[PATCH] malloc: Rewrite with explicit TLS access using __thread

[Florian Weimer <fweimer at redhat dot com>]

The generated assembly is basically the same.  The only differences are
due to the changed variable name and changed line numbers in asserts.

Florian

2015-10-16  Florian Weimer  <fweimer@redhat.com>

	malloc: Rewrite with explicit TLS access using __thread.
	* sysdeps/generic/malloc-machine.h (tsd_key_t, tsd_key_create)
	(tsd_setspecific, tsd_getspecific): Remove.
	* sysdeps/mach/hurd/malloc-machine.h (tsd_key_t, tsd_key_create)
	(tsd_setspecific, tsd_getspecific): Likewise.
	* sysdeps/nptl/malloc-machine.h (tsd_key_t, tsd_key_create)
	(tsd_setspecific, tsd_getspecific): Likewise.
	* malloc/arena.c (thread_arena): New TLS variable.
	(arena_key): Remove variable.
	(arena_get): Use thread_arena.
	(arena_lookup): Remove macro.
	(malloc_atfork, free_atfork, ptmalloc_lock_all)
	(ptmalloc_unlock_all, ptmalloc_unlock_all2, ptmalloc_init)
	(_int_new_arena, get_free_list, reused_arena)
	(arena_thread_freeres): Use thread_arena.
	* manual/memory.texi (Basic Allocation): Remove arena_lookup,
	tsd_getspecific, tsd_setspecific from safety annotations.
	(Allocating Cleared Space): Remove arena_lookup from safety
	annotations.

diff --git a/malloc/arena.c b/malloc/arena.c
index cb45a04..caf718d 100644
--- a/malloc/arena.c
+++ b/malloc/arena.c
@@ -64,9 +64,12 @@ extern int sanity_check_heap_info_alignment[(sizeof (heap_info)
                                              + 2 * SIZE_SZ) % MALLOC_ALIGNMENT
                                             ? -1 : 1];
 
-/* Thread specific data */
+/* Thread specific data.  */
+
+static __thread mstate thread_arena attribute_tls_model_ie;
+
+/* Arena free list.  */
 
-static tsd_key_t arena_key;
 static mutex_t list_lock = MUTEX_INITIALIZER;
 static size_t narenas = 1;
 static mstate free_list;
@@ -89,15 +92,10 @@ int __malloc_initialized = -1;
    in the new arena. */
 
 #define arena_get(ptr, size) do { \
-      arena_lookup (ptr);						      \
+      ptr = thread_arena;						      \
       arena_lock (ptr, size);						      \
   } while (0)
 
-#define arena_lookup(ptr) do { \
-      void *vptr = NULL;						      \
-      ptr = (mstate) tsd_getspecific (arena_key, vptr);			      \
-  } while (0)
-
 #define arena_lock(ptr, size) do {					      \
       if (ptr && !arena_is_corrupt (ptr))				      \
         (void) mutex_lock (&ptr->mutex);				      \
@@ -138,11 +136,9 @@ ATFORK_MEM;
 static void *
 malloc_atfork (size_t sz, const void *caller)
 {
-  void *vptr = NULL;
   void *victim;
 
-  tsd_getspecific (arena_key, vptr);
-  if (vptr == ATFORK_ARENA_PTR)
+  if (thread_arena == ATFORK_ARENA_PTR)
     {
       /* We are the only thread that may allocate at all.  */
       if (save_malloc_hook != malloc_check)
@@ -172,7 +168,6 @@ malloc_atfork (size_t sz, const void *caller)
 static void
 free_atfork (void *mem, const void *caller)
 {
-  void *vptr = NULL;
   mstate ar_ptr;
   mchunkptr p;                          /* chunk corresponding to mem */
 
@@ -188,8 +183,7 @@ free_atfork (void *mem, const void *caller)
     }
 
   ar_ptr = arena_for_chunk (p);
-  tsd_getspecific (arena_key, vptr);
-  _int_free (ar_ptr, p, vptr == ATFORK_ARENA_PTR);
+  _int_free (ar_ptr, p, thread_arena == ATFORK_ARENA_PTR);
 }
 
 
@@ -212,9 +206,7 @@ ptmalloc_lock_all (void)
 
   if (mutex_trylock (&list_lock))
     {
-      void *my_arena;
-      tsd_getspecific (arena_key, my_arena);
-      if (my_arena == ATFORK_ARENA_PTR)
+      if (thread_arena == ATFORK_ARENA_PTR)
         /* This is the same thread which already locks the global list.
            Just bump the counter.  */
         goto out;
@@ -234,8 +226,8 @@ ptmalloc_lock_all (void)
   __malloc_hook = malloc_atfork;
   __free_hook = free_atfork;
   /* Only the current thread may perform malloc/free calls now. */
-  tsd_getspecific (arena_key, save_arena);
-  tsd_setspecific (arena_key, ATFORK_ARENA_PTR);
+  save_arena = thread_arena;
+  thread_arena = ATFORK_ARENA_PTR;
 out:
   ++atfork_recursive_cntr;
 }
@@ -251,7 +243,7 @@ ptmalloc_unlock_all (void)
   if (--atfork_recursive_cntr != 0)
     return;
 
-  tsd_setspecific (arena_key, save_arena);
+  thread_arena = save_arena;
   __malloc_hook = save_malloc_hook;
   __free_hook = save_free_hook;
   for (ar_ptr = &main_arena;; )
@@ -279,7 +271,7 @@ ptmalloc_unlock_all2 (void)
   if (__malloc_initialized < 1)
     return;
 
-  tsd_setspecific (arena_key, save_arena);
+  thread_arena = save_arena;
   __malloc_hook = save_malloc_hook;
   __free_hook = save_free_hook;
   free_list = NULL;
@@ -372,8 +364,7 @@ ptmalloc_init (void)
     __morecore = __failing_morecore;
 #endif
 
-  tsd_key_create (&arena_key, NULL);
-  tsd_setspecific (arena_key, (void *) &main_arena);
+  thread_arena = &main_arena;
   thread_atfork (ptmalloc_lock_all, ptmalloc_unlock_all, ptmalloc_unlock_all2);
   const char *s = NULL;
   if (__glibc_likely (_environ != NULL))
@@ -761,7 +752,7 @@ _int_new_arena (size_t size)
   set_head (top (a), (((char *) h + h->size) - ptr) | PREV_INUSE);
 
   LIBC_PROBE (memory_arena_new, 2, a, size);
-  tsd_setspecific (arena_key, (void *) a);
+  thread_arena = a;
   mutex_init (&a->mutex);
   (void) mutex_lock (&a->mutex);
 
@@ -794,7 +785,7 @@ get_free_list (void)
         {
           LIBC_PROBE (memory_arena_reuse_free_list, 1, result);
           (void) mutex_lock (&result->mutex);
-          tsd_setspecific (arena_key, (void *) result);
+	  thread_arena = result;
         }
     }
 
@@ -847,7 +838,7 @@ reused_arena (mstate avoid_arena)
 
 out:
   LIBC_PROBE (memory_arena_reuse, 2, result, avoid_arena);
-  tsd_setspecific (arena_key, (void *) result);
+  thread_arena = result;
   next_to_use = result->next;
 
   return result;
@@ -934,9 +925,8 @@ arena_get_retry (mstate ar_ptr, size_t bytes)
 static void __attribute__ ((section ("__libc_thread_freeres_fn")))
 arena_thread_freeres (void)
 {
-  void *vptr = NULL;
-  mstate a = tsd_getspecific (arena_key, vptr);
-  tsd_setspecific (arena_key, NULL);
+  mstate a = thread_arena;
+  thread_arena = NULL;
 
   if (a != NULL)
     {
diff --git a/manual/memory.texi b/manual/memory.texi
index 0729e70..cea2cd7 100644
--- a/manual/memory.texi
+++ b/manual/memory.texi
@@ -332,8 +332,6 @@ this function is in @file{stdlib.h}.
 @c __libc_malloc @asulock @aculock @acsfd @acsmem
 @c  force_reg ok
 @c  *malloc_hook unguarded
-@c  arena_lookup ok
-@c   tsd_getspecific ok, TLS
 @c  arena_lock @asulock @aculock @acsfd @acsmem
 @c   mutex_lock @asulock @aculock
 @c   arena_get2 @asulock @aculock @acsfd @acsmem
@@ -341,7 +339,6 @@ this function is in @file{stdlib.h}.
 @c     mutex_lock (list_lock) dup @asulock @aculock
 @c     mutex_unlock (list_lock) dup @aculock
 @c     mutex_lock (arena lock) dup @asulock @aculock [returns locked]
-@c     tsd_setspecific ok, TLS
 @c    __get_nprocs ext ok @acsfd
 @c    NARENAS_FROM_NCORES ok
 @c    catomic_compare_and_exchange_bool_acq ok
@@ -835,7 +832,6 @@ is declared in @file{stdlib.h}.
 @c  *__malloc_hook dup unguarded
 @c  memset dup ok
 @c  arena_get @asulock @aculock @acsfd @acsmem
-@c   arena_lookup dup ok
 @c   arena_lock dup @asulock @aculock @acsfd @acsmem
 @c  top dup ok
 @c  chunksize dup ok
diff --git a/sysdeps/generic/malloc-machine.h b/sysdeps/generic/malloc-machine.h
index 10f6e72..802d1f5 100644
--- a/sysdeps/generic/malloc-machine.h
+++ b/sysdeps/generic/malloc-machine.h
@@ -40,13 +40,6 @@ typedef int mutex_t;
 # define mutex_unlock(m)        (*(m) = 0)
 # define MUTEX_INITIALIZER      (0)
 
-typedef void *tsd_key_t;
-# define tsd_key_create(key, destr) do {} while(0)
-# define tsd_setspecific(key, data) ((key) = (data))
-# define tsd_getspecific(key, vptr) (vptr = (key))
-
-# define thread_atfork(prepare, parent, child) do {} while(0)
-
 #endif /* !defined mutex_init */
 
 #ifndef atomic_full_barrier
diff --git a/sysdeps/mach/hurd/malloc-machine.h b/sysdeps/mach/hurd/malloc-machine.h
index 1fdbd3d..9221d1b 100644
--- a/sysdeps/mach/hurd/malloc-machine.h
+++ b/sysdeps/mach/hurd/malloc-machine.h
@@ -52,16 +52,6 @@
 /* No we're *not* using pthreads.  */
 #define __pthread_initialize ((void (*)(void))0)
 
-/* thread specific data for glibc */
-
-#include <libc-tsd.h>
-
-typedef int tsd_key_t[1];	/* no key data structure, libc magic does it */
-__libc_tsd_define (static, void *, MALLOC)	/* declaration/common definition */
-#define tsd_key_create(key, destr)	((void) (key))
-#define tsd_setspecific(key, data)	__libc_tsd_set (void *, MALLOC, (data))
-#define tsd_getspecific(key, vptr)	((vptr) = __libc_tsd_get (void *, MALLOC))
-
 /* madvise is a stub on Hurd, so don't bother calling it.  */
 
 #include <sys/mman.h>
diff --git a/sysdeps/nptl/malloc-machine.h b/sysdeps/nptl/malloc-machine.h
index c0ec49e..8dea606 100644
--- a/sysdeps/nptl/malloc-machine.h
+++ b/sysdeps/nptl/malloc-machine.h
@@ -58,16 +58,6 @@ extern void *__dso_handle __attribute__ ((__weak__));
   __linkin_atfork (&atfork_mem)
 #endif
 
-/* thread specific data for glibc */
-
-#include <libc-tsd.h>
-
-typedef int tsd_key_t[1];	/* no key data structure, libc magic does it */
-__libc_tsd_define (static, void *, MALLOC)	/* declaration/common definition */
-#define tsd_key_create(key, destr)	((void) (key))
-#define tsd_setspecific(key, data)	__libc_tsd_set (void *, MALLOC, (data))
-#define tsd_getspecific(key, vptr)	((vptr) = __libc_tsd_get (void *, MALLOC))
-
 #include <sysdeps/generic/malloc-machine.h>
 
 #endif /* !defined(_MALLOC_MACHINE_H) */