* Huge dynamically linked program does not run on mips-linux
@ 2003-10-22 8:11 Atsushi Nemoto
2003-10-29 7:32 ` Atsushi Nemoto
0 siblings, 1 reply; 16+ messages in thread
From: Atsushi Nemoto @ 2003-10-22 8:11 UTC (permalink / raw)
To: linux-mips, binutils
I have a problem that my huge dynamically linked program cause
SIGSEGV or SIGBUS immediately after running from main() on mips-linux.
Digging into this problem, I found that GOT entries are corrupted.
Here are some informations from readelf and objdump:
$ mips-linux-readelf -a myapp
...
Relocation section '.rel.dyn' at offset 0xe3b20 contains 426 entries:
Offset Info Type Sym.Value Sym. Name
...
100c47bc 003f7f03 R_MIPS_REL32 004e4990 getrlimit
...
Symbol table '.dynsym' contains 17742 entries:
Num: Value Size Type Bind Vis Ndx Name
...
16255: 004e4990 0 FUNC GLOBAL DEFAULT UND getrlimit
...
Symbol table '.symtab' contains 26849 entries:
Num: Value Size Type Bind Vis Ndx Name
...
7664: 004e4990 0 FUNC GLOBAL DEFAULT UND getrlimit
...
$ mips-linux-objdump -s -j .got myapp
...
Contents of section .got:
...
100c47b0 100c5cba 00c07c54 100c5c5c 004e4990 ..\...|T..\\.NI.
...
0x004e4990 is "stub" routine to jump into libc's getrlimit.
0x100c47bc must contain 0x004e4990 at run-time, but when the signal
sent the value is 0x009c9320. It is 0x004e4990+0x004e4990.
The outout from objdump shows 0x004e4990 is already in .got and the
output from readelf shows 0x004e4990 will be added to it by dynamic
loader. Is my understanding right?
Could anyone tell me why binutils generate such informations? Or is
this dynamic loader issue?
I'm using binutils 2.14, gcc-3.3.2, glibc-2.2.5. I tried gcc-3.3.1,
glibc-2.3.2, uClibc-0.9.21, binutils-2.14.90.0.6 and binutils-cvs but
no lock.
My program is huge enough so that older binutils causes "relocation
truncated to fit" error.
The program can work well if statically linked. Other (relatively
small) dynamically linked programs can work well also.
Thanks.
---
Atsushi Nemoto
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-22 8:11 Huge dynamically linked program does not run on mips-linux Atsushi Nemoto
@ 2003-10-29 7:32 ` Atsushi Nemoto
2003-10-29 18:14 ` Jun Sun
0 siblings, 1 reply; 16+ messages in thread
From: Atsushi Nemoto @ 2003-10-29 7:32 UTC (permalink / raw)
To: linux-mips, binutils
>>>>> On Wed, 22 Oct 2003 17:11:18 +0900 (JST), Atsushi Nemoto <anemo@mba.ocn.ne.jp> said:
anemo> I have a problem that my huge dynamically linked program cause
anemo> SIGSEGV or SIGBUS immediately after running from main() on
anemo> mips-linux.
anemo> Digging into this problem, I found that GOT entries are
anemo> corrupted.
...
anemo> My program is huge enough so that older binutils causes
anemo> "relocation truncated to fit" error.
More information. My program's .got size exceeds 64K. It seems the
corruption does not happen if .got size is smaller then 64K.
$ mips-linux-readelf -e myapp
...
Section Headers:
[Nr] Name Type Addr Off Size ES Flg Lk Inf Al
...
[21] .got PROGBITS 100b15d0 a075d0 013a04 04 WAp 0 0 16
---
Atsushi Nemoto
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-29 7:32 ` Atsushi Nemoto
@ 2003-10-29 18:14 ` Jun Sun
2003-10-29 18:15 ` Daniel Jacobowitz
0 siblings, 1 reply; 16+ messages in thread
From: Jun Sun @ 2003-10-29 18:14 UTC (permalink / raw)
To: Atsushi Nemoto; +Cc: linux-mips, binutils, jsun
On Wed, Oct 29, 2003 at 04:32:01PM +0900, Atsushi Nemoto wrote:
> >>>>> On Wed, 22 Oct 2003 17:11:18 +0900 (JST), Atsushi Nemoto <anemo@mba.ocn.ne.jp> said:
> anemo> I have a problem that my huge dynamically linked program cause
> anemo> SIGSEGV or SIGBUS immediately after running from main() on
> anemo> mips-linux.
>
> anemo> Digging into this problem, I found that GOT entries are
> anemo> corrupted.
> ...
> anemo> My program is huge enough so that older binutils causes
> anemo> "relocation truncated to fit" error.
>
> More information. My program's .got size exceeds 64K. It seems the
> corruption does not happen if .got size is smaller then 64K.
>
> $ mips-linux-readelf -e myapp
> ...
> Section Headers:
> [Nr] Name Type Addr Off Size ES Flg Lk Inf Al
> ...
> [21] .got PROGBITS 100b15d0 a075d0 013a04 04 WAp 0 0 16
>
Isn't this a known problem in binutils? IIRC, someone is working or has
added "--big-got" support.
Jun
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-29 18:14 ` Jun Sun
@ 2003-10-29 18:15 ` Daniel Jacobowitz
2003-10-30 1:53 ` Thiemo Seufer
2003-10-30 12:54 ` Atsushi Nemoto
0 siblings, 2 replies; 16+ messages in thread
From: Daniel Jacobowitz @ 2003-10-29 18:15 UTC (permalink / raw)
To: Jun Sun; +Cc: Atsushi Nemoto, linux-mips, binutils
On Wed, Oct 29, 2003 at 10:14:00AM -0800, Jun Sun wrote:
> On Wed, Oct 29, 2003 at 04:32:01PM +0900, Atsushi Nemoto wrote:
> > >>>>> On Wed, 22 Oct 2003 17:11:18 +0900 (JST), Atsushi Nemoto <anemo@mba.ocn.ne.jp> said:
> > anemo> I have a problem that my huge dynamically linked program cause
> > anemo> SIGSEGV or SIGBUS immediately after running from main() on
> > anemo> mips-linux.
> >
> > anemo> Digging into this problem, I found that GOT entries are
> > anemo> corrupted.
> > ...
> > anemo> My program is huge enough so that older binutils causes
> > anemo> "relocation truncated to fit" error.
> >
> > More information. My program's .got size exceeds 64K. It seems the
> > corruption does not happen if .got size is smaller then 64K.
> >
> > $ mips-linux-readelf -e myapp
> > ...
> > Section Headers:
> > [Nr] Name Type Addr Off Size ES Flg Lk Inf Al
> > ...
> > [21] .got PROGBITS 100b15d0 a075d0 013a04 04 WAp 0 0 16
> >
>
> Isn't this a known problem in binutils? IIRC, someone is working or has
> added "--big-got" support.
Atsushi-san's program would not even link with a binutils that didn't
support multiple GOTs; I guess that something is going wrong with that
support.
I don't suppose you could provide a testcase?
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-29 18:15 ` Daniel Jacobowitz
@ 2003-10-30 1:53 ` Thiemo Seufer
2003-10-30 2:25 ` Eric Christopher
2003-10-30 12:54 ` Atsushi Nemoto
1 sibling, 1 reply; 16+ messages in thread
From: Thiemo Seufer @ 2003-10-30 1:53 UTC (permalink / raw)
To: Jun Sun, Atsushi Nemoto, linux-mips, binutils
Daniel Jacobowitz wrote:
[snip]
> > > anemo> My program is huge enough so that older binutils causes
> > > anemo> "relocation truncated to fit" error.
> > >
> > > More information. My program's .got size exceeds 64K. It seems the
> > > corruption does not happen if .got size is smaller then 64K.
> > >
> > > $ mips-linux-readelf -e myapp
> > > ...
> > > Section Headers:
> > > [Nr] Name Type Addr Off Size ES Flg Lk Inf Al
> > > ...
> > > [21] .got PROGBITS 100b15d0 a075d0 013a04 04 WAp 0 0 16
> > >
> >
> > Isn't this a known problem in binutils? IIRC, someone is working or has
> > added "--big-got" support.
>
> Atsushi-san's program would not even link with a binutils that didn't
> support multiple GOTs; I guess that something is going wrong with that
> support.
When building python-qt3 on debian unstable I get an oversize GOT and:
/usr/bin/ld: BFD 2.14.90.0.6 20030820 Debian GNU/Linux assertion fail ../../bfd/elfxx-mips.c:2287
Seems like multi-GOT is broken for this case.
Thiemo
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-30 1:53 ` Thiemo Seufer
@ 2003-10-30 2:25 ` Eric Christopher
2003-11-04 5:21 ` Atsushi Nemoto
0 siblings, 1 reply; 16+ messages in thread
From: Eric Christopher @ 2003-10-30 2:25 UTC (permalink / raw)
To: Thiemo Seufer; +Cc: Jun Sun, Atsushi Nemoto, linux-mips, binutils
> When building python-qt3 on debian unstable I get an oversize GOT and:
>
> /usr/bin/ld: BFD 2.14.90.0.6 20030820 Debian GNU/Linux assertion fail ../../bfd/elfxx-mips.c:2287
>
> Seems like multi-GOT is broken for this case.
Judging from the line offset from today's sources to that error message
it might have been fixed with rsandifo's last patch that fixed an
uninitialized variable problem. (global_gotno)
-eric
--
Eric Christopher <echristo@redhat.com>
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-29 18:15 ` Daniel Jacobowitz
2003-10-30 1:53 ` Thiemo Seufer
@ 2003-10-30 12:54 ` Atsushi Nemoto
2003-11-04 16:17 ` Daniel Jacobowitz
1 sibling, 1 reply; 16+ messages in thread
From: Atsushi Nemoto @ 2003-10-30 12:54 UTC (permalink / raw)
To: dan; +Cc: jsun, linux-mips, binutils
[-- Attachment #1: Type: Text/Plain, Size: 8124 bytes --]
>>>>> On Wed, 29 Oct 2003 13:15:17 -0500, Daniel Jacobowitz <dan@debian.org> said:
dan> Atsushi-san's program would not even link with a binutils that
dan> didn't support multiple GOTs; I guess that something is going
dan> wrong with that support.
Yes, I guess too.
dan> I don't suppose you could provide a testcase?
I wrote a short script to generate a testcase. Running attached awk
script create src0.c, src1.c, ..., src3.c.
$ mips-linux-gcc -c src[0-4].c
$ mips-linux-gcc -o bigapp src[0-4].o
This bigapp program compiled with binutils-2.14, gcc-3.3.2, uClibc
0.9.21 is corrupted.
Here is some outputs from readelf:
$ mips-linux-readelf -S bigapp
There are 31 section headers, starting at offset 0x36ad78:
Section Headers:
[Nr] Name Type Addr Off Size ES Flg Lk Inf Al
[ 0] NULL 00000000 000000 000000 00 0 0 0
[ 1] .interp PROGBITS 004000f4 0000f4 00000d 00 A 0 0 1
[ 2] .reginfo MIPS_REGINFO 00400104 000104 000018 18 A 0 0 4
[ 3] .dynamic DYNAMIC 0040011c 00011c 0000d8 08 A 6 0 4
[ 4] .hash HASH 004001f4 0001f4 028784 04 A 5 0 4
[ 5] .dynsym DYNSYM 00428978 028978 061c40 10 A 6 1 4
[ 6] .dynstr STRTAB 0048a5b8 08a5b8 041da5 00 A 0 0 1
[ 7] .init PROGBITS 004cc370 0cc370 0000ac 00 AX 0 0 4
[ 8] .text PROGBITS 004cc420 0cc420 1b7d60 00 AX 0 0 16
[ 9] .fini PROGBITS 00684180 284180 00005c 00 AX 0 0 4
[10] .rodata PROGBITS 006841e0 2841e0 000010 00 A 0 0 16
[11] .data PROGBITS 10000000 285000 000010 00 WA 0 0 16
[12] .rld_map PROGBITS 10000010 285010 000004 00 WA 0 0 4
[13] .eh_frame PROGBITS 10000014 285014 000004 00 A 0 0 4
[14] .ctors PROGBITS 10000018 285018 000008 00 WA 0 0 4
[15] .dtors PROGBITS 10000020 285020 000008 00 WA 0 0 4
[16] .jcr PROGBITS 10000028 285028 000004 00 WA 0 0 4
[17] .got PROGBITS 10000030 285030 02245c 04 WAp 0 0 16
[18] .sbss NOBITS 1002248c 2a7490 000000 00 WAp 0 0 1
[19] .bss NOBITS 10022490 2a7490 000020 00 WA 0 0 16
[20] .comment PROGBITS 00000000 2a7490 0000a2 00 0 0 1
[21] .debug_aranges MIPS_DWARF 00000000 2a7538 000020 00 0 0 8
[22] .debug_info MIPS_DWARF 00000000 2a7558 000069 00 0 0 1
[23] .debug_abbrev MIPS_DWARF 00000000 2a75c1 000014 00 0 0 1
[24] .debug_line MIPS_DWARF 00000000 2a75d5 000044 00 0 0 1
[25] .pdr PROGBITS 00000000 2a761c 0c3660 00 0 0 4
[26] .rel.dyn REL 004cc360 0cc360 000010 08 A 5 0 16
[27] .mdebug.abi32 PROGBITS 00000000 36ac7c 000000 00 0 0 1
[28] .shstrtab STRTAB 00000000 36ac7c 0000fb 00 0 0 1
[29] .symtab SYMTAB 00000000 36b250 0620a0 10 30 45 4
[30] .strtab STRTAB 00000000 3cd2f0 041ee7 00 0 0 1
Key to Flags:
W (write), A (alloc), X (execute), M (merge), S (strings)
I (info), L (link order), G (group), x (unknown)
O (extra OS processing required) o (OS specific), p (processor specific)
$ mips-linux-readelf -r bigapp
Relocation section '.rel.dyn' at offset 0xcc360 contains 2 entries:
Offset Info Type Sym.Value Sym. Name
00000000 00000000 R_MIPS_NONE
1001f4f0 00314703 R_MIPS_REL32 004cc490 getpid
$ mips-linux-readelf -x 17 bigapp | grep 1001f4f0
0x1001f4f0 004cc490 005ea570 00655ba0 0062c9e0 .L...^.p.e[..b..
$ mips-linux-readelf -s bigapp | grep getpid
12615: 004cc490 0 FUNC GLOBAL DEFAULT UND getpid
4109: 004cc490 0 FUNC GLOBAL DEFAULT UND getpid
Strangely the bigapp works correct if I compiled it with glibc 2.2.5.
Here is some outputs from readelf on bigapp.glibc:
$ mips-linux-readelf -S bigapp.glibc
There are 36 section headers, starting at offset 0x378644:
Section Headers:
[Nr] Name Type Addr Off Size ES Flg Lk Inf Al
[ 0] NULL 00000000 000000 000000 00 0 0 0
[ 1] .interp PROGBITS 00400114 000114 00000d 00 A 0 0 1
[ 2] .note.ABI-tag NOTE 00400130 000130 000020 00 A 0 0 16
[ 3] .reginfo MIPS_REGINFO 00400150 000150 000018 18 A 0 0 4
[ 4] .dynamic DYNAMIC 00400168 000168 0000f0 08 A 7 0 4
[ 5] .hash HASH 00400258 000258 028764 04 A 6 0 4
[ 6] .dynsym DYNSYM 004289bc 0289bc 061bc0 10 A 7 1 4
[ 7] .dynstr STRTAB 0048a57c 08a57c 041d6d 00 A 0 0 1
[ 8] .gnu.version VERSYM 004cc2ea 0cc2ea 00c378 02 A 6 0 2
[ 9] .gnu.version_r VERNEED 004d8664 0d8664 000020 00 A 7 1 4
[10] .init PROGBITS 004d86a0 0d86a0 0000ac 00 AX 0 0 4
[11] .text PROGBITS 004d8750 0d8750 1b7d40 00 AX 0 0 16
[12] .fini PROGBITS 00690490 290490 00005c 00 AX 0 0 4
[13] .rodata PROGBITS 006904f0 2904f0 000020 00 A 0 0 16
[14] .data PROGBITS 10000000 291000 000020 00 WA 0 0 16
[15] .rld_map PROGBITS 10000020 291020 000004 00 WA 0 0 4
[16] .eh_frame PROGBITS 10000024 291024 000004 00 A 0 0 4
[17] .ctors PROGBITS 10000028 291028 000008 00 WA 0 0 4
[18] .dtors PROGBITS 10000030 291030 000008 00 WA 0 0 4
[19] .jcr PROGBITS 10000038 291038 000004 00 WA 0 0 4
[20] .got PROGBITS 10000040 291040 022460 04 WAp 0 0 16
[21] .sbss NOBITS 100224a0 2b34a0 000000 00 WAp 0 0 1
[22] .bss NOBITS 100224a0 2b34a0 000020 00 WA 0 0 16
[23] .comment PROGBITS 00000000 2b34a0 0000b4 00 0 0 1
[24] .debug_aranges MIPS_DWARF 00000000 2b3558 000058 00 0 0 8
[25] .debug_pubnames MIPS_DWARF 00000000 2b35b0 000025 00 0 0 1
[26] .debug_info MIPS_DWARF 00000000 2b35d5 000c55 00 0 0 1
[27] .debug_abbrev MIPS_DWARF 00000000 2b422a 00012b 00 0 0 1
[28] .debug_line MIPS_DWARF 00000000 2b4355 00028e 00 0 0 1
[29] .debug_str MIPS_DWARF 00000000 2b45e3 000960 01 MS 0 0 1
[30] .pdr PROGBITS 00000000 2b4f44 0c35c0 00 0 0 4
[31] .mdebug.abi32 PROGBITS 00000000 378504 000000 00 0 0 1
[32] .rel.dyn REL 004d8690 0d8690 000010 08 A 6 0 16
[33] .shstrtab STRTAB 00000000 378504 000140 00 0 0 1
[34] .symtab SYMTAB 00000000 378be4 062200 10 35 59 4
[35] .strtab STRTAB 00000000 3dade4 04204e 00 0 0 1
Key to Flags:
W (write), A (alloc), X (execute), M (merge), S (strings)
I (info), L (link order), G (group), x (unknown)
O (extra OS processing required) o (OS specific), p (processor specific)
$ mips-linux-readelf -r bigapp.glibc
Relocation section '.rel.dyn' at offset 0xd8690 contains 2 entries:
Offset Info Type Sym.Value Sym. Name
00000000 00000000 R_MIPS_NONE
1001c3c0 00340403 R_MIPS_REL32 00000000 getpid
$ mips-linux-readelf -x 20 bigapp.glibc | grep 1001c3c0
0x1001c3c0 00000000 0060d3e0 0063cd10 00608f70 .....`...c...`.p
$ mips-linux-readelf -s bigapp.glibc | grep getpid
13316: 00000000 0 FUNC GLOBAL DEFAULT UND getpid@GLIBC_2.0 (2)
2943: 00000000 0 FUNC GLOBAL DEFAULT UND getpid@@GLIBC_2.0
Is there other informations I can provide?
---
Atsushi Nemoto
[-- Attachment #2: mkbigapp --]
[-- Type: Text/Plain, Size: 688 bytes --]
#!/bin/awk -f
BEGIN {
nfile = 5;
nfunc = 5000;
for (j = 0; j < nfile; j++) {
fname = "src" j ".c";
printf("/* %s */\n", fname) > fname;
for (i = 0; i < nfunc; i++) {
printf("void func%d_%d(void) {}\n", j, i, i) >> fname;
}
printf("int func%d(void) {\n", j) >> fname;
for (i = 0; i < nfunc; i++) {
printf("func%d_%d();\n", j, i) >> fname;
}
printf("return getpid();\n") >> fname;
printf("}\n") >> fname;
if (j == 0) {
printf("int main(int argc, char **argv) {\n") > fname;
for (i = 0; i < nfile; i++) {
printf("printf(\"%%d\\n\", func%d());\n", i) >> fname;
}
printf("return 0;\n") >> fname;
printf("}\n") >> fname;
}
close(fname);
}
}
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-30 2:25 ` Eric Christopher
@ 2003-11-04 5:21 ` Atsushi Nemoto
2003-11-04 8:05 ` Eric Christopher
0 siblings, 1 reply; 16+ messages in thread
From: Atsushi Nemoto @ 2003-11-04 5:21 UTC (permalink / raw)
To: echristo; +Cc: ica2_ts, jsun, linux-mips, binutils
>>>>> On Wed, 29 Oct 2003 18:25:04 -0800, Eric Christopher <echristo@redhat.com> said:
>> When building python-qt3 on debian unstable I get an oversize GOT
>> and:
>>
>> /usr/bin/ld: BFD 2.14.90.0.6 20030820 Debian GNU/Linux assertion
>> fail ../../bfd/elfxx-mips.c:2287
>>
>> Seems like multi-GOT is broken for this case.
eric> Judging from the line offset from today's sources to that error
eric> message it might have been fixed with rsandifo's last patch that
eric> fixed an uninitialized variable problem. (global_gotno)
I tried binutils-2.14.90.0.7 (based on binutils 2003 1029 in CVS) but
my problem did no solved. It seems something is still wrong.
---
Atsushi Nemoto
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-04 5:21 ` Atsushi Nemoto
@ 2003-11-04 8:05 ` Eric Christopher
2003-11-04 11:02 ` Atsushi Nemoto
0 siblings, 1 reply; 16+ messages in thread
From: Eric Christopher @ 2003-11-04 8:05 UTC (permalink / raw)
To: Atsushi Nemoto; +Cc: Thiemo Seufer, jsun, linux-mips, binutils
> eric> Judging from the line offset from today's sources to that error
> eric> message it might have been fixed with rsandifo's last patch that
> eric> fixed an uninitialized variable problem. (global_gotno)
>
> I tried binutils-2.14.90.0.7 (based on binutils 2003 1029 in CVS) but
> my problem did no solved. It seems something is still wrong.
This was plain mips-linux? Not mips64-linux?
And where would I find the sources?
-eric
--
Eric Christopher <echristo@redhat.com>
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-04 8:05 ` Eric Christopher
@ 2003-11-04 11:02 ` Atsushi Nemoto
2003-11-04 17:53 ` Eric Christopher
0 siblings, 1 reply; 16+ messages in thread
From: Atsushi Nemoto @ 2003-11-04 11:02 UTC (permalink / raw)
To: echristo; +Cc: ica2_ts, jsun, linux-mips, binutils
>>>>> On Tue, 04 Nov 2003 00:05:56 -0800, Eric Christopher <echristo@redhat.com> said:
>> I tried binutils-2.14.90.0.7 (based on binutils 2003 1029 in CVS)
>> but my problem did no solved. It seems something is still wrong.
eric> This was plain mips-linux? Not mips64-linux?
Yes. mips-linux and mipsel-linux target (host is i386). Both target
generate broken binary for my test program.
eric> And where would I find the sources?
I'm using plain binutils 2.14 and gcc 3.3.2 from gnu.org FTP site,
binutils 2.14.90.0.7 from
http://www.kernel.org/pub/linux/devel/binutils/.
---
Atsushi Nemoto
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-10-30 12:54 ` Atsushi Nemoto
@ 2003-11-04 16:17 ` Daniel Jacobowitz
0 siblings, 0 replies; 16+ messages in thread
From: Daniel Jacobowitz @ 2003-11-04 16:17 UTC (permalink / raw)
To: Atsushi Nemoto; +Cc: jsun, linux-mips, binutils, aoliva
On Thu, Oct 30, 2003 at 09:54:53PM +0900, Atsushi Nemoto wrote:
> >>>>> On Wed, 29 Oct 2003 13:15:17 -0500, Daniel Jacobowitz <dan@debian.org> said:
> dan> Atsushi-san's program would not even link with a binutils that
> dan> didn't support multiple GOTs; I guess that something is going
> dan> wrong with that support.
>
> Yes, I guess too.
>
> dan> I don't suppose you could provide a testcase?
>
> I wrote a short script to generate a testcase. Running attached awk
> script create src0.c, src1.c, ..., src3.c.
>
> $ mips-linux-gcc -c src[0-4].c
> $ mips-linux-gcc -o bigapp src[0-4].o
>
> This bigapp program compiled with binutils-2.14, gcc-3.3.2, uClibc
> 0.9.21 is corrupted.
>
> Here is some outputs from readelf:
>
> $ mips-linux-readelf -S bigapp
> There are 31 section headers, starting at offset 0x36ad78:
>
> Section Headers:
> [Nr] Name Type Addr Off Size ES Flg Lk Inf Al
> [ 0] NULL 00000000 000000 000000 00 0 0 0
> [ 1] .interp PROGBITS 004000f4 0000f4 00000d 00 A 0 0 1
> [ 2] .reginfo MIPS_REGINFO 00400104 000104 000018 18 A 0 0 4
> [ 3] .dynamic DYNAMIC 0040011c 00011c 0000d8 08 A 6 0 4
> [ 4] .hash HASH 004001f4 0001f4 028784 04 A 5 0 4
> [ 5] .dynsym DYNSYM 00428978 028978 061c40 10 A 6 1 4
> [ 6] .dynstr STRTAB 0048a5b8 08a5b8 041da5 00 A 0 0 1
> [ 7] .init PROGBITS 004cc370 0cc370 0000ac 00 AX 0 0 4
> [ 8] .text PROGBITS 004cc420 0cc420 1b7d60 00 AX 0 0 16
> [ 9] .fini PROGBITS 00684180 284180 00005c 00 AX 0 0 4
> [10] .rodata PROGBITS 006841e0 2841e0 000010 00 A 0 0 16
> [11] .data PROGBITS 10000000 285000 000010 00 WA 0 0 16
> [12] .rld_map PROGBITS 10000010 285010 000004 00 WA 0 0 4
> [13] .eh_frame PROGBITS 10000014 285014 000004 00 A 0 0 4
> [14] .ctors PROGBITS 10000018 285018 000008 00 WA 0 0 4
> [15] .dtors PROGBITS 10000020 285020 000008 00 WA 0 0 4
> [16] .jcr PROGBITS 10000028 285028 000004 00 WA 0 0 4
> [17] .got PROGBITS 10000030 285030 02245c 04 WAp 0 0 16
> [18] .sbss NOBITS 1002248c 2a7490 000000 00 WAp 0 0 1
> [19] .bss NOBITS 10022490 2a7490 000020 00 WA 0 0 16
> [20] .comment PROGBITS 00000000 2a7490 0000a2 00 0 0 1
> [21] .debug_aranges MIPS_DWARF 00000000 2a7538 000020 00 0 0 8
> [22] .debug_info MIPS_DWARF 00000000 2a7558 000069 00 0 0 1
> [23] .debug_abbrev MIPS_DWARF 00000000 2a75c1 000014 00 0 0 1
> [24] .debug_line MIPS_DWARF 00000000 2a75d5 000044 00 0 0 1
> [25] .pdr PROGBITS 00000000 2a761c 0c3660 00 0 0 4
> [26] .rel.dyn REL 004cc360 0cc360 000010 08 A 5 0 16
> [27] .mdebug.abi32 PROGBITS 00000000 36ac7c 000000 00 0 0 1
> [28] .shstrtab STRTAB 00000000 36ac7c 0000fb 00 0 0 1
> [29] .symtab SYMTAB 00000000 36b250 0620a0 10 30 45 4
> [30] .strtab STRTAB 00000000 3cd2f0 041ee7 00 0 0 1
> Key to Flags:
> W (write), A (alloc), X (execute), M (merge), S (strings)
> I (info), L (link order), G (group), x (unknown)
> O (extra OS processing required) o (OS specific), p (processor specific)
> $ mips-linux-readelf -r bigapp
>
> Relocation section '.rel.dyn' at offset 0xcc360 contains 2 entries:
> Offset Info Type Sym.Value Sym. Name
> 00000000 00000000 R_MIPS_NONE
> 1001f4f0 00314703 R_MIPS_REL32 004cc490 getpid
> $ mips-linux-readelf -x 17 bigapp | grep 1001f4f0
> 0x1001f4f0 004cc490 005ea570 00655ba0 0062c9e0 .L...^.p.e[..b..
> $ mips-linux-readelf -s bigapp | grep getpid
> 12615: 004cc490 0 FUNC GLOBAL DEFAULT UND getpid
> 4109: 004cc490 0 FUNC GLOBAL DEFAULT UND getpid
Now this is very strange. Please bear with me; I have the bad habit of
starting my email before I've fully figured out the problem...
MIPS, at least MIPS/Linux, has a very odd ELF ABI. The following isn't
based on documentation, just on my reading of the glibc loader. The
.got section is relocated implicitly; dynamic relocations aren't even
_used_ normally, in the single-GOT case.
First we have reserved entres; two of them. Then local got entries;
DT_MIPS_LOCAL_GOTNO of them. They get the library base added. Then
global entries. We start in the dynsym table at DT_MIPS_GOTSYM, and
count up to just under DT_MIPS_SYMTABNO; GOT entries are set to the
address of the referenced symbol.
The loader does not require or expect any dynamic relocations to refer
to the .got section. It also doesn't seem to care what's there, at
least not for external functions - defined functions and sections are
another matter. So it doesn't expect the relocation to be there.
Now, here's the peculiar thing:
drow@nevyn:~/mips-got% readelf -Ds app | sort -n | grep UND
55 3777: 00000000 0 NOTYPE WEAK DEFAULT UND __gmon_start__
768 1005: 00000000 0 NOTYPE WEAK DEFAULT UND _Jv_RegisterClasses
5231 8511: 00000000 0 FUNC GLOBAL DEFAULT UND __libc_start_main
13350 4887: 00000000 0 FUNC GLOBAL DEFAULT UND getpid
22116 8185: 00000000 0 FUNC GLOBAL DEFAULT UND printf
drow@nevyn:~/mips-got% readelf -Dr app
'REL' relocation section at offset 0x4d8700 contains 24 bytes:
Offset Info Type Sym.Value Sym. Name
00000000 00000000 R_MIPS_NONE
1001c53c 00342603 R_MIPS_REL32 00000000 getpid
100196c4 00566403 R_MIPS_REL32 00000000 printf
i.e. there are five undefined symbols, and I'm guessing that the ones
in the first created GOT don't get REL32 relocations emitted.
A little poking in binutils reveals that this is deliberate. This
comment, from the multigot patch in January:
/* A pointer to the primary got, i.e., the one that's going to get
the implicit relocations from DT_MIPS_LOCAL_GOTNO and
DT_MIPS_GOTSYM. */
struct mips_got_info *primary;
There's only one problem - it doesn't work the way that comment
suggests. got[2] through got[DT_MIPS_LOCAL_GOTNO - 1] are relocated
by l_addr, as local GOT entries. Then got[DT_MIPS_LOCAL_GOTNO] through
got[DT_MIPS_SYMTABNO - 1] are relocated as global GOT entries. So
DT_MIPS_SYMTABNO is probably intended to be the number of symbols _used
by the primary GOT_. Not the total size of .dynsym.
Reading the comments in elfxx-mips.c it looks like this shouldn't
even be happening; the global symbols should end up in the "master
GOT". And there should be stubs. But in my tests the stubs don't get
built, and the value in the GOT for getpid points to func0_4663. And I
can't find code that ought to create them. So I'm going to have to
punt to Alexandre for details. A quick potential fix might be changing
DT_MIPS_SYMTABNO.
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-04 11:02 ` Atsushi Nemoto
@ 2003-11-04 17:53 ` Eric Christopher
2003-11-06 16:54 ` Atsushi Nemoto
2003-11-07 16:40 ` Thiemo Seufer
0 siblings, 2 replies; 16+ messages in thread
From: Eric Christopher @ 2003-11-04 17:53 UTC (permalink / raw)
To: Atsushi Nemoto; +Cc: Thiemo Seufer, jsun, linux-mips, binutils
On Tue, 2003-11-04 at 03:02, Atsushi Nemoto wrote:
> >>>>> On Tue, 04 Nov 2003 00:05:56 -0800, Eric Christopher <echristo@redhat.com> said:
> >> I tried binutils-2.14.90.0.7 (based on binutils 2003 1029 in CVS)
> >> but my problem did no solved. It seems something is still wrong.
>
> eric> This was plain mips-linux? Not mips64-linux?
>
> Yes. mips-linux and mipsel-linux target (host is i386). Both target
> generate broken binary for my test program.
>
> eric> And where would I find the sources?
>
> I'm using plain binutils 2.14 and gcc 3.3.2 from gnu.org FTP site,
> binutils 2.14.90.0.7 from
> http://www.kernel.org/pub/linux/devel/binutils/.
I'm using mainline gcc, but I meant the python-qt sources you were
compiling.
-eric
--
Eric Christopher <echristo@redhat.com>
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-04 17:53 ` Eric Christopher
@ 2003-11-06 16:54 ` Atsushi Nemoto
2003-12-01 5:07 ` Atsushi Nemoto
2003-11-07 16:40 ` Thiemo Seufer
1 sibling, 1 reply; 16+ messages in thread
From: Atsushi Nemoto @ 2003-11-06 16:54 UTC (permalink / raw)
To: echristo; +Cc: ica2_ts, jsun, linux-mips, binutils
>>>>> On Tue, 04 Nov 2003 09:53:06 -0800, Eric Christopher <echristo@redhat.com> said:
echristo> I'm using mainline gcc, but I meant the python-qt sources
echristo> you were compiling.
The link error of phyton-qt was reported by Thiemo Seufer. I have not
tried it.
My problem is runtime failure, not link error. So it may be a
different problem.
---
Atsushi Nemoto
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-04 17:53 ` Eric Christopher
2003-11-06 16:54 ` Atsushi Nemoto
@ 2003-11-07 16:40 ` Thiemo Seufer
2003-11-07 20:51 ` Eric Christopher
1 sibling, 1 reply; 16+ messages in thread
From: Thiemo Seufer @ 2003-11-07 16:40 UTC (permalink / raw)
To: Eric Christopher; +Cc: Atsushi Nemoto, jsun, linux-mips, binutils
Eric Christopher wrote:
[snip]
> > Yes. mips-linux and mipsel-linux target (host is i386). Both target
> > generate broken binary for my test program.
> >
> > eric> And where would I find the sources?
> >
> > I'm using plain binutils 2.14 and gcc 3.3.2 from gnu.org FTP site,
> > binutils 2.14.90.0.7 from
> > http://www.kernel.org/pub/linux/devel/binutils/.
>
> I'm using mainline gcc, but I meant the python-qt sources you were
> compiling.
It was python-qt-3.8 from debian unstable, compiled with
"gcc (GCC) 3.3.2 (Debian)" and binutils
"2.14.90.0.7 20031029 Debian GNU/Linux"
An attempt to link with CVS ld shows the same BFD assertion.
Thiemo
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-07 16:40 ` Thiemo Seufer
@ 2003-11-07 20:51 ` Eric Christopher
0 siblings, 0 replies; 16+ messages in thread
From: Eric Christopher @ 2003-11-07 20:51 UTC (permalink / raw)
To: Thiemo Seufer; +Cc: Atsushi Nemoto, jsun, linux-mips, binutils
> > I'm using mainline gcc, but I meant the python-qt sources you were
> > compiling.
>
> It was python-qt-3.8 from debian unstable, compiled with
> "gcc (GCC) 3.3.2 (Debian)" and binutils
> "2.14.90.0.7 20031029 Debian GNU/Linux"
>
> An attempt to link with CVS ld shows the same BFD assertion.
OK. Well, the one machine I have has current sources for gcc and
binutils so I'll be trying that :)
Pointer to a tarball with the sources?
-eric
--
Eric Christopher <echristo@redhat.com>
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: Huge dynamically linked program does not run on mips-linux
2003-11-06 16:54 ` Atsushi Nemoto
@ 2003-12-01 5:07 ` Atsushi Nemoto
0 siblings, 0 replies; 16+ messages in thread
From: Atsushi Nemoto @ 2003-12-01 5:07 UTC (permalink / raw)
To: linux-mips
>>>>> On Fri, 07 Nov 2003 01:54:21 +0900 (JST), Atsushi Nemoto <anemo@mba.ocn.ne.jp> said:
anemo> My problem is runtime failure, not link error. So it may be a
anemo> different problem.
My problem was solved by recent fix in binutils.
The patch is:
http://sources.redhat.com/ml/binutils/2003-11/msg00469.html
and is already in binutils CVS. Thank you.
---
Atsushi Nemoto
^ permalink raw reply [flat|nested] 16+ messages in thread
end of thread, other threads:[~2003-12-01 5:03 UTC | newest]
Thread overview: 16+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-10-22 8:11 Huge dynamically linked program does not run on mips-linux Atsushi Nemoto
2003-10-29 7:32 ` Atsushi Nemoto
2003-10-29 18:14 ` Jun Sun
2003-10-29 18:15 ` Daniel Jacobowitz
2003-10-30 1:53 ` Thiemo Seufer
2003-10-30 2:25 ` Eric Christopher
2003-11-04 5:21 ` Atsushi Nemoto
2003-11-04 8:05 ` Eric Christopher
2003-11-04 11:02 ` Atsushi Nemoto
2003-11-04 17:53 ` Eric Christopher
2003-11-06 16:54 ` Atsushi Nemoto
2003-12-01 5:07 ` Atsushi Nemoto
2003-11-07 16:40 ` Thiemo Seufer
2003-11-07 20:51 ` Eric Christopher
2003-10-30 12:54 ` Atsushi Nemoto
2003-11-04 16:17 ` Daniel Jacobowitz
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.