[PATCH] Reserve a few .dynamic entries

[PATCH] Reserve a few .dynamic entries

[Jakub Jelinek]

Hi!

As .dynamic section is usually located in between sections which may have
relocated relocations against, it is not possible after final link to insert
new .dynamic entries which e.g. prelinking needs to do.

The following patch inserts 6 DT_NULL ElfW(Dyn) entries at the end of
.dynamic (iff .dynamic section is created), so that post-link utilities can
just put the tags they need into those slots and don't have to copy the
whole .dynamic somewhere else (which would require new PT_LOAD segment
usually). It wastes 48 resp. 96 bytes if no prelinking or any other
post-link optimization is done.

Do you agree with this? Can I commit it?

2001-06-06  Jakub Jelinek  <jakub@redhat.com>

	* scripttempl/elf.sc (DYNAMIC): Reserve 6 ElfW(Dyn) entries for
	post-linking tools.

--- ld/scripttempl/elf.sc.jj	Sat Sep  2 22:43:22 2000
+++ ld/scripttempl/elf.sc	Wed Jun  6 17:59:17 2001
@@ -66,7 +66,7 @@ test -z "${ALIGNMENT}" && ALIGNMENT="${E
 test "$LD_FLAG" = "N" && DATA_ADDR=.
 INTERP=".interp   ${RELOCATING-0} : { *(.interp) 	}"
 PLT=".plt    ${RELOCATING-0} : { *(.plt)	}"
-DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) }"
+DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) ${RELOCATING+FILL(0) . += 6 * ${ELFSIZE} / 4; } }"
 RODATA=".rodata ${RELOCATING-0} : { *(.rodata) ${RELOCATING+*(.rodata.*)} ${RELOCATING+*(.gnu.linkonce.r.*)} }"
 SBSS2=".sbss2 ${RELOCATING-0} : { *(.sbss2) ${RELOCATING+*(.sbss2.*)} ${RELOCATING+*(.gnu.linkonce.sb2.*)} }"
 SDATA2=".sdata2 ${RELOCATING-0} : { *(.sdata2) ${RELOCATING+*(.sdata2.*)} ${RELOCATING+*(.gnu.linkonce.s2.*)} }"

	Jakub

Re: [PATCH] Reserve a few .dynamic entries

[Hans-Peter Nilsson]

> Date: Wed, 6 Jun 2001 18:13:38 +0200
> From: Jakub Jelinek <jakub@redhat.com>

> +DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) ${RELOCATING+FILL(0) . += 6 * ${ELFSIZE} / 4; } }"

I think there should be a shell-variable (perhaps default on)
that a ld emulation can (un)set, and which should be tested when
adding that padding.  And a comment.

brgds, H-P

[PATCH] Reserve a few .dynamic entries (take 2)

[Jakub Jelinek]

On Wed, Jun 06, 2001 at 07:15:40PM +0200, Hans-Peter Nilsson wrote:
> > Date: Wed, 6 Jun 2001 18:13:38 +0200
> > From: Jakub Jelinek <jakub@redhat.com>
> 
> > +DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) ${RELOCATING+FILL(0) . += 6 * ${ELFSIZE} / 4; } }"
> 
> I think there should be a shell-variable (perhaps default on)
> that a ld emulation can (un)set, and which should be tested when
> adding that padding.  And a comment.

Ok, is this better (ie. can I commit it)?
ld emulations which don't want the reserved entries can just set
DYNAMIC_PAD=" "

2001-06-08  Jakub Jelinek  <jakub@redhat.com>

	* scripttempl/elf.sc (DYNAMIC_PAD): New variable.
	(DYNAMIC): Use it to reserve few dynamic entries for post-linking
	tools.

--- ld/scripttempl/elf.sc.jj	Sat Sep  2 22:43:22 2000
+++ ld/scripttempl/elf.sc	Fri Jun  8 20:07:12 2001
@@ -32,6 +32,8 @@
 # 	combination of .init sections.
 #	FINI_START, FINI_END - statements just before and just after
 # 	combination of .fini sections.
+#	DYNAMIC_PAD - script to insert pad at end of .dynamic section.
+#		Setting this to " " will result in no .dynamic padding.
 #
 # When adding sections, do note that the names of some sections are used
 # when specifying the start address of the next.
@@ -66,7 +68,14 @@ test -z "${ALIGNMENT}" && ALIGNMENT="${E
 test "$LD_FLAG" = "N" && DATA_ADDR=.
 INTERP=".interp   ${RELOCATING-0} : { *(.interp) 	}"
 PLT=".plt    ${RELOCATING-0} : { *(.plt)	}"
-DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) }"
+test -z "$DYNAMIC_PAD" && test -n "$RELOCATING" && DYNAMIC_PAD="
+    /* Leave some space in .dynamic section, so that
+       post-linking utilities can add new dynamic tags.  */
+    FILL(0) . += 6 * ${ELFSIZE} / 4;"
+DYNAMIC=".dynamic     ${RELOCATING-0} :
+  {
+    *(.dynamic) $DYNAMIC_PAD
+  }"
 RODATA=".rodata ${RELOCATING-0} : { *(.rodata) ${RELOCATING+*(.rodata.*)} ${RELOCATING+*(.gnu.linkonce.r.*)} }"
 SBSS2=".sbss2 ${RELOCATING-0} : { *(.sbss2) ${RELOCATING+*(.sbss2.*)} ${RELOCATING+*(.gnu.linkonce.sb2.*)} }"
 SDATA2=".sdata2 ${RELOCATING-0} : { *(.sdata2) ${RELOCATING+*(.sdata2.*)} ${RELOCATING+*(.gnu.linkonce.s2.*)} }"


	Jakub

Re: [PATCH] Reserve a few .dynamic entries (take 2)

[Hans-Peter Nilsson]

> Date: Fri, 8 Jun 2001 20:16:57 +0200
> From: Jakub Jelinek <jakub@redhat.com>

> Ok, is this better (ie. can I commit it)?

FWIW, it looks good to me, but I'm not an approver.

brgds, H-P

Re: [PATCH] Reserve a few .dynamic entries

[Hans-Peter Nilsson]

> Date: Wed, 6 Jun 2001 18:13:38 +0200
> From: Jakub Jelinek <jakub@redhat.com>

> +DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) ${RELOCATING+FILL(0) . += 6 * ${ELFSIZE} / 4; } }"

Oh, and I think you need tp rewrite that to not have a '}' in
the '+'-part.  IIRC it breaks with solaris /bin/sh.  Ask Alex
Oliva.

brgds, H-P
PS.  Looking forward to that prelinking stuff.

Re: [PATCH] Reserve a few .dynamic entries

[Ben Elliston]

>>>>> "Hans-Peter" == Hans-Peter Nilsson <hans-peter.nilsson@axis.com> writes:

  >> Date: Wed, 6 Jun 2001 18:13:38 +0200
  >> From: Jakub Jelinek <jakub@redhat.com>

  >> +DYNAMIC=".dynamic     ${RELOCATING-0} : { *(.dynamic) ${RELOCATING+FILL(0) . += 6 * ${ELFSIZE} / 4; } }"

  Hans-Peter> Oh, and I think you need tp rewrite that to not have a '}' in
  Hans-Peter> the '+'-part.  IIRC it breaks with solaris /bin/sh.  Ask Alex
  Hans-Peter> Oliva.

Here is a snippet of a patch that explains this situation.  Hopefully
the comment is sufficient.

  -OTHER_RELOCATING_SECTIONS=${OTHER_RELOCATING_SECTIONS-'
  +
  +# Do not use the ${varname-${varname}} construct here; there will be
  +# problems with some shells (e.g. on Solaris) since there is a bug
  +# that trigs when the incoming macro value contains a "}".
  +test -z "$OTHER_RELOCATING_SECTIONS" && OTHER_RELOCATING_SECTIONS='
    .foo 0 : { *(.foo) }
    .stack 0x30000 : { _stack = .; *(.stack) }
  -'}
  +'

Ben

Re: [PATCH] Reserve a few .dynamic entries

[Ulrich Drepper]

Jakub Jelinek <jakub@redhat.com> writes:

> Do you agree with this? Can I commit it?

I think it's no good idea to do this unconditionally.

-- 
---------------.                          ,-.   1325 Chesapeake Terrace
Ulrich Drepper  \    ,-------------------'   \  Sunnyvale, CA 94089 USA
Red Hat          `--' drepper at redhat.com   `------------------------