Re: [PATCH] [BZ #20985] robust mutexes: Fix broken x86 assembly by removing it

[Torvald Riegel <triegel at redhat dot com>]

On Thu, 2016-12-22 at 12:23 +0100, Torvald Riegel wrote:
> On Thu, 2016-12-22 at 12:06 +0100, Florian Weimer wrote:
> > On 12/22/2016 11:22 AM, Torvald Riegel wrote:
> > >     	(LLL_ROBUST_MUTEX_LOCK_MODIFIER): New.
> > 
> > This needs to be documented in the code.  It seems the code always 
> > defines it as 0, so it's not clear to me why it makes sense to define it 
> > at all (twice even).
> 
> This should be defined differently when pthread_mutex_cond_lock.c
> includes the file; I just noticed that it doesn't do that.
> 
> Is the patch OK besides that (ie, so after fixing that oversight and
> adding a suitable comment to the definition of
> LLL_ROBUST_MUTEX_LOCK_MODIFIER?).

In particular, with this change:

diff --git a/nptl/pthread_mutex_cond_lock.c b/nptl/pthread_mutex_cond_lock.c
index 6ea6b05..bd7149e 100644
--- a/nptl/pthread_mutex_cond_lock.c
+++ b/nptl/pthread_mutex_cond_lock.c
@@ -11,7 +11,9 @@
   lll_cond_trylock ((mutex)->__data.__lock)
 #define LLL_MUTEX_TRYLOCK_ELISION(mutex) LLL_MUTEX_TRYLOCK(mutex)
 
-#define LLL_ROBUST_MUTEX_LOCK_MODIFIER 0
+/* We need to assume that there are other threads blocked on the futex.
+   See __pthread_mutex_lock_full for further details.  */
+#define LLL_ROBUST_MUTEX_LOCK_MODIFIER FUTEX_WAITERS
 #define __pthread_mutex_lock internal_function __pthread_mutex_cond_lock
 #define __pthread_mutex_lock_full __pthread_mutex_cond_lock_full
 #define NO_INCR
diff --git a/nptl/pthread_mutex_lock.c b/nptl/pthread_mutex_lock.c
index dff735e..4dbf47d 100644
--- a/nptl/pthread_mutex_lock.c
+++ b/nptl/pthread_mutex_lock.c
@@ -36,6 +36,8 @@
 #define lll_trylock_elision(a,t) lll_trylock(a)
 #endif
 
+/* Some of the following definitions differ when pthread_mutex_cond_lock.c
+   includes this file.  */
 #ifndef LLL_MUTEX_LOCK
 # define LLL_MUTEX_LOCK(mutex) \
   lll_lock ((mutex)->__data.__lock, PTHREAD_MUTEX_PSHARED (mutex))