I just killed quite some time debugging a weird lockup I had with a
third party program under NetBSD.
Long story short... That program overrides malloc and returns memory
that is only 8 bytes aligned. That creates problems for our pthreads,
as rwlock code assumes that pthread_self() is 16-byte aligned and uses
the lower bits for the flags (see RW_THREAD in pthread_int.h and code
in pthread_rwlock.c).
The two approaches I can think of are to use posix_memalign+memset
instead of calloc and to use calloc and manually re-align the result.
So either
newthread = calloc(1, __pthread_st_size + ~RW_THREAD);
if ((uintptr_t)newthread & ~RW_THREAD) {
newthread = (pthread_t)(((uintptr_t)(newthread)
+ ~RW_THREAD) & RW_THREAD);
}
or
posix_memalign((void **)&newthread, -RW_THREAD, __pthread_st_size);
memset(newthread, 0, __pthread_st_size);
The latter seems a bit prettier, but many older programs that override
malloc don't override posix_memalign, so we will still use our own
malloc for this. Not sure if this is a problem. Debugging mallocs
probably all properly override posix_memalign.
Ideas?
-uwe
