WSL: pthread_mutexattr_init() returns ENOSYS
Might as well file this one too. This first came up in #486 which (naturally) isn’t getting a lot of love over on User Voice. Besides PA, the construct also rears its head in a large popular software package here. Turns out that lock_impl_posix.cc ends up in a library that is in turn used by the gn meta build system. Which, if you close your eyes, you could frame as a CLI development use case.
Build 14905 fails with: pthread_mutexattr_init: Operation not supported. I made some minor mods from the original so it compiles cleanly.
gcc -o mutex_priority mutex_priority.c -pthread
/********************************************************
* An example source module to accompany...
*
* "Using POSIX Threads: Programming with Pthreads"
* by Brad nichols, Dick Buttlar, Jackie Farrell
* O'Reilly & Associates, Inc.
*
********************************************************
* mutex_priority.c
*/
#include <pthread.h>
#include <stdlib.h>
#include <string.h>
#include <sched.h>
#include <unistd.h>
#include <stdio.h>
#ifndef _POSIX_THREAD_PRIO_INHERIT
#error "This system does not support priority inherit protection in mutex"
#endif
pthread_mutex_t m1;
pthread_mutexattr_t mutexattr_prioinherit;
extern int
main(void)
{
int rtn;
int mutex_protocol;
/* What is the default protocol on the host? */
if ((rtn = pthread_mutexattr_init(&mutexattr_prioinherit)) != 0)
fprintf(stderr,"pthread_mutexattr_init: %s",strerror(rtn)),exit(1);
if ((rtn = pthread_mutexattr_getprotocol(
&mutexattr_prioinherit, &mutex_protocol)) != 0)
fprintf(stderr,"pthread_mutexattr_getprotocol: %s",strerror(rtn)),exit(1);
if (mutex_protocol == PTHREAD_PRIO_PROTECT)
printf("Default mutex protocol is PTHREAD_PRIO_PROTECT\n");
else if (mutex_protocol == PTHREAD_PRIO_INHERIT)
printf("Default mutex protocol is PTHREAD_PRIO_INHERIT\n");
else if (mutex_protocol == PTHREAD_PRIO_NONE)
printf("Default mutex protocol is PTHREAD_PRIO_NONE\n");
else
printf("Default mutex protocol is unrecognized\n");
/* Set this mutex attribute to INHERIT */
if ((rtn = pthread_mutexattr_setprotocol(
&mutexattr_prioinherit, PTHREAD_PRIO_INHERIT)) != 0)
fprintf(stderr,"pthread_mutexattr_setprotocol: %s\n",strerror(rtn)),exit(1);
/* Initialize a mutex with the attribute object */
if ((rtn = pthread_mutex_init(&m1,&mutexattr_prioinherit)) != 0)
fprintf(stderr,"pthread_mutexattr_init: %s\n",strerror(rtn)),exit(1);
if ((rtn = pthread_mutex_lock(&m1)) != 0)
fprintf(stderr,"pthread_mutex_lock: %s\n",strerror(rtn)),exit(1);
if ((rtn = pthread_mutex_unlock(&m1)) != 0)
fprintf(stderr,"pthread_mutex_unlock: %s\n",strerror(rtn)),exit(1);
return 0;
}
About this issue
- Original URL
- State: closed
- Created 10 years ago
- Reactions: 7
- Comments: 22 (1 by maintainers)
Marking this as a bug so we can track the request for FUTEX_UNLOCK_PI, FUTEX_CMP_REQUEUE_PRIVATE, and proper zombie support.
The pi-aware futex operations are now available in the Skip-ahead 17627 build.
I’ve checked in a change that adds support for FUTEX_REQUEUE and FUTEX_CMP_REQUEUE (and the _PRIVATE versions obviously). Priority-inheritance futex operations are still on the backlog.
Pinging this, and renewing my query from #996 (message) whether this can be faked for the time being. I assume mapping priority semantics to NT isn’t trivial, else it would have happened already.
So, my humble request is to tell userspace they are getting priority wakeup semantics, but give them normal semantics. If an application is depending on priority semantics it might not work. Maybe. But doing this won’t break anything that isn’t already broken by virtue of the failed system call. Here’s a typical pattern I see in straces out of (I think)
glibcpthreads,It just falls back. Much other code hard fails on the
ENOSYSthough.While I am here, a quick correction on my original issue submission above, where I cited GN. That was incorrect. It looks like GN’s beef is with #1353. Still, the
futex ENOSYSthing must be showing up in your telemetry.