diff mbox series

GAS fix section alignment for aarch64-pe

Message ID CACVBkZ+kM2xcpwk5zHQ4bqDV7dbPjZ6cMU9tR-h62+tDTgi3Tw@mail.gmail.com
State Unresolved
Headers
Series GAS fix section alignment for aarch64-pe |

Checks

Context Check Description
snail/binutils-gdb-check warning Git am fail log

Commit Message

Zac Walker Nov. 7, 2022, 5:53 p.m. UTC 
  For the aarch64-pe target, section alignment for various values of
'.align'  causes writing of coff files to fail.
This patch performs aligning in md_section_align as opposed to the
original approach in aarch64_handle_align. The change is specific to
the aarch64-pe target.

Thanks,

Zac

---
 gas/config/tc-aarch64.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

    of an rs_align_code fragment.
@@ -8353,7 +8362,11 @@ aarch64_handle_align (fragS * fragP)

   if (noop_size)
     memcpy (p, aarch64_noop, noop_size);
+
+/* COFF sections may have unaligned fr_var and then be aligned in
md_section_align */
+#ifndef OBJ_COFF
   fragP->fr_var = noop_size;
+#endif
 }

 /* Perform target specific initialisation of a frag.

Comments

Alan Modra Nov. 7, 2022, 10:58 p.m. UTC | #1 
On Mon, Nov 07, 2022 at 06:53:35PM +0100, Zac Walker via Binutils wrote:
> For the aarch64-pe target, section alignment for various values of
> '.align'  causes writing of coff files to fail.
> This patch performs aligning in md_section_align as opposed to the
> original approach in aarch64_handle_align. The change is specific to
> the aarch64-pe target.

This can't be correct since HANDLE_ALIGN and md_section_align perform
quite different functions.  I don't think you should be changing
aarch64_handle_align.  For example:

 .text
 .p2align 4
 nop
 .p2align 3
 nop

The second p2align needs to continue emitting nops.

(Also, please fix your email client to *not* format lines.  Wrapping
lines and changing white space breaks patches.)
Frager, Neal via Binutils Nov. 8, 2022, 6:40 p.m. UTC | #2 
Hi Alan, Thanks for the feedback. 

You are correct - looks like I was over-focused on getting the write method working for various failing tests. Your example is
useful to show missing noops. From what I can see the changes to md_section_align seem necessary for COFF but unrelated to the write
issue. (Assuming the fix is HANDLE_ALIGN specific.) 

I will re-work the patch and do more testing. Might take a few days.

Also, I will double check my email settings. Every email client I have tried (except git send-email) seems to break formatting one
way or another. 

Cheers,

Zac


-----Original Message-----
From: Alan Modra <amodra@gmail.com> 
Sent: 07 November 2022 23:58
To: Zac Walker <zac.walker@linaro.org>
Cc: binutils@sourceware.org
Subject: Re: [PATCH] GAS fix section alignment for aarch64-pe

On Mon, Nov 07, 2022 at 06:53:35PM +0100, Zac Walker via Binutils wrote:
> For the aarch64-pe target, section alignment for various values of 
> '.align'  causes writing of coff files to fail.
> This patch performs aligning in md_section_align as opposed to the 
> original approach in aarch64_handle_align. The change is specific to 
> the aarch64-pe target.

This can't be correct since HANDLE_ALIGN and md_section_align perform quite different functions.  I don't think you should be
changing aarch64_handle_align.  For example:

 .text
 .p2align 4
 nop
 .p2align 3
 nop

The second p2align needs to continue emitting nops.

(Also, please fix your email client to *not* format lines.  Wrapping lines and changing white space breaks patches.)

--
Alan Modra
Australia Development Lab, IBM
diff mbox series

Patch

diff --git a/gas/config/tc-aarch64.c b/gas/config/tc-aarch64.c
index f6fa158583..44ad510209 100644
--- a/gas/config/tc-aarch64.c
+++ b/gas/config/tc-aarch64.c
@@ -8290,13 +8290,22 @@  md_estimate_size_before_relax (fragS * fragp,
segT segtype ATTRIBUTE_UNUSED)
   return 4;
 }

-/* Round up a section size to the appropriate boundary. */
-
+/* Round up a COFF section size not needed for ELF */
+#ifdef OBJ_COFF
+valueT
+md_section_align (segT segment, valueT size)
+{
+  int align = bfd_section_alignment (segment);
+  valueT mask = ((valueT) 1 << align) - 1;
+  return (size + mask) & ~mask;
+}
+#else /* !OBJ_COFF */
 valueT
 md_section_align (segT segment ATTRIBUTE_UNUSED, valueT size)
 {
   return size;
 }
+#endif /* !OBJ_COFF */

 /* This is called from HANDLE_ALIGN in write.c. Fill in the contents