Fix blocking pthread_join. [BZ #23137]

On s390 (31bit) if glibc is build with -Os, pthread_join sometimes
blocks indefinitely. This is e.g. observable with
testcase intl/tst-gettext6.

pthread_join is calling lll_wait_tid(tid), which performs the futex-wait
syscall in a loop as long as tid != 0 (thread is alive).

On s390 (and build with -Os), tid is loaded from memory before
comparing against zero and then the tid is loaded a second time
in order to pass it to the futex-wait-syscall.
If the thread exits in between, then the futex-wait-syscall is
called with the value zero and it waits until a futex-wake occurs.
As the thread is already exited, there won't be a futex-wake.

In lll_wait_tid, the tid is stored to the local variable __tid,
which is then used as argument for the futex-wait-syscall.
But unfortunately the compiler is allowed to reload the value
from memory.

With this patch, the tid is loaded with atomic_load_acquire.
Then the compiler is not allowed to reload the value for __tid from memory.

ChangeLog:

	[BZ #23137]
	* sysdeps/nptl/lowlevellock.h (lll_wait_tid):
	Use atomic_load_acquire to load __tid.

diff --git a/ChangeLog b/ChangeLog
index 1f1cf0a0497e62758300e9d01ca821c51ae829fd..a081ea41955748a594642d0bc2b1038172248a0b 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+2018-05-04  Stefan Liebler  <stli@linux.vnet.ibm.com>
+
+       [BZ #23137]
+       * sysdeps/nptl/lowlevellock.h (lll_wait_tid):
+       Use atomic_load_acquire to load __tid.
+
 2018-05-02  H.J. Lu  <hongjiu.lu@intel.com>
 
        * sysdeps/unix/sysv/linux/x86_64/swapcontext.S (__swapcontext):

diff --git a/sysdeps/nptl/lowlevellock.h b/sysdeps/nptl/lowlevellock.h
index 8326e2805c74431ed5be45fbb0095ba15b67c3b6..bfbda99940c037229d85252eba7c61d4a102343c 100644 (file)
--- a/sysdeps/nptl/lowlevellock.h
+++ b/sysdeps/nptl/lowlevellock.h
@@ -181,11 +181,14 @@ extern int __lll_timedlock_wait (int *futex, const struct timespec *,
    thread ID while the clone is running and is reset to zero by the kernel
    afterwards.  The kernel up to version 3.16.3 does not use the private futex
    operations for futex wake-up when the clone terminates.  */
-#define lll_wait_tid(tid) \
-  do {                                 \
-    __typeof (tid) __tid;              \
-    while ((__tid = (tid)) != 0)       \
-      lll_futex_wait (&(tid), __tid, LLL_SHARED);\
+#define lll_wait_tid(tid)                              \
+  do {                                                 \
+    __typeof (tid) __tid;                              \
+    /* We need acquire MO here so that we synchronize  \
+       with the kernel's store to 0 when the clone     \
+       terminates. (see above)  */                     \
+    while ((__tid = atomic_load_acquire (&(tid))) != 0)        \
+      lll_futex_wait (&(tid), __tid, LLL_SHARED);      \
   } while (0)
 
 extern int __lll_timedwait_tid (int *, const struct timespec *)