Fix atomic_full_barrier on x86 and x86_64.

This fixes BZ #17403 by defining atomic_full_barrier,
atomic_read_barrier, and atomic_write_barrier on x86 and x86_64.  A full
barrier is implemented through an atomic idempotent modification to the
stack and not through using mfence because the latter can supposedly be
somewhat slower due to having to provide stronger guarantees wrt.
self-modifying code, for example.

4 files changed, 35 insertions, 12 deletions

diff --git a/ChangeLog b/ChangeLog
index d2132be616..6c18edd5c5 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,11 @@
+2015-06-23  Torvald Riegel  <triegel@redhat.com>
+
+	[BZ #17403]
+	* sysdeps/x86_64/bits/atomic.h: (atomic_full_barrier,
+	atomic_read_barrier, atomic_write_barrier): Define.
+	* sysdeps/i386/i486/bits/atomic.h (atomic_full_barrier,
+	atomic_read_barrier, atomic_write_barrier): Define.
+
 2015-06-23  Joseph Myers  <joseph@codesourcery.com>
 
 	* math/Makefile [$(PERL) != no]
diff --git a/NEWS b/NEWS
index fcffaead68..a46efd69bc 100644
--- a/NEWS
+++ b/NEWS
@@ -12,18 +12,19 @@ Version 2.22
   438, 4719, 6792, 13028, 13064, 14094, 14841, 14906, 14958, 15319, 15467,
   15790, 15969, 16159, 16339, 16350, 16351, 16352, 16353, 16361, 16512,
   16560, 16704, 16783, 16850, 17053, 17090, 17195, 17269, 17293, 17322,
-  17523, 17542, 17569, 17581, 17588, 17596, 17620, 17621, 17628, 17631,
-  17692, 17711, 17715, 17776, 17779, 17792, 17836, 17912, 17916, 17930,
-  17932, 17944, 17949, 17964, 17965, 17967, 17969, 17977, 17978, 17987,
-  17991, 17996, 17998, 17999, 18007, 18019, 18020, 18029, 18030, 18032,
-  18034, 18036, 18038, 18039, 18042, 18043, 18046, 18047, 18049, 18068,
-  18080, 18093, 18100, 18104, 18110, 18111, 18116, 18125, 18128, 18138,
-  18185, 18196, 18197, 18206, 18210, 18211, 18217, 18219, 18220, 18221,
-  18234, 18244, 18247, 18287, 18319, 18324, 18333, 18346, 18371, 18397,
-  18409, 18410, 18412, 18418, 18422, 18434, 18444, 18468, 18469, 18470,
-  18479, 18483, 18495, 18496, 18497, 18498, 18507, 18512, 18513, 18519,
-  18520, 18522, 18527, 18528, 18529, 18530, 18532, 18533, 18534, 18536,
-  18539, 18540, 18542, 18544, 18545, 18546, 18547, 18553, 18558, 18569.
+  17403, 17523, 17542, 17569, 17581, 17588, 17596, 17620, 17621, 17628,
+  17631, 17692, 17711, 17715, 17776, 17779, 17792, 17836, 17912, 17916,
+  17930, 17932, 17944, 17949, 17964, 17965, 17967, 17969, 17977, 17978,
+  17987, 17991, 17996, 17998, 17999, 18007, 18019, 18020, 18029, 18030,
+  18032, 18034, 18036, 18038, 18039, 18042, 18043, 18046, 18047, 18049,
+  18068, 18080, 18093, 18100, 18104, 18110, 18111, 18116, 18125, 18128,
+  18138, 18185, 18196, 18197, 18206, 18210, 18211, 18217, 18219, 18220,
+  18221, 18234, 18244, 18247, 18287, 18319, 18324, 18333, 18346, 18371,
+  18397, 18409, 18410, 18412, 18418, 18422, 18434, 18444, 18468, 18469,
+  18470, 18479, 18483, 18495, 18496, 18497, 18498, 18507, 18512, 18513,
+  18519, 18520, 18522, 18527, 18528, 18529, 18530, 18532, 18533, 18534,
+  18536, 18539, 18540, 18542, 18544, 18545, 18546, 18547, 18553, 18558,
+  18569.
 
 * Cache information can be queried via sysconf() function on s390 e.g. with
   _SC_LEVEL1_ICACHE_SIZE as argument.
diff --git a/sysdeps/i386/i486/bits/atomic.h b/sysdeps/i386/i486/bits/atomic.h
index 01c2b942b3..59165bec94 100644
--- a/sysdeps/i386/i486/bits/atomic.h
+++ b/sysdeps/i386/i486/bits/atomic.h
@@ -535,3 +535,10 @@ typedef uintmax_t uatomic_max_t;
 #define atomic_or(mem, mask) __arch_or_body (LOCK_PREFIX, mem, mask)
 
 #define catomic_or(mem, mask) __arch_or_body (__arch_cprefix, mem, mask)
+
+/* We don't use mfence because it is supposedly slower due to having to
+   provide stronger guarantees (e.g., regarding self-modifying code).  */
+#define atomic_full_barrier() \
+    __asm __volatile (LOCK_PREFIX "orl $0, (%%esp)" ::: "memory")
+#define atomic_read_barrier() __asm ("" ::: "memory")
+#define atomic_write_barrier() __asm ("" ::: "memory")
diff --git a/sysdeps/x86_64/bits/atomic.h b/sysdeps/x86_64/bits/atomic.h
index 5d08146cd7..203d92c20d 100644
--- a/sysdeps/x86_64/bits/atomic.h
+++ b/sysdeps/x86_64/bits/atomic.h
@@ -472,3 +472,10 @@ typedef uintmax_t uatomic_max_t;
 #define atomic_or(mem, mask) __arch_or_body (LOCK_PREFIX, mem, mask)
 
 #define catomic_or(mem, mask) __arch_or_body (__arch_cprefix, mem, mask)
+
+/* We don't use mfence because it is supposedly slower due to having to
+   provide stronger guarantees (e.g., regarding self-modifying code).  */
+#define atomic_full_barrier() \
+    __asm __volatile (LOCK_PREFIX "orl $0, (%%rsp)" ::: "memory")
+#define atomic_read_barrier() __asm ("" ::: "memory")
+#define atomic_write_barrier() __asm ("" ::: "memory")