Adjust code to support AMD64, on AMD64, thread needs to set fsbase by

itself before it can execute any other code, so new thread should be
created with all signals are masked until after fsbase is set.

3 files changed, 31 insertions, 7 deletions

diff --git a/lib/libthr/arch/i386/i386/_setcurthread.c b/lib/libthr/arch/i386/i386/_setcurthread.c
index 0ff4ff6..4e07579 100644
--- a/lib/libthr/arch/i386/i386/_setcurthread.c
+++ b/lib/libthr/arch/i386/i386/_setcurthread.c
@@ -47,13 +47,16 @@ struct tcb {
 	struct tcb		*tcb_self;	/* required by rtld */
 	void			*tcb_dtv;	/* required by rtld */
 	struct pthread		*tcb_thread;
+	int			tcb_ldt;
 };
 
 void
 _retire_thread(void *entry)
 {
-	_rtld_free_tls(entry, sizeof(struct tcb), 16);
-	/* XXX free ldt descriptor here */
+	struct tcb *tcb = (struct tcb *)entry;
+
+	i386_set_ldt(tcb->tcb_ldt, NULL, 1);
+	_rtld_free_tls(tcb, sizeof(struct tcb), 16);
 }
 
 void *
@@ -66,6 +69,10 @@ _set_curthread(ucontext_t *uc, struct pthread *thr, int *err)
 
 	*err = 0;
 
+	if (uc == NULL && thr->arch_id != NULL) {
+		return (thr->arch_id);
+	}
+
 	if (uc == NULL) {
 		__asm __volatile("movl %%gs:0, %0" : "=r" (oldtls));
 	} else {
@@ -104,7 +111,7 @@ _set_curthread(ucontext_t *uc, struct pthread *thr, int *err)
 	ldt_index = i386_set_ldt(LDT_AUTO_ALLOC, &desc, 1);
 	if (ldt_index == -1)
 		abort();
-
+	tcb->tcb_ldt = ldt_index;
 	/*
 	 * Set up our gs with the index into the ldt for this entry.
 	 */
diff --git a/lib/libthr/thread/thr_create.c b/lib/libthr/thread/thr_create.c
index 38e483f..c5ad8c9 100644
--- a/lib/libthr/thread/thr_create.c
+++ b/lib/libthr/thread/thr_create.c
@@ -99,9 +99,10 @@ _pthread_create(pthread_t * thread, const pthread_attr_t * attr,
 
 	/* Initialise the machine context: */
 	getcontext(&new_thread->ctx);
+	new_thread->savedsig = new_thread->ctx.uc_sigmask;
 	new_thread->ctx.uc_stack.ss_sp = new_thread->stack;
 	new_thread->ctx.uc_stack.ss_size = pattr->stacksize_attr;
-	makecontext(&new_thread->ctx, _thread_start, 0);
+	makecontext(&new_thread->ctx, (void (*)(void))_thread_start, 1, new_thread);
 	new_thread->arch_id = _set_curthread(&new_thread->ctx, new_thread, &ret);
 	if (ret != 0) {
 		if (pattr->stackaddr_attr == NULL) {
@@ -145,7 +146,11 @@ _pthread_create(pthread_t * thread, const pthread_attr_t * attr,
 	 */
 	if (pattr->suspend == PTHREAD_CREATE_SUSPENDED)
 		new_thread->flags |= PTHREAD_FLAGS_SUSPENDED;
+	/* new thread inherits signal mask in kernel */
+	_thread_sigblock();
 	ret = thr_create(&new_thread->ctx, &new_thread->thr_id, flags);
+	/* restore my signal mask */
+	_thread_sigunblock();
 	if (ret != 0) {
 		_thread_printf(STDERR_FILENO, "thr_create() == %d\n", ret);
 		PANIC("thr_create");
@@ -160,12 +165,24 @@ _pthread_create(pthread_t * thread, const pthread_attr_t * attr,
 }
 
 void
-_thread_start(void)
+_thread_start(pthread_t td)
 {
+	int ret;
+
+	/*
+	 * for AMD64, we need to set fsbase by thread itself, before
+	 * fsbase is set, we can not run any other code, for example
+	 * signal code.
+	 */
+	_set_curthread(NULL, td, &ret);
+
+	/* restore signal mask inherited before */
+	__sys_sigprocmask(SIG_SETMASK, &td->savedsig, NULL);
+
 	if ((curthread->flags & PTHREAD_FLAGS_SUSPENDED) != 0)
 		_thread_suspend(curthread, NULL);
-	pthread_exit(curthread->start_routine(curthread->arg));
 
+	pthread_exit(curthread->start_routine(curthread->arg));
 	/* This point should never be reached. */
 	PANIC("Thread has resumed after exit");
 }
diff --git a/lib/libthr/thread/thr_private.h b/lib/libthr/thread/thr_private.h
index 45c7fe4..06caf96 100644
--- a/lib/libthr/thread/thr_private.h
+++ b/lib/libthr/thread/thr_private.h
@@ -797,7 +797,7 @@ void    _thread_cleanupspecific(void);
 void    _thread_dump_info(void);
 void    _thread_init(void);
 void	_thread_printf(int fd, const char *, ...);
-void    _thread_start(void);
+void    _thread_start(pthread_t td);
 void	_thread_seterrno(pthread_t, int);
 void	_thread_enter_cancellation_point(void);
 void	_thread_leave_cancellation_point(void);