# \[Bug #2724\] fork from other than the main thread causes wrong pthread condition on NetBSD **URL:** \ **Category:** Development\ **Created:** [February 8, 2010, 7:12pm UTC](https://www.ruby-forum.com/t/bug-2724-fork-from-other-than-the-main-thread-causes-wrong-pthread-condition-on-netbsd/183131 "2010-02-08T19:12:44Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![Yui\_NARUSE](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/y/8edcca/32.png) [@Yui\_NARUSE](https://www.ruby-forum.com/u/Yui_NARUSE)\ **Post date:** [February 8, 2010, 7:12pm UTC](https://www.ruby-forum.com/t/bug-2724-fork-from-other-than-the-main-thread-causes-wrong-pthread-condition-on-netbsd/183131/1 "2010-02-08T19:12:44Z")
Bug #2724: fork from other than the main thread causes wrong pthread condition on NetBSD [http://redmine.ruby-lang.org/issues/show/2724](http://redmine.ruby-lang.org/issues/show/2724) e$B5/I\<\]$,3NG’$5$l$F$$$^$9!#e(B e$B8e=R$N%Q%C%A$,$“$^$j$K%”%l$J$N$G$3$NLdBj$Oe(B Third Party’s Issue e$B$H$7!“e(B Ruby e$BB&$G$O=$@5$rF~$l$J$$;v$H$7$^$9$,!”\>pJs$N6&M-$H5-O?$N$?$a$Ke(B e$B$3$3$K%A%1%C%H$r@Z$C$F$\*$-$^$9!#e(B e$B$J$\*!“$3$Ne(B workaround e$B$N:n@.$K$Oe(B @\_enamie$B$5$s$N=u$1$,$”$j$^$7$?!#e(B # Index: thread\_pthread.c — thread\_pthread.c (revision 26615) +++ thread\_pthread.c (working copy) @@ -17,6 +17,93 @@ #include \ #endif +#if defined( **NetBSD\_Version** ) && **NetBSD\_Version** \>= 500000000 +/\* Hack for NetBSD 5.0.x’s broken pthread-\>pt\_lid _/ +/_ Copied from /src/lib/libpthread/pthread\_int.h _/ +#define BROKEN\_PTHREAD\_T\_PT\_LID +#include \ +#include \ +#include \ + +#define PTHREAD\_KEYS\_MAX 256 +#define PTHREAD\_\_UNPARK\_MAX 32 + +/_ - - The size of this structure needs to be no larger than struct - - \_\_pthread\_cleanup\_store, defined in pthread.h. - \*/ +struct pt\_clean\_t { - ``` PTQ_ENTRY(pt_clean_t) ptc_next; ``` - ``` void (*ptc_cleanup)(void *); ``` - ``` void *ptc_arg; ``` +}; + +struct pthread\_lock\_ops { - ``` void (*plo_init)(__cpu_simple_lock_t *); ``` - ``` int (*plo_try)(__cpu_simple_lock_t *); ``` - ``` void (*plo_unlock)(__cpu_simple_lock_t *); ``` - ``` void (*plo_lock)(__cpu_simple_lock_t *); ``` +}; + +struct \_\_pthread\_st { - ``` pthread_t pt_self; /* Must be first. */ ``` - ``` unsigned int pt_magic; /* Magic number */ ``` - ``` int pt_state; /* running, blocked, etc. */ ``` - ``` pthread_mutex_t pt_lock; /* lock on state */ ``` - ``` int pt_flags; /* see PT_FLAG_* below */ ``` - ``` int pt_cancel; /* Deferred cancellation */ ``` - ``` int pt_errno; /* Thread-specific errno. */ ``` - ``` stack_t pt_stack; /* Our stack */ ``` - ``` void *pt_exitval; /* Read by pthread_join() */ ``` - ``` char *pt_name; /* Thread's name, set by the ``` app. \*/ - ``` int pt_willpark; /* About to park */ ``` - ``` lwpid_t pt_unpark; /* Unpark this when parking */ ``` - ``` struct pthread_lock_ops pt_lockops;/* Cached to avoid PIC ``` overhead \*/ - ``` pthread_mutex_t *pt_droplock; /* Drop this lock if cancelled ``` \*/ - ``` pthread_cond_t pt_joiners; /* Threads waiting to join. */ ``` - - ``` /* Threads to defer waking, usually until ``` pthread\_mutex\_unlock(). \*/ - ``` lwpid_t pt_waiters[PTHREAD__UNPARK_MAX]; ``` - ``` size_t pt_nwaiters; ``` - - ``` /* Stack of cancellation cleanup handlers and their arguments */ ``` - ``` PTQ_HEAD(, pt_clean_t) pt_cleanup_stack; ``` - - ``` /* LWP ID and entry on the list of all threads. */ ``` - ``` lwpid_t pt_lid; ``` - ``` RB_ENTRY(__pthread_st) pt_alltree; ``` - ``` PTQ_ENTRY(__pthread_st) pt_allq; ``` - ``` PTQ_ENTRY(__pthread_st) pt_deadq; ``` - - ``` /* ``` - ``` * General synchronization data. We try to align, as threads ``` - ``` * on other CPUs will access this data frequently. ``` - ``` */ ``` - ``` int pt_dummy1 __aligned(128); ``` - ``` struct lwpctl *pt_lwpctl; /* Kernel/user comms area */ ``` - ``` volatile int pt_blocking; /* Blocking in userspace */ ``` - ``` volatile int pt_rwlocked; /* Handed rwlock successfully */ ``` - ``` volatile int pt_signalled; /* Received ``` pthread\_cond\_signal() \*/ - ``` volatile int pt_mutexwait; /* Waiting to acquire mutex */ ``` - ``` void * volatile pt_mutexnext; /* Next thread in chain */ ``` - ``` void * volatile pt_sleepobj; /* Object slept on */ ``` - ``` PTQ_ENTRY(__pthread_st) pt_sleep; ``` - ``` void (*pt_early)(void *); ``` - ``` int pt_dummy2 __aligned(128); ``` - - ``` /* Thread-specific data. Large so it sits close to the end. */ ``` - ``` int pt_havespecific; ``` - ``` void *pt_specific[PTHREAD_KEYS_MAX]; ``` - - ``` /* ``` - ``` * Context for thread creation. At the end as it's cached ``` - ``` * and then only ever passed to _lwp_create(). ``` - ``` */ ``` - ``` ucontext_t pt_uc; ``` +}; +#endif /\* **NetBSD** \*/ + + static void native\_mutex\_lock(pthread\_mutex\_t \*lock); static void native\_mutex\_unlock(pthread\_mutex\_t \*lock); static int native\_mutex\_trylock(pthread\_mutex\_t \*lock); @@ -833,6 +920,9 @@ native\_reset\_timer\_thread(void) { timer\_thread\_id = 0; +#ifdef BROKEN\_PTHREAD\_T\_PT\_LID - ((struct \_\_pthread\_st \*)pthread\_self())-\>pt\_lid = \_lwp\_self(); +#endif } #ifdef HAVE\_SIGALTSTACK ---
**Author:** ![Yui\_NARUSE](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/y/8edcca/32.png) [@Yui\_NARUSE](https://www.ruby-forum.com/u/Yui_NARUSE)\ **Post date:** [April 6, 2010, 12:20pm UTC](https://www.ruby-forum.com/t/bug-2724-fork-from-other-than-the-main-thread-causes-wrong-pthread-condition-on-netbsd/183131/2 "2010-04-06T12:20:27Z")
e$B%A%1%C%He(B #2724 e$B$,99?7$5$l$^$7$?!#e(B (by Michael G.) I have made a patch (which was based upon this problem report) to NetBSD-current two weeks ago. I do not recommend this patch be accepted, but instead be marked as a “OS vendor problem” and fixed as such. ## I have requested a pull-up request to the NetBSD release engineers as well, for NetBSD 5.x, which should make it in the next NetBSD 5.x release on that path. [http://redmine.ruby-lang.org/issues/show/2724](http://redmine.ruby-lang.org/issues/show/2724) ---
**Author:** ![Yui\_NARUSE](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/y/8edcca/32.png) [@Yui\_NARUSE](https://www.ruby-forum.com/u/Yui_NARUSE)\ **Post date:** [April 6, 2010, 12:50pm UTC](https://www.ruby-forum.com/t/bug-2724-fork-from-other-than-the-main-thread-causes-wrong-pthread-condition-on-netbsd/183131/3 "2010-04-06T12:50:54Z")
e$B%A%1%C%He(B #2724 e$B$,99?7$5$l$^$7$?!#e(B (by Usaku NAKAMURA) Original reporter Naruse says that “This is a problem of OS itself. Following patch is only a sample of the workaround. We shouldn’t apply this patch to ruby. I leave it only the record for the sharing of knowledge.” So, this ticket has been already marked as “Third Party’s Issue”. ## BTW, it’s very good news that this problem is fixed at NetBSD side. Thank you for giving a useful information! [http://redmine.ruby-lang.org/issues/show/2724](http://redmine.ruby-lang.org/issues/show/2724)