emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Damien Deville]

Hi emacs-devel,

On FreeBSD 6.2 with an up to date checkout of emacs-unicode-2 CVS temacs 
coredump while bootstraping.

`/bin/pwd`/temacs --batch --load loadup bootstrap
Segmentation fault (core dumped)
gmake[2]: *** [bootstrap-emacs] Error 139

Launching gdb on the core shows the following trace:

(gdb) bt
#0  0x28a007a0 in pthread_once () from /lib/libpthread.so.2
#1  0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
#2  0x081a6ed1 in malloc (size=68) at gmalloc.c:969
#3  0x28a00c0d in pthread_mutex_init () from /lib/libpthread.so.2
#4  0x28a00daf in _pthread_mutex_init () from /lib/libpthread.so.2
#5  0x28a0a8a5 in pthread_setconcurrency () from /lib/libpthread.so.2
#6  0x28a045be in _pthread_mutex_lock () from /lib/libpthread.so.2
#7  0x28a007ca in pthread_once () from /lib/libpthread.so.2
#8  0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
#9  0x081a6ed1 in malloc (size=68) at gmalloc.c:969
#10 0x28a00c0d in pthread_mutex_init () from /lib/libpthread.so.2
#11 0x28a00daf in _pthread_mutex_init () from /lib/libpthread.so.2
#12 0x28a0a8a5 in pthread_setconcurrency () from /lib/libpthread.so.2
#13 0x28a045be in _pthread_mutex_lock () from /lib/libpthread.so.2
#14 0x28a007ca in pthread_once () from /lib/libpthread.so.2
#15 0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
#16 0x081a6ed1 in malloc (size=68) at gmalloc.c:969
#17 0x28a00c0d in pthread_mutex_init () from /lib/libpthread.so.2
#18 0x28a00daf in _pthread_mutex_init () from /lib/libpthread.so.2
#19 0x28a0a8a5 in pthread_setconcurrency () from /lib/libpthread.so.2
#20 0x28a045be in _pthread_mutex_lock () from /lib/libpthread.so.2
#21 0x28a007ca in pthread_once () from /lib/libpthread.so.2
#22 0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
---Type <return> to continue, or q <return> to quit---

The rest of the backtrace contains the same recursive call to 
__malloc_initialize()


Thanks in advance,
Damien

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan Djärv]

I've checked in a fix in CVS HEAD.  It should propagate to unicode2, please
try it then.

	Jan D.

Damien Deville skrev:
> Hi emacs-devel,
> 
> On FreeBSD 6.2 with an up to date checkout of emacs-unicode-2 CVS temacs
> coredump while bootstraping.
> 
> `/bin/pwd`/temacs --batch --load loadup bootstrap
> Segmentation fault (core dumped)
> gmake[2]: *** [bootstrap-emacs] Error 139
> 
> Launching gdb on the core shows the following trace:
> 
> (gdb) bt
> #0  0x28a007a0 in pthread_once () from /lib/libpthread.so.2
> #1  0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
> #2  0x081a6ed1 in malloc (size=68) at gmalloc.c:969
> #3  0x28a00c0d in pthread_mutex_init () from /lib/libpthread.so.2
> #4  0x28a00daf in _pthread_mutex_init () from /lib/libpthread.so.2
> #5  0x28a0a8a5 in pthread_setconcurrency () from /lib/libpthread.so.2
> #6  0x28a045be in _pthread_mutex_lock () from /lib/libpthread.so.2
> #7  0x28a007ca in pthread_once () from /lib/libpthread.so.2
> #8  0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
> #9  0x081a6ed1 in malloc (size=68) at gmalloc.c:969
> #10 0x28a00c0d in pthread_mutex_init () from /lib/libpthread.so.2
> #11 0x28a00daf in _pthread_mutex_init () from /lib/libpthread.so.2
> #12 0x28a0a8a5 in pthread_setconcurrency () from /lib/libpthread.so.2
> #13 0x28a045be in _pthread_mutex_lock () from /lib/libpthread.so.2
> #14 0x28a007ca in pthread_once () from /lib/libpthread.so.2
> #15 0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
> #16 0x081a6ed1 in malloc (size=68) at gmalloc.c:969
> #17 0x28a00c0d in pthread_mutex_init () from /lib/libpthread.so.2
> #18 0x28a00daf in _pthread_mutex_init () from /lib/libpthread.so.2
> #19 0x28a0a8a5 in pthread_setconcurrency () from /lib/libpthread.so.2
> #20 0x28a045be in _pthread_mutex_lock () from /lib/libpthread.so.2
> #21 0x28a007ca in pthread_once () from /lib/libpthread.so.2
> #22 0x081a6ea1 in __malloc_initialize () at gmalloc.c:615
> ---Type <return> to continue, or q <return> to quit---
> 
> The rest of the backtrace contains the same recursive call to
> __malloc_initialize()
> 
> 
> Thanks in advance,
> Damien
> 
> 
> _______________________________________________
> Emacs-devel mailing list
> Emacs-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/emacs-devel

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Herbert J. Skuhra]

Jan Djärv skrev:
> I've checked in a fix in CVS HEAD.  It should propagate to unicode2, please
> try it then.
>
> 	Jan D.

Thanks. But the problem persists in CVS HEAD:

% ./configure --prefix=/usr/local/emacs --with-gtk --build=i386-freebsd
% make bootstrap

Build error:
./temacs --batch --load loadup bootstrap
*** Signal 11

(gdb) run
Starting program: /home/herbert/cvs/tmp/emacs/src/temacs --batch --load
loadup bootstrap

Program received signal SIGSEGV, Segmentation fault.
0x289fcd86 in _pthread_mutex_init () from /lib/libpthread.so.2

#0  0x289fcd86 in _pthread_mutex_init () from /lib/libpthread.so.2
#1  0x28a068a5 in pthread_setconcurrency () from /lib/libpthread.so.2
#2  0x28a007aa in pthread_mutex_lock () from /lib/libpthread.so.2
#3  0x0818c465 in _malloc_internal (size=68) at gmalloc.c:953
#4  0x289fcc0d in pthread_mutex_init () from /lib/libpthread.so.2
#5  0x289fcdaf in _pthread_mutex_init () from /lib/libpthread.so.2
#6  0x28a068a5 in pthread_setconcurrency () from /lib/libpthread.so.2
#7  0x28a007aa in pthread_mutex_lock () from /lib/libpthread.so.2
#8  0x0818c465 in _malloc_internal (size=68) at gmalloc.c:953
#9  0x289fcc0d in pthread_mutex_init () from /lib/libpthread.so.2
#10 0x289fcdaf in _pthread_mutex_init () from /lib/libpthread.so.2
#11 0x28a068a5 in pthread_setconcurrency () from /lib/libpthread.so.2
#12 0x28a007aa in pthread_mutex_lock () from /lib/libpthread.so.2
#13 0x0818c465 in _malloc_internal (size=68) at gmalloc.c:953
#14 0x289fcc0d in pthread_mutex_init () from /lib/libpthread.so.2
...
...
...

When "--with-x-toolkit=motif" is used, make bootstrap is OK.
Hmm, Openmotif doesn't look that bad! :)

BR
Herbert

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Tue, 31 Jul 2007 21:09:55 +0200, "Herbert J. Skuhra" <h.skuhra@gmail.com> said:

> Jan Djärv skrev:
>> I've checked in a fix in CVS HEAD.  It should propagate to unicode2, please
>> try it then.
>> 
>> Jan D.

> Thanks. But the problem persists in CVS HEAD:

> % ./configure --prefix=/usr/local/emacs --with-gtk --build=i386-freebsd
> % make bootstrap

> Build error:
> ./temacs --batch --load loadup bootstrap
> *** Signal 11

A comment in src/lib/libc/stdlib/malloc.c (FreeBSD Current) says:

/*
 * Mutexes based on spinlocks.  We can't use normal pthread mutexes, because
 * they require malloc()ed memory.
 */

So it seems to be impossible to make src/gmalloc.c thread-safe only
using pthread mutexes.

I'm not sure if spinlocks are open to user applications, but anyway it
would be simpler to define SYSTEM_MALLOC when HAVE_GTK_AND_PTHREAD &&
!DOUG_LEA_MALLOC, and abandon emacs_blocked_malloc etc. on such
platforms.  I don't care if my changes made to src/gmalloc.c for
thread-safety is reverted.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan Djärv]

YAMAMOTO Mitsuharu skrev:

> 
> A comment in src/lib/libc/stdlib/malloc.c (FreeBSD Current) says:
> 
> /*
>  * Mutexes based on spinlocks.  We can't use normal pthread mutexes, because
>  * they require malloc()ed memory.
>  */
> 
> So it seems to be impossible to make src/gmalloc.c thread-safe only
> using pthread mutexes.

Totally thread safe is impossible, but for Emacs we don't need that.  It 
should be OK to initialize gmalloc in a non-thread safe way because it is done 
in main (or possibly before) before any threads are created.

> 
> I'm not sure if spinlocks are open to user applications, but anyway it
> would be simpler to define SYSTEM_MALLOC when HAVE_GTK_AND_PTHREAD &&
> !DOUG_LEA_MALLOC, and abandon emacs_blocked_malloc etc. on such
> platforms.  I don't care if my changes made to src/gmalloc.c for
> thread-safety is reverted.

That is always a possibility.  I'll look into both.

	Jan D.

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Wed, 01 Aug 2007 11:20:26 +0200, Jan Djärv <jan.h.d@swipnet.se> said:

> YAMAMOTO Mitsuharu skrev:

>> 
>> A comment in src/lib/libc/stdlib/malloc.c (FreeBSD Current) says:
>> 
>> /* * Mutexes based on spinlocks.  We can't use normal pthread
>> mutexes, because * they require malloc()ed memory.  */
>> 
>> So it seems to be impossible to make src/gmalloc.c thread-safe only
>> using pthread mutexes.

> Totally thread safe is impossible, but for Emacs we don't need that.
> It should be OK to initialize gmalloc in a non-thread safe way
> because it is done in main (or possibly before) before any threads
> are created.

I didn't mean the initialization problem with respect to pthread_once.
The followup report shows that there's another infinite recursion like
pthread_mutex_lock -> -> malloc -> -> pthread_mutex_lock -> ... when a
certain (statically initialized) mutex is used first.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Wed, 01 Aug 2007 18:35:40 +0900, YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> said:

>> Totally thread safe is impossible, but for Emacs we don't need
>> that.  It should be OK to initialize gmalloc in a non-thread safe
>> way because it is done in main (or possibly before) before any
>> threads are created.

> I didn't mean the initialization problem with respect to
> pthread_once.  The followup report shows that there's another
> infinite recursion like pthread_mutex_lock -> -> malloc -> ->
> pthread_mutex_lock -> ... when a certain (statically initialized)
> mutex is used first.

Ah, maybe I misunderstood what you mean.  It might be possible to call
pthread_mutex_init for all malloc mutexes at the (non-thread-safe)
initialization stage with temporarily disabling the use of the
mutexes.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Wed, 01 Aug 2007 19:07:47 +0900, YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> said:

> It might be possible to call pthread_mutex_init for all malloc
> mutexes at the (non-thread-safe) initialization stage with
> temporarily disabling the use of the mutexes.

Could you try if this works?  As usual, I can't test it myself.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Index: src/gmalloc.c
===================================================================
RCS file: /cvsroot/emacs/emacs/src/gmalloc.c,v
retrieving revision 1.24
diff -c -p -r1.24 gmalloc.c
*** src/gmalloc.c	29 Jul 2007 10:12:21 -0000	1.24
--- src/gmalloc.c	1 Aug 2007 10:35:05 -0000
*************** extern void _free_internal_nolock PP ((_
*** 242,249 ****
  
  #ifdef USE_PTHREAD
  extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
! #define LOCK()     pthread_mutex_lock (&_malloc_mutex)
! #define UNLOCK()   pthread_mutex_unlock (&_malloc_mutex)
  #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
  #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
  #else
--- 242,256 ----
  
  #ifdef USE_PTHREAD
  extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
! extern int _malloc_mutex_disabled_p;
! #define LOCK() \
!   do { \
!     if (!_malloc_mutex_disabled_p) pthread_mutex_lock (&_malloc_mutex); \
!   } while (0)
! #define UNLOCK() \
!   do { \
!     if (!_malloc_mutex_disabled_p) pthread_mutex_unlock (&_malloc_mutex); \
!   } while (0)
  #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
  #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
  #else
*************** register_heapinfo ()
*** 563,568 ****
--- 570,576 ----
  #ifdef USE_PTHREAD
  pthread_mutex_t _malloc_mutex = PTHREAD_MUTEX_INITIALIZER;
  pthread_mutex_t _aligned_blocks_mutex = PTHREAD_MUTEX_INITIALIZER;
+ int _malloc_mutex_disabled_p;
  #endif
  
  static void
*************** __malloc_initialize ()
*** 617,622 ****
--- 625,641 ----
  
    malloc_initialize_1 ();
  
+   /* Some pthread implementations call malloc for statically
+      initialized mutexes when they are used first.  To avoid such a
+      situation, we initialize mutexes here with temporarily disabling
+      the use of mutexes.  */
+ #ifdef USE_PTHREAD
+   _malloc_mutex_disabled_p = 1;
+   pthread_mutex_init (&_malloc_mutex, NULL);
+   pthread_mutex_init (&_aligned_blocks_mutex, NULL);
+   _malloc_mutex_disabled_p = 0;
+ #endif
+ 
    return __malloc_initialized;
  }

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan Djärv]

YAMAMOTO Mitsuharu skrev:

> Ah, maybe I misunderstood what you mean.  It might be possible to call
> pthread_mutex_init for all malloc mutexes at the (non-thread-safe)
> initialization stage with temporarily disabling the use of the
> mutexes.

Exactly.

	Jan D.

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan Djärv]

YAMAMOTO Mitsuharu skrev:
>>>>>> On Wed, 01 Aug 2007 19:07:47 +0900, YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> said:
> 
>> It might be possible to call pthread_mutex_init for all malloc
>> mutexes at the (non-thread-safe) initialization stage with
>> temporarily disabling the use of the mutexes.
> 
> Could you try if this works?  As usual, I can't test it myself.

I think we should call LOCK/UNLOCK on them also just to be sure.
You never know what those BDS guys are up to :-)

	Jan D.

> 
> 				     YAMAMOTO Mitsuharu
> 				mituharu@math.s.chiba-u.ac.jp
> 
> Index: src/gmalloc.c
> ===================================================================
> RCS file: /cvsroot/emacs/emacs/src/gmalloc.c,v
> retrieving revision 1.24
> diff -c -p -r1.24 gmalloc.c
> *** src/gmalloc.c	29 Jul 2007 10:12:21 -0000	1.24
> --- src/gmalloc.c	1 Aug 2007 10:35:05 -0000
> *************** extern void _free_internal_nolock PP ((_
> *** 242,249 ****
>   
>   #ifdef USE_PTHREAD
>   extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
> ! #define LOCK()     pthread_mutex_lock (&_malloc_mutex)
> ! #define UNLOCK()   pthread_mutex_unlock (&_malloc_mutex)
>   #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
>   #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
>   #else
> --- 242,256 ----
>   
>   #ifdef USE_PTHREAD
>   extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
> ! extern int _malloc_mutex_disabled_p;
> ! #define LOCK() \
> !   do { \
> !     if (!_malloc_mutex_disabled_p) pthread_mutex_lock (&_malloc_mutex); \
> !   } while (0)
> ! #define UNLOCK() \
> !   do { \
> !     if (!_malloc_mutex_disabled_p) pthread_mutex_unlock (&_malloc_mutex); \
> !   } while (0)
>   #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
>   #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
>   #else
> *************** register_heapinfo ()
> *** 563,568 ****
> --- 570,576 ----
>   #ifdef USE_PTHREAD
>   pthread_mutex_t _malloc_mutex = PTHREAD_MUTEX_INITIALIZER;
>   pthread_mutex_t _aligned_blocks_mutex = PTHREAD_MUTEX_INITIALIZER;
> + int _malloc_mutex_disabled_p;
>   #endif
>   
>   static void
> *************** __malloc_initialize ()
> *** 617,622 ****
> --- 625,641 ----
>   
>     malloc_initialize_1 ();
>   
> +   /* Some pthread implementations call malloc for statically
> +      initialized mutexes when they are used first.  To avoid such a
> +      situation, we initialize mutexes here with temporarily disabling
> +      the use of mutexes.  */
> + #ifdef USE_PTHREAD
> +   _malloc_mutex_disabled_p = 1;
> +   pthread_mutex_init (&_malloc_mutex, NULL);
> +   pthread_mutex_init (&_aligned_blocks_mutex, NULL);
> +   _malloc_mutex_disabled_p = 0;
> + #endif
> + 
>     return __malloc_initialized;
>   }
>   
> 
> 
> _______________________________________________
> Emacs-devel mailing list
> Emacs-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/emacs-devel

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan Djärv]

Jan Djärv skrev:
> 
> 
> I think we should call LOCK/UNLOCK on them also just to be sure.
> You never know what those BDS guys are up to :-)

----------------------------^^^---  BSD obviously.

	Jan D.

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Miles Bader]

Jan Djärv <jan.h.d@swipnet.se> writes:
>> I think we should call LOCK/UNLOCK on them also just to be sure.
>> You never know what those BDS guys are up to :-)
>
> ----------------------------^^^---  BSD obviously.

You've clearly done too much LDS...

-Miles
-- 
I'd rather be consing.

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan D.]

YAMAMOTO Mitsuharu wrote:
>>>>>> On Wed, 01 Aug 2007 19:07:47 +0900, YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> said:
>>>>>>             
>
>   
>> It might be possible to call pthread_mutex_init for all malloc
>> mutexes at the (non-thread-safe) initialization stage with
>> temporarily disabling the use of the mutexes.
>>     
>
> Could you try if this works?  As usual, I can't test it myself.
>   

I tried it and temacs runs OK.  However, it seems that the mutex 
initialization is lost, so it has to be done again with the dumped 
emacs.  pthread_self also allocates memory, so even if this is fixed we 
still get a loop.  I also tried using system malloc on FreeBSD 6.2, but 
that didn't work.  Emacs crashed on first realloc (heap corrupt).  I 
guess undumping on FreeBSD with system malloc is broken (or perhaps 
impossible).

There are spin locks available on BSD and GNU/Linux, see 
pthread_spin_init.  But that don't help for the pthread_self issue.  
Sigh, back to the drawing board.

    Jan D.

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Wed, 01 Aug 2007 20:27:06 +0200, "Jan D." <jan.h.d@swipnet.se> said:

> I tried it and temacs runs OK.  However, it seems that the mutex 
> initialization is lost, so it has to be done again with the dumped
> emacs.  pthread_self also allocates memory, so even if this is fixed
> we still get a loop.

How about deferring mutex initializations until an interactive session
starts?  Could you try the patch below?

> I also tried using system malloc on FreeBSD 6.2, but that didn't
> work.  Emacs crashed on first realloc (heap corrupt).  I guess
> undumping on FreeBSD with system malloc is broken (or perhaps
> impossible).

Too bad...

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Index: src/emacs.c
===================================================================
RCS file: /cvsroot/emacs/emacs/src/emacs.c,v
retrieving revision 1.404
diff -c -p -r1.404 emacs.c
*** src/emacs.c	26 Jul 2007 05:27:51 -0000	1.404
--- src/emacs.c	6 Aug 2007 08:40:40 -0000
*************** main (argc, argv
*** 1164,1169 ****
--- 1164,1176 ----
        setpgrp ();
  #endif
  #endif
+ #if defined (HAVE_GTK_AND_PTHREAD) && !defined (SYSTEM_MALLOC) && !defined (DOUG_LEA_MALLOC)
+       {
+ 	extern void malloc_enable_thread P_ ((void));
+ 
+ 	malloc_enable_thread ();
+       }
+ #endif
      }
  
    init_signals ();
Index: src/gmalloc.c
===================================================================
RCS file: /cvsroot/emacs/emacs/src/gmalloc.c,v
retrieving revision 1.24
diff -c -p -r1.24 gmalloc.c
*** src/gmalloc.c	29 Jul 2007 10:12:21 -0000	1.24
--- src/gmalloc.c	6 Aug 2007 08:40:40 -0000
*************** extern __ptr_t memalign PP ((__malloc_si
*** 136,141 ****
--- 136,145 ----
  extern __ptr_t valloc PP ((__malloc_size_t __size));
  #endif
  
+ #ifdef USE_PTHREAD
+ /* Set up mutexes and make malloc etc. thread-safe.  */
+ extern void malloc_enable_thread PP ((void));
+ #endif
  
  #ifdef _MALLOC_INTERNAL
  
*************** extern void _free_internal_nolock PP ((_
*** 242,249 ****
  
  #ifdef USE_PTHREAD
  extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
! #define LOCK()     pthread_mutex_lock (&_malloc_mutex)
! #define UNLOCK()   pthread_mutex_unlock (&_malloc_mutex)
  #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
  #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
  #else
--- 246,260 ----
  
  #ifdef USE_PTHREAD
  extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
! extern int _malloc_thread_enabled_p;
! #define LOCK() \
!   do { \
!     if (_malloc_thread_enabled_p) pthread_mutex_lock (&_malloc_mutex); \
!   } while (0)
! #define UNLOCK() \
!   do { \
!     if (_malloc_thread_enabled_p) pthread_mutex_unlock (&_malloc_mutex); \
!   } while (0)
  #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
  #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
  #else
*************** register_heapinfo ()
*** 563,568 ****
--- 574,596 ----
  #ifdef USE_PTHREAD
  pthread_mutex_t _malloc_mutex = PTHREAD_MUTEX_INITIALIZER;
  pthread_mutex_t _aligned_blocks_mutex = PTHREAD_MUTEX_INITIALIZER;
+ int _malloc_thread_enabled_p;
+ 
+ /* Set up mutexes and make malloc etc. thread-safe.  */
+ void
+ malloc_enable_thread ()
+ {
+   if (_malloc_thread_enabled_p)
+     return;
+ 
+   /* Some pthread implementations call malloc for statically
+      initialized mutexes when they are used first.  To avoid such a
+      situation, we initialize mutexes here while their use is
+      disabled in malloc etc.  */
+   pthread_mutex_init (&_malloc_mutex, NULL);
+   pthread_mutex_init (&_aligned_blocks_mutex, NULL);
+   _malloc_thread_enabled_p = 1;
+ }
  #endif
  
  static void

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Mon, 06 Aug 2007 17:47:55 +0900, YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp> said:

> How about deferring mutex initializations until an interactive
> session starts?  Could you try the patch below?

(UN)LOCK_ALIGNED_BLOCKS was not conditionalized.  Please try this one
instead.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp

Index: src/emacs.c
===================================================================
RCS file: /cvsroot/emacs/emacs/src/emacs.c,v
retrieving revision 1.404
diff -c -p -r1.404 emacs.c
*** src/emacs.c	26 Jul 2007 05:27:51 -0000	1.404
--- src/emacs.c	6 Aug 2007 09:34:26 -0000
*************** main (argc, argv
*** 1164,1169 ****
--- 1164,1176 ----
        setpgrp ();
  #endif
  #endif
+ #if defined (HAVE_GTK_AND_PTHREAD) && !defined (SYSTEM_MALLOC) && !defined (DOUG_LEA_MALLOC)
+       {
+ 	extern void malloc_enable_thread P_ ((void));
+ 
+ 	malloc_enable_thread ();
+       }
+ #endif
      }
  
    init_signals ();
Index: src/gmalloc.c
===================================================================
RCS file: /cvsroot/emacs/emacs/src/gmalloc.c,v
retrieving revision 1.24
diff -c -p -r1.24 gmalloc.c
*** src/gmalloc.c	29 Jul 2007 10:12:21 -0000	1.24
--- src/gmalloc.c	6 Aug 2007 09:34:26 -0000
*************** extern __ptr_t memalign PP ((__malloc_si
*** 136,141 ****
--- 136,145 ----
  extern __ptr_t valloc PP ((__malloc_size_t __size));
  #endif
  
+ #ifdef USE_PTHREAD
+ /* Set up mutexes and make malloc etc. thread-safe.  */
+ extern void malloc_enable_thread PP ((void));
+ #endif
  
  #ifdef _MALLOC_INTERNAL
  
*************** extern void _free_internal_nolock PP ((_
*** 242,251 ****
  
  #ifdef USE_PTHREAD
  extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
! #define LOCK()     pthread_mutex_lock (&_malloc_mutex)
! #define UNLOCK()   pthread_mutex_unlock (&_malloc_mutex)
! #define LOCK_ALIGNED_BLOCKS()     pthread_mutex_lock (&_aligned_blocks_mutex)
! #define UNLOCK_ALIGNED_BLOCKS()   pthread_mutex_unlock (&_aligned_blocks_mutex)
  #else
  #define LOCK()
  #define UNLOCK()
--- 246,272 ----
  
  #ifdef USE_PTHREAD
  extern pthread_mutex_t _malloc_mutex, _aligned_blocks_mutex;
! extern int _malloc_thread_enabled_p;
! #define LOCK()					\
!   do {						\
!     if (_malloc_thread_enabled_p)		\
!       pthread_mutex_lock (&_malloc_mutex);	\
!   } while (0)
! #define UNLOCK()				\
!   do {						\
!     if (_malloc_thread_enabled_p)		\
!       pthread_mutex_unlock (&_malloc_mutex);	\
!   } while (0)
! #define LOCK_ALIGNED_BLOCKS()				\
!   do {							\
!     if (_malloc_thread_enabled_p)			\
!       pthread_mutex_lock (&_aligned_blocks_mutex);	\
!   } while (0)
! #define UNLOCK_ALIGNED_BLOCKS()				\
!   do {							\
!     if (_malloc_thread_enabled_p)			\
!       pthread_mutex_unlock (&_aligned_blocks_mutex);	\
!   } while (0)
  #else
  #define LOCK()
  #define UNLOCK()
*************** register_heapinfo ()
*** 563,568 ****
--- 584,630 ----
  #ifdef USE_PTHREAD
  pthread_mutex_t _malloc_mutex = PTHREAD_MUTEX_INITIALIZER;
  pthread_mutex_t _aligned_blocks_mutex = PTHREAD_MUTEX_INITIALIZER;
+ int _malloc_thread_enabled_p;
+ 
+ static void
+ malloc_atfork_handler_prepare ()
+ {
+   LOCK ();
+   LOCK_ALIGNED_BLOCKS ();
+ }
+ 
+ static void
+ malloc_atfork_handler_parent ()
+ {
+   UNLOCK_ALIGNED_BLOCKS ();
+   UNLOCK ();
+ }
+ 
+ static void
+ malloc_atfork_handler_child ()
+ {
+   UNLOCK_ALIGNED_BLOCKS ();
+   UNLOCK ();
+ }
+ 
+ /* Set up mutexes and make malloc etc. thread-safe.  */
+ void
+ malloc_enable_thread ()
+ {
+   if (_malloc_thread_enabled_p)
+     return;
+ 
+   /* Some pthread implementations call malloc for statically
+      initialized mutexes when they are used first.  To avoid such a
+      situation, we initialize mutexes here while their use is
+      disabled in malloc etc.  */
+   pthread_mutex_init (&_malloc_mutex, NULL);
+   pthread_mutex_init (&_aligned_blocks_mutex, NULL);
+   pthread_atfork (malloc_atfork_handler_prepare,
+ 		  malloc_atfork_handler_parent,
+ 		  malloc_atfork_handler_child);
+   _malloc_thread_enabled_p = 1;
+ }
  #endif
  
  static void

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Herbert J. Skuhra]

On Mon, Aug 06, 2007 at 07:04:08PM +0900, YAMAMOTO Mitsuharu wrote:
> >>>>> On Mon, 06 Aug 2007 17:47:55 +0900, YAMAMOTO Mitsuharu
> >>>>> <mituharu@math.s.chiba-u.ac.jp> said:
> 
> > How about deferring mutex initializations until an interactive
> > session starts?  Could you try the patch below?
> 
> (UN)LOCK_ALIGNED_BLOCKS was not conditionalized.  Please try this one
> instead.

Arigatou.

With this patch the emacs build completes again on FreeBSD 6.2-STABLE!

---
This is GNU Emacs 22.1.50.1 (i386-pc-freebsd, GTK+ Version 2.10.14)
 of 2007-08-06 
---

Regards,
Herbert

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Ryan Yeske]

"Herbert J. Skuhra" <h.skuhra@gmail.com> writes:

>> (UN)LOCK_ALIGNED_BLOCKS was not conditionalized.  Please try this one
>> instead.
>
> Arigatou.
>
> With this patch the emacs build completes again on FreeBSD 6.2-STABLE!

And on OpenBSD 4.1/i386, which was having similar build problems before.

Thanks!

Ryan

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[Jan Djärv]

Ryan Yeske skrev:
> "Herbert J. Skuhra" <h.skuhra@gmail.com> writes:
> 
>>> (UN)LOCK_ALIGNED_BLOCKS was not conditionalized.  Please try this one
>>> instead.
>> Arigatou.
>>
>> With this patch the emacs build completes again on FreeBSD 6.2-STABLE!
> 
> And on OpenBSD 4.1/i386, which was having similar build problems before.
> 
> Thanks!
> 

This should go into both HEAD and EMACS_22_BASE.

	Jan D.

Re: emacs-unicode-2 bootstrap on FreeBSD (temacs coredump)

[YAMAMOTO Mitsuharu]

>>>>> On Tue, 07 Aug 2007 10:05:37 +0200, Jan Djärv <jan.h.d@swipnet.se> said:

> Ryan Yeske skrev:
>> "Herbert J. Skuhra" <h.skuhra@gmail.com> writes:
>> 
>>>> (UN)LOCK_ALIGNED_BLOCKS was not conditionalized.  Please try this
>>>> one instead.
>>> Arigatou.
>>> 
>>> With this patch the emacs build completes again on FreeBSD
>>> 6.2-STABLE!
>> 
>> And on OpenBSD 4.1/i386, which was having similar build problems
>> before.
>> 
>> Thanks!
>> 

> This should go into both HEAD and EMACS_22_BASE.

Done.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp