https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109540
--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
The description is a bit confusing, but the issue is that we define:
typedef struct timespec __gthread_time_t;
and then use that in several functions like this:
static inline int
__gthread_cond_timedwait (__gthread_cond_t *__cond, __gthread_mutex_t *__mutex,
const __gthread_time_t *__abs_timeout)
{
return __gthrw_(pthread_cond_timedwait) (__cond, __mutex, __abs_timeout);
}
If libc uses a 64-bit time_t in struct timespec then we need to use
__pthread_cond_timedwait64 instead of pthread_cond_timedwait, because the
latter expects a struct containing a 32-bit time_t instead of a 64-bit one.
The weak alias referenced by __gthrw_ is defined like so:
static __typeof(pthread_cond_timedwait) __gthrw_pthread_cond_timedwait
__attribute__ ((__weakref__("pthread_cond_timedwait"), __copy__
(pthread_cond_timedwait)));
Where the _TIME_BITS=64 declaration of pthread_cond_timedwait is:
extern int pthread_cond_timedwait (pthread_cond_t *__restrict __cond,
pthread_mutex_t *__restrict __mutex, const struct timespec *__restrict
__abstime) __asm__ ("" "__pthread_cond_timedwait64")
The reported issue is that weakref("pthread_cond_timedwait") binds to a symbol
of that name, rather than to __pthread_cond_timedwait64, so we end up passing a
64-bit timespec to the 32-bit pthread_cond_timedwait.
However, that's not what I observe when I test it.
If I compile with 32-bit time_t for i686 using GCC 12 and
glibc-2.36-9.fc37.i686 then a call to __gthread_cond_timedwait will resolve to
glibc's pthread_cond_timedwait which expects a 32-bit timespec:
(gdb)
13 __gthread_cond_timedwait(&cond, &mutex, &ts);
(gdb) step
__gthread_cond_timedwait (__cond=0x804c060 <cond>, __mutex=0x804c090 <mutex>,
__abs_timeout=0xffffc7e8)
at /usr/include/c++/12/x86_64-redhat-linux/32/bits/gthr-default.h:872
872 return __gthrw_(pthread_cond_timedwait) (__cond, __mutex,
__abs_timeout);
(gdb)
___pthread_cond_timedwait (cond=0x804c060 <cond>, mutex=0x804c090 <mutex>,
abstime=0xffffc7e8) at pthread_cond_wait.c:655
655 {
(gdb) l
650 libc_hidden_def (__pthread_cond_timedwait64)
651
652 int
653 ___pthread_cond_timedwait (pthread_cond_t *cond, pthread_mutex_t
*mutex,
654 const struct timespec *abstime)
655 {
656 struct __timespec64 ts64 = valid_timespec_to_timespec64 (*abstime);
657
658 return __pthread_cond_timedwait64 (cond, mutex, &ts64);
659 }
If I recompile the same code with -D_TIME_BITS=64 -D_FILE_OFFSET_BITS=64 then
the call to __gthread_cond_timedwait resolves to glibc's
__pthread_cond_timedwait64:
(gdb)
13 __gthread_cond_timedwait(&cond, &mutex, &ts);
(gdb) step
__gthread_cond_timedwait (__cond=0x804c060 <cond>, __mutex=0x804c090 <mutex>,
__abs_timeout=0xffffc7e0)
at /usr/include/c++/12/x86_64-redhat-linux/32/bits/gthr-default.h:872
872 return __gthrw_(pthread_cond_timedwait) (__cond, __mutex,
__abs_timeout);
(gdb)
___pthread_cond_timedwait64 (cond=0x804c060 <cond>, mutex=0x804c090 <mutex>,
abstime=0xffffc7e0) at pthread_cond_wait.c:632
632 {
(gdb) l
627
628 /* See __pthread_cond_wait_common. */
629 int
630 ___pthread_cond_timedwait64 (pthread_cond_t *cond, pthread_mutex_t
*mutex,
631 const struct __timespec64 *abstime)
632 {
633 /* Check parameter validity. This should also tell the compiler that
634 it can assume that abstime is not NULL. */
635 if (! valid_nanoseconds (abstime->tv_nsec))
636 return EINVAL;
So it seems to work fine. Do you actually observe a bug, or are you just
speculating from reading the wiki page and headers?
