Async-signal safe TLS.

ChangeLog:

2014-01-03  Andrew Hunter  <ahh@google.com>

	* elf/dl-open.c (): New comment.
	* elf/dl-reloc.c (_dl_try_allocate_static_tls): Use
	atomic_compare_and_exchange_bool_acq
	(_dl_allocate_static_tls): Block signals.
	* elf/dl-tls.c (allocate_and_init): Return void.
	(_dl_update_slotinfo): Block signals, use atomic update.

nptl/ChangeLog:

2014-01-03  Andrew Hunter  <ahh@google.com>

	* nptl/Makefile (tst-tls7): New test.
	* nptl/tst-tls7.c: New file.
	* nptl/tst-tls7mod.c: New file.
	* nptl/allocatestack.c (init_one_static_tls): Use atomic barrier.

3 files changed, 114 insertions, 39 deletions

diff --git a/elf/dl-open.c b/elf/dl-open.c
index a9ca6b3..ea222d0 100644
--- a/elf/dl-open.c
+++ b/elf/dl-open.c
@@ -548,7 +548,10 @@ cannot load any more object with static TLS"));
 	     generation of the DSO we are allocating data for.  */
 	  _dl_update_slotinfo (imap->l_tls_modid);
 #endif
-
+	  /* We do this iteration under a signal mask in dl-reloc; why not
+	     here?  Because these symbols are new and dlopen hasn't
+	     returned yet.  So we can't possibly be racing with a TLS
+	     access to them from another thread.  */
 	  GL(dl_init_static_tls) (imap);
 	  assert (imap->l_need_tls_init == 0);
 	}
diff --git a/elf/dl-reloc.c b/elf/dl-reloc.c
index 1f66fcc..81ee47e 100644
--- a/elf/dl-reloc.c
+++ b/elf/dl-reloc.c
@@ -16,8 +16,10 @@
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
+#include <atomic.h>
 #include <errno.h>
 #include <libintl.h>
+#include <signal.h>
 #include <stdlib.h>
 #include <unistd.h>
 #include <ldsodefs.h>
@@ -70,8 +72,6 @@ _dl_try_allocate_static_tls (struct link_map *map)
 
   size_t offset = GL(dl_tls_static_used) + (freebytes - n * map->l_tls_align
 					    - map->l_tls_firstbyte_offset);
-
-  map->l_tls_offset = GL(dl_tls_static_used) = offset;
 #elif TLS_DTV_AT_TP
   /* dl_tls_static_used includes the TCB at the beginning.  */
   size_t offset = (((GL(dl_tls_static_used)
@@ -83,7 +83,36 @@ _dl_try_allocate_static_tls (struct link_map *map)
   if (used > GL(dl_tls_static_size))
     goto fail;
 
-  map->l_tls_offset = offset;
+#else
+# error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+#endif
+  /* We've computed the new value we want, now try to install it.  */
+  ptrdiff_t val;
+  if ((val = map->l_tls_offset) == NO_TLS_OFFSET)
+    {
+      /* l_tls_offset starts out at NO_TLS_OFFSET, and all attempts to
+	 change it go from NO_TLS_OFFSET to some other value.  We use
+	 compare_and_exchange to ensure only one attempt succeeds.  We
+	 don't actually need any memory ordering here, but _acq is the
+	 weakest available.  */
+      (void ) atomic_compare_and_exchange_bool_acq (&map->l_tls_offset,
+						    offset,
+						    NO_TLS_OFFSET);
+      val = map->l_tls_offset;
+      assert (val != NO_TLS_OFFSET);
+    }
+  if (val != offset)
+    {
+      /* We'd like to set a static offset for this section, but another
+	 thread has already used a dynamic TLS block for it.  Since we can
+	 only use static offsets if everyone does (and it's not practical
+	 to move that thread's dynamic block), we have to fail.  */
+      goto fail;
+    }
+  /* We installed the value; now update the globals.  */
+#if TLS_TCB_AT_TP
+  GL(dl_tls_static_used) = offset;
+#elif TLS_DTV_AT_TP
   map->l_tls_firstbyte_offset = GL(dl_tls_static_used);
   GL(dl_tls_static_used) = used;
 #else
@@ -114,8 +143,17 @@ void
 internal_function __attribute_noinline__
 _dl_allocate_static_tls (struct link_map *map)
 {
-  if (map->l_tls_offset == FORCED_DYNAMIC_TLS_OFFSET
-      || _dl_try_allocate_static_tls (map))
+  /* We wrap this in a signal mask because it has to iterate all threads
+     (including this one) and update this map's TLS entry. A signal handler
+     accessing TLS would try to do the same update and break.  */
+  sigset_t old;
+  _dl_mask_all_signals (&old);
+  int err = -1;
+  if (map->l_tls_offset != FORCED_DYNAMIC_TLS_OFFSET)
+    err = _dl_try_allocate_static_tls (map);
+
+  _dl_unmask_signals (&old);
+  if (err != 0)
     {
       _dl_signal_error (0, map->l_name, NULL, N_("\
 cannot allocate memory in static TLS block"));
diff --git a/elf/dl-tls.c b/elf/dl-tls.c
index c1802e7..50ec876 100644
--- a/elf/dl-tls.c
+++ b/elf/dl-tls.c
@@ -17,6 +17,7 @@
    <http://www.gnu.org/licenses/>.  */
 
 #include <assert.h>
+#include <atomic.h>
 #include <errno.h>
 #include <libintl.h>
 #include <signal.h>
@@ -533,19 +534,21 @@ rtld_hidden_def (_dl_deallocate_tls)
 # endif
 
 
-static void *
-allocate_and_init (struct link_map *map)
+static void
+allocate_and_init (dtv_t *dtv, struct link_map *map)
 {
   void *newp;
   newp = __signal_safe_memalign (map->l_tls_align, map->l_tls_blocksize);
   if (newp == NULL)
     oom ();
 
-  /* Initialize the memory.  */
+  /* Initialize the memory. Since this is our thread's space, we are
+     under a signal mask, and no one has touched this section before,
+     we can safely just overwrite whatever's there.  */
   memset (__mempcpy (newp, map->l_tls_initimage, map->l_tls_initimage_size),
 	  '\0', map->l_tls_blocksize - map->l_tls_initimage_size);
 
-  return newp;
+  dtv->pointer.val = newp;
 }
 
 
@@ -587,7 +590,15 @@ _dl_update_slotinfo (unsigned long int req_modid)
 	 the entry we need.  */
       size_t new_gen = listp->slotinfo[idx].gen;
       size_t total = 0;
-
+      sigset_t old;
+
+      _dl_mask_all_signals (&old);
+      /* We use the signal mask as a lock against reentrancy here.
+         Check that a signal taken before the lock didn't already
+         update us.  */
+      dtv = THREAD_DTV ();
+      if (dtv[0].counter >= listp->slotinfo[idx].gen)
+        goto out;
       /* We have to look through the entire dtv slotinfo list.  */
       listp =  GL(dl_tls_dtv_slotinfo_list);
       do
@@ -699,6 +710,8 @@ _dl_update_slotinfo (unsigned long int req_modid)
 
       /* This will be the new maximum generation counter.  */
       dtv[0].counter = new_gen;
+   out:
+      _dl_unmask_signals (&old);
     }
 
   return the_map;
@@ -724,39 +737,60 @@ tls_get_addr_tail (GET_ADDR_ARGS, dtv_t *dtv, struct link_map *the_map)
 
       the_map = listp->slotinfo[idx].map;
     }
-
- again:
-  /* Make sure that, if a dlopen running in parallel forces the
-     variable into static storage, we'll wait until the address in the
-     static TLS block is set up, and use that.  If we're undecided
-     yet, make sure we make the decision holding the lock as well.  */
-  if (__builtin_expect (the_map->l_tls_offset
-			!= FORCED_DYNAMIC_TLS_OFFSET, 0))
+  sigset_t old;
+  _dl_mask_all_signals (&old);
+
+  /* As with update_slotinfo, we use the sigmask as a check against
+     reentrancy.  */
+  if (dtv[GET_ADDR_MODULE].pointer.val != TLS_DTV_UNALLOCATED)
+    goto out;
+
+  /* Synchronize against a parallel dlopen() forcing this variable
+     into static storage.  If that happens, we have to be more careful
+     about initializing the area, as that dlopen() will be iterating
+     the threads to do so itself.  */
+  ptrdiff_t offset;
+  if ((offset = the_map->l_tls_offset) == NO_TLS_OFFSET)
     {
-      __rtld_lock_lock_recursive (GL(dl_load_lock));
-      if (__builtin_expect (the_map->l_tls_offset == NO_TLS_OFFSET, 1))
-	{
-	  the_map->l_tls_offset = FORCED_DYNAMIC_TLS_OFFSET;
-	  __rtld_lock_unlock_recursive (GL(dl_load_lock));
-	}
-      else
+      /* l_tls_offset starts out at NO_TLS_OFFSET, and all attempts to
+	 change it go from NO_TLS_OFFSET to some other value.  We use
+	 compare_and_exchange to ensure only one attempt succeeds.  We
+	 don't actually need any memory ordering here, but _acq is the
+	 weakest available.  */
+      (void) atomic_compare_and_exchange_bool_acq (&the_map->l_tls_offset,
+						   FORCED_DYNAMIC_TLS_OFFSET,
+						   NO_TLS_OFFSET);
+      offset = the_map->l_tls_offset;
+      assert (offset != NO_TLS_OFFSET);
+    }
+  if (offset == FORCED_DYNAMIC_TLS_OFFSET)
+    {
+      allocate_and_init (&dtv[GET_ADDR_MODULE], the_map);
+    }
+  else
+    {
+      void **pp = &dtv[GET_ADDR_MODULE].pointer.val;
+      while (atomic_forced_read (*pp) == TLS_DTV_UNALLOCATED)
 	{
-	  __rtld_lock_unlock_recursive (GL(dl_load_lock));
-	  if (__builtin_expect (the_map->l_tls_offset
-				!= FORCED_DYNAMIC_TLS_OFFSET, 1))
-	    {
-	      void *p = dtv[GET_ADDR_MODULE].pointer.val;
-	      if (__builtin_expect (p == TLS_DTV_UNALLOCATED, 0))
-		goto again;
-
-	      return (char *) p + GET_ADDR_OFFSET;
-	    }
+	  /* for lack of a better (safe) thing to do, just spin.
+	    Someone else (not us; it's done under a signal mask) set
+	    this map to a static TLS offset, and they'll iterate all
+	    threads to initialize it.  They'll eventually write
+	    to pointer.val, at which point we know they've fully
+	    completed initialization.  */
+	  atomic_delay ();
 	}
+      /* Make sure we've picked up their initialization of the actual
+	 block; this pairs against the write barrier in
+	 init_one_static_tls, guaranteeing that we see their write of
+	 the tls_initimage into the static region.  */
+      atomic_read_barrier ();
     }
-  void *p = dtv[GET_ADDR_MODULE].pointer.val = allocate_and_init (the_map);
-  dtv[GET_ADDR_MODULE].pointer.is_static = false;
+out:
+  assert (dtv[GET_ADDR_MODULE].pointer.val != TLS_DTV_UNALLOCATED);
+  _dl_unmask_signals (&old);
 
-  return (char *) p + GET_ADDR_OFFSET;
+  return (char *) dtv[GET_ADDR_MODULE].pointer.val + GET_ADDR_OFFSET;
 }