fix arm __tlsdesc_dynamic when built as thumb code without __ARM_ARCH>=5

[musl] / src / thread / pthread_create.c

diff --git a/src/thread/pthread_create.c b/src/thread/pthread_create.c
index 03cdea0..5d00d76 100644 (file)
--- a/src/thread/pthread_create.c
+++ b/src/thread/pthread_create.c
@@ -15,29 +15,40 @@ weak_alias(dummy_0, __release_ptc);
 weak_alias(dummy_0, __pthread_tsd_run_dtors);
 weak_alias(dummy_0, __do_orphaned_stdio_locks);
 weak_alias(dummy_0, __dl_thread_cleanup);
+weak_alias(dummy_0, __membarrier_init);
+
+static int tl_lock_count;
+static int tl_lock_waiters;
 
 void __tl_lock(void)
 {
-       if (!a_cas(&__thread_list_lock, 0, 1)) return;
-       do {
-               a_cas(&__thread_list_lock, 1, 2);
-               __futexwait(&__thread_list_lock, 2, 0);
-       } while (a_cas(&__thread_list_lock, 0, 2));
+       int tid = __pthread_self()->tid;
+       int val = __thread_list_lock;
+       if (val == tid) {
+               tl_lock_count++;
+               return;
+       }
+       while ((val = a_cas(&__thread_list_lock, 0, tid)))
+               __wait(&__thread_list_lock, &tl_lock_waiters, val, 0);
 }
 
 void __tl_unlock(void)
 {
-       if (a_swap(&__thread_list_lock, 0)==2)
-               __wake(&__thread_list_lock, 1, 0);
+       if (tl_lock_count) {
+               tl_lock_count--;
+               return;
+       }
+       a_store(&__thread_list_lock, 0);
+       if (tl_lock_waiters) __wake(&__thread_list_lock, 1, 0);
 }
 
 void __tl_sync(pthread_t td)
 {
        a_barrier();
-       if (!__thread_list_lock) return;
-       a_cas(&__thread_list_lock, 1, 2);
-       __wait(&__thread_list_lock, 0, 2, 0);
-       __wake(&__thread_list_lock, 1, 0);
+       int val = __thread_list_lock;
+       if (!val) return;
+       __wait(&__thread_list_lock, &tl_lock_waiters, val, 0);
+       if (tl_lock_waiters) __wake(&__thread_list_lock, 1, 0);
 }
 
 _Noreturn void __pthread_exit(void *result)
@@ -161,23 +172,20 @@ void __do_cleanup_pop(struct __ptcb *cb)
 struct start_args {
        void *(*start_func)(void *);
        void *start_arg;
-       pthread_attr_t *attr;
-       volatile int *perr;
+       volatile int control;
        unsigned long sig_mask[_NSIG/8/sizeof(long)];
 };
 
 static int start(void *p)
 {
        struct start_args *args = p;
-       if (args->attr) {
-               pthread_t self = __pthread_self();
-               int ret = -__syscall(SYS_sched_setscheduler, self->tid,
-                       args->attr->_a_policy, &args->attr->_a_prio);
-               if (a_swap(args->perr, ret)==-2)
-                       __wake(args->perr, 1, 1);
-               if (ret) {
-                       self->detach_state = DT_DETACHED;
-                       __pthread_exit(0);
+       int state = args->control;
+       if (state) {
+               if (a_cas(&args->control, 1, 2)==1)
+                       __wait(&args->control, 0, 2, 1);
+               if (args->control) {
+                       __syscall(SYS_set_tid_address, &args->control);
+                       return 0;
                }
        }
        __syscall(SYS_rt_sigprocmask, SIG_SETMASK, &args->sig_mask, 0, _NSIG/8);
@@ -201,8 +209,6 @@ weak_alias(dummy, __pthread_tsd_size);
 static void *dummy_tsd[1] = { 0 };
 weak_alias(dummy_tsd, __pthread_tsd_main);
 
-volatile int __block_new_threads = 0;
-
 static FILE *volatile dummy_file = 0;
 weak_alias(dummy_file, __stdin_used);
 weak_alias(dummy_file, __stdout_used);
@@ -224,7 +230,6 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
                | CLONE_PARENT_SETTID | CLONE_CHILD_CLEARTID | CLONE_DETACHED;
        pthread_attr_t attr = { 0 };
        sigset_t set;
-       volatile int err = -1;
 
        if (!libc.can_do_threads) return ENOSYS;
        self = __pthread_self();
@@ -237,6 +242,7 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
                init_file_lock(__stderr_used);
                __syscall(SYS_rt_sigprocmask, SIG_UNBLOCK, SIGPT_SET, 0, _NSIG/8);
                self->tsd = (void **)__pthread_tsd_main;
+               __membarrier_init();
                libc.threaded = 1;
        }
        if (attrp && !c11) attr = *attrp;
@@ -247,8 +253,6 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
                attr._a_guardsize = __default_guardsize;
        }
 
-       if (__block_new_threads) __wait(&__block_new_threads, 0, 1, 1);
-
        if (attr._a_stackaddr) {
                size_t need = libc.tls_size + __pthread_tsd_size;
                size = attr._a_stacksize;
@@ -307,6 +311,7 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
        }
        new->robust_list.head = &new->robust_list.head;
        new->CANARY = self->CANARY;
+       new->sysinfo = self->sysinfo;
 
        /* Setup argument structure for the new thread on its stack.
         * It's safe to access from the caller only until the thread
@@ -316,13 +321,7 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
        struct start_args *args = (void *)stack;
        args->start_func = entry;
        args->start_arg = arg;
-       if (attr._a_sched) {
-               args->attr = &attr;
-               args->perr = &err;
-       } else {
-               args->attr = 0;
-               args->perr = 0;
-       }
+       args->control = attr._a_sched ? 1 : 0;
 
        /* Application signals (but not the synccall signal) must be
         * blocked before the thread list lock can be taken, to ensure
@@ -340,29 +339,36 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
        libc.threads_minus_1++;
        ret = __clone((c11 ? start_c11 : start), stack, flags, args, &new->tid, TP_ADJ(new), &__thread_list_lock);
 
-       /* If clone succeeded, new thread must be linked on the thread
-        * list before unlocking it, even if scheduling may still fail. */
+       /* All clone failures translate to EAGAIN. If explicit scheduling
+        * was requested, attempt it before unlocking the thread list so
+        * that the failed thread is never exposed and so that we can
+        * clean up all transient resource usage before returning. */
+       if (ret < 0) {
+               ret = -EAGAIN;
+       } else if (attr._a_sched) {
+               ret = __syscall(SYS_sched_setscheduler,
+                       new->tid, attr._a_policy, &attr._a_prio);
+               if (a_swap(&args->control, ret ? 3 : 0)==2)
+                       __wake(&args->control, 1, 1);
+               if (ret)
+                       __wait(&args->control, 0, 3, 0);
+       }
+
        if (ret >= 0) {
                new->next = self->next;
                new->prev = self;
                new->next->prev = new;
                new->prev->next = new;
+       } else {
+               libc.threads_minus_1--;
        }
        __tl_unlock();
        __restore_sigs(&set);
        __release_ptc();
 
        if (ret < 0) {
-               libc.threads_minus_1--;
                if (map) __munmap(map, size);
-               return EAGAIN;
-       }
-
-       if (attr._a_sched) {
-               if (a_cas(&err, -1, -2)==-1)
-                       __wait(&err, 0, -2, 1);
-               ret = err;
-               if (ret) return ret;
+               return -ret;
        }
 
        *res = new;