allocatestack.c changes for TLS_DTV_AT_TP

Here are the changes that I have made locally to allocatestack.c and
pthread_create.c to support the TLS_DTV_AT_TP model, which we use
(slightly modified :) on PPC.  As it says in the comments, it probably
needs stronger alignment applied in the calculation of pd.

Comments?

Paul.

diff -urN nptl-0.28/nptl/allocatestack.c libc/nptl/allocatestack.c
--- nptl-0.28/nptl/allocatestack.c	2003-03-02 08:39:31.000000000 +1100
+++ libc/nptl/allocatestack.c	2003-03-11 21:32:09.000000000 +1100
@@ -51,7 +51,11 @@
 # define MINIMAL_REST_STACK	4096
 #endif
 
-
+#if TLS_TCB_AT_TP
+# define TLS_TCB(pd)	(pd)
+#elif TLS_DTV_AT_TP
+# define TLS_TCB(pd)	((void *)((pd) + 1))
+#endif
 
 
 /* Cache handling for not-yet free stacks.  */
@@ -157,11 +161,12 @@
   result->nextevent = NULL;
 
   /* Clear the DTV.  */
-  dtv_t *dtv = GET_DTV (result);
+  dtv_t *dtv = GET_DTV (TLS_TCB(result));
   memset (dtv, '\0', (dtv[-1].counter + 1) * sizeof (dtv_t));
 
   /* Re-initialize the TLS.  */
-  return _dl_allocate_tls_init (result);
+  _dl_allocate_tls_init (TLS_TCB(result));
+  return result;
 }
 
 
@@ -198,7 +203,7 @@
 	      stack_cache_actsize -= curr->stackblock_size;
 
 	      /* Free the memory associated with the ELF TLS.  */
-	      _dl_deallocate_tls (curr, false);
+	      _dl_deallocate_tls (TLS_TCB(curr), false);
 
 	      /* Remove this block.  This should never fail.  If it
 		 does something is really wrong.  */
@@ -250,8 +255,17 @@
 	 size...  We do not allocate guard pages if the user provided
 	 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 - adj)
 			       & ~(__alignof (struct pthread) - 1)) - 1;
+#else
+      /* Place the static TLS area at the end of the stack,
+	 preceded by the thread descriptor.  */
+      /* XXX probably need more alignment */
+      pd = (struct pthread *)(((uintptr_t) attr->stackaddr - adj
+			       - TLS_TCB_SIZE - __static_tls_size)
+			      & ~(__alignof (struct pthread) - 1)) - 1;
+#endif
 
       /* The user provided stack memory needs to be cleared.  */
       memset (pd, '\0', sizeof (struct pthread));
@@ -282,7 +296,7 @@
 #endif
 
       /* Allocate the DTV for this thread.  */
-      if (_dl_allocate_tls (pd) == NULL)
+      if (_dl_allocate_tls (TLS_TCB(pd)) == NULL)
 	/* Something went wrong.  */
 	return errno;
 
@@ -363,8 +377,16 @@
 # define coloring 0
 #endif
 
+#if TLS_TCB_AT_TP
 	  /* Place the thread descriptor at the end of the stack.  */
 	  pd = (struct pthread *) ((char *) mem + size - coloring) - 1;
+#else
+	  /* Place the static TLS area at the end of the stack,
+	     preceded by the thread descriptor.  */
+	  /* XXX probably need more alignment */
+	  pd = (struct pthread *)((char *) mem + size - TLS_TCB_SIZE
+				  - __static_tls_size) - 1;
+#endif
 
 	  /* Remember the stack-related values.  */
 	  pd->stackblock = mem;
@@ -390,7 +412,7 @@
 #endif
 
 	  /* Allocate the DTV for this thread.  */
-	  if (_dl_allocate_tls (pd) == NULL)
+	  if (_dl_allocate_tls (TLS_TCB(pd)) == NULL)
 	    {
 	      /* Something went wrong.  */
 	      int err = errno;
@@ -466,7 +488,9 @@
   /* The stack begins before the TCB and the static TLS block.  */
   *stack = ((char *) (pd + 1) - __static_tls_size);
 #else
-# error "Implement me"
+  /* The top of the stack is at the start of the pthread struct.
+     This assumes that the stack grows downwards.  */
+  *stack = (char *) pd;
 #endif
 
   return 0;
@@ -495,7 +519,7 @@
     (void) queue_stack (pd);
   else
     /* Free the memory associated with the ELF TLS.  */
-    _dl_deallocate_tls (pd, false);
+    _dl_deallocate_tls (TLS_TCB(pd), false);
 
   lll_unlock (stack_cache_lock);
 }
diff -urN nptl-0.28/nptl/pthread_create.c libc/nptl/pthread_create.c
--- nptl-0.28/nptl/pthread_create.c	2003-02-23 20:00:24.000000000 +1100
+++ libc/nptl/pthread_create.c	2003-03-11 19:47:20.000000000 +1100
@@ -351,6 +351,8 @@
 #ifdef TLS_TCB_AT_TP
   /* Self-reference.  */
   pd->header.data.tcb = pd;
+#elif TLS_DTV_AT_TP
+  pd->header.data.tcb = pd + 1;
 #endif
 
   /* Store the address of the start routine and the parameter.  Since