Re: pkg/39012 (boehm-gc coredumps on GC_INIT under hpcmips (mipsel))

[David Holland <dholland-pbugs%netbsd.org@localhost>]

The following reply was made to PR pkg/39012; it has been noted by GNATS.

From: David Holland <dholland-pbugs%netbsd.org@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: pkg/39012 (boehm-gc coredumps on GC_INIT under hpcmips (mipsel))
Date: Wed, 31 Oct 2012 20:04:52 +0000

 These two mails were never entered into gnats:
 
    ------
 
 From: ben%wongs.net@localhost
 To: recht%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, 
pkgsrc-bugs%netbsd.org@localhost,
        ben%wongs.net@localhost
 Subject: Re: pkg/39012 (boehm-gc coredumps on GC_INIT under hpcmips (mipsel))
 Date: Mon, 7 Jul 2008 06:15:42 -0700
 
 Additional information:
 
 I have tried building boehm-gc from the latest pkgsrc and can now
 provide a more precise backtrace.
 
   Script started on Mon Jul  7 05:30:12 2008
   bash-3.2# gdb ./a.out
   GNU gdb 5.3nb1
   Copyright 2002 Free Software Foundation, Inc.
   This GDB was configured as "mipsel--netbsd"...
   (gdb) run
   Starting program: /usr/pkgsrc/devel/boehm-gc/work/gc-7.0/a.out
 
   Program received signal SIGSEGV, Segmentation fault.
   0x7df456e8 in GC_find_limit_with_bound () at os_dep.c:917
   917                  GC_noop1((word)(*result));
   (gdb) c
   Continuing.
 
   Program received signal SIGSEGV, Segmentation fault.
   0x7df456e8 in GC_find_limit_with_bound () at os_dep.c:917
   917                  GC_noop1((word)(*result));
   (gdb) c
   Continuing.
 
   Program received signal SIGSEGV, Segmentation fault.
   0x7df3b920 in GC_FirstDLOpenedLinkMap () at dyn_load.c:455
   455                          = ((struct r_debug *)(dp->d_un.d_ptr))->r_map;
 
   (gdb) list
   450      if( cachedResult == 0 ) {
   451          int tag;
   452          for( dp = _DYNAMIC; (tag = dp->d_tag) != 0; dp++ ) {
   453              if( tag == DT_DEBUG ) {
   454                  struct link_map *lm
   455                          = ((struct r_debug *)(dp->d_un.d_ptr))->r_map;
   456                  if( lm != 0 ) cachedResult = lm->l_next; /*
 might be NIL */
   457                  break;
   458              }
   459          }
 
   (gdb) bt
   #0  0x7df3b920 in GC_FirstDLOpenedLinkMap () at dyn_load.c:455
   #1  0x7df3b9f0 in GC_register_dynamic_libraries () at dyn_load.c:475
   #2  0x7df42cd0 in GC_push_roots () from ./.libs/libgc.so.1
   #3  0x7df422d4 in GC_mark_some () from ./.libs/libgc.so.1
   #4  0x7df3809c in GC_stopped_mark () from ./.libs/libgc.so.1
   #5  0x7df384d8 in GC_try_to_collect_inner () from ./.libs/libgc.so.1
   #6  0x00400adc in main (argc=1, argv=0x7fffdca4, envp=0x7fffdcac) at
 gctest.c:7
   (gdb) quit
   The program is running.  Exit anyway? (y or n) y
   bash-3.2# exit
 
   Script done on Mon Jul  7 05:31:04 2008
 
 Note that the first two segfaults are red herrings. Apparently some
 OSes (including NetBSD?) do not have a native way to find the
 beginning and end of the main data segment, so boehm-gc must perform a
 search at startup which causes segmentation faults.
 
 The true bug appears to be releated to the dynamic library tracing
 routines in dyn_load.c. The following page has helpful information
 about the porting changes necessary to enable dynamic library support:
 
   http://www.hpl.hp.com/personal/Hans_Boehm/gc/porting.html
 
 The page also mentions that if DYNAMIC_LOADING is undefined in
 gcconfig.h, tracing of dynamic library data will be disabled. This
 might be a quick, if rather unsatisfying, way to get boehm-gc
 "working" in NetBSD/mipsel.
 
 --Ben
 
 
 From: ben%wongs.net@localhost
 To: recht%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, 
pkgsrc-bugs%netbsd.org@localhost,
        ben%wongs.net@localhost
 Subject: Re: pkg/39012 (boehm-gc coredumps on GC_INIT under hpcmips (mipsel))
 Date: Sun, 20 Jul 2008 05:36:54 -0700
 
 As I suspected, a quick work around is to simply disable
 DYNAMIC_LOADING in the Boehm garbage collector. I have recompiled
 gc-7.0 and verified that libgc works as expected (or, at least, it
 doesn't dump core like the one currently available in pkgsrc).
 
 I have a patch (below) which skips the #define of DYNAMIC_LOADING when
 NETBSD and MIPSEL are defined. While this is not the optimal solution,
 it is more correct then what is currently shipping. By the way, this
 patch to boehm-gc will make the w3m package, and perhaps others,
 compile cleanly instead of breaking.
 
        --Ben
 
 --- include/private/gcconfig.h.orig     2007-06-28 17:00:09.000000000 -0700
 +++ include/private/gcconfig.h  2008-07-20 05:25:28.000000000 -0700
 @@ -1397,7 +1397,13 @@
          extern int etext[];
  #       define DATASTART GC_data_start
  #       define NEED_FIND_LIMIT
 -#       define DYNAMIC_LOADING
 +/* Sadly, DYNAMIC_LOADING is broken for NetBSD/mipsel as of July 2008 */
 +#      ifdef __MIPSEL__
 +#        warning "DYNAMIC_LOADING support is broken for little endian MIPS"
 +#        undef DYNAMIC_LOADING
 +#      else
 +#         define DYNAMIC_LOADING
 +#      endif /* __MIPSEL__ */
  #     else
  #       define DATASTART ((ptr_t) 0x10000000)
  #       define STACKBOTTOM ((ptr_t) 0x7ffff000)
 
 
 On 7/7/08, ben%wongs.net@localhost <ben%wongs.net@localhost> wrote:
 > Additional information:
 > [[snipped by dholland]]