1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/thread_sync.c

1463 lines
32 KiB
C
Raw Normal View History

/* included by thread.c */
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
#include "ccan/list/list.h"
static VALUE rb_cMutex, rb_cQueue, rb_cSizedQueue, rb_cConditionVariable;
static VALUE rb_eClosedQueueError;
/* sync_waiter is always on-stack */
struct sync_waiter {
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
rb_thread_t *th;
struct list_node node;
};
#define MUTEX_ALLOW_TRAP FL_USER1
static int
wakeup_one(struct list_head *head)
{
struct sync_waiter *cur = 0, *next = 0;
list_for_each_safe(head, cur, next, node) {
list_del_init(&cur->node);
if (cur->th->status != THREAD_KILLED) {
rb_threadptr_interrupt(cur->th);
cur->th->status = THREAD_RUNNABLE;
return TRUE;
}
}
return FALSE;
}
static void
wakeup_all(struct list_head *head)
{
struct sync_waiter *cur = 0, *next = 0;
list_for_each_safe(head, cur, next, node) {
list_del_init(&cur->node);
if (cur->th->status != THREAD_KILLED) {
rb_threadptr_interrupt(cur->th);
cur->th->status = THREAD_RUNNABLE;
}
}
}
/* Mutex */
typedef struct rb_mutex_struct {
struct rb_thread_struct volatile *th;
struct rb_mutex_struct *next_mutex;
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
struct list_head waitq; /* protected by GVL */
} rb_mutex_t;
#if defined(HAVE_WORKING_FORK)
static void rb_mutex_abandon_all(rb_mutex_t *mutexes);
static void rb_mutex_abandon_keeping_mutexes(rb_thread_t *th);
static void rb_mutex_abandon_locking_mutex(rb_thread_t *th);
#endif
static const char* rb_mutex_unlock_th(rb_mutex_t *mutex, rb_thread_t volatile *th);
/*
* Document-class: Mutex
*
* Mutex implements a simple semaphore that can be used to coordinate access to
* shared data from multiple concurrent threads.
*
* Example:
*
* semaphore = Mutex.new
*
* a = Thread.new {
* semaphore.synchronize {
* # access shared resource
* }
* }
*
* b = Thread.new {
* semaphore.synchronize {
* # access shared resource
* }
* }
*
*/
#define GetMutexPtr(obj, tobj) \
TypedData_Get_Struct((obj), rb_mutex_t, &mutex_data_type, (tobj))
#define mutex_mark NULL
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
static size_t
rb_mutex_num_waiting(rb_mutex_t *mutex)
{
struct sync_waiter *w = 0;
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
size_t n = 0;
list_for_each(&mutex->waitq, w, node) {
n++;
}
return n;
}
static void
mutex_free(void *ptr)
{
rb_mutex_t *mutex = ptr;
if (mutex->th) {
/* rb_warn("free locked mutex"); */
const char *err = rb_mutex_unlock_th(mutex, mutex->th);
if (err) rb_bug("%s", err);
}
ruby_xfree(ptr);
}
static size_t
mutex_memsize(const void *ptr)
{
return sizeof(rb_mutex_t);
}
static const rb_data_type_t mutex_data_type = {
"mutex",
{mutex_mark, mutex_free, mutex_memsize,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY
};
VALUE
rb_obj_is_mutex(VALUE obj)
{
if (rb_typeddata_is_kind_of(obj, &mutex_data_type)) {
return Qtrue;
}
else {
return Qfalse;
}
}
static VALUE
mutex_alloc(VALUE klass)
{
VALUE obj;
rb_mutex_t *mutex;
obj = TypedData_Make_Struct(klass, rb_mutex_t, &mutex_data_type, mutex);
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
list_head_init(&mutex->waitq);
return obj;
}
/*
* call-seq:
* Mutex.new -> mutex
*
* Creates a new Mutex
*/
static VALUE
mutex_initialize(VALUE self)
{
return self;
}
VALUE
rb_mutex_new(void)
{
return mutex_alloc(rb_cMutex);
}
/*
* call-seq:
* mutex.locked? -> true or false
*
* Returns +true+ if this lock is currently held by some thread.
*/
VALUE
rb_mutex_locked_p(VALUE self)
{
rb_mutex_t *mutex;
GetMutexPtr(self, mutex);
return mutex->th ? Qtrue : Qfalse;
}
static void
mutex_locked(rb_thread_t *th, VALUE self)
{
rb_mutex_t *mutex;
GetMutexPtr(self, mutex);
if (th->keeping_mutexes) {
mutex->next_mutex = th->keeping_mutexes;
}
th->keeping_mutexes = mutex;
}
/*
* call-seq:
* mutex.try_lock -> true or false
*
* Attempts to obtain the lock and returns immediately. Returns +true+ if the
* lock was granted.
*/
VALUE
rb_mutex_trylock(VALUE self)
{
rb_mutex_t *mutex;
VALUE locked = Qfalse;
GetMutexPtr(self, mutex);
if (mutex->th == 0) {
rb_thread_t *th = GET_THREAD();
mutex->th = th;
locked = Qtrue;
mutex_locked(th, self);
}
return locked;
}
/*
* At maximum, only one thread can use cond_timedwait and watch deadlock
* periodically. Multiple polling thread (i.e. concurrent deadlock check)
* introduces new race conditions. [Bug #6278] [ruby-core:44275]
*/
static const rb_thread_t *patrol_thread = NULL;
/*
* call-seq:
* mutex.lock -> self
*
* Attempts to grab the lock and waits if it isn't available.
* Raises +ThreadError+ if +mutex+ was locked by the current thread.
*/
VALUE
rb_mutex_lock(VALUE self)
{
rb_thread_t *th = GET_THREAD();
rb_mutex_t *mutex;
GetMutexPtr(self, mutex);
/* When running trap handler */
if (!FL_TEST_RAW(self, MUTEX_ALLOW_TRAP) &&
th->ec->interrupt_mask & TRAP_INTERRUPT_MASK) {
rb_raise(rb_eThreadError, "can't be called from trap context");
}
if (rb_mutex_trylock(self) == Qfalse) {
struct sync_waiter w;
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
if (mutex->th == th) {
rb_raise(rb_eThreadError, "deadlock; recursive locking");
}
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
w.th = th;
while (mutex->th != th) {
enum rb_thread_status prev_status = th->status;
struct timespec *timeout = 0;
struct timespec ts = { 0, 100000000 }; /* 100ms */
th->status = THREAD_STOPPED_FOREVER;
th->locking_mutex = self;
th->vm->sleeper++;
/*
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
* Carefully! while some contended threads are in native_sleep(),
* vm->sleeper is unstable value. we have to avoid both deadlock
* and busy loop.
*/
if ((vm_living_thread_num(th->vm) == th->vm->sleeper) &&
!patrol_thread) {
timeout = &ts;
patrol_thread = th;
}
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
list_add_tail(&mutex->waitq, &w.node);
native_sleep(th, timeout); /* release GVL */
list_del(&w.node);
if (!mutex->th) {
mutex->th = th;
}
if (patrol_thread == th)
patrol_thread = NULL;
th->locking_mutex = Qfalse;
if (mutex->th && timeout && !RUBY_VM_INTERRUPTED(th->ec)) {
rb_check_deadlock(th->vm);
}
if (th->status == THREAD_STOPPED_FOREVER) {
th->status = prev_status;
}
th->vm->sleeper--;
if (mutex->th == th) mutex_locked(th, self);
RUBY_VM_CHECK_INTS_BLOCKING(th->ec);
}
}
return self;
}
/*
* call-seq:
* mutex.owned? -> true or false
*
* Returns +true+ if this lock is currently held by current thread.
*/
VALUE
rb_mutex_owned_p(VALUE self)
{
VALUE owned = Qfalse;
rb_thread_t *th = GET_THREAD();
rb_mutex_t *mutex;
GetMutexPtr(self, mutex);
if (mutex->th == th)
owned = Qtrue;
return owned;
}
static const char *
rb_mutex_unlock_th(rb_mutex_t *mutex, rb_thread_t volatile *th)
{
const char *err = NULL;
if (mutex->th == 0) {
err = "Attempt to unlock a mutex which is not locked";
}
else if (mutex->th != th) {
err = "Attempt to unlock a mutex which is locked by another thread";
}
else {
struct sync_waiter *cur = 0, *next = 0;
rb_mutex_t *volatile *th_mutex = &th->keeping_mutexes;
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
mutex->th = 0;
list_for_each_safe(&mutex->waitq, cur, next, node) {
list_del_init(&cur->node);
switch (cur->th->status) {
case THREAD_RUNNABLE: /* from someone else calling Thread#run */
case THREAD_STOPPED_FOREVER: /* likely (rb_mutex_lock) */
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
rb_threadptr_interrupt(cur->th);
goto found;
case THREAD_STOPPED: /* probably impossible */
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
rb_bug("unexpected THREAD_STOPPED");
case THREAD_KILLED:
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
/* not sure about this, possible in exit GC? */
rb_bug("unexpected THREAD_KILLED");
continue;
}
}
found:
while (*th_mutex != mutex) {
th_mutex = &(*th_mutex)->next_mutex;
}
*th_mutex = mutex->next_mutex;
mutex->next_mutex = NULL;
}
return err;
}
/*
* call-seq:
* mutex.unlock -> self
*
* Releases the lock.
* Raises +ThreadError+ if +mutex+ wasn't locked by the current thread.
*/
VALUE
rb_mutex_unlock(VALUE self)
{
const char *err;
rb_mutex_t *mutex;
GetMutexPtr(self, mutex);
err = rb_mutex_unlock_th(mutex, GET_THREAD());
if (err) rb_raise(rb_eThreadError, "%s", err);
return self;
}
#if defined(HAVE_WORKING_FORK)
static void
rb_mutex_abandon_keeping_mutexes(rb_thread_t *th)
{
if (th->keeping_mutexes) {
rb_mutex_abandon_all(th->keeping_mutexes);
}
th->keeping_mutexes = NULL;
}
static void
rb_mutex_abandon_locking_mutex(rb_thread_t *th)
{
rb_mutex_t *mutex;
if (!th->locking_mutex) return;
GetMutexPtr(th->locking_mutex, mutex);
if (mutex->th == th)
rb_mutex_abandon_all(mutex);
th->locking_mutex = Qfalse;
}
static void
rb_mutex_abandon_all(rb_mutex_t *mutexes)
{
rb_mutex_t *mutex;
while (mutexes) {
mutex = mutexes;
mutexes = mutex->next_mutex;
mutex->th = 0;
mutex->next_mutex = 0;
reduce rb_mutex_t size from 160 to 80 bytes on 64-bit Instead of relying on a native condition variable and mutex for every Ruby Mutex object, use a doubly linked-list to implement a waiter queue in the Mutex. The immediate benefit of this is reducing the size of every Mutex object, as some projects have many objects requiring synchronization. In the future, this technique using a linked-list and on-stack list node (struct mutex_waiter) should allow us to easily transition to M:N threading model, as we can avoid the native thread dependency to implement Mutex. We already do something similar for autoload in variable.c, and this was inspired by the Linux kernel wait queue (as ccan/list is inspired by the Linux kernel linked-list). Finaly, there are big performance improvements for Mutex benchmarks, especially in contended cases: measure target: real name |trunk |built ----------------|------:|------: loop_whileloop2 | 0.149| 0.148 vm2_mutex* | 0.893| 0.651 vm_thread_mutex1| 0.809| 0.624 vm_thread_mutex2| 2.608| 0.628 vm_thread_mutex3| 28.227| 0.881 Speedup ratio: compare with the result of `trunk' (greater is better) name |built ----------------|------: loop_whileloop2 | 1.002 vm2_mutex* | 1.372 vm_thread_mutex1| 1.297 vm_thread_mutex2| 4.149 vm_thread_mutex3| 32.044 Tested on AMD FX-8320 8-core at 3.5GHz * thread_sync.c (struct mutex_waiter): new on-stack struct (struct rb_mutex_struct): remove native lock/cond, use ccan/list (rb_mutex_num_waiting): new function for debug_deadlock_check (mutex_free): remove native_*_destroy (mutex_alloc): initialize waitq, remove native_*_initialize (rb_mutex_trylock): remove native_mutex_{lock,unlock} (lock_func): remove (lock_interrupt): remove (rb_mutex_lock): rewrite waiting path to use native_sleep + ccan/list (rb_mutex_unlock_th): rewrite to wake up from native_sleep using rb_threadptr_interrupt (rb_mutex_abandon_all): empty waitq * thread.c (debug_deadlock_check): update for new struct (rb_check_deadlock): ditto [ruby-core:80913] [Feature #13517] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58604 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2017-05-07 20:18:53 -04:00
list_head_init(&mutex->waitq);
}
}
#endif
static VALUE
rb_mutex_sleep_forever(VALUE time)
{
rb_thread_sleep_deadly_allow_spurious_wakeup();
return Qnil;
}
static VALUE
rb_mutex_wait_for(VALUE time)
{
struct timespec *t = (struct timespec*)time;
sleep_timespec(GET_THREAD(), *t, 0); /* permit spurious check */
return Qnil;
}
VALUE
rb_mutex_sleep(VALUE self, VALUE timeout)
{
time_t beg, end;
struct timeval t;
if (!NIL_P(timeout)) {
t = rb_time_interval(timeout);
}
rb_mutex_unlock(self);
beg = time(0);
if (NIL_P(timeout)) {
rb_ensure(rb_mutex_sleep_forever, Qnil, rb_mutex_lock, self);
}
else {
struct timespec ts;
VALUE tsp = (VALUE)timespec_for(&ts, &t);
rb_ensure(rb_mutex_wait_for, tsp, rb_mutex_lock, self);
}
end = time(0) - beg;
return INT2FIX(end);
}
/*
* call-seq:
* mutex.sleep(timeout = nil) -> number
*
* Releases the lock and sleeps +timeout+ seconds if it is given and
* non-nil or forever. Raises +ThreadError+ if +mutex+ wasn't locked by
* the current thread.
*
* When the thread is next woken up, it will attempt to reacquire
* the lock.
*
* Note that this method can wakeup without explicit Thread#wakeup call.
* For example, receiving signal and so on.
*/
static VALUE
mutex_sleep(int argc, VALUE *argv, VALUE self)
{
VALUE timeout;
rb_scan_args(argc, argv, "01", &timeout);
return rb_mutex_sleep(self, timeout);
}
/*
* call-seq:
* mutex.synchronize { ... } -> result of the block
*
* Obtains a lock, runs the block, and releases the lock when the block
* completes. See the example under +Mutex+.
*/
VALUE
rb_mutex_synchronize(VALUE mutex, VALUE (*func)(VALUE arg), VALUE arg)
{
rb_mutex_lock(mutex);
return rb_ensure(func, arg, rb_mutex_unlock, mutex);
}
/*
* call-seq:
* mutex.synchronize { ... } -> result of the block
*
* Obtains a lock, runs the block, and releases the lock when the block
* completes. See the example under +Mutex+.
*/
static VALUE
rb_mutex_synchronize_m(VALUE self, VALUE args)
{
if (!rb_block_given_p()) {
rb_raise(rb_eThreadError, "must be called with a block");
}
return rb_mutex_synchronize(self, rb_yield, Qundef);
}
void rb_mutex_allow_trap(VALUE self, int val)
{
Check_TypedStruct(self, &mutex_data_type);
if (val)
FL_SET_RAW(self, MUTEX_ALLOW_TRAP);
else
FL_UNSET_RAW(self, MUTEX_ALLOW_TRAP);
}
/* Queue */
#define queue_waitq(q) UNALIGNED_MEMBER_PTR(q, waitq)
PACKED_STRUCT_UNALIGNED(struct rb_queue {
struct list_head waitq;
const VALUE que;
int num_waiting;
});
#define szqueue_waitq(sq) UNALIGNED_MEMBER_PTR(sq, q.waitq)
#define szqueue_pushq(sq) UNALIGNED_MEMBER_PTR(sq, pushq)
PACKED_STRUCT_UNALIGNED(struct rb_szqueue {
struct rb_queue q;
int num_waiting_push;
struct list_head pushq;
long max;
});
static void
queue_mark(void *ptr)
{
struct rb_queue *q = ptr;
/* no need to mark threads in waitq, they are on stack */
rb_gc_mark(q->que);
}
static size_t
queue_memsize(const void *ptr)
{
return sizeof(struct rb_queue);
}
static const rb_data_type_t queue_data_type = {
"queue",
{queue_mark, RUBY_TYPED_DEFAULT_FREE, queue_memsize,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY|RUBY_TYPED_WB_PROTECTED
};
static VALUE
queue_alloc(VALUE klass)
{
VALUE obj;
struct rb_queue *q;
obj = TypedData_Make_Struct(klass, struct rb_queue, &queue_data_type, q);
list_head_init(queue_waitq(q));
return obj;
}
static struct rb_queue *
queue_ptr(VALUE obj)
{
struct rb_queue *q;
TypedData_Get_Struct(obj, struct rb_queue, &queue_data_type, q);
return q;
}
#define QUEUE_CLOSED FL_USER5
static void
szqueue_mark(void *ptr)
{
struct rb_szqueue *sq = ptr;
queue_mark(&sq->q);
}
static size_t
szqueue_memsize(const void *ptr)
{
return sizeof(struct rb_szqueue);
}
static const rb_data_type_t szqueue_data_type = {
"sized_queue",
{szqueue_mark, RUBY_TYPED_DEFAULT_FREE, szqueue_memsize,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY|RUBY_TYPED_WB_PROTECTED
};
static VALUE
szqueue_alloc(VALUE klass)
{
struct rb_szqueue *sq;
VALUE obj = TypedData_Make_Struct(klass, struct rb_szqueue,
&szqueue_data_type, sq);
list_head_init(szqueue_waitq(sq));
list_head_init(szqueue_pushq(sq));
return obj;
}
static struct rb_szqueue *
szqueue_ptr(VALUE obj)
{
struct rb_szqueue *sq;
TypedData_Get_Struct(obj, struct rb_szqueue, &szqueue_data_type, sq);
return sq;
}
static VALUE
ary_buf_new(void)
{
return rb_ary_tmp_new(1);
}
static VALUE
check_array(VALUE obj, VALUE ary)
{
if (!RB_TYPE_P(ary, T_ARRAY)) {
rb_raise(rb_eTypeError, "%+"PRIsVALUE" not initialized", obj);
}
return ary;
}
static long
queue_length(VALUE self, struct rb_queue *q)
{
return RARRAY_LEN(check_array(self, q->que));
}
static int
queue_closed_p(VALUE self)
{
return FL_TEST_RAW(self, QUEUE_CLOSED) != 0;
}
NORETURN(static void raise_closed_queue_error(VALUE self));
static void
raise_closed_queue_error(VALUE self)
{
rb_raise(rb_eClosedQueueError, "queue closed");
}
static VALUE
queue_closed_result(VALUE self, struct rb_queue *q)
{
assert(queue_length(self, q) == 0);
return Qnil;
}
/*
* Document-class: Queue
*
* The Queue class implements multi-producer, multi-consumer queues.
* It is especially useful in threaded programming when information
* must be exchanged safely between multiple threads. The Queue class
* implements all the required locking semantics.
*
* The class implements FIFO type of queue. In a FIFO queue, the first
* tasks added are the first retrieved.
*
* Example:
*
* queue = Queue.new
*
* producer = Thread.new do
* 5.times do |i|
* sleep rand(i) # simulate expense
* queue << i
* puts "#{i} produced"
* end
* end
*
* consumer = Thread.new do
* 5.times do |i|
* value = queue.pop
* sleep rand(i/2) # simulate expense
* puts "consumed #{value}"
* end
* end
*
*/
/*
* Document-method: Queue::new
*
* Creates a new queue instance.
*/
static VALUE
rb_queue_initialize(VALUE self)
{
struct rb_queue *q = queue_ptr(self);
RB_OBJ_WRITE(self, &q->que, ary_buf_new());
list_head_init(queue_waitq(q));
return self;
}
static VALUE
queue_do_push(VALUE self, struct rb_queue *q, VALUE obj)
{
if (queue_closed_p(self)) {
raise_closed_queue_error(self);
}
rb_ary_push(check_array(self, q->que), obj);
wakeup_one(queue_waitq(q));
return self;
}
/*
* Document-method: Queue#close
* call-seq:
* close
*
* Closes the queue. A closed queue cannot be re-opened.
*
* After the call to close completes, the following are true:
*
* - +closed?+ will return true
*
* - +close+ will be ignored.
*
* - calling enq/push/<< will raise an exception.
*
* - when +empty?+ is false, calling deq/pop/shift will return an object
* from the queue as usual.
*
* ClosedQueueError is inherited from StopIteration, so that you can break loop block.
*
* Example:
*
* q = Queue.new
* Thread.new{
* while e = q.deq # wait for nil to break loop
* # ...
* end
* }
* q.close
*/
static VALUE
rb_queue_close(VALUE self)
{
struct rb_queue *q = queue_ptr(self);
if (!queue_closed_p(self)) {
FL_SET(self, QUEUE_CLOSED);
wakeup_all(queue_waitq(q));
}
return self;
}
/*
* Document-method: Queue#closed?
* call-seq: closed?
*
* Returns +true+ if the queue is closed.
*/
static VALUE
rb_queue_closed_p(VALUE self)
{
return queue_closed_p(self) ? Qtrue : Qfalse;
}
/*
* Document-method: Queue#push
* call-seq:
* push(object)
* enq(object)
* <<(object)
*
* Pushes the given +object+ to the queue.
*/
static VALUE
rb_queue_push(VALUE self, VALUE obj)
{
return queue_do_push(self, queue_ptr(self), obj);
}
static VALUE
queue_sleep(VALUE arg)
{
rb_thread_sleep_deadly_allow_spurious_wakeup();
return Qnil;
}
struct queue_waiter {
struct sync_waiter w;
union {
struct rb_queue *q;
struct rb_szqueue *sq;
} as;
};
static VALUE
queue_sleep_done(VALUE p)
{
struct queue_waiter *qw = (struct queue_waiter *)p;
list_del(&qw->w.node);
qw->as.q->num_waiting--;
return Qfalse;
}
static VALUE
szqueue_sleep_done(VALUE p)
{
struct queue_waiter *qw = (struct queue_waiter *)p;
list_del(&qw->w.node);
qw->as.sq->num_waiting_push--;
return Qfalse;
}
static VALUE
queue_do_pop(VALUE self, struct rb_queue *q, int should_block)
{
check_array(self, q->que);
while (RARRAY_LEN(q->que) == 0) {
if (!should_block) {
rb_raise(rb_eThreadError, "queue empty");
}
else if (queue_closed_p(self)) {
return queue_closed_result(self, q);
}
else {
struct queue_waiter qw;
assert(RARRAY_LEN(q->que) == 0);
assert(queue_closed_p(self) == 0);
qw.w.th = GET_THREAD();
qw.as.q = q;
list_add_tail(&qw.as.q->waitq, &qw.w.node);
qw.as.q->num_waiting++;
rb_ensure(queue_sleep, Qfalse, queue_sleep_done, (VALUE)&qw);
}
}
return rb_ary_shift(q->que);
}
static int
queue_pop_should_block(int argc, const VALUE *argv)
{
int should_block = 1;
rb_check_arity(argc, 0, 1);
if (argc > 0) {
should_block = !RTEST(argv[0]);
}
return should_block;
}
/*
* Document-method: Queue#pop
* call-seq:
* pop(non_block=false)
* deq(non_block=false)
* shift(non_block=false)
*
* Retrieves data from the queue.
*
* If the queue is empty, the calling thread is suspended until data is pushed
* onto the queue. If +non_block+ is true, the thread isn't suspended, and
* +ThreadError+ is raised.
*/
static VALUE
rb_queue_pop(int argc, VALUE *argv, VALUE self)
{
int should_block = queue_pop_should_block(argc, argv);
return queue_do_pop(self, queue_ptr(self), should_block);
}
/*
* Document-method: Queue#empty?
* call-seq: empty?
*
* Returns +true+ if the queue is empty.
*/
static VALUE
rb_queue_empty_p(VALUE self)
{
return queue_length(self, queue_ptr(self)) == 0 ? Qtrue : Qfalse;
}
/*
* Document-method: Queue#clear
*
* Removes all objects from the queue.
*/
static VALUE
rb_queue_clear(VALUE self)
{
struct rb_queue *q = queue_ptr(self);
rb_ary_clear(check_array(self, q->que));
return self;
}
/*
* Document-method: Queue#length
* call-seq:
* length
* size
*
* Returns the length of the queue.
*/
static VALUE
rb_queue_length(VALUE self)
{
return LONG2NUM(queue_length(self, queue_ptr(self)));
}
/*
* Document-method: Queue#num_waiting
*
* Returns the number of threads waiting on the queue.
*/
static VALUE
rb_queue_num_waiting(VALUE self)
{
struct rb_queue *q = queue_ptr(self);
return INT2NUM(q->num_waiting);
}
/*
* Document-class: SizedQueue
*
* This class represents queues of specified size capacity. The push operation
* may be blocked if the capacity is full.
*
* See Queue for an example of how a SizedQueue works.
*/
/*
* Document-method: SizedQueue::new
* call-seq: new(max)
*
* Creates a fixed-length queue with a maximum size of +max+.
*/
static VALUE
rb_szqueue_initialize(VALUE self, VALUE vmax)
{
long max;
struct rb_szqueue *sq = szqueue_ptr(self);
max = NUM2LONG(vmax);
if (max <= 0) {
rb_raise(rb_eArgError, "queue size must be positive");
}
RB_OBJ_WRITE(self, &sq->q.que, ary_buf_new());
list_head_init(szqueue_waitq(sq));
list_head_init(szqueue_pushq(sq));
sq->max = max;
return self;
}
/*
* Document-method: SizedQueue#close
* call-seq:
* close
*
* Similar to Queue#close.
*
* The difference is behavior with waiting enqueuing threads.
*
* If there are waiting enqueuing threads, they are interrupted by
* raising ClosedQueueError('queue closed').
*/
static VALUE
rb_szqueue_close(VALUE self)
{
if (!queue_closed_p(self)) {
struct rb_szqueue *sq = szqueue_ptr(self);
FL_SET(self, QUEUE_CLOSED);
wakeup_all(szqueue_waitq(sq));
wakeup_all(szqueue_pushq(sq));
}
return self;
}
/*
* Document-method: SizedQueue#max
*
* Returns the maximum size of the queue.
*/
static VALUE
rb_szqueue_max_get(VALUE self)
{
return LONG2NUM(szqueue_ptr(self)->max);
}
/*
* Document-method: SizedQueue#max=
* call-seq: max=(number)
*
* Sets the maximum size of the queue to the given +number+.
*/
static VALUE
rb_szqueue_max_set(VALUE self, VALUE vmax)
{
long max = NUM2LONG(vmax);
long diff = 0;
struct rb_szqueue *sq = szqueue_ptr(self);
if (max <= 0) {
rb_raise(rb_eArgError, "queue size must be positive");
}
if (max > sq->max) {
diff = max - sq->max;
}
sq->max = max;
while (diff-- > 0 && wakeup_one(szqueue_pushq(sq))) {
/* keep waking more up */
}
return vmax;
}
static int
szqueue_push_should_block(int argc, const VALUE *argv)
{
int should_block = 1;
rb_check_arity(argc, 1, 2);
if (argc > 1) {
should_block = !RTEST(argv[1]);
}
return should_block;
}
/*
* Document-method: SizedQueue#push
* call-seq:
* push(object, non_block=false)
* enq(object, non_block=false)
* <<(object)
*
* Pushes +object+ to the queue.
*
* If there is no space left in the queue, waits until space becomes
* available, unless +non_block+ is true. If +non_block+ is true, the
* thread isn't suspended, and +ThreadError+ is raised.
*/
static VALUE
rb_szqueue_push(int argc, VALUE *argv, VALUE self)
{
struct rb_szqueue *sq = szqueue_ptr(self);
int should_block = szqueue_push_should_block(argc, argv);
while (queue_length(self, &sq->q) >= sq->max) {
if (!should_block) {
rb_raise(rb_eThreadError, "queue full");
}
else if (queue_closed_p(self)) {
goto closed;
}
else {
struct queue_waiter qw;
struct list_head *pushq = szqueue_pushq(sq);
qw.w.th = GET_THREAD();
qw.as.sq = sq;
list_add_tail(pushq, &qw.w.node);
sq->num_waiting_push++;
rb_ensure(queue_sleep, Qfalse, szqueue_sleep_done, (VALUE)&qw);
}
}
if (queue_closed_p(self)) {
closed:
raise_closed_queue_error(self);
}
return queue_do_push(self, &sq->q, argv[0]);
}
static VALUE
szqueue_do_pop(VALUE self, int should_block)
{
struct rb_szqueue *sq = szqueue_ptr(self);
VALUE retval = queue_do_pop(self, &sq->q, should_block);
if (queue_length(self, &sq->q) < sq->max) {
wakeup_one(szqueue_pushq(sq));
}
return retval;
}
/*
* Document-method: SizedQueue#pop
* call-seq:
* pop(non_block=false)
* deq(non_block=false)
* shift(non_block=false)
*
* Retrieves data from the queue.
*
* If the queue is empty, the calling thread is suspended until data is pushed
* onto the queue. If +non_block+ is true, the thread isn't suspended, and
* +ThreadError+ is raised.
*/
static VALUE
rb_szqueue_pop(int argc, VALUE *argv, VALUE self)
{
int should_block = queue_pop_should_block(argc, argv);
return szqueue_do_pop(self, should_block);
}
/*
* Document-method: SizedQueue#clear
*
* Removes all objects from the queue.
*/
static VALUE
rb_szqueue_clear(VALUE self)
{
struct rb_szqueue *sq = szqueue_ptr(self);
rb_ary_clear(check_array(self, sq->q.que));
wakeup_all(szqueue_pushq(sq));
return self;
}
static VALUE
rb_szqueue_length(VALUE self)
{
struct rb_szqueue *sq = szqueue_ptr(self);
return LONG2NUM(queue_length(self, &sq->q));
}
/*
* Document-method: SizedQueue#num_waiting
*
* Returns the number of threads waiting on the queue.
*/
static VALUE
rb_szqueue_num_waiting(VALUE self)
{
struct rb_szqueue *sq = szqueue_ptr(self);
return INT2NUM(sq->q.num_waiting + sq->num_waiting_push);
}
/*
* Document-method: SizedQueue#empty?
* call-seq: empty?
*
* Returns +true+ if the queue is empty.
*/
static VALUE
rb_szqueue_empty_p(VALUE self)
{
struct rb_szqueue *sq = szqueue_ptr(self);
return queue_length(self, &sq->q) == 0 ? Qtrue : Qfalse;
}
/* ConditionalVariable */
/* TODO: maybe this can be IMEMO */
struct rb_condvar {
struct list_head waitq;
};
/*
* Document-class: ConditionVariable
*
* ConditionVariable objects augment class Mutex. Using condition variables,
* it is possible to suspend while in the middle of a critical section until a
* resource becomes available.
*
* Example:
*
* mutex = Mutex.new
* resource = ConditionVariable.new
*
* a = Thread.new {
* mutex.synchronize {
* # Thread 'a' now needs the resource
* resource.wait(mutex)
* # 'a' can now have the resource
* }
* }
*
* b = Thread.new {
* mutex.synchronize {
* # Thread 'b' has finished using the resource
* resource.signal
* }
* }
*/
static size_t
condvar_memsize(const void *ptr)
{
return sizeof(struct rb_condvar);
}
static const rb_data_type_t cv_data_type = {
"condvar",
{0, RUBY_TYPED_DEFAULT_FREE, condvar_memsize,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY|RUBY_TYPED_WB_PROTECTED
};
static struct rb_condvar *
condvar_ptr(VALUE self)
{
struct rb_condvar *cv;
TypedData_Get_Struct(self, struct rb_condvar, &cv_data_type, cv);
return cv;
}
static VALUE
condvar_alloc(VALUE klass)
{
struct rb_condvar *cv;
VALUE obj;
obj = TypedData_Make_Struct(klass, struct rb_condvar, &cv_data_type, cv);
list_head_init(&cv->waitq);
return obj;
}
/*
* Document-method: ConditionVariable::new
*
* Creates a new condition variable instance.
*/
static VALUE
rb_condvar_initialize(VALUE self)
{
struct rb_condvar *cv = condvar_ptr(self);;
list_head_init(&cv->waitq);
return self;
}
struct sleep_call {
VALUE mutex;
VALUE timeout;
};
static ID id_sleep;
static VALUE
do_sleep(VALUE args)
{
struct sleep_call *p = (struct sleep_call *)args;
return rb_funcallv(p->mutex, id_sleep, 1, &p->timeout);
}
static VALUE
delete_from_waitq(struct sync_waiter *w)
{
list_del(&w->node);
return Qnil;
}
/*
* Document-method: ConditionVariable#wait
* call-seq: wait(mutex, timeout=nil)
*
* Releases the lock held in +mutex+ and waits; reacquires the lock on wakeup.
*
* If +timeout+ is given, this method returns after +timeout+ seconds passed,
* even if no other thread doesn't signal.
*/
static VALUE
rb_condvar_wait(int argc, VALUE *argv, VALUE self)
{
struct rb_condvar *cv = condvar_ptr(self);
VALUE mutex, timeout;
struct sleep_call args;
struct sync_waiter w;
rb_scan_args(argc, argv, "11", &mutex, &timeout);
args.mutex = mutex;
args.timeout = timeout;
w.th = GET_THREAD();
list_add_tail(&cv->waitq, &w.node);
rb_ensure(do_sleep, (VALUE)&args, delete_from_waitq, (VALUE)&w);
return self;
}
/*
* Document-method: ConditionVariable#signal
*
* Wakes up the first thread in line waiting for this lock.
*/
static VALUE
rb_condvar_signal(VALUE self)
{
struct rb_condvar *cv = condvar_ptr(self);
wakeup_one(&cv->waitq);
return self;
}
/*
* Document-method: ConditionVariable#broadcast
*
* Wakes up all threads waiting for this lock.
*/
static VALUE
rb_condvar_broadcast(VALUE self)
{
struct rb_condvar *cv = condvar_ptr(self);
wakeup_all(&cv->waitq);
return self;
}
/* :nodoc: */
static VALUE
undumpable(VALUE obj)
{
rb_raise(rb_eTypeError, "can't dump %"PRIsVALUE, rb_obj_class(obj));
UNREACHABLE;
}
static VALUE
define_thread_class(VALUE outer, const char *name, VALUE super)
{
VALUE klass = rb_define_class_under(outer, name, super);
rb_define_const(rb_cObject, name, klass);
return klass;
}
static void
Init_thread_sync(void)
{
#if 0
rb_cMutex = rb_define_class("Mutex", rb_cObject); /* teach rdoc Mutex */
rb_cConditionVariable = rb_define_class("ConditionVariable", rb_cObject); /* teach rdoc ConditionVariable */
rb_cQueue = rb_define_class("Queue", rb_cObject); /* teach rdoc Queue */
rb_cSizedQueue = rb_define_class("SizedQueue", rb_cObject); /* teach rdoc SizedQueue */
#endif
#define DEFINE_CLASS(name, super) \
rb_c##name = define_thread_class(rb_cThread, #name, rb_c##super)
/* Mutex */
DEFINE_CLASS(Mutex, Object);
rb_define_alloc_func(rb_cMutex, mutex_alloc);
rb_define_method(rb_cMutex, "initialize", mutex_initialize, 0);
rb_define_method(rb_cMutex, "locked?", rb_mutex_locked_p, 0);
rb_define_method(rb_cMutex, "try_lock", rb_mutex_trylock, 0);
rb_define_method(rb_cMutex, "lock", rb_mutex_lock, 0);
rb_define_method(rb_cMutex, "unlock", rb_mutex_unlock, 0);
rb_define_method(rb_cMutex, "sleep", mutex_sleep, -1);
rb_define_method(rb_cMutex, "synchronize", rb_mutex_synchronize_m, 0);
rb_define_method(rb_cMutex, "owned?", rb_mutex_owned_p, 0);
/* Queue */
DEFINE_CLASS(Queue, Object);
rb_define_alloc_func(rb_cQueue, queue_alloc);
rb_eClosedQueueError = rb_define_class("ClosedQueueError", rb_eStopIteration);
rb_define_method(rb_cQueue, "initialize", rb_queue_initialize, 0);
rb_undef_method(rb_cQueue, "initialize_copy");
rb_define_method(rb_cQueue, "marshal_dump", undumpable, 0);
rb_define_method(rb_cQueue, "close", rb_queue_close, 0);
rb_define_method(rb_cQueue, "closed?", rb_queue_closed_p, 0);
rb_define_method(rb_cQueue, "push", rb_queue_push, 1);
rb_define_method(rb_cQueue, "pop", rb_queue_pop, -1);
rb_define_method(rb_cQueue, "empty?", rb_queue_empty_p, 0);
rb_define_method(rb_cQueue, "clear", rb_queue_clear, 0);
rb_define_method(rb_cQueue, "length", rb_queue_length, 0);
rb_define_method(rb_cQueue, "num_waiting", rb_queue_num_waiting, 0);
rb_define_alias(rb_cQueue, "enq", "push");
rb_define_alias(rb_cQueue, "<<", "push");
rb_define_alias(rb_cQueue, "deq", "pop");
rb_define_alias(rb_cQueue, "shift", "pop");
rb_define_alias(rb_cQueue, "size", "length");
DEFINE_CLASS(SizedQueue, Queue);
rb_define_alloc_func(rb_cSizedQueue, szqueue_alloc);
rb_define_method(rb_cSizedQueue, "initialize", rb_szqueue_initialize, 1);
rb_define_method(rb_cSizedQueue, "close", rb_szqueue_close, 0);
rb_define_method(rb_cSizedQueue, "max", rb_szqueue_max_get, 0);
rb_define_method(rb_cSizedQueue, "max=", rb_szqueue_max_set, 1);
rb_define_method(rb_cSizedQueue, "push", rb_szqueue_push, -1);
rb_define_method(rb_cSizedQueue, "pop", rb_szqueue_pop, -1);
rb_define_method(rb_cSizedQueue, "empty?", rb_szqueue_empty_p, 0);
rb_define_method(rb_cSizedQueue, "clear", rb_szqueue_clear, 0);
rb_define_method(rb_cSizedQueue, "length", rb_szqueue_length, 0);
rb_define_method(rb_cSizedQueue, "num_waiting", rb_szqueue_num_waiting, 0);
rb_define_alias(rb_cSizedQueue, "enq", "push");
rb_define_alias(rb_cSizedQueue, "<<", "push");
rb_define_alias(rb_cSizedQueue, "deq", "pop");
rb_define_alias(rb_cSizedQueue, "shift", "pop");
rb_define_alias(rb_cSizedQueue, "size", "length");
/* CVar */
DEFINE_CLASS(ConditionVariable, Object);
rb_define_alloc_func(rb_cConditionVariable, condvar_alloc);
id_sleep = rb_intern("sleep");
rb_define_method(rb_cConditionVariable, "initialize", rb_condvar_initialize, 0);
rb_undef_method(rb_cConditionVariable, "initialize_copy");
rb_define_method(rb_cConditionVariable, "marshal_dump", undumpable, 0);
rb_define_method(rb_cConditionVariable, "wait", rb_condvar_wait, -1);
rb_define_method(rb_cConditionVariable, "signal", rb_condvar_signal, 0);
rb_define_method(rb_cConditionVariable, "broadcast", rb_condvar_broadcast, 0);
rb_provide("thread.rb");
}