---
 nptl/allocatestack.c         |   35 +++++++++++++++++++++++++----------
 nptl/pthread_attr_getstack.c |    4 ++++
 nptl/pthread_attr_setstack.c |    8 ++++++++
 nptl/pthread_create.c        |   19 ++++++++++++++++---
 nptl/pthread_getattr_np.c    |    4 ++++
 5 files changed, 57 insertions(+), 13 deletions(-)

--- a/nptl/allocatestack.c
+++ b/nptl/allocatestack.c
@@ -361,6 +361,15 @@
   if (__builtin_expect (attr->flags & ATTR_FLAG_STACKADDR, 0))
     {
       uintptr_t adj;
+#if _STACK_GROWS_DOWN
+      char * stackaddr = (char *) attr->stackaddr;
+#else
+      /* Assume the same layout as the _STACK_GROWS_DOWN case, 
+	 with struct pthread at the top of the stack block. 
+	 Later we adjust the guard location and stack address 
+	 to match the _STACK_GROWS_UP case.  */
+      char * stackaddr = (char *) attr->stackaddr + attr->stacksize;
+#endif
 
       /* If the user also specified the size of the stack make sure it
 	 is large enough.  */
@@ -370,11 +379,11 @@
 
       /* Adjust stack size for alignment of the TLS block.  */
 #if TLS_TCB_AT_TP
-      adj = ((uintptr_t) attr->stackaddr - TLS_TCB_SIZE)
+      adj = ((uintptr_t) stackaddr - TLS_TCB_SIZE)
 	    & __static_tls_align_m1;
       assert (size > adj + TLS_TCB_SIZE);
 #elif TLS_DTV_AT_TP
-      adj = ((uintptr_t) attr->stackaddr - __static_tls_size)
+      adj = ((uintptr_t) stackaddr - __static_tls_size)
 	    & __static_tls_align_m1;
       assert (size > adj);
 #endif
@@ -384,10 +393,10 @@
 	 the stack.  It is the user's responsibility to do this if it
 	 is wanted.  */
 #if TLS_TCB_AT_TP
-      pd = (struct pthread *) ((uintptr_t) attr->stackaddr
+      pd = (struct pthread *) ((uintptr_t) stackaddr
 			       - TLS_TCB_SIZE - adj);
 #elif TLS_DTV_AT_TP
-      pd = (struct pthread *) (((uintptr_t) attr->stackaddr
+      pd = (struct pthread *) (((uintptr_t) stackaddr
 				- __static_tls_size - adj)
 			       - TLS_PRE_TCB_SIZE);
 #endif
@@ -399,7 +408,7 @@
       pd->specific[0] = pd->specific_1stblock;
 
       /* Remember the stack-related values.  */
-      pd->stackblock = (char *) attr->stackaddr - size;
+      pd->stackblock = (char *) stackaddr - size;
       pd->stackblock_size = size;
 
       /* This is a user-provided stack.  It will not be queued in the
@@ -625,7 +634,7 @@
 	  char *guard = mem + (((size - guardsize) / 2) & ~pagesize_m1);
 #elif _STACK_GROWS_DOWN
 	  char *guard = mem;
-# elif _STACK_GROWS_UP
+#elif _STACK_GROWS_UP
 	  char *guard = (char *) (((uintptr_t) pd - guardsize) & ~pagesize_m1);
 #endif
 	  if (mprotect (guard, guardsize, PROT_NONE) != 0)
@@ -675,9 +684,13 @@
 			prot) != 0)
 	    goto mprot_error;
 #elif _STACK_GROWS_UP
-	  if (mprotect ((char *) pd - pd->guardsize,
-			pd->guardsize - guardsize, prot) != 0)
-	    goto mprot_error;
+	  char *new_guard = (char *) (((uintptr_t) pd - guardsize) & ~pagesize_m1);
+	  char *old_guard = (char *) (((uintptr_t) pd - pd->guardsize) & ~pagesize_m1);
+	  /* The guard size difference might be > 0, but once rounded
+	     to the nearest page the size difference might be zero.  */
+	  if (old_guard - new_guard > 0)
+	    if (mprotect (old_guard, new_guard - old_guard, prot) != 0)
+	      goto mprot_error;
 #endif
 
 	  pd->guardsize = guardsize;
@@ -720,8 +733,10 @@
 #elif _STACK_GROWS_DOWN
   *stack = stacktop;
 #elif _STACK_GROWS_UP
+  /* We don't use stacktop. In _STACK_GROWS_UP the start
+     of the stack is simply stackblock (lowest address of
+     the stored block of memory for the stack).  */
   *stack = pd->stackblock;
-  assert (*stack > 0);
 #endif
 
   return 0;
--- a/nptl/pthread_attr_getstack.c
+++ b/nptl/pthread_attr_getstack.c
@@ -32,7 +32,11 @@
   iattr = (struct pthread_attr *) attr;
 
   /* Store the result.  */
+#ifdef _STACK_GROWS_DOWN
   *stackaddr = (char *) iattr->stackaddr - iattr->stacksize;
+#else
+  *stackaddr = (char *) iattr->stackaddr;
+#endif
   *stacksize = iattr->stacksize;
 
   return 0;
--- a/nptl/pthread_attr_setstack.c
+++ b/nptl/pthread_attr_setstack.c
@@ -47,7 +47,11 @@
 #endif
 
   iattr->stacksize = stacksize;
+#if _STACK_GROWS_DOWN
   iattr->stackaddr = (char *) stackaddr + stacksize;
+#else
+  iattr->stackaddr = (char *) stackaddr;
+#endif
   iattr->flags |= ATTR_FLAG_STACKADDR;
 
   return 0;
@@ -80,7 +84,11 @@
 #  endif
 
   iattr->stacksize = stacksize;
+#if _STACK_GROWS_DOWN
   iattr->stackaddr = (char *) stackaddr + stacksize;
+#else
+  iattr->stackaddr = (char *) stackaddr;
+#endif
   iattr->flags |= ATTR_FLAG_STACKADDR;
 
   return 0;
--- a/nptl/pthread_create.c
+++ b/nptl/pthread_create.c
@@ -399,12 +399,25 @@
 #ifdef _STACK_GROWS_DOWN
   char *sp = CURRENT_STACK_FRAME;
   size_t freesize = (sp - (char *) pd->stackblock) & ~pagesize_m1;
-#else
-# error "to do"
-#endif
   assert (freesize < pd->stackblock_size);
   if (freesize > PTHREAD_STACK_MIN)
     __madvise (pd->stackblock, freesize - PTHREAD_STACK_MIN, MADV_DONTNEED);
+#else
+  /* Page aligned start of memory to free (higher than or equal 
+     to current sp plus the minimum stack size).  */
+  void *freeblock = (void*)((size_t)(CURRENT_STACK_FRAME 
+				     + PTHREAD_STACK_MIN 
+				     + pagesize_m1) 
+				    & ~pagesize_m1);
+  char *free_end = (char *) (((uintptr_t) pd - pd->guardsize) & ~pagesize_m1);
+  /* Is there any space to free?  */
+  if (free_end > (char *)freeblock)
+    {
+      size_t freesize = (size_t)(free_end - (char *)freeblock);
+      assert (freesize < pd->stackblock_size);
+      __madvise (freeblock, freesize, MADV_DONTNEED);
+    }
+#endif
 
   /* If the thread is detached free the TCB.  */
   if (IS_DETACHED (pd))
--- a/nptl/pthread_getattr_np.c
+++ b/nptl/pthread_getattr_np.c
@@ -60,7 +60,11 @@
   if (__builtin_expect (thread->stackblock != NULL, 1))
     {
       iattr->stacksize = thread->stackblock_size;
+#ifdef _STACK_GROWS_DOWN
       iattr->stackaddr = (char *) thread->stackblock + iattr->stacksize;
+#else
+      iattr->stackaddr = (char *) thread->stackblock;
+#endif
     }
   else
     {