Re: Bug#413787: libc0.3: TLS patch

[Thomas Schwinge]

Hello!

On Wed, Mar 07, 2007 at 10:51:21AM +0100, Samuel Thibault wrote:
> Thomas Schwinge, le Wed 07 Mar 2007 10:32:00 +0100, a ?crit :
> > On Wed, Mar 07, 2007 at 03:06:22AM +0100, Samuel Thibault wrote:
> > > Thanks to Barry's long glibc builds, here is at last a patch that builds
> > > a tls/__thread -enabled and working glibc.
> > 
> > Great!
> 
> BTW, by "working", I mean that simple programs do work, event with
> __thread variables.

:-)

I'm just about to start a build of glibc-2_5-branch, even though I won't
have the time to really have a look at it.


> > Can you comment on the issues mentioned at
> > <http://savannah.gnu.org/bugs/?17644#comment2>?
> 
> Mmm, which issues?

The Jeroen patch (<http://www.dekkers.cx/hurd/glibc-tls.patch>)...

#v+
Index: sysdeps/mach/hurd/dl-sysdep.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/mach/hurd/dl-sysdep.c,v
retrieving revision 1.78
diff -u -p -r1.78 dl-sysdep.c
--- sysdeps/mach/hurd/dl-sysdep.c       6 Sep 2004 21:31:38 -0000       1.78
+++ sysdeps/mach/hurd/dl-sysdep.c       18 Sep 2005 13:34:50 -0000
@@ -63,8 +63,10 @@ hp_timing_t _dl_cpuclock_offset;
 struct hurd_startup_data *_dl_hurd_data;
 
 /* This is used only within ld.so, via dl-minimal.c's __errno_location.  */
+#ifndef USE___THREAD
 #undef errno
 int errno attribute_hidden;
+#endif
 
 /* Defining these variables here avoids the inclusion of hurdsig.c.  */
 unsigned long int __hurd_sigthread_stack_base;
Index: sysdeps/mach/hurd/fork.c
===================================================================
RCS file: /cvs/glibc/libc/sysdeps/mach/hurd/fork.c,v
retrieving revision 1.56
diff -u -p -r1.56 fork.c
--- sysdeps/mach/hurd/fork.c    27 Oct 2004 07:47:10 -0000      1.56
+++ sysdeps/mach/hurd/fork.c    18 Sep 2005 13:34:51 -0000
@@ -537,7 +537,7 @@ __fork (void)
        LOSE;
 #endif
 
-      if (err = __thread_set_state (thread, MACHINE_THREAD_STATE_FLAVOR,
+      if (err = __thread_set_state (thread, i386_REGS_SEGS_STATE,
                                    (natural_t *) &state, statecount))
        LOSE;
 
#v-

... and his comments...

#v+
| <Jeroen> http://www.dekkers.cx/hurd/glibc-tls.patch
| <Jeroen> the fork fix isn't the right one
| <Jeroen> because the define is i386-specific
| <Jeroen> we should probably define a MACHINE_THREAD_STATE_FLAVOR2 to
| i386_REGS_SEGS_STATE or something, roland will probably know the
| cleanest solution
| <Jeroen> with that patch you can run programs
| <Jeroen> but dlopen() doesn't work
| <Jeroen> maybe other things don't work either
| <Jeroen> but the basic TLS things work :) 
#v-

Or isn't that relevant anymore?


Regards,
 Thomas

signature.asc
Description: Digital signature