patch for i386-pe section header [Re: section bug?]

patch for i386-pe section header [Re: section bug?]

[Mumit Khan]

SONE Takeshi <ts1@cma.co.jp> writes:
> In Cygwin32 and Mingw32 environment with binutils 2.9.4, this code
> crashes:
> 
> int a __attribute__((section("mysec"))) = 0;
> main()
> {
> 	a = 1;
> }
> 
> gcc looks doing right thing:
> 
> .section	mysec,"w"
> 	.align 4
> _a:
> 	.long 0
> 
> I run dumpbin.exe (from MS) on the output and it says that mysec in the
> .o file is marked Read Only, and in the .exe file it is marked Code and
> Read Only.

The following should fix this bug. 

Sat Jun 26 18:25:30 1999  Mumit Khan  <khan@xraylith.wisc.edu>

	* peicode.h (coff_swap_scnhdr_out): Mark user-defined writable 
	sections as writable.

Index: bfd/peicode.h
===================================================================
RCS file: /cvs/binutils/binutils/bfd/peicode.h,v
retrieving revision 1.3
diff -u -3 -p -r1.3 peicode.h
--- bfd/peicode.h	1999/05/25 11:37:40	1.3
+++ bfd/peicode.h	1999/06/26 23:22:14
@@ -1234,6 +1234,8 @@ coff_swap_scnhdr_out (abfd, in, out)
 		| IMAGE_SCN_MEM_SHARED | IMAGE_SCN_MEM_READ);
     else if (strcmp (scnhdr_int->s_name, ".rsrc")  == 0)
       flags |= IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_SHARED;
+    else if (! (flags & SEC_READONLY))
+      flags |= IMAGE_SCN_MEM_WRITE;
     else
       flags |= IMAGE_SCN_MEM_READ;
 
Regards,
Mumit

Re: patch for i386-pe section header [Re: section bug?]

[Mumit Khan]

Ian Lance Taylor <ian@zembu.com> writes:
> 
> In the second new line, shouldn't that be
>     flags |= IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_WRITE
> ?
> 
> I would almost just check that in, but I thought I'd better check to
> see if I was missing something.
> 

And I also forgot to diff the whole tree. Damn. Now it's tested on both
NT and W9x.

bfd/ChangeLog:
Sat Jun 26 21:09:44 1999  Mumit Khan  <khan@xraylith.wisc.edu>

	* peicode.h (coff_swap_scnhdr_out): Mark user-defined writable 
	sections as writable.

gas/ChangeLog:
Sat Jun 26 21:09:44 1999  Mumit Khan  <khan@xraylith.wisc.edu>
	* config/obj-coff.c (obj_coff_section): Mark writable sections
	as data.

Index: bfd/peicode.h
===================================================================
RCS file: /cvs/binutils/binutils/bfd/peicode.h,v
retrieving revision 1.3
diff -u -3 -p -r1.3 peicode.h
--- bfd/peicode.h	1999/05/25 11:37:40	1.3
+++ bfd/peicode.h	1999/06/27 02:08:58
@@ -1235,7 +1235,11 @@ coff_swap_scnhdr_out (abfd, in, out)
     else if (strcmp (scnhdr_int->s_name, ".rsrc")  == 0)
       flags |= IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_SHARED;
     else
-      flags |= IMAGE_SCN_MEM_READ;
+      {
+	flags |= IMAGE_SCN_MEM_READ;
+	if (! (flags & SEC_READONLY))
+	  flags |= IMAGE_SCN_MEM_WRITE;
+      }
 
     bfd_h_put_32(abfd, flags, (bfd_byte *) scnhdr_ext->s_flags);
   }
Index: gas/config/obj-coff.c
===================================================================
RCS file: /cvs/binutils/binutils/gas/config/obj-coff.c,v
retrieving revision 1.5
diff -u -3 -p -r1.5 obj-coff.c
--- gas/config/obj-coff.c	1999/06/19 14:04:44	1.5
+++ gas/config/obj-coff.c	1999/06/27 02:08:59
@@ -1247,7 +1247,7 @@ obj_coff_section (ignore)
 		case 'b': flags |= SEC_ALLOC; flags &=~ SEC_LOAD; break;
 		case 'n': flags &=~ SEC_LOAD; break;
 		case 'd':
-		case 'w': flags &=~ SEC_READONLY; break;
+		case 'w': flags |= SEC_DATA; flags &=~ SEC_READONLY; break;
 		case 'x': flags |= SEC_CODE; break;
 		case 'r': flags |= SEC_READONLY; break;
 
Regards,
Mumit

section bug?

[SONE Takeshi]

In Cygwin32 and Mingw32 environment with binutils 2.9.4, this code
crashes:

int a __attribute__((section("mysec"))) = 0;
main()
{
	a = 1;
}

gcc looks doing right thing:

.section	mysec,"w"
	.align 4
_a:
	.long 0

I run dumpbin.exe (from MS) on the output and it says that mysec in the
.o file is marked Read Only, and in the .exe file it is marked Code and
Read Only.

--
SONE Takeshi                                      ^[$B$=$M^[(B ^[$B$?$1$7^[(B
mailto:ts1@cma.co.jp                    Office Craftsman Arts
http://www.cma.co.jp/~ts1/

Re: patch for i386-pe section header [Re: section bug?]

[Ian Lance Taylor]

   Date: Sat, 26 Jun 1999 18:31:14 -0500
   From: Mumit Khan <khan@xraylith.wisc.EDU>

   The following should fix this bug. 

   Sat Jun 26 18:25:30 1999  Mumit Khan  <khan@xraylith.wisc.edu>

	   * peicode.h (coff_swap_scnhdr_out): Mark user-defined writable 
	   sections as writable.

   Index: bfd/peicode.h
   ===================================================================
   RCS file: /cvs/binutils/binutils/bfd/peicode.h,v
   retrieving revision 1.3
   diff -u -3 -p -r1.3 peicode.h
   --- bfd/peicode.h	1999/05/25 11:37:40	1.3
   +++ bfd/peicode.h	1999/06/26 23:22:14
   @@ -1234,6 +1234,8 @@ coff_swap_scnhdr_out (abfd, in, out)
		   | IMAGE_SCN_MEM_SHARED | IMAGE_SCN_MEM_READ);
	else if (strcmp (scnhdr_int->s_name, ".rsrc")  == 0)
	  flags |= IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_SHARED;
   +    else if (! (flags & SEC_READONLY))
   +      flags |= IMAGE_SCN_MEM_WRITE;
	else
	  flags |= IMAGE_SCN_MEM_READ;

In the second new line, shouldn't that be
    flags |= IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_WRITE
?

I would almost just check that in, but I thought I'd better check to
see if I was missing something.

Ian

Re: patch for i386-pe section header [Re: section bug?]

[Mumit Khan]

On 26 Jun 1999, Ian Lance Taylor wrote:

>
> In the second new line, shouldn't that be
>     flags |= IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_WRITE
> ?
>
> I would almost just check that in, but I thought I'd better check to
> see if I was missing something.

For correctness yes, but never noticed it since NT doesn't seem to care. 
Btw, I'm just about to fix a few other section attribute things in the 
same routine, so let's wait till I send an update. It'll also give me 
time to run the testsuite + test W9x, where something always goes wrong.

Regards,
Mumit