triton.c

#include <signal.h>
#include <errno.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <limits.h>
#include <sys/resource.h>

#include "triton_p.h"
#include "memdebug.h"

#if __WORDSIZE == 32
#define WORKER_STACK_SIZE 64*1024
#else
#define WORKER_STACK_SIZE 128*1024
#endif

int thread_count = 2;
int max_events = 64;

static spinlock_t threads_lock = SPINLOCK_INITIALIZER;
static LIST_HEAD(threads);
static LIST_HEAD(sleep_threads);

static LIST_HEAD(ctx_queue);

static spinlock_t ctx_list_lock = SPINLOCK_INITIALIZER;
static LIST_HEAD(ctx_list);

static LIST_HEAD(init_list);

static int terminate;
static int need_terminate;

static int need_config_reload;
static void (*config_reload_notify)(int);

static mempool_t *ctx_pool;
static mempool_t *call_pool;

struct triton_stat_t __export triton_stat;

static struct timeval ru_utime;
static struct timeval ru_stime;
static struct timespec ru_timestamp;
static int ru_refs;
static void ru_update(struct triton_timer_t *);
static struct triton_timer_t ru_timer = {
	.period = 1000,
	.expire = ru_update,
};
struct triton_context_t default_ctx;

static struct triton_context_t __thread *this_ctx;

#define log_debug2(fmt, ...)

void triton_thread_wakeup(struct _triton_thread_t *thread)
{
	log_debug2("wake up thread %p\n", thread);
	pthread_kill(thread->thread, SIGUSR1);
}

static void __config_reload(void (*notify)(int))
{
	struct _triton_thread_t *t;
	int r;

	log_debug2("config_reload: enter\n");
	r = conf_reload(NULL);
	notify(r);

	spin_lock(&threads_lock);
	need_config_reload = 0;
	list_for_each_entry(t, &threads, entry)
		triton_thread_wakeup(t);
	spin_unlock(&threads_lock);
	log_debug2("config_reload: exit\n");
}

static void ctx_thread(struct _triton_context_t *ctx);
static void* triton_thread(struct _triton_thread_t *thread)
{
	sigset_t set;
	int sig;

	sigfillset(&set);
	sigdelset(&set, SIGKILL);
	sigdelset(&set, SIGSTOP);
	pthread_sigmask(SIG_BLOCK, &set, NULL);

	sigemptyset(&set);
	sigaddset(&set, SIGUSR1);
	sigaddset(&set, SIGQUIT);

	pthread_mutex_lock(&thread->sleep_lock);
	pthread_mutex_unlock(&thread->sleep_lock);

	while (1) {
		spin_lock(&threads_lock);
		if (!list_empty(&ctx_queue) && !need_config_reload && triton_stat.thread_active <= thread_count) {
			thread->ctx = list_entry(ctx_queue.next, typeof(*thread->ctx), entry2);
			log_debug2("thread: %p: dequeued ctx %p\n", thread, thread->ctx);
			list_del(&thread->ctx->entry2);
			spin_unlock(&threads_lock);
			spin_lock(&thread->ctx->lock);
			thread->ctx->thread = thread;
			thread->ctx->queued = 0;
			spin_unlock(&thread->ctx->lock);
			__sync_sub_and_fetch(&triton_stat.context_pending, 1);
		} else {
			if (triton_stat.thread_count > thread_count + triton_stat.context_sleeping) {
				__sync_sub_and_fetch(&triton_stat.thread_active, 1);
				__sync_sub_and_fetch(&triton_stat.thread_count, 1);
				list_del(&thread->entry);
				spin_unlock(&threads_lock);
				pthread_detach(pthread_self());
				log_debug2("thread: %p: exit\n", thread);
				_free(thread);
				return NULL;
			}
			log_debug2("thread: %p: sleeping\n", thread);
			if (!terminate)
				list_add(&thread->entry2, &sleep_threads);
			
			if (__sync_sub_and_fetch(&triton_stat.thread_active, 1) == 0 && need_config_reload) {
				spin_unlock(&threads_lock);
				__config_reload(config_reload_notify);
			} else
				spin_unlock(&threads_lock);
			
			if (terminate) {
				spin_lock(&threads_lock);
				list_del(&thread->entry);
				spin_unlock(&threads_lock);
				return NULL;
			}

			//printf("thread %p: enter sigwait\n", thread);
			sigwait(&set, &sig);
			//printf("thread %p: exit sigwait\n", thread);

			spin_lock(&threads_lock);
			__sync_add_and_fetch(&triton_stat.thread_active, 1);
			if (!thread->ctx) {
				list_del(&thread->entry2);
				spin_unlock(&threads_lock);
				continue;
			}
			spin_unlock(&threads_lock);
		}

cont:
		log_debug2("thread %p: ctx=%p %p\n", thread, thread->ctx, thread->ctx ? thread->ctx->thread : NULL);
		this_ctx = thread->ctx->ud;
		if (thread->ctx->ud->before_switch)
			thread->ctx->ud->before_switch(thread->ctx->ud, thread->ctx->bf_arg);

		log_debug2("thread %p: switch to %p\n", thread, thread->ctx);
		ctx_thread(thread->ctx);
		log_debug2("thread %p: switch from %p %p\n", thread, thread->ctx, thread->ctx->thread);

		spin_lock(&thread->ctx->lock);
		if (thread->ctx->pending) {
			spin_unlock(&thread->ctx->lock);
			goto cont;
		}
		thread->ctx->thread = NULL;
		spin_unlock(&thread->ctx->lock);

		if (thread->ctx->need_free) {
			log_debug2("- context %p removed\n", thread->ctx);
			thread->ctx->ud->tpd = NULL;
			mempool_free(thread->ctx);
		}

		thread->ctx = NULL;
	}
}

static void ctx_thread(struct _triton_context_t *ctx)
{
	struct _triton_md_handler_t *h;
	struct _triton_timer_t *t;
	struct _triton_ctx_call_t *call;
	uint64_t tt;

	log_debug2("ctx %p %p: enter\n", ctx, ctx->thread);
	if (ctx->need_close) {
		if (ctx->ud->close)
			ctx->ud->close(ctx->ud);
		ctx->need_close = 0;
	}

	while (1) {
		spin_lock(&ctx->lock);
		if (!list_empty(&ctx->pending_timers)) {
			t = list_entry(ctx->pending_timers.next, typeof(*t), entry2);
			list_del(&t->entry2);
			t->pending = 0;
			spin_unlock(&ctx->lock);
			__sync_sub_and_fetch(&triton_stat.timer_pending, 1);
			read(t->fd, &tt, sizeof(tt));
			if (t->ud)
				t->ud->expire(t->ud);
			continue;
		}
		if (!list_empty(&ctx->pending_handlers)) {
			h = list_entry(ctx->pending_handlers.next, typeof(*h), entry2);
			list_del(&h->entry2);
			h->pending = 0;
			spin_unlock(&ctx->lock);
			__sync_sub_and_fetch(&triton_stat.md_handler_pending, 1);
			if (h->trig_epoll_events & (EPOLLIN | EPOLLERR | EPOLLHUP))
				if (h->ud && h->ud->read)
					if (h->ud->read(h->ud))
						continue;
			if (h->trig_epoll_events & (EPOLLOUT | EPOLLERR | EPOLLHUP))
				if (h->ud && h->ud->write)
					if (h->ud->write(h->ud))
						continue;
			h->trig_epoll_events = 0;
			continue;
		}
		if (!list_empty(&ctx->pending_calls)) {
			call = list_entry(ctx->pending_calls.next, typeof(*call), entry);
			list_del(&call->entry);
			spin_unlock(&ctx->lock);
			call->func(call->arg);
			mempool_free(call);
			continue;
		}
		ctx->pending = 0;
		spin_unlock(&ctx->lock);
		break;	
	}

	log_debug2("ctx %p %p: exit\n", ctx, ctx->thread);
}

struct _triton_thread_t *create_thread()
{
	pthread_attr_t attr;
	struct _triton_thread_t *thread = _malloc(sizeof(*thread));
	if (!thread) {
		triton_log_error("out of memory");
		return NULL;
	}

	pthread_attr_init(&attr);
	pthread_attr_setstacksize(&attr, WORKER_STACK_SIZE);

	memset(thread, 0, sizeof(*thread));
	pthread_mutex_init(&thread->sleep_lock, NULL);
	pthread_cond_init(&thread->sleep_cond, NULL);
	pthread_mutex_lock(&thread->sleep_lock);
	if (pthread_create(&thread->thread, &attr, (void*(*)(void*))triton_thread, thread)) {
		triton_log_error("pthread_create: %s", strerror(errno));
		return NULL;
	}

	__sync_add_and_fetch(&triton_stat.thread_count, 1);
	__sync_add_and_fetch(&triton_stat.thread_active, 1);

	return thread;
}

int triton_queue_ctx(struct _triton_context_t *ctx)
{
	ctx->pending = 1;
	if (ctx->thread || ctx->queued || ctx->init)
		return 0;

	spin_lock(&threads_lock);
	if (list_empty(&sleep_threads) || need_config_reload || triton_stat.thread_active > thread_count) {
		if (ctx->priority)
			list_add(&ctx->entry2, &ctx_queue);
		else
			list_add_tail(&ctx->entry2, &ctx_queue);
		spin_unlock(&threads_lock);
		ctx->queued = 1;
		log_debug2("ctx %p: queued\n", ctx);
		__sync_add_and_fetch(&triton_stat.context_pending, 1);
		return 0;
	}

	ctx->thread = list_entry(sleep_threads.next, typeof(*ctx->thread), entry2);
	ctx->thread->ctx = ctx;
	log_debug2("ctx %p: assigned to thread %p\n", ctx, ctx->thread);
	list_del(&ctx->thread->entry2);
	spin_unlock(&threads_lock);

	return 1;
}

int __export triton_context_register(struct triton_context_t *ud, void *bf_arg)
{
	struct _triton_context_t *ctx = mempool_alloc(ctx_pool);

	log_debug2("ctx %p: register\n", ctx);
	if (!ctx)
		return -1;

	memset(ctx, 0, sizeof(*ctx));
	ctx->ud = ud;
	ctx->bf_arg = bf_arg;
	ctx->init = 1;
	spinlock_init(&ctx->lock);
	INIT_LIST_HEAD(&ctx->handlers);
	INIT_LIST_HEAD(&ctx->timers);
	INIT_LIST_HEAD(&ctx->pending_handlers);
	INIT_LIST_HEAD(&ctx->pending_timers);
	INIT_LIST_HEAD(&ctx->pending_calls);

	ud->tpd = ctx;

	spin_lock(&ctx_list_lock);
	list_add_tail(&ctx->entry, &ctx_list);
	spin_unlock(&ctx_list_lock);

	__sync_add_and_fetch(&triton_stat.context_sleeping, 1);
	__sync_add_and_fetch(&triton_stat.context_count, 1);

	return 0;
}

void __export triton_context_unregister(struct triton_context_t *ud)
{
	struct _triton_context_t *ctx = (struct _triton_context_t *)ud->tpd;
	struct _triton_ctx_call_t *call;
	struct _triton_thread_t *t;

	log_debug2("ctx %p: unregister\n", ctx);

	while (!list_empty(&ctx->pending_calls)) {
		call = list_entry(ctx->pending_calls.next, typeof(*call), entry);
		list_del(&call->entry);
		mempool_free(call);
	}

	if (!list_empty(&ctx->handlers)) {
		triton_log_error("BUG:ctx:triton_unregister_ctx: handlers is not empty");
		{
			struct _triton_md_handler_t *h;
			list_for_each_entry(h, &ctx->handlers, entry)
				if (h->ud)
					printf("%p\n", h->ud);
		}
		abort();
	}
	if (!list_empty(&ctx->pending_handlers)) {
		triton_log_error("BUG:ctx:triton_unregister_ctx: pending_handlers is not empty");
		abort();
	}
	if (!list_empty(&ctx->timers)) {
		triton_log_error("BUG:ctx:triton_unregister_ctx: timers is not empty");
		abort();
	}
	if (!list_empty(&ctx->pending_timers)) {
		triton_log_error("BUG:ctx:triton_unregister_ctx: pending_timers is not empty");
		abort();
	}

	ctx->need_free = 1;
	spin_lock(&ctx_list_lock);
	list_del(&ctx->entry);
	if (__sync_sub_and_fetch(&triton_stat.context_count, 1) == 1) {
		if (need_terminate)
			terminate = 1;
	}
	spin_unlock(&ctx_list_lock);
	
	if (terminate) {
		list_for_each_entry(t, &threads, entry)
			triton_thread_wakeup(t);
	}
}

void __export triton_context_set_priority(struct triton_context_t *ud, int prio)
{
	struct _triton_context_t *ctx = (struct _triton_context_t *)ud->tpd;
	
	ctx->priority = prio > 0;
}

void __export triton_context_schedule()
{
	struct _triton_context_t *ctx = (struct _triton_context_t *)this_ctx->tpd;
	struct _triton_thread_t *t = NULL;
	
	log_debug2("ctx %p: enter schedule\n", ctx);
	__sync_add_and_fetch(&triton_stat.context_sleeping, 1);
	__sync_sub_and_fetch(&triton_stat.thread_active, 1);
	pthread_mutex_lock(&ctx->thread->sleep_lock);
	while (1) {
		if (ctx->wakeup) {
			ctx->wakeup = 0;
			break;
		} else {
			if (!t && triton_stat.thread_count <= thread_count + triton_stat.context_sleeping) {
				t = create_thread();
				spin_lock(&threads_lock);
				list_add_tail(&t->entry, &threads);
				spin_unlock(&threads_lock);
				pthread_mutex_unlock(&t->sleep_lock);
			}
			pthread_cond_wait(&ctx->thread->sleep_cond, &ctx->thread->sleep_lock);
		}
	}
	pthread_mutex_unlock(&ctx->thread->sleep_lock);
	__sync_sub_and_fetch(&triton_stat.context_sleeping, 1);
	__sync_add_and_fetch(&triton_stat.thread_active, 1);
	log_debug2("ctx %p: exit schedule\n", ctx);
}

struct triton_context_t __export *triton_context_self(void)
{
	return this_ctx;
}

void triton_context_print(void)
{
	struct _triton_context_t *ctx;

	list_for_each_entry(ctx, &ctx_list, entry)
		printf("%p\n", ctx);
}

void __export triton_context_wakeup(struct triton_context_t *ud)
{
	struct _triton_context_t *ctx = (struct _triton_context_t *)ud->tpd;
	int r = 0;

	log_debug2("ctx %p: wakeup\n", ctx);

	if (ctx->init) {
		__sync_sub_and_fetch(&triton_stat.context_sleeping, 1);
		spin_lock(&ctx->lock);
		ctx->init = 0;
		if (ctx->pending)
			r = triton_queue_ctx(ctx);
		spin_unlock(&ctx->lock);
		if (r)
			triton_thread_wakeup(ctx->thread);
		return;
	}

	pthread_mutex_lock(&ctx->thread->sleep_lock);
	ctx->wakeup = 1;
	pthread_cond_signal(&ctx->thread->sleep_cond);
	pthread_mutex_unlock(&ctx->thread->sleep_lock);
}

int __export triton_context_call(struct triton_context_t *ud, void (*func)(void *), void *arg)
{
	struct _triton_context_t *ctx = (struct _triton_context_t *)ud->tpd;
	struct _triton_ctx_call_t *call = mempool_alloc(call_pool);
	int r;

	if (!call)
		return -1;
	
	call->func = func;
	call->arg = arg;

	spin_lock(&ctx->lock);
	list_add_tail(&call->entry, &ctx->pending_calls);
	r = triton_queue_ctx(ctx);
	spin_unlock(&ctx->lock);

	if (r)
		triton_thread_wakeup(ctx->thread);

	return 0;
}

void __export triton_cancel_call(struct triton_context_t *ud, void (*func)(void *))
{
	struct _triton_context_t *ctx = (struct _triton_context_t *)ud->tpd;
	struct list_head *pos, *n;
	struct _triton_ctx_call_t *call;

	list_for_each_safe(pos, n, &ctx->pending_calls) {
		call = list_entry(pos, typeof(*call), entry);
		if (call->func != func)
			continue;
		list_del(&call->entry);
		mempool_free(call);
	}
}

void __export triton_collect_cpu_usage(void)
{
	struct rusage rusage;

	if (__sync_fetch_and_add(&ru_refs, 1) == 0) {
		triton_timer_add(NULL, &ru_timer, 0);
		getrusage(RUSAGE_SELF, &rusage);
		clock_gettime(CLOCK_MONOTONIC, &ru_timestamp);
		ru_utime = rusage.ru_utime;
		ru_stime = rusage.ru_stime;
		triton_stat.cpu = 0;
	}
}

void __export triton_stop_collect_cpu_usage(void)
{
	if (__sync_sub_and_fetch(&ru_refs, 1) == 0)
		triton_timer_del(&ru_timer);
}

static void ru_update(struct triton_timer_t *t)
{
	struct timespec ts;
	struct rusage rusage;
	unsigned int dt;
	unsigned int val;

	getrusage(RUSAGE_SELF, &rusage);
	clock_gettime(CLOCK_MONOTONIC, &ts);

	dt = (ts.tv_sec - ru_timestamp.tv_sec) * 1000000 + (ts.tv_nsec - ru_timestamp.tv_nsec) / 1000000;
	val = (double)((rusage.ru_utime.tv_sec - ru_utime.tv_sec) * 1000000 + (rusage.ru_utime.tv_usec - ru_utime.tv_usec) + 
	      (rusage.ru_stime.tv_sec - ru_stime.tv_sec) * 1000000 + (rusage.ru_stime.tv_usec - ru_stime.tv_usec)) / dt * 100;

	triton_stat.cpu = val;

	ru_timestamp = ts;
	ru_utime = rusage.ru_utime;
	ru_stime = rusage.ru_stime;
}

void __export triton_register_init(int order, void (*func)(void))
{
	struct _triton_init_t *i1, *i = _malloc(sizeof(*i));
	struct list_head *p = init_list.prev;


	i->order = order;
	i->func = func;

	while (p != &init_list) {
		i1 = list_entry(p, typeof(*i1), entry);
		if (order > i1->order)
			break;
		p = p->prev;
	}
	list_add(&i->entry, p);
}

int __export triton_init(const char *conf_file)
{
	ctx_pool = mempool_create(sizeof(struct _triton_context_t));
	call_pool = mempool_create(sizeof(struct _triton_ctx_call_t));

	if (conf_load(conf_file))
		return -1;

	if (log_init())
		return -1;

	if (md_init())
		return -1;

	if (timer_init())
		return -1;

	if (event_init())
		return -1;

	triton_context_register(&default_ctx, NULL);

	return 0;
}

int __export triton_load_modules(const char *mod_sect)
{
	struct _triton_init_t *i;

	if (load_modules(mod_sect))
		return -1;

	while (!list_empty(&init_list)) {
		i = list_entry(init_list.next, typeof(*i), entry);
		i->func();
		list_del(&i->entry);
		_free(i);
	}
	
	return 0;
}

void __export triton_conf_reload(void (*notify)(int))
{
	spin_lock(&threads_lock);
	need_config_reload = 1;
	config_reload_notify = notify;
	if (triton_stat.thread_active == 0) {
		spin_unlock(&threads_lock);
		__config_reload(notify);
	} else
		spin_unlock(&threads_lock);
}

void __export triton_run()
{
	struct _triton_thread_t *t;
	int i;
	char *opt;
	struct timespec ts;

	opt = conf_get_opt("core", "thread-count");
	if (opt && atoi(opt) > 0)
		thread_count = atoi(opt);

	for(i = 0; i < thread_count; i++) {
		t = create_thread();
		if (!t)
			_exit(-1);

		list_add_tail(&t->entry, &threads);
		pthread_mutex_unlock(&t->sleep_lock);
	}

	clock_gettime(CLOCK_MONOTONIC, &ts);
	triton_stat.start_time = ts.tv_sec;

	md_run();
	timer_run();

	triton_context_wakeup(&default_ctx);
}

void __export triton_terminate()
{
	struct _triton_context_t *ctx;
	int r;

	need_terminate = 1;

	spin_lock(&ctx_list_lock);
	list_for_each_entry(ctx, &ctx_list, entry) {
		spin_lock(&ctx->lock);
		ctx->need_close = 1;
		r = triton_queue_ctx(ctx);
		if (r)
			triton_thread_wakeup(ctx->thread);
		spin_unlock(&ctx->lock);
	}
	spin_unlock(&ctx_list_lock);

	while (1) {
		spin_lock(&threads_lock);
		if (list_empty(&threads)) {
			spin_unlock(&threads_lock);
			break;
		}
		spin_unlock(&threads_lock);
		sleep(1);
	}
	
	md_terminate();
	timer_terminate();
}