pthread: wake cond/sem waiters when cancelled

[derfsss]

Problem

pthread_cancel() of a thread blocked in pthread_cond_wait / pthread_cond_timedwait / sem_wait (which is built on the cond wait) is never acted upon: the target never wakes, and a subsequent pthread_join on it blocks forever.

The cause: pthread_cancel() signals the target's dedicated cancel signal (inf->cancel_signal_mask), but _pthread_cond_timedwait() does not include that signal in its Wait() mask — it only waits on the condvar waiter bit, the optional timer bit and SIGBREAKF_CTRL_C.

Repro

pthread_create(&t, NULL, fn_that_cond_waits_forever, NULL);
sleep(1);
pthread_cancel(t);      /* returns 0 */
pthread_join(t, &res);  /* hangs forever */

Fix

Include the dedicated cancel signal in the wait mask and run pthread_testcancel() when woken by it — exactly as is already done for the SIGBREAKF_CTRL_C fallback. The existing cancellation cleanup handler (CondWaitCleanupHandler) already handles waiter-node removal and mutex reacquisition for this path, so cancellation semantics match POSIX (mutex is reacquired before cleanup handlers run).

Validation

Tested on AmigaOS 4.1 FE (QEMU amigaone) with a clean GNUmakefile.os4 build of this branch: cancel + join of a cond-parked thread and of a sem-parked thread both complete with PTHREAD_CANCELED. Found while porting OpenJDK 17 to AmigaOS 4, where JVM daemon threads parked in pthread_cond_wait could never be cancelled at VM shutdown.

🤖 Generated with Claude Code

[afxgroup]

Can you please try all pthreads examples in test_library/threads folder? just go in test_library and compile everything with make. We need to be sure that we don't broke anything on that examples