* ld bug? @ 2001-10-29 14:35 Vassili Karpov 2001-10-30 23:30 ` H . J . Lu 0 siblings, 1 reply; 14+ messages in thread From: Vassili Karpov @ 2001-10-29 14:35 UTC (permalink / raw) To: binutils LD discards ELF visibility attributes when making relocateable objects. Example: a.s == .protected symbol b.s == dummy: --- as -o a.o a.s as -o b.o b.s ld -r -o c.o a.o b.o a.o: 4: 00000000 0 NOTYPE GLOBAL PROTECTED UND symbol c.o: 8: 00000000 0 NOTYPE GLOBAL DEFAULT UND symbol -- mailto:malc@pulsesoft.com ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: ld bug? 2001-10-29 14:35 ld bug? Vassili Karpov @ 2001-10-30 23:30 ` H . J . Lu 2001-10-30 23:33 ` H . J . Lu 0 siblings, 1 reply; 14+ messages in thread From: H . J . Lu @ 2001-10-30 23:30 UTC (permalink / raw) To: Vassili Karpov; +Cc: binutils On Tue, Oct 30, 2001 at 01:36:04AM +0300, Vassili Karpov wrote: > LD discards ELF visibility attributes when making relocateable objects. > > Example: > a.s > == > .protected symbol > > b.s > == > dummy: > --- > as -o a.o a.s > as -o b.o b.s > ld -r -o c.o a.o b.o > > a.o: 4: 00000000 0 NOTYPE GLOBAL PROTECTED UND symbol > c.o: 8: 00000000 0 NOTYPE GLOBAL DEFAULT UND symbol > -- It is ok since symbol is undefined. There is no such a thing as protected, undefined symbol. H.J. ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: ld bug? 2001-10-30 23:30 ` H . J . Lu @ 2001-10-30 23:33 ` H . J . Lu 2001-10-30 23:42 ` H . J . Lu 0 siblings, 1 reply; 14+ messages in thread From: H . J . Lu @ 2001-10-30 23:33 UTC (permalink / raw) To: Vassili Karpov; +Cc: binutils On Tue, Oct 30, 2001 at 11:30:13PM -0800, H . J . Lu wrote: > On Tue, Oct 30, 2001 at 01:36:04AM +0300, Vassili Karpov wrote: > > LD discards ELF visibility attributes when making relocateable objects. > > > > Example: > > a.s > > == > > .protected symbol > > > > b.s > > == > > dummy: > > --- > > as -o a.o a.s > > as -o b.o b.s > > ld -r -o c.o a.o b.o > > > > a.o: 4: 00000000 0 NOTYPE GLOBAL PROTECTED UND symbol > > c.o: 8: 00000000 0 NOTYPE GLOBAL DEFAULT UND symbol > > -- > > It is ok since symbol is undefined. There is no such a thing as > protected, undefined symbol. I take it back. It is a bug. I will fix it. H.J. ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: ld bug? 2001-10-30 23:33 ` H . J . Lu @ 2001-10-30 23:42 ` H . J . Lu 2001-10-31 6:27 ` Nick Clifton 0 siblings, 1 reply; 14+ messages in thread From: H . J . Lu @ 2001-10-30 23:42 UTC (permalink / raw) To: Vassili Karpov; +Cc: binutils On Tue, Oct 30, 2001 at 11:33:09PM -0800, H . J . Lu wrote: > On Tue, Oct 30, 2001 at 11:30:13PM -0800, H . J . Lu wrote: > > On Tue, Oct 30, 2001 at 01:36:04AM +0300, Vassili Karpov wrote: > > > LD discards ELF visibility attributes when making relocateable objects. > > > > > > Example: > > > a.s > > > == > > > .protected symbol > > > > > > b.s > > > == > > > dummy: > > > --- > > > as -o a.o a.s > > > as -o b.o b.s > > > ld -r -o c.o a.o b.o > > > > > > a.o: 4: 00000000 0 NOTYPE GLOBAL PROTECTED UND symbol > > > c.o: 8: 00000000 0 NOTYPE GLOBAL DEFAULT UND symbol > > > -- > > > > It is ok since symbol is undefined. There is no such a thing as > > protected, undefined symbol. > > I take it back. It is a bug. I will fix it. > > Here it is. H.J. --- 2001-10-30 H.J. Lu <hjl@gnu.org> * elflink.h (elf_link_output_extsym): Don't clear the visibility field for relocateable. Index: elflink.h =================================================================== RCS file: /work/cvs/gnu/binutils/bfd/elflink.h,v retrieving revision 1.87 diff -u -p -r1.87 elflink.h --- elflink.h 2001/10/28 17:38:27 1.87 +++ elflink.h 2001/10/31 07:40:35 @@ -5877,7 +5877,8 @@ elf_link_output_extsym (h, data) /* If a symbol is not defined locally, we clear the visibility field. */ - if ((h->elf_link_hash_flags & ELF_LINK_HASH_DEF_REGULAR) == 0) + if (! finfo->info->relocateable + && (h->elf_link_hash_flags & ELF_LINK_HASH_DEF_REGULAR) == 0) sym.st_other ^= ELF_ST_VISIBILITY (sym.st_other); /* If this symbol should be put in the .dynsym section, then put it ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: ld bug? 2001-10-30 23:42 ` H . J . Lu @ 2001-10-31 6:27 ` Nick Clifton [not found] ` <mailpost.1004538510.13276@postal.sibyte.com> 0 siblings, 1 reply; 14+ messages in thread From: Nick Clifton @ 2001-10-31 6:27 UTC (permalink / raw) To: H . J . Lu; +Cc: Vassili Karpov, binutils Hi H.J. > 2001-10-30 H.J. Lu <hjl@gnu.org> > > * elflink.h (elf_link_output_extsym): Don't clear the visibility > field for relocateable. Approved. Cheers Nick ^ permalink raw reply [flat|nested] 14+ messages in thread
[parent not found: <mailpost.1004538510.13276@postal.sibyte.com>]
[parent not found: <yov5wv1bpxbb.fsf@broadcom.com>]
* PATCH: Add the "ld -r" test for ELF visibility. [not found] ` <yov5wv1bpxbb.fsf@broadcom.com> @ 2001-10-31 10:37 ` H . J . Lu 2001-10-31 11:12 ` Hans-Peter Nilsson 0 siblings, 1 reply; 14+ messages in thread From: H . J . Lu @ 2001-10-31 10:37 UTC (permalink / raw) To: cgd; +Cc: nickc, binutils On Wed, Oct 31, 2001 at 08:51:36AM -0800, cgd@broadcom.com wrote: > nickc@cambridge.redhat.com ("Nick Clifton") writes: > > > 2001-10-30 H.J. Lu <hjl@gnu.org> > > > > > > * elflink.h (elf_link_output_extsym): Don't clear the visibility > > > field for relocateable. > > > > Approved. > > So, shouldn't we even _try_ to add test cases to the ld testsuite to > handle things like this? > Here is the new test. H.J. ---- 2001-10-31 H.J. Lu <hjl@gnu.org> * ld-elfvsb/elfvsb.exp: Add the "ld -r" test. * ld-elfvsb/visibility.s: New. --- ld-elfvsb/elfvsb.exp.vsb Wed Jun 6 15:07:50 2001 +++ ld-elfvsb/elfvsb.exp Wed Oct 31 10:29:04 2001 @@ -42,6 +42,39 @@ if { [istarget *-*-linux*aout*] \ return } +if { ![ld_assemble $as $srcdir/$subdir/visibility.s tmpdir/visibility.o] + || ![ld_simple_link $ld tmpdir/relocateable.o "-r tmpdir/visibility.o"] } { + unresolved "visibility relocateable" +} else { + global READELF + + catch "exec $READELF -s tmpdir/relocateable.o > tmpdir/relocateable.output" got + if ![string match "" $got] then { + send_log "$got\n" + verbose "$got" + fail "visibility relocateable" + } else { + catch "exec grep protected tmpdir/relocateable.output | grep PROTECTED" got + if [regexp "PROTECTED" $got] then { + pass "visibility relocateable (protected)" + } else { + fail "visibility relocateable (protected)" + } + catch "exec grep hidden tmpdir/relocateable.output | grep HIDDEN" got + if [regexp "HIDDEN" $got] then { + pass "visibility relocateable (hidden)" + } else { + fail "visibility relocateable (hidden)" + } + catch "exec grep internal tmpdir/relocateable.output | grep INTERNAL" got + if [regexp "INTERNAL" $got] then { + pass "visibility relocateable (internal)" + } else { + fail "visibility relocateable (internal)" + } + } +} + set tmpdir tmpdir set SHCFLAG "" --- ld-elfvsb/visibility.s.vsb Wed Oct 31 09:17:47 2001 +++ ld-elfvsb/visibility.s Wed Oct 31 09:18:16 2001 @@ -0,0 +1,3 @@ + .protected protected + .hidden hidden + .internal internal ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 10:37 ` PATCH: Add the "ld -r" test for ELF visibility H . J . Lu @ 2001-10-31 11:12 ` Hans-Peter Nilsson 2001-10-31 12:07 ` H . J . Lu 2001-10-31 13:20 ` H . J . Lu 0 siblings, 2 replies; 14+ messages in thread From: Hans-Peter Nilsson @ 2001-10-31 11:12 UTC (permalink / raw) To: H . J . Lu; +Cc: cgd, nickc, binutils On Wed, 31 Oct 2001, H . J . Lu wrote: > Here is the new test. > > * ld-elfvsb/elfvsb.exp: Add the "ld -r" test. > > * ld-elfvsb/visibility.s: New. This looks like it could use the run_dump_test framework. See ld-linkonce for an example (change "objdump" to "readelf", and weed out everything but the HIDDEN and PROTECTED lines with #... and #pass). Much less footwork per test. brgds, H-P ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 11:12 ` Hans-Peter Nilsson @ 2001-10-31 12:07 ` H . J . Lu 2001-10-31 13:20 ` H . J . Lu 1 sibling, 0 replies; 14+ messages in thread From: H . J . Lu @ 2001-10-31 12:07 UTC (permalink / raw) To: Hans-Peter Nilsson; +Cc: cgd, nickc, binutils On Wed, Oct 31, 2001 at 02:12:10PM -0500, Hans-Peter Nilsson wrote: > On Wed, 31 Oct 2001, H . J . Lu wrote: > > Here is the new test. > > > > * ld-elfvsb/elfvsb.exp: Add the "ld -r" test. > > > > * ld-elfvsb/visibility.s: New. > > This looks like it could use the run_dump_test framework. See > ld-linkonce for an example (change "objdump" to "readelf", and > weed out everything but the HIDDEN and PROTECTED lines with #... > and #pass). Much less footwork per test. > I am not sure if it will work too well in this case. The problem is I don't care the exact output. I only look for 7: 00000000 0 NOTYPE GLOBAL HIDDEN UND hidden 8: 00000000 0 NOTYPE GLOBAL INTERNAL UND internal 9: 00000000 0 NOTYPE GLOBAL PROTECTED UND protected Different targets may have different indexes and additional informations. How do I filter them out? H.J. ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 11:12 ` Hans-Peter Nilsson 2001-10-31 12:07 ` H . J . Lu @ 2001-10-31 13:20 ` H . J . Lu 2001-10-31 13:31 ` Hans-Peter Nilsson 1 sibling, 1 reply; 14+ messages in thread From: H . J . Lu @ 2001-10-31 13:20 UTC (permalink / raw) To: Hans-Peter Nilsson; +Cc: cgd, nickc, binutils On Wed, Oct 31, 2001 at 02:12:10PM -0500, Hans-Peter Nilsson wrote: > On Wed, 31 Oct 2001, H . J . Lu wrote: > > Here is the new test. > > > > * ld-elfvsb/elfvsb.exp: Add the "ld -r" test. > > > > * ld-elfvsb/visibility.s: New. > > This looks like it could use the run_dump_test framework. See > ld-linkonce for an example (change "objdump" to "readelf", and > weed out everything but the HIDDEN and PROTECTED lines with #... > and #pass). Much less footwork per test. > How about this patch? Tested on Linux/x86 and Linux/alpha. H.J. ---- --- ld-elfvsb/define.d.vsb Wed Oct 31 12:39:27 2001 +++ ld-elfvsb/define.d Wed Oct 31 13:14:16 2001 @@ -0,0 +1,13 @@ +#source: undef.s +#source: define.s +#ld: -r +#readelf: -s +#target: *-*-* + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL HIDDEN . hidden + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL INTERNAL . internal + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED . protected +#... --- ld-elfvsb/define.s.vsb Wed Oct 31 12:39:24 2001 +++ ld-elfvsb/define.s Wed Oct 31 12:39:07 2001 @@ -0,0 +1,6 @@ + .globl protected +protected: + .globl hidden +hidden: + .globl internal +internal: --- ld-elfvsb/elfvsb.exp.vsb Wed Jun 6 15:07:50 2001 +++ ld-elfvsb/elfvsb.exp Wed Oct 31 12:04:59 2001 @@ -42,6 +42,13 @@ if { [istarget *-*-linux*aout*] \ return } +set test_list [lsort [glob -nocomplain $srcdir/$subdir/*.d]] +foreach t $test_list { + # We need to strip the ".d", but can leave the dirname. + verbose [file rootname $t] + run_dump_test [file rootname $t] +} + set tmpdir tmpdir set SHCFLAG "" --- ld-elfvsb/undef.d.vsb Wed Oct 31 11:49:47 2001 +++ ld-elfvsb/undef.d Wed Oct 31 13:14:27 2001 @@ -0,0 +1,12 @@ +#source: undef.s +#ld: -r +#readelf: -s +#target: *-*-* + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL HIDDEN UND hidden + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL INTERNAL UND internal + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED UND protected +#... --- ld-elfvsb/undef.s.vsb Wed Oct 31 11:47:24 2001 +++ ld-elfvsb/undef.s Wed Oct 31 11:47:24 2001 @@ -0,0 +1,3 @@ + .protected protected + .hidden hidden + .internal internal ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 13:20 ` H . J . Lu @ 2001-10-31 13:31 ` Hans-Peter Nilsson 2001-10-31 13:35 ` cgd 2001-10-31 13:53 ` H . J . Lu 0 siblings, 2 replies; 14+ messages in thread From: Hans-Peter Nilsson @ 2001-10-31 13:31 UTC (permalink / raw) To: H . J . Lu; +Cc: cgd, nickc, binutils On Wed, 31 Oct 2001, H . J . Lu wrote: > How about this patch? Tested on Linux/x86 and Linux/alpha. That's what I was thinking of, except (for both tests)... > +#target: *-*-* The target-specifier is optional; you don't need it for *-*-*. > + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED . protected > +#... Not sure why I think so, but I think the last "#..." should better be "#pass". Perhaps it's because things then look a bit clearer when the test fails. Another caveat is that the order of the symbols might change, the test fail, but the output still be valid. Perhaps different tests, each selecting one line, if it's important. brgds, H-P ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 13:31 ` Hans-Peter Nilsson @ 2001-10-31 13:35 ` cgd 2001-10-31 13:53 ` H . J . Lu 1 sibling, 0 replies; 14+ messages in thread From: cgd @ 2001-10-31 13:35 UTC (permalink / raw) To: Hans-Peter Nilsson; +Cc: hjl, binutils "Hans-Peter Nilsson" <hp@bitrange.com> writes: > > + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED . protected > > +#... > > Not sure why I think so, but I think the last "#..." should > better be "#pass". Perhaps it's because things then look a bit > clearer when the test fails. Actually, currently, AFAIK, a trailing #... at end of file _doesn't_ match additional output in the objdump dump output. (I posted a patch to make it so, it was quasi-approved, but I didn't get around to applying it... and now that i know about #pass I don't know that it's really a good idea.) Also, e.g. for: .: [0-9a-fA-F]* 0 NOTYPE GLOBAL HIDDEN . hidden I think i'd suggest: +[0-9]+: [0-9a-fA-F]+ 0 NOTYPE GLOBAL HIDDEN +[0-9]+ hidden (more tolerant of different section numbers) chris ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 13:31 ` Hans-Peter Nilsson 2001-10-31 13:35 ` cgd @ 2001-10-31 13:53 ` H . J . Lu 2001-10-31 15:09 ` Hans-Peter Nilsson 1 sibling, 1 reply; 14+ messages in thread From: H . J . Lu @ 2001-10-31 13:53 UTC (permalink / raw) To: Hans-Peter Nilsson; +Cc: cgd, nickc, binutils On Wed, Oct 31, 2001 at 04:31:13PM -0500, Hans-Peter Nilsson wrote: > On Wed, 31 Oct 2001, H . J . Lu wrote: > > How about this patch? Tested on Linux/x86 and Linux/alpha. > > That's what I was thinking of, except (for both tests)... > > > +#target: *-*-* > > The target-specifier is optional; you don't need it for *-*-*. > > > + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED . protected > > +#... > > Not sure why I think so, but I think the last "#..." should > better be "#pass". Perhaps it's because things then look a bit > clearer when the test fails. > > Another caveat is that the order of the symbols might change, > the test fail, but the output still be valid. Perhaps different > tests, each selecting one line, if it's important. > How about this one? H.J. ---- 2001-10-31 H.J. Lu <hjl@gnu.org> * ld-elfvsb/elfvsb.exp: Add the "ld -r" test. * ld-elfvsb/hidden0.d: New. * ld-elfvsb/hidden1.d: New. * ld-elfvsb/internal0.d: New. * ld-elfvsb/internal1.d: New. * ld-elfvsb/protected0.d: New. * ld-elfvsb/protected1.d: New. * ld-elfvsb/undef.d: New. * ld-elfvsb/undef.s: New. --- ld-elfvsb/define.s.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/define.s Wed Oct 31 13:40:44 2001 @@ -0,0 +1,6 @@ + .globl protected +protected: + .globl hidden +hidden: + .globl internal +internal: --- ld-elfvsb/elfvsb.exp.vsb Wed Jun 6 15:07:50 2001 +++ ld-elfvsb/elfvsb.exp Wed Oct 31 12:04:59 2001 @@ -42,6 +42,13 @@ if { [istarget *-*-linux*aout*] \ return } +set test_list [lsort [glob -nocomplain $srcdir/$subdir/*.d]] +foreach t $test_list { + # We need to strip the ".d", but can leave the dirname. + verbose [file rootname $t] + run_dump_test [file rootname $t] +} + set tmpdir tmpdir set SHCFLAG "" --- ld-elfvsb/hidden0.d.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/hidden0.d Wed Oct 31 13:49:11 2001 @@ -0,0 +1,10 @@ +#source: define.s +#source: undef.s +#ld: -r +#readelf: -s + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL HIDDEN . hidden +#pass --- ld-elfvsb/hidden1.d.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/hidden1.d Wed Oct 31 13:48:26 2001 @@ -0,0 +1,9 @@ +#source: undef.s +#ld: -r +#readelf: -s + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL HIDDEN UND hidden +#pass --- ld-elfvsb/internal0.d.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/internal0.d Wed Oct 31 13:48:37 2001 @@ -0,0 +1,10 @@ +#source: define.s +#source: undef.s +#ld: -r +#readelf: -s + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL INTERNAL . internal +#pass --- ld-elfvsb/internal1.d.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/internal1.d Wed Oct 31 13:48:46 2001 @@ -0,0 +1,9 @@ +#source: undef.s +#ld: -r +#readelf: -s + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL INTERNAL UND internal +#pass --- ld-elfvsb/protected0.d.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/protected0.d Wed Oct 31 13:48:57 2001 @@ -0,0 +1,10 @@ +#source: define.s +#source: undef.s +#ld: -r +#readelf: -s + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED . protected +#pass --- ld-elfvsb/protected1.d.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/protected1.d Wed Oct 31 13:49:04 2001 @@ -0,0 +1,9 @@ +#source: undef.s +#ld: -r +#readelf: -s + +Symbol table '.symtab' contains .* entries: + Num: Value[ ]+Size Type Bind Vis Ndx Name +#... + .: [0-9a-fA-F]* 0 NOTYPE GLOBAL PROTECTED UND protected +#pass --- ld-elfvsb/undef.s.vsb Wed Oct 31 13:45:07 2001 +++ ld-elfvsb/undef.s Wed Oct 31 13:40:44 2001 @@ -0,0 +1,3 @@ + .protected protected + .hidden hidden + .internal internal ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: PATCH: Add the "ld -r" test for ELF visibility. 2001-10-31 13:53 ` H . J . Lu @ 2001-10-31 15:09 ` Hans-Peter Nilsson 0 siblings, 0 replies; 14+ messages in thread From: Hans-Peter Nilsson @ 2001-10-31 15:09 UTC (permalink / raw) To: H . J . Lu; +Cc: cgd, nickc, binutils On Wed, 31 Oct 2001, H . J . Lu wrote: > How about this one? Looks good to me, FWIW. brgds, H-P ^ permalink raw reply [flat|nested] 14+ messages in thread
[parent not found: <1060095548.9686.69.camel@brick.watson.ibm.com>]
* Re: ld bug? [not found] <1060095548.9686.69.camel@brick.watson.ibm.com> @ 2003-08-06 1:26 ` Alan Modra 0 siblings, 0 replies; 14+ messages in thread From: Alan Modra @ 2003-08-06 1:26 UTC (permalink / raw) To: Michal Ostrowski, binutils On Tue, Aug 05, 2003 at 10:59:09AM -0400, Michal Ostrowski wrote: > Hi, > > I have a bunch of object files that when linked together (to build > "troff") result in certain symbols being assigned the same location in > memory. > > This was achieved with gcc-3.2.3 (c++) and, > > powerpc64-linux-ld -v > GNU ld version 2.14.90 20030805 > > The problem is that we have a couple of symbols as follow: > > 84: 0000000020015c10 8 NOTYPE LOCAL DEFAULT 26 soft_hyphen_char > 2986: 0000000020015c18 8 NOTYPE GLOBAL DEFAULT 26 global_special_fonts > > > However, we also have (from .rela.dyn): > 000020015c10 004700000013 R_PPC_COPY 0000000020015c10 _ZTVN10__cxxabiv117__c + 0 > > This causes ld.so to copy 88 bytes of data to the location of > soft_hyphen_char (and the symbols that follow it). [snip] Hmm, yes. A ld map shows .bss 0x0000000010085b30 0x4248 *(.dynbss) *fill* 0x0000000010085b30 0x0 00 *(.bss .bss.* .gnu.linkonce.b.*) .bss 0x0000000010085b30 0x1 ./crtbegin.o *fill* 0x0000000010085b31 0x7 00 .bss 0x0000000010085b38 0xc8 ./env.o Note the zero .dynbss allocation. It should be 312 bytes in this case. The problem was uncoverd by this patch: 2003-07-10 Alan Modra <amodra@bigpond.net.au> (ppc64_elf_build_stubs): Check for stub sections in stub bfd by testing section flags. Fixed like this. * elflink.c (_bfd_elf_create_dynamic_sections): Mark .dynbss with SEC_LINKER_CREATED. Index: bfd/elflink.c =================================================================== RCS file: /cvs/src/src/bfd/elflink.c,v retrieving revision 1.42 diff -u -p -r1.42 elflink.c --- bfd/elflink.c 4 Aug 2003 09:39:46 -0000 1.42 +++ bfd/elflink.c 6 Aug 2003 01:22:35 -0000 @@ -309,7 +309,7 @@ _bfd_elf_create_dynamic_sections (bfd *a section into the .bss section of the final image. */ s = bfd_make_section (abfd, ".dynbss"); if (s == NULL - || ! bfd_set_section_flags (abfd, s, SEC_ALLOC)) + || ! bfd_set_section_flags (abfd, s, SEC_ALLOC | SEC_LINKER_CREATED)) return FALSE; /* The .rel[a].bss section holds copy relocs. This section is not -- Alan Modra IBM OzLabs - Linux Technology Centre ^ permalink raw reply [flat|nested] 14+ messages in thread
end of thread, other threads:[~2003-08-06 1:26 UTC | newest]
Thread overview: 14+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-10-29 14:35 ld bug? Vassili Karpov
2001-10-30 23:30 ` H . J . Lu
2001-10-30 23:33 ` H . J . Lu
2001-10-30 23:42 ` H . J . Lu
2001-10-31 6:27 ` Nick Clifton
[not found] ` <mailpost.1004538510.13276@postal.sibyte.com>
[not found] ` <yov5wv1bpxbb.fsf@broadcom.com>
2001-10-31 10:37 ` PATCH: Add the "ld -r" test for ELF visibility H . J . Lu
2001-10-31 11:12 ` Hans-Peter Nilsson
2001-10-31 12:07 ` H . J . Lu
2001-10-31 13:20 ` H . J . Lu
2001-10-31 13:31 ` Hans-Peter Nilsson
2001-10-31 13:35 ` cgd
2001-10-31 13:53 ` H . J . Lu
2001-10-31 15:09 ` Hans-Peter Nilsson
[not found] <1060095548.9686.69.camel@brick.watson.ibm.com>
2003-08-06 1:26 ` ld bug? Alan Modra
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).