gmane.comp.sysutils.elfutils.devel

FYI small test environment tweak

Mark Wielaard — 2012-05-11T11:08:56

I pushed the following:

commit dbbc67a97ace66c3adc299f26261fdec99d0c0d6
Author: Mark Wielaard <mjw< at >redhat.com>
Date:   Fri May 11 12:57:57 2012 +0200

    TESTS_ENVIRONMENT set LC_ALL and LANG to C.
    
    Some tests rely on sort order and output of tools that might be localized.
    
    Reported-by: Paweł Sikora <pluto< at >pld-linux.org>

diff --git a/tests/ChangeLog b/tests/ChangeLog
index 4c6e9f7..841742d 100644
--- a/tests/ChangeLog
+++ b/tests/ChangeLog
< at >< at > -1,3 +1,7 < at >< at >
+2012-05-11  Mark Wielaard  <mjw< at >redhat.com>
+
+       * Makefile.am (TESTS_ENVIRONMENT): Set LC_ALL and LANG to C.
+
 2012-05-07  Mark Wielaard  <mjw< at >redhat.com>
 
        * low_high_pc.c: Allow highpc == lowpc for CU DIEs for buggy GCC.
diff --git a/tests/Makefile.am b/tests/Makefile.am
index 61247e4..660791c 100644
--- a/tests/Makefile.am
+++ b/tests/Makefile.am
< at >< at > -169,13 +169,14 < at >< at > EXTRA_DIST = run-arextract.sh run-arsymtest.sh \
 
 installed_TESTS_ENVIRONMENT = libdir=$(DESTDIR)$(libdir) \
                              bindir=$(DES

Re: Strip on mips

Mark Wielaard — 2012-05-10T07:32:17

Thanks for the ping. Again I got distracted by 1001 other things. Sigh.
Sorry. There isn't anything difficult to this as far as I understand,
but I still need to get a positive OK from legal. I'll contact them
later today to get this resolved.

Apologies for being so slow/distracted,

Mark
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: Strip on mips

Kurt Roeckx — 2012-05-09T20:09:18

Any news on this?


Kurt

_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: elflint: (zero sized section) alloc flag set but section not inany loaded segment

Mark Wielaard — 2012-05-09T09:31:14

That does make sense, and works for me. I have checked in the following:

diff --git a/src/ChangeLog b/src/ChangeLog
index cae5e6b..af1d92b 100644
--- a/src/ChangeLog
+++ b/src/ChangeLog
< at >< at > -1,3 +1,8 < at >< at >
+2012-05-09  Roland McGrath  <roland< at >hack.frob.com>
+
+       * elflint (check_sections): Allow zero sized sections at (filesz) end
+       of segment. And make check overflow-proofed.
+
 2012-04-24  Mark Wielaard  <mjw< at >redhat.com>
 
        * readelf.c (print_ops): Add DW_OP_GNU_push_tls_address,
diff --git a/src/elflint.c b/src/elflint.c
index a03caac..7c56069 100644
--- a/src/elflint.c
+++ b/src/elflint.c
< at >< at > -3716,8 +3716,10 < at >< at > section [%2zu] '%s' is both executable and writable\n"),
                    || (phdr->p_type == PT_TLS
                        && (shdr->sh_flags & SHF_TLS) != 0))
                && phdr->p_offset <= shdr->sh_offset
-               && (phdr->p_offset + phdr->p_filesz > shdr->sh_offset
-                   || (phdr->p_offset + phdr->p_memsz > shdr->sh_offset
+               && ((sh

Re: Function symbol retention after stripping

Douglas, William — 2012-05-07T21:07:13


Ah that is an interesting, thanks for the link!
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: Function symbol retention after stripping

Douglas, William — 2012-05-07T21:04:24

That's what I'm seeing right now when I try it. That is very confusing
as on a different system last week I had strip -g -f when you don't
have access to the debuginfo file it outputs showing ?? for the
function names but if I readelf the binaries on this system I clearly
see functions in the symtab table.  I'll have to get back on that
system tomorrow and see what I did/missed, thanks.
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: Function symbol retention after stripping

Roland McGrath — 2012-05-07T19:35:30

It sounds like you want strip -g, which has existed for many years.
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: Function symbol retention after stripping

David Smith — 2012-05-07T19:19:20



William,

I can't answer for the elfutils library itself, but you might want to
look at/help out with a proposed new F18 feature called 'MiniDebugInfo',
which sounds similar to what you want to do.

<http://news.gmane.org/gmane.linux.redhat.fedora.devel>
<https://fedoraproject.org/wiki/Features/MiniDebugInfo>

Function symbol retention after stripping

Douglas, William — 2012-05-07T18:51:24

Hi all,

I'm trying to get gbd backtraces to give function names without having
a package's debuginfo installed. Would elfutils take a patch allowing
strip to take an option for keeping the minimal amount of information
needed to do this?

I'm still trying to figure out how to do this exactly but right now I
think I would be keeping SHT_SYMTAB and SHT_DYNSYM function symbols
but I haven't looked at this code before and I'm still trying to
figure out how it all comes together. Any advice would be appreciated
if you'd be okay with this functionality in elfutils.


Thanks,
William
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

FYI: tests/low_high_pc.c: Allow highpc == lowpc for CU DIEs forbuggy GCC.

Mark Wielaard — 2012-05-07T11:34:59

I pushed the following. Although the bug is fixed in GCC 4.7 that has
only been released last month. Most distros are still based on 4.6 or
lower.

commit d9bc75fc20f0ce5064cac192a518af3c7192e96d
Author: Mark Wielaard <mjw< at >redhat.com>
Date:   Mon May 7 13:29:52 2012 +0200

    tests/low_high_pc.c: Allow highpc == lowpc for CU DIEs for buggy GCC.
    
    GCC < 4.7 had a bug where no code CUs got a highpc == lowpc.
    Allow that, because it is not the main purpose of this test.

diff --git a/tests/ChangeLog b/tests/ChangeLog
index abc7340..4c6e9f7 100644
--- a/tests/ChangeLog
+++ b/tests/ChangeLog
< at >< at > -1,3 +1,7 < at >< at >
+2012-05-07  Mark Wielaard  <mjw< at >redhat.com>
+
+* low_high_pc.c: Allow highpc == lowpc for CU DIEs for buggy GCC.
+
 2012-04-27  Mark Wielaard  <mjw< at >redhat.com>
 
 * Makefile.am (TESTS): Add run-low_high_pc.sh
diff --git a/tests/low_high_pc.c b/tests/low_high_pc.c
index 3054046..6fc8b15 100644
--- a/tests/low_high_pc.c
+++ b/tests/low_high_pc.c
< at >< at > -72,9 +72,12 < at >< at > handle_die (Dwarf_Die *die, void *

Re: FYI: readelf.c (print_ops): Add missing DW_OP_GNU operands.

Roland McGrath — 2012-04-28T06:03:35

Agreed.  We held off before because we thought the C++ interface was
going to become usable soon enough, and we have it there already in
fairly nice ways.  But it never did, and there are C consumers to
consider too.  Incidentally, it looks like gcc's libiberty just added
exactly this in the last few days.

We need to merge the generator stuff into the implementation first,
since obviously we don't want just to move the manually-maintained
readelf.c tables into manually-maintained libdw tables.

Then we just have to decide on the exact interfaces, which shouldn't
be too hard.


Thanks,
Roland
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: PATCH: dwarf_highpc: Handle DW_AT_high_pc being a constantoffset from DW_AT_low_pc.

Roland McGrath — 2012-04-28T05:56:10

Typo in NEWS.

We use '== NULL' for pointer tests.

No real need for an else block after a conditional return.
Just fall through.

Go ahead and commit it with those nits fixed and happy tests.

I'm glad to know you're paying attention to the relocate branch.


Thanks,
Roland


_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: PATCH: dwarf_highpc: Handle DW_AT_high_pc being a constantoffset from DW_AT_low_pc.

Mark Wielaard — 2012-04-27T23:00:35

OK, added.


Done.


Done.


Also done.

Rewritten patch attached.


I'll take a look. This actually came from studying the relocation branch
and realizing a lot of relocations came in low_pc/high_pc pairs.

Thanks,

Mark
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: elflint: (zero sized section) alloc flag set but section not inany loaded segment

Roland McGrath — 2012-04-27T22:43:31

But it *is* clearly inside the second PT_LOAD segment.
It's just that the check is:

  && (phdr->p_offset + phdr->p_filesz > shdr->sh_offset
      || (phdr->p_offset + phdr->p_memsz > shdr->sh_offset
  && shdr->sh_type == SHT_NOBITS)))

and it's on the cusp between p_filesz and p_memsz.  So:

  && (phdr->p_offset + phdr->p_filesz > shdr->sh_offset
      || (phdr->p_offset + phdr->p_memsz > shdr->sh_offset
  && shdr->sh_type == SHT_NOBITS)
      || (phdr->p_offset + phdr->p_filesz == shdr->sh_offset
  && shdr->sh_size == 0)))

or perhaps better overflow-proofed:

  && ((shdr->sh_offset - phdr->p_offset <= phdr->p_filesz
       && (shdr->sh_offset - phdr->p_offset < phdr->p_filesz
   || shdr->sh_size == 0))
      || (shdr->sh_offset - phdr->p_offset < phdr->p_memsz
  && shdr->sh_type == SHT_NOBITS)))

I think we should accept these corner cases but not accept wild sh_addr or
sh_offset values just because they are technically meaningless with a zero
sh_size.


Thanks,
Roland

_________________

Re: PATCH: dwarf_highpc: Handle DW_AT_high_pc being a constant offsetfrom DW_AT_low_pc.

Roland McGrath — 2012-04-27T21:44:14

The functionality change is fine.  Needs a NEWS item.

Don't make the same dwarf_attr call twice.  If it fails,
return immediately.

I'm not sure you should use dwarf_formaddr for a case that might not work,
since we get an extra __libdw_seterrno call when the user's call succeeds.
I don't know off hand if we have clear precedent about that issue.  But
here it's easy enough just to check attr->form for DW_FORM_addr.

If dwarf_formudata fails I don't think we should return with
DWARF_E_NO_CONSTANT, since the user asked for an address, not a constant.
Probably simplest just to add a
__libdw_seterrno (DWARF_E_NO_ADDR);
in that error case.

Also, please look at the relocate branch for how this can be handled there,
so the branch (including new interfaces) doesn't regress from trunk in
subtle ways.  I think the only change needed is dwarf_ranges_relocatable,
but that seems like it could be nontrivial.


Thanks,
Roland
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >list

PATCH: dwarf_highpc: Handle DW_AT_high_pc being a constant offsetfrom DW_AT_low_pc.

Mark Wielaard — 2012-04-27T11:11:27

Hi,

The DWARF spec says (since version 4) that DW_AT_high_pc can be
represented by a constant form.

        If the value of the DW_AT_high_pc is of class address, it is the
        relocated address of the first location past the last
        instruction associated with the entity; if it is of class
        constant, the value is an unsigned integer offset which when
        added to the low PC gives the address of the first location past
        the last instruction associated with the entity.

I have a patch for gcc to encode DW_AT_high_pc this way (which saves a
lot of relocations) and jakub has a patch for dwz to encode
DW_AT_high_pc in the smallest possible constant form, which can save ~1%
on the size of debuginfo.

The attached patch adds a testcase for this and makes dwarf_highpc ()
handle the constant form.

Does this look OK?

Thanks,

Mark
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/el

Re: elflint: (zero sized section) alloc flag set but section not inany loaded segment

Mark Wielaard — 2012-04-27T08:39:50

Here is an example. The issue is section 24, which is the last PROGBITS
alloc one, and is of size zero, it shares its offset with the NOBITS
section right after it.

There are 38 section headers, starting at offset 0xb638:

Section Headers:
[Nr] Name                 Type         Addr             Off      Size     ES Flags Lk Inf Al
[ 0]                      NULL         0000000000000000 00000000 00000000  0        0   0  0
[ 1] .interp              PROGBITS     0000000000400200 00000200 0000001c  0 A      0   0  1
[ 2] .note.ABI-tag        NOTE         000000000040021c 0000021c 00000020  0 A      0   0  4
[ 3] .dynsym              DYNSYM       0000000000400240 00000240 00000438 24 A      4   1  8
[ 4] .dynstr              STRTAB       0000000000400678 00000678 0000026b  0 A      0   0  1
[ 5] .hash                HASH         00000000004008e8 000008e8 00000150  4 A      3   0  8
[ 6] .gnu.version         GNU_versym   0000000000400a38 00000a38 0000005a  2 A      3   0  2
[ 7] .gnu.version_r       GNU_verneed

Re: elflint: (zero sized section) alloc flag set but section not inany loaded segment

Roland McGrath — 2012-04-26T16:50:20

Can you show an example of eu-readelf -Sl output for an affected binary?
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

Re: FYI: readelf.c (print_ops): Add missing DW_OP_GNU operands.

Petr Machata — 2012-04-26T16:25:42


That would be great.  But I'm thinking whether we don't need (in
addition to publishing the generator macros) a stringifying interface as
part of ABI.  The specific task of reporting dwarf tag names in (error)
messages seems to come up in all kinds of clients.  Without a library
entry point for this, the client will at best be able to write something
like "DW_TAG_unknown_0x12345 unsupported".  It's a bit better than the
current approach, where everyone has their own hand-crafted switch, but
you would still link the known values statically to each client.

Thanks,
PM
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

FYI: Removed old libdwarf test show-ciefde.

Mark Wielaard — 2012-04-26T10:28:10

Hi,

elfutils libdwarf was removed a long time ago and this testcase was
never actually ran anyway (which is why nobody noticed it was broken).
I just removed it.

commit bdad8e39aede889dbac95944a07eaebdebe4ad89
Author: Mark Wielaard <mjw< at >redhat.com>
Date:   Thu Apr 26 12:24:33 2012 +0200

    Removed old libdwarf test show-ciefde.

diff --git a/tests/ChangeLog b/tests/ChangeLog
index c38ede5..ffe61d5 100644
--- a/tests/ChangeLog
+++ b/tests/ChangeLog
< at >< at > -1,3 +1,9 < at >< at >
+2012-04-26  Mark Wielaard  <mjw< at >redhat.com>
+
+       * Makefile.am (EXTRA_DIST): Remove run-show-ciefde.sh.
+       * run-show-ciefde.sh: Removed old libdwarf test.
+       * show-ciefde.c: Likewise.

_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel

elflint: (zero sized section) alloc flag set but section not in anyloaded segment

Mark Wielaard — 2012-04-26T09:58:52

Hi,

Against development versions of gcc/binutils elflint gives some
complaints like:

section [24] '.tm_clone_table': alloc flag set but section not in any
loaded segment

This is because that section's size is zero. So it is somewhat
reasonable IMHO that the linker doesn't put it in any segment.

Is the following patch to elflint OK?

Thanks,

Mark
_______________________________________________
elfutils-devel mailing list
elfutils-devel< at >lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/elfutils-devel