[PATCH] Two more vDSO issues

[PATCH] Two more vDSO issues

[Jakub Jelinek]

Hi!

1) l->ld needs to be relocated as well (I did not catch this up in my
   previous testing, because vDSO was mapped both in the new place and
   in the old location)
2) if kernel provides just AT_SYSINFO_EHDR but not AT_SYSINFO, ld.so would
   segfault.

2004-02-27  Jakub Jelinek  <jakub@redhat.com>

	* elf/rtld.c (dl_main): Adjust l->l_ld of the vDSO by l->l_addr.
	* sysdeps/generic/dl-sysdep.c (_dl_sysdep_start): Only set
	GL(dl_sysinfo) if non-zero.

--- libc/elf/rtld.c.jj	2004-02-27 13:51:54.000000000 +0100
+++ libc/elf/rtld.c	2004-02-27 14:56:05.000000000 +0100
@@ -1246,6 +1246,7 @@ ERROR: ld.so: object '%s' from %s cannot
 	  l->l_map_start = (ElfW(Addr)) GL(dl_sysinfo_dso);
 	  l->l_addr = l->l_map_start - l->l_addr;
 	  l->l_map_end += l->l_addr;
+	  l->l_ld = (void *) ((ElfW(Addr)) l->l_ld + l->l_addr);
 	  elf_get_dynamic_info (l, dyn_temp);
 	  _dl_setup_hash (l);
 	  l->l_relocated = 1;
--- libc/sysdeps/generic/dl-sysdep.c.jj	2004-02-27 17:08:24.000000000 +0100
+++ libc/sysdeps/generic/dl-sysdep.c	2004-02-27 17:18:39.992891935 +0100
@@ -201,7 +201,7 @@ _dl_sysdep_start (void **start_argptr,
 
 #if defined NEED_DL_SYSINFO
   /* Only set the sysinfo value if we also have the vsyscall DSO.  */
-  if (GL(dl_sysinfo_dso) != 0)
+  if (GL(dl_sysinfo_dso) != 0 && new_sysinfo)
     GL(dl_sysinfo) = new_sysinfo;
 #endif
 

	Jakub

Re: [PATCH] Two more vDSO issues

[Thorsten Kukuk]

On Fri, Feb 27, Jakub Jelinek wrote:

> Hi!
> 
> 1) l->ld needs to be relocated as well (I did not catch this up in my
>    previous testing, because vDSO was mapped both in the new place and
>    in the old location)
> 2) if kernel provides just AT_SYSINFO_EHDR but not AT_SYSINFO, ld.so would
>    segfault.

Even with this patches, glibc always seg.faults if compiled with
NPTL support on IA64 when calling rpcgen the first time during build.

It works fine with linuxthreads and I don't see problems on other
architectures.

  Thorsten

-- 
Thorsten Kukuk       http://www.suse.de/~kukuk/        kukuk@suse.de
SuSE Linux AG        Maxfeldstr. 5                 D-90409 Nuernberg
--------------------------------------------------------------------    
Key fingerprint = A368 676B 5E1B 3E46 CFCE  2D97 F8FD 4E23 56C6 FB4B

Re: [PATCH] Two more vDSO issues

[Ulrich Drepper]

Applied.

-- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖

Re: [PATCH] Two more vDSO issues

[Andreas Schwab]

Thorsten Kukuk <kukuk@suse.de> writes:

> On Fri, Feb 27, Jakub Jelinek wrote:
>
>> Hi!
>> 
>> 1) l->ld needs to be relocated as well (I did not catch this up in my
>>    previous testing, because vDSO was mapped both in the new place and
>>    in the old location)
>> 2) if kernel provides just AT_SYSINFO_EHDR but not AT_SYSINFO, ld.so would
>>    segfault.
>
> Even with this patches, glibc always seg.faults if compiled with
> NPTL support on IA64 when calling rpcgen the first time during build.

Note that it only crashes when starting a program by explicitly calling
the dynamic linker, it appears that __thread_self->private is not
correctly set.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH] Two more vDSO issues

[Andreas Schwab]

Jakub Jelinek <jakub@redhat.com> writes:

> 2) if kernel provides just AT_SYSINFO_EHDR but not AT_SYSINFO, ld.so would
>    segfault.

Here's another bugfix:

2004-03-08  Andreas Schwab  <schwab@suse.de>

	* sysdeps/generic/dl-sysdep.c (_dl_sysdep_start): Don't truncate
	sysinfo pointer value.

--- sysdeps/generic/dl-sysdep.c.~1.110.~	2004-03-06 22:46:14.000000000 +0100
+++ sysdeps/generic/dl-sysdep.c	2004-03-08 14:36:01.934030411 +0100
@@ -99,7 +99,7 @@ _dl_sysdep_start (void **start_argptr,
 # endif
 #endif
 #ifdef NEED_DL_SYSINFO
-  ElfW(Word) new_sysinfo = 0;
+  uintptr_t new_sysinfo = 0;
 #endif
 
   __libc_stack_end = DL_STACK_END (start_argptr);

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH] Two more vDSO issues

[Ulrich Drepper]

Andreas Schwab wrote:

>>Even with this patches, glibc always seg.faults if compiled with
>>NPTL support on IA64 when calling rpcgen the first time during build.
> 
> 
> Note that it only crashes when starting a program by explicitly calling
> the dynamic linker, it appears that __thread_self->private is not
> correctly set.

I compiled on ia64 just yesterday and all worked beautifully.  No idea
what you run into.

-- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖

Re: [PATCH] Two more vDSO issues

[Andreas Schwab]

Ulrich Drepper <drepper@redhat.com> writes:

> Andreas Schwab wrote:
>
>>>Even with this patches, glibc always seg.faults if compiled with
>>>NPTL support on IA64 when calling rpcgen the first time during build.
>> 
>> 
>> Note that it only crashes when starting a program by explicitly calling
>> the dynamic linker, it appears that __thread_self->private is not
>> correctly set.
>
> I compiled on ia64 just yesterday and all worked beautifully.  No idea
> what you run into.

You didn't have USE_DL_SYSINFO defined.  The bug is obvious.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [PATCH] Two more vDSO issues

[Ulrich Drepper]

Andreas Schwab wrote:

> 	* sysdeps/generic/dl-sysdep.c (_dl_sysdep_start): Don't truncate
> 	sysinfo pointer value.

Applied.

-- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖