For the amd64 we need to do some extra stack alignment fixups. Otherwise

we can end up with some threads with a non-16-byte-aligned stack.  This
causes some interesting side effects, including general protection
faults leading to a SIGBUS when doing floating point or varargs.  This
should be just a verbose NOP for the other platforms.

Approved by:  re (scottl)

3 files changed, 27 insertions, 6 deletions

diff --git a/lib/libc_r/uthread/uthread_create.c b/lib/libc_r/uthread/uthread_create.c
index 2ea87b2..356bc07 100644
--- a/lib/libc_r/uthread/uthread_create.c
+++ b/lib/libc_r/uthread/uthread_create.c
@@ -73,6 +73,9 @@ _pthread_create(pthread_t *thread, const pthread_attr_t *attr,
 	pthread_t       new_thread;
 	pthread_attr_t	pattr;
 	void           *stack;
+#if !defined(__ia64__)
+	u_long 		stackp;
+#endif
 
 	if (thread == NULL)
 		return(EINVAL);
@@ -145,10 +148,12 @@ _pthread_create(pthread_t *thread, const pthread_attr_t *attr,
 			SET_RETURN_ADDR_JB(new_thread->ctx.jb, _thread_start);
 
 #if !defined(__ia64__)
+			stackp = (long)new_thread->stack + pattr->stacksize_attr - sizeof(double);
+#if defined(__amd64__)
+			stackp &= ~0xFUL;
+#endif
 			/* The stack starts high and builds down: */
-			SET_STACK_JB(new_thread->ctx.jb,
-			    (long)new_thread->stack + pattr->stacksize_attr
-			    - sizeof(double));
+			SET_STACK_JB(new_thread->ctx.jb, stackp);
 #else
 			SET_STACK_JB(new_thread->ctx.jb,
 			    (long)new_thread->stack, pattr->stacksize_attr);
diff --git a/lib/libc_r/uthread/uthread_init.c b/lib/libc_r/uthread/uthread_init.c
index 5367d0b..d3a2b45 100644
--- a/lib/libc_r/uthread/uthread_init.c
+++ b/lib/libc_r/uthread/uthread_init.c
@@ -208,6 +208,9 @@ _thread_init(void)
 	size_t		len;
 	int		mib[2];
 	int		sched_stack_size;	/* Size of scheduler stack. */
+#if !defined(__ia64__)
+	u_long		stackp;
+#endif
 
 	struct clockinfo clockinfo;
 	struct sigaction act;
@@ -374,8 +377,11 @@ _thread_init(void)
 		/* Setup the context for the scheduler: */
 		_setjmp(_thread_kern_sched_jb);
 #if !defined(__ia64__)
-		SET_STACK_JB(_thread_kern_sched_jb, _thread_kern_sched_stack +
-		    sched_stack_size - sizeof(double));
+		stackp = (long)_thread_kern_sched_stack + sched_stack_size - sizeof(double);
+#if defined(__amd64__)
+		stackp &= ~0xFUL;
+#endif
+		SET_STACK_JB(_thread_kern_sched_jb, stackp);
 #else
 		SET_STACK_JB(_thread_kern_sched_jb, _thread_kern_sched_stack,
 		    sched_stack_size);
diff --git a/lib/libc_r/uthread/uthread_sig.c b/lib/libc_r/uthread/uthread_sig.c
index e02ccc3..3617b36 100644
--- a/lib/libc_r/uthread/uthread_sig.c
+++ b/lib/libc_r/uthread/uthread_sig.c
@@ -1048,13 +1048,20 @@ thread_sigframe_add(struct pthread *thread, int sig, int has_args)
 	 * Leave a little space on the stack and round down to the
 	 * nearest aligned word:
 	 */
+#if defined(__amd64__)
+	stackp -= 128;		/* Skip over 128 byte red-zone */
+#endif
 	stackp -= sizeof(double);
+#if defined(__amd64__)
+	stackp &= ~0xFUL;
+#else
 	stackp &= ~0x3UL;
 #endif
+#endif
 
 	/* Allocate room on top of the stack for a new signal frame: */
 	stackp -= sizeof(struct pthread_signal_frame);
-#if defined(__ia64__)
+#if defined(__ia64__) || defined(__amd64__)
 	stackp &= ~0xFUL;
 #endif
 
@@ -1087,6 +1094,9 @@ thread_sigframe_add(struct pthread *thread, int sig, int has_args)
 	 */
 #if !defined(__ia64__)
 	stackp -= sizeof(double);
+#if defined(__amd64__)
+	stackp &= ~0xFUL;
+#endif
 #endif
 	_setjmp(thread->ctx.jb);
 #if !defined(__ia64__)