Re: Help with some pkgsrc stuff

[Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>]

:* Mono applications build mostly now after some bugfixes for both mono and 
:  DragonFly itself. Applications even run, but die after some seconds of 
:  work with following backtrace:
:
:  Core was generated by `mono'.
:  Program terminated with signal 4, Illegal instruction.
:  #0  0x28265a8c in symlook_list (name=0x804f74b "waitpid",
:      hash=226539140, objlist=0x28286408, defobj_out=0xbf9fe1d0,
:      in_plt=1 '\001', dlp=0xbf9fe1d4)
:      at /home/hasso/dragonfly-src/libexec/rtld-elf/rtld.c:2412

    Upload the core file to leaf, I'll take a look at it to
    try to figure out what the illegal instruction is.

:* After applying the attached patch it's possible to compile www/firefox3
:  with jemalloc support (would be a good test to compare it with our
:  malloc). The problem is that it goes into infinite loop, seems:
:
:  Core was generated by `firefox-bin'.
:  Program terminated with signal 11, Segmentation fault.
:  #0  0x2809af80 in malloc_init_hard () at jemalloc.c:5194
:
:  5194    {
:  (gdb) bt
:  #0  0x2809af80 in malloc_init_hard () at jemalloc.c:5194
:  #1  0x2809c6c1 in malloc (size=224) at jemalloc.c:5183
:  #2  0x2809379e in _thr_alloc (curthread=0x0)
:      at /home/hasso/dragonfly-src/lib/libthread_xu/thread/thr_list.c:171
:  #3  0x2809201c in _libpthread_init (curthread=0x0)
:      at /home/hasso/dragonfly-src/lib/libthread_xu/thread/thr_init.c:215
:  #4  0x2809167a in __pthread_mutex_lock (m=0x2809ef74)
:      at /home/hasso/dragonfly-src/lib/libthread_xu/thread/thr_private.h:791
:  #5  0x2809af9e in malloc_init_hard () at jemalloc.c:1254
:  #6  0x2809c6c1 in malloc (size=224) at jemalloc.c:5183
:  #7  0x2809379e in _thr_alloc (curthread=0x0)
:...
:Any help to track these problems down is welcome.
:
:-- 
:Hasso Tepper

    A pthread/malloc init loop.

    There is some additional fragility here, since the mutex itself
    probably has to be allocated as well.

    Our malloc's use the _SPINLOCK API which allows libthread_xu 
    to allocate the initial spinlocks needed for bootstrapping malloc
    out of a static array.  It looks like jemalloc is trying to use
    core POSIX mutexes right off the bat.

    Someone else will have to dig into the jemalloc code, though, I just
    don't have time.

						-Matt