jan. Adding a similar override to "tc-arm.c" fixes it:
#ifdef OBJ_ELF
...
#else
{ "word", cons, 4},
#endif
I will do some testing on other pseudo types in case I am missing more.
Zac
-----Original Message-----
From: Jan Beulich <jbeulich@suse.com>
Sent: 18 November 2022 10:09
To: zac.walker@linaro.org
Cc: 'Nick Clifton' <nickc@redhat.com>; binutils@sourceware.org; 'Alan Modra' <amodra@gmail.com>
Subject: Re: aarch64-pe can't fill 16 bytes in section .text
On 18.11.2022 09:52, Zac Walker via Binutils wrote:
> Patch merged just now. (Was a holiday here yesterday)
>
> I am still investigating another alignment problem so might be another follow up patch. This code from OpenBLAS:
>
> data_ar:
> .word 0x3e44fae6
> data_ai:
> .word 0x3d320fa2
> start:
> ldr s20, data_ar
> ldr s21, data_ai
>
> Produces:
>
> zscal-min.s:2: Warning: value 0x3e44fae6 truncated to 0xfae6
> zscal-min.s:4: Warning: value 0x3d320fa2 truncated to 0xfa2
> zscal-min.s:7: Error: pc-relative load offset not word aligned
>
> Works ok with clang. I am not sure if .word data should be aligned by default in gas or this is just a difference in behaviour.
.word is overridden by tc-aarch64.c only for ELF. So in your COFF (aiui) case .word emits just two bytes instead of the expected four. Hence also the warnings, not just the error.
Jan
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2022-11-18 9:35 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-14 17:24 [PATCH V2] GAS fix alignment for aarch64-pe Zac Walker
2022-11-15 10:50 ` Nick Clifton
2022-11-15 11:23 ` Zac Walker
2022-11-15 21:53 ` aarch64-pe can't fill 16 bytes in section .text Alan Modra
2022-11-16 17:03 ` Zac Walker
2022-11-17 3:01 ` Alan Modra
2022-11-18 8:52 ` zac.walker
2022-11-18 9:08 ` Jan Beulich
2022-11-18 9:35 ` zac.walker
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).