diff --git a/htl/pt-create.c b/htl/pt-create.c
index fac61f1b22..8a735d99cb 100644
--- a/htl/pt-create.c
+++ b/htl/pt-create.c
@@ -177,7 +177,9 @@ __pthread_create_internal (struct __pthread **thread,
       err = ENOMEM;
       goto failed_thread_tls_alloc;
     }
+#if TLS_TCB_AT_TP
   pthread->tcb->tcb = pthread->tcb;
+#endif
 
   /* And initialize the rest of the machine context.  This may include
      additional machine- and system-specific initializations that
diff --git a/sysdeps/htl/dl-thread_gscope_wait.c b/sysdeps/htl/dl-thread_gscope_wait.c
index 90a9a79850..ee0a316527 100644
--- a/sysdeps/htl/dl-thread_gscope_wait.c
+++ b/sysdeps/htl/dl-thread_gscope_wait.c
@@ -20,6 +20,18 @@
 #include <pthread.h>
 #include <htl/pt-internal.h>
 
+static inline int *
+thread_gscope_flag (struct __pthread *t)
+{
+#if TLS_TCB_AT_TP
+  return &t->tcb->gscope_flag;
+#elif TLS_DTV_AT_TP
+  return &((tcbprehead_t *) t->tcb - 1)->gscope_flag;
+#else
+# error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+#endif
+}
+
 void
 __thread_gscope_wait (void)
 {
@@ -33,10 +45,10 @@ __thread_gscope_wait (void)
   for (i = 0; i < GL (dl_pthread_num_threads); ++i)
     {
       t = GL (dl_pthread_threads[i]);
-      if (t == NULL || t->tcb->gscope_flag == THREAD_GSCOPE_FLAG_UNUSED)
+      if (t == NULL || *thread_gscope_flag (t) == THREAD_GSCOPE_FLAG_UNUSED)
         continue;
 
-      gscope_flagp = &t->tcb->gscope_flag;
+      gscope_flagp = thread_gscope_flag (t);
 
       /* We have to wait until this thread is done with the global
          scope.  First tell the thread that we are waiting and
diff --git a/sysdeps/mach/hurd/htl/pt-sysdep.c b/sysdeps/mach/hurd/htl/pt-sysdep.c
index 270e77531e..5372cbf77f 100644
--- a/sysdeps/mach/hurd/htl/pt-sysdep.c
+++ b/sysdeps/mach/hurd/htl/pt-sysdep.c
@@ -100,7 +100,16 @@ _init_routine (void *stack)
      to the new stack.  Pretend it wasn't allocated so that it remains
      valid if the main thread terminates.  */
   thread->stack = 0;
+#if TLS_TCB_AT_TP
   thread->tcb = THREAD_SELF;
+#elif TLS_DTV_AT_TP
+  /* Assuming THREAD_SELF is implemented as subtracting TLS_PRE_TCB_SIZE
+     from the value of a thread pointer regsiter, this should optimize
+     down to simply reading that register.  */
+  thread->tcb = (tcbhead_t *) (((char *) THREAD_SELF) + TLS_PRE_TCB_SIZE);
+#else
+# error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+#endif
 
 #ifndef PAGESIZE
   __pthread_default_attr.__guardsize = __vm_page_size;