X-Git-Url: https://gerrit.automotivelinux.org/gerrit/gitweb?a=blobdiff_plain;f=src%2Fjobs.c;h=417f7eac1fab792720c9e1a9acba6f1a19436ba1;hb=e9c6530ee21bb3345741044e8bf87fc6a73252e9;hp=ed812f819ac322b5a4a02b82a3313b572863974d;hpb=bfffa28ba0483b49770326c4f567950ea80d0fb6;p=src%2Fapp-framework-binder.git

diff --git a/src/jobs.c b/src/jobs.c
index ed812f81..417f7eac 100644
--- a/src/jobs.c
+++ b/src/jobs.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2016, 2017 "IoT.bzh"
+ * Copyright (C) 2016, 2017, 2018 "IoT.bzh"
  * Author José Bollo <jose.bollo@iot.bzh>
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,32 +17,43 @@
 
 #define _GNU_SOURCE
 
+#if defined(NO_JOBS_WATCHDOG)
+#   define HAS_WATCHDOG 0
+#else
+#   define HAS_WATCHDOG 1
+#endif
+
 #include <stdlib.h>
 #include <stdint.h>
 #include <unistd.h>
 #include <signal.h>
+#include <string.h>
 #include <time.h>
 #include <sys/syscall.h>
 #include <pthread.h>
 #include <errno.h>
 #include <assert.h>
+#include <sys/eventfd.h>
 
 #include <systemd/sd-event.h>
+#include "fdev.h"
+#if HAS_WATCHDOG
+#include <systemd/sd-daemon.h>
+#endif
 
 #include "jobs.h"
 #include "sig-monitor.h"
 #include "verbose.h"
 
-#if 0
-#define _alert_ "do you really want to remove monitoring?"
-#define sig_monitor_init_timeouts()  ((void)0)
-#define sig_monitor_clean_timeouts() ((void)0)
-#define sig_monitor(to,cb,arg)       (cb(0,arg))
+#if defined(REMOVE_SYSTEMD_EVENT)
+#include "fdev-epoll.h"
 #endif
 
 #define EVENT_TIMEOUT_TOP  	((uint64_t)-1)
 #define EVENT_TIMEOUT_CHILD	((uint64_t)10000)
 
+struct thread;
+
 /** Internal shortcut for callback */
 typedef void (*job_cb_t)(int, void*);
 
@@ -50,7 +61,7 @@ typedef void (*job_cb_t)(int, void*);
 struct job
 {
 	struct job *next;    /**< link to the next job enqueued */
-	void *group;         /**< group of the request */
+	const void *group;   /**< group of the request */
 	job_cb_t callback;   /**< processing callback */
 	void *arg;           /**< argument */
 	int timeout;         /**< timeout in second for processing the request */
@@ -59,30 +70,33 @@ struct job
 };
 
 /** Description of handled event loops */
-struct events
+struct evloop
 {
-	struct events *next;
-	struct sd_event *event;
-	uint64_t timeout;
-	unsigned used: 1;
-	unsigned runs: 1;
+	unsigned state;        /**< encoded state */
+	int efd;               /**< event notification */
+	struct sd_event *sdev; /**< the systemd event loop */
+	struct fdev *fdev;     /**< handling of events */
+	struct thread *holder; /**< holder of the evloop */
 };
 
+#define EVLOOP_STATE_WAIT           1U
+#define EVLOOP_STATE_RUN            2U
+
 /** Description of threads */
 struct thread
 {
 	struct thread *next;   /**< next thread of the list */
 	struct thread *upper;  /**< upper same thread */
+	struct thread *nholder;/**< next holder for evloop */
+	pthread_cond_t *cwhold;/**< condition wait for holding */
 	struct job *job;       /**< currently processed job */
-	struct events *events; /**< currently processed job */
 	pthread_t tid;         /**< the thread id */
-	unsigned stop: 1;      /**< stop requested */
-	unsigned lowered: 1;   /**< has a lower same thread */
-	unsigned waits: 1;     /**< is waiting? */
+	volatile unsigned stop: 1;      /**< stop requested */
+	volatile unsigned waits: 1;     /**< is waiting? */
 };
 
 /**
- * Description of synchonous callback
+ * Description of synchronous callback
  */
 struct sync
 {
@@ -100,22 +114,28 @@ struct sync
 static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
 static pthread_cond_t  cond = PTHREAD_COND_INITIALIZER;
 
-/* count allowed, started and waiting threads */
+/* count allowed, started and running threads */
 static int allowed = 0; /** allowed count of threads */
 static int started = 0; /** started count of threads */
-static int waiting = 0; /** waiting count of threads */
+static int running = 0; /** running count of threads */
 static int remains = 0; /** allowed count of waiting jobs */
-static int nevents = 0; /** count of events */
 
 /* list of threads */
 static struct thread *threads;
-static _Thread_local struct thread *current;
+static _Thread_local struct thread *current_thread;
 
 /* queue of pending jobs */
 static struct job *first_job;
-static struct events *first_events;
 static struct job *free_jobs;
 
+/* event loop */
+static struct evloop evloop;
+
+#if defined(REMOVE_SYSTEMD_EVENT)
+static struct fdev_epoll *fdevepoll;
+static int waitevt;
+#endif
+
 /**
  * Create a new job with the given parameters
  * @param group    the group of the job
@@ -125,7 +145,7 @@ static struct job *free_jobs;
  * @return the created job unblock or NULL when no more memory
  */
 static struct job *job_create(
-		void *group,
+		const void *group,
 		int timeout,
 		job_cb_t callback,
 		void *arg)
@@ -137,12 +157,12 @@ static struct job *job_create(
 	if (job)
 		free_jobs = job->next;
 	else {
-		/* allocation  without blocking */
+		/* allocation without blocking */
 		pthread_mutex_unlock(&mutex);
 		job = malloc(sizeof *job);
 		pthread_mutex_lock(&mutex);
 		if (!job) {
-			errno = -ENOMEM;
+			errno = ENOMEM;
 			goto end;
 		}
 	}
@@ -164,7 +184,7 @@ end:
  */
 static void job_add(struct job *job)
 {
-	void *group;
+	const void *group;
 	struct job *ijob, **pjob;
 
 	/* prepare to add */
@@ -197,18 +217,6 @@ static inline struct job *job_get()
 	return job;
 }
 
-/**
- * Get the next events to process or NULL if none.
- * @return the first events that isn't running or NULL
- */
-static inline struct events *events_get()
-{
-	struct events *events = first_events;
-	while (events && events->used)
-		events = events->next;
-	return events;
-}
-
 /**
  * Releases the processed 'job': removes it
  * from the list of jobs and unblock the first
@@ -218,7 +226,7 @@ static inline struct events *events_get()
 static inline void job_release(struct job *job)
 {
 	struct job *ijob, **pjob;
-	void *group;
+	const void *group;
 
 	/* first unqueue the job */
 	pjob = &first_job;
@@ -260,6 +268,23 @@ static void job_cancel(int signum, void *arg)
 	job->callback(SIGABRT, job->arg);
 }
 
+#if defined(REMOVE_SYSTEMD_EVENT)
+/**
+ * Gets a fdev_epoll item.
+ * @return a fdev_epoll or NULL in case of error
+ */
+static struct fdev_epoll *get_fdevepoll()
+{
+	struct fdev_epoll *result;
+
+	result = fdevepoll;
+	if (!result)
+		result = fdevepoll = fdev_epoll_create();
+
+	return result;
+}
+#endif
+
 /**
  * Monitored normal callback for events.
  * This function is called by the monitor
@@ -270,13 +295,141 @@ static void job_cancel(int signum, void *arg)
  *               flow
  * @param arg     the events to run
  */
-static void events_call(int signum, void *arg)
+static void evloop_run(int signum, void *arg)
+{
+	int rc;
+	struct sd_event *se;
+
+	if (!signum) {
+		se = evloop.sdev;
+		rc = sd_event_prepare(se);
+		if (rc < 0) {
+			errno = -rc;
+			CRITICAL("sd_event_prepare returned an error (state: %d): %m", sd_event_get_state(se));
+			abort();
+		} else {
+			if (rc == 0) {
+				rc = sd_event_wait(se, (uint64_t)(int64_t)-1);
+				if (rc < 0) {
+					errno = -rc;
+					ERROR("sd_event_wait returned an error (state: %d): %m", sd_event_get_state(se));
+				}
+			}
+			evloop.state = EVLOOP_STATE_RUN;
+			if (rc > 0) {
+				rc = sd_event_dispatch(se);
+				if (rc < 0) {
+					errno = -rc;
+					ERROR("sd_event_dispatch returned an error (state: %d): %m", sd_event_get_state(se));
+				}
+			}
+		}
+	}
+}
+
+/**
+ * Internal callback for evloop management.
+ * The effect of this function is hidden: it exits
+ * the waiting poll if any.
+ */
+static void evloop_on_efd_event()
 {
-	struct events *events = arg;
-	if (!signum)
-		sd_event_run(events->event, events->timeout);
+	uint64_t x;
+	read(evloop.efd, &x, sizeof x);
 }
 
+/**
+ * wakeup the event loop if needed by sending
+ * an event.
+ */
+static void evloop_wakeup()
+{
+	uint64_t x;
+
+	if (evloop.state & EVLOOP_STATE_WAIT) {
+		x = 1;
+		write(evloop.efd, &x, sizeof x);
+	}
+}
+
+/**
+ * Release the currently held event loop
+ */
+static void evloop_release()
+{
+	struct thread *nh, *ct = current_thread;
+
+	if (evloop.holder == ct) {
+		nh = ct->nholder;
+		evloop.holder = nh;
+		if (nh)
+			pthread_cond_signal(nh->cwhold);
+	}
+}
+
+/**
+ * get the eventloop for the current thread
+ */
+static int evloop_get()
+{
+	struct thread *ct = current_thread;
+
+	if (evloop.holder)
+		return evloop.holder == ct;
+
+	ct->nholder = NULL;
+	evloop.holder = ct;
+	return 1;
+}
+
+/**
+ * acquire the eventloop for the current thread
+ */
+static void evloop_acquire()
+{
+	struct thread **pwait, *ct;
+	pthread_cond_t cond;
+
+	/* try to get the evloop */
+	if (!evloop_get()) {
+		/* failed, init waiting state */
+		ct = current_thread;
+		ct->nholder = NULL;
+		ct->cwhold = &cond;
+		pthread_cond_init(&cond, NULL);
+
+		/* queue current thread in holder list */
+		pwait = &evloop.holder;
+		while (*pwait)
+			pwait = &(*pwait)->nholder;
+		*pwait = ct;
+
+		/* wake up the evloop */
+		evloop_wakeup();
+
+		/* wait to acquire the evloop */
+		pthread_cond_wait(&cond, &mutex);
+		pthread_cond_destroy(&cond);
+	}
+}
+
+#if defined(REMOVE_SYSTEMD_EVENT)
+/**
+ * Monitored normal loop for waiting events.
+ * @param signum 0 on normal flow or the number
+ *               of the signal that interrupted the normal
+ *               flow
+ * @param arg     the events to run
+ */
+static void monitored_wait_and_dispatch(int signum, void *arg)
+{
+	struct fdev_epoll *fdev_epoll = arg;
+	if (!signum) {
+		fdev_epoll_wait_and_dispatch(fdev_epoll, -1);
+	}
+}
+#endif
+
 /**
  * Main processing loop of threads processing jobs.
  * The loop must be called with the mutex locked
@@ -286,35 +439,29 @@ static void events_call(int signum, void *arg)
  */
 static void thread_run(volatile struct thread *me)
 {
-	struct thread **prv, *thr;
+	struct thread **prv;
 	struct job *job;
-	struct events *events;
-	uint64_t evto;
 
 	/* initialize description of itself and link it in the list */
 	me->tid = pthread_self();
 	me->stop = 0;
-	me->lowered = 0;
 	me->waits = 0;
-	me->upper = current;
-	if (current) {
-		current->lowered = 1;
-		evto = EVENT_TIMEOUT_CHILD;
-		me->events = current->events;
-	} else {
+	me->upper = current_thread;
+	if (!current_thread) {
 		started++;
 		sig_monitor_init_timeouts();
-		evto = EVENT_TIMEOUT_TOP;
-		me->events = NULL;
 	}
 	me->next = threads;
 	threads = (struct thread*)me;
-	current = (struct thread*)me;
+	current_thread = (struct thread*)me;
 
 	/* loop until stopped */
 	while (!me->stop) {
+		/* release the current event loop */
+		evloop_release();
+
 		/* get a job */
-		job = job_get(first_job);
+		job = job_get();
 		if (job) {
 			/* prepare running the job */
 			remains++; /* increases count of job that can wait */
@@ -328,55 +475,63 @@ static void thread_run(volatile struct thread *me)
 
 			/* release the run job */
 			job_release(job);
+#if !defined(REMOVE_SYSTEMD_EVENT)
+
+
 
-			/* release event if any */
-			events = me->events;
-			if (events) {
-				events->used = 0;
-				me->events = NULL;
+		/* no job, check event loop wait */
+		} else if (evloop_get()) {
+			if (evloop.state != 0) {
+				/* busy ? */
+				CRITICAL("Can't enter dispatch while in dispatch!");
+				abort();
 			}
+			/* run the events */
+			evloop.state = EVLOOP_STATE_RUN|EVLOOP_STATE_WAIT;
+			pthread_mutex_unlock(&mutex);
+			sig_monitor(0, evloop_run, NULL);
+			pthread_mutex_lock(&mutex);
+			evloop.state = 0;
 		} else {
-			/* no job, check events */
-			events = me->events;
-			if (!events || events->runs)
-				events = events_get();
-			if (events) {
-				/* run the events */
-				events->used = 1;
-				events->runs = 1;
-				events->timeout = evto;
-				me->events = events;
-				pthread_mutex_unlock(&mutex);
-				sig_monitor(0, events_call, events);
-				pthread_mutex_lock(&mutex);
-				events->used = 0;
-				events->runs = 0;
-				me->events = NULL;
-				thr = me->upper;
-				while (thr && thr->events == events) {
-					thr->events = NULL;
-					thr = thr->upper;
-				}
-			} else {
-				/* no job and not events */
-				waiting++;
-				me->waits = 1;
-				pthread_cond_wait(&cond, &mutex);
-				me->waits = 0;
-				waiting--;
-			}
+			/* no job and no event loop */
+			running--;
+			if (!running)
+				ERROR("Entering job deep sleep! Check your bindings.");
+			me->waits = 1;
+			pthread_cond_wait(&cond, &mutex);
+			me->waits = 0;
+			running++;
+#else
+		} else if (waitevt) {
+			/* no job and not events */
+			running--;
+			if (!running)
+				ERROR("Entering job deep sleep! Check your bindings.");
+			me->waits = 1;
+			pthread_cond_wait(&cond, &mutex);
+			me->waits = 0;
+			running++;
+		} else {
+			/* wait for events */
+			waitevt = 1;
+			pthread_mutex_unlock(&mutex);
+			sig_monitor(0, monitored_wait_and_dispatch, get_fdevepoll());
+			pthread_mutex_lock(&mutex);
+			waitevt = 0;
+#endif
 		}
 	}
 
+	/* release the event loop */
+	evloop_release();
+
 	/* unlink the current thread and cleanup */
 	prv = &threads;
 	while (*prv != me)
 		prv = &(*prv)->next;
 	*prv = me->next;
-	current = me->upper;
-	if (current) {
-		current->lowered = 0;
-	} else {
+	current_thread = me->upper;
+	if (!current_thread) {
 		sig_monitor_clean_timeouts();
 		started--;
 	}
@@ -392,7 +547,9 @@ static void *thread_main(void *data)
 	struct thread me;
 
 	pthread_mutex_lock(&mutex);
+	running++;
 	thread_run(&me);
+	running--;
 	pthread_mutex_unlock(&mutex);
 	return NULL;
 }
@@ -433,7 +590,7 @@ static int start_one_thread()
  * @return 0 in case of success or -1 in case of error
  */
 int jobs_queue(
-		void *group,
+		const void *group,
 		int timeout,
 		void (*callback)(int, void*),
 		void *arg)
@@ -460,7 +617,7 @@ int jobs_queue(
 	}
 
 	/* start a thread if needed */
-	if (waiting == 0 && started < allowed) {
+	if (running == started && started < allowed) {
 		/* all threads are busy and a new can be started */
 		rc = start_one_thread();
 		if (rc < 0 && started == 0) {
@@ -515,7 +672,7 @@ static void call_cb(int signum, void *closure)
  * @see jobs_call, jobs_enter, jobs_leave
  */
 static int do_sync(
-		void *group,
+		const void *group,
 		int timeout,
 		void (*sync_cb)(int signum, void *closure),
 		struct sync *sync
@@ -555,11 +712,11 @@ static int do_sync(
  *                 of interrupted flow, the context 'closure' as given and
  *                 a 'jobloop' reference that must be used when the job is
  *                 terminated to unlock the current execution flow.
- * @param arg the argument to the callback
+ * @param closure the argument to the callback
  * @return 0 on success or -1 in case of error
  */
 int jobs_enter(
-		void *group,
+		const void *group,
 		int timeout,
 		void (*callback)(int signum, void *closure, struct jobloop *jobloop),
 		void *closure
@@ -591,6 +748,8 @@ int jobs_leave(struct jobloop *jobloop)
 		t->stop = 1;
 		if (t->waits)
 			pthread_cond_broadcast(&cond);
+		else
+			evloop_wakeup();
 	}
 	pthread_mutex_unlock(&mutex);
 	return -!t;
@@ -611,7 +770,7 @@ int jobs_leave(struct jobloop *jobloop)
  * @return 0 in case of success or -1 in case of error
  */
 int jobs_call(
-		void *group,
+		const void *group,
 		int timeout,
 		void (*callback)(int, void*),
 		void *arg)
@@ -624,74 +783,147 @@ int jobs_call(
 	return do_sync(group, timeout, call_cb, &sync);
 }
 
+/**
+ * Internal callback for evloop management.
+ * The effect of this function is hidden: it exits
+ * the waiting poll if any. Then it wakes up a thread
+ * awaiting the evloop using signal.
+ */
+static int on_evloop_efd(sd_event_source *s, int fd, uint32_t revents, void *userdata)
+{
+	evloop_on_efd_event();
+	return 1;
+}
+
+/* temporary hack */
+#if !defined(REMOVE_SYSTEMD_EVENT)
+__attribute__((unused))
+#endif
+static void evloop_callback(void *arg, uint32_t event, struct fdev *fdev)
+{
+	sig_monitor(0, evloop_run, arg);
+}
+
 /**
  * Gets a sd_event item for the current thread.
  * @return a sd_event or NULL in case of error
  */
-struct sd_event *jobs_get_sd_event()
+static struct sd_event *get_sd_event_locked()
 {
-	struct events *events;
-	struct thread *me;
 	int rc;
 
-	pthread_mutex_lock(&mutex);
-
-	/* search events on stack */
-	me = current;
-	while (me && !me->events)
-		me = me->upper;
-	if (me)
-		/* return the stacked events */
-		events = me->events;
-	else {
-		/* search an available events */
-		events = events_get();
-		if (!events) {
-			/* not found, check if creation possible */
-			if (nevents >= allowed) {
-				ERROR("not possible to add a new event");
-				events = NULL;
-			} else {
-				events = malloc(sizeof *events);
-				if (events && (rc = sd_event_new(&events->event)) >= 0) {
-					if (nevents < started || start_one_thread() >= 0) {
-						events->used = 0;
-						events->runs = 0;
-						events->next = first_events;
-						first_events = events;
-					} else {
-						ERROR("can't start thread for events");
-						sd_event_unref(events->event);
-						free(events);
-						events = NULL;
-					}
-				} else {
-					if (!events) {
-						ERROR("out of memory");
-						errno = ENOMEM;
-					} else {
-						free(events);
-						ERROR("creation of sd_event failed: %m");
-						events = NULL;
-						errno = -rc;
-					} 
-				}
-			}
+	/* creates the evloop on need */
+	if (!evloop.sdev) {
+		/* start the creation */
+		evloop.state = 0;
+		/* creates the eventfd for waking up polls */
+		evloop.efd = eventfd(0, EFD_CLOEXEC|EFD_SEMAPHORE);
+		if (evloop.efd < 0) {
+			ERROR("can't make eventfd for events");
+			goto error1;
 		}
-		if (events) {
-			me = current;
-			if (me) {
-				events->used = 1;
-				me->events = events;
-			} else {
-				WARNING("event returned for unknown thread!");
-			}
+		/* create the systemd event loop */
+		rc = sd_event_new(&evloop.sdev);
+		if (rc < 0) {
+			ERROR("can't make new event loop");
+			goto error2;
+		}
+		/* put the eventfd in the event loop */
+		rc = sd_event_add_io(evloop.sdev, NULL, evloop.efd, EPOLLIN, on_evloop_efd, NULL);
+		if (rc < 0) {
+			ERROR("can't register eventfd");
+#if !defined(REMOVE_SYSTEMD_EVENT)
+			sd_event_unref(evloop.sdev);
+			evloop.sdev = NULL;
+error2:
+			close(evloop.efd);
+error1:
+			return NULL;
+		}
+#else
+			goto error3;
 		}
+		/* handle the event loop */
+		evloop.fdev = fdev_epoll_add(get_fdevepoll(), sd_event_get_fd(evloop.sdev));
+		if (!evloop.fdev) {
+			ERROR("can't create fdev");
+error3:
+			sd_event_unref(evloop.sdev);
+error2:
+			close(evloop.efd);
+error1:
+			memset(&evloop, 0, sizeof evloop);
+			return NULL;
+		}
+		fdev_set_autoclose(evloop.fdev, 0);
+		fdev_set_events(evloop.fdev, EPOLLIN);
+		fdev_set_callback(evloop.fdev, evloop_callback, NULL);
+#endif
+	}
+
+	/* acquire the event loop */
+	evloop_acquire();
+
+	return evloop.sdev;
+}
+
+/**
+ * Gets a sd_event item for the current thread.
+ * @return a sd_event or NULL in case of error
+ */
+struct sd_event *jobs_get_sd_event()
+{
+	struct sd_event *result;
+	struct thread lt;
+
+	/* ensure an existing thread environment */
+	if (!current_thread) {
+		memset(&lt, 0, sizeof lt);
+		current_thread = &lt;
 	}
+
+	/* process */
+	pthread_mutex_lock(&mutex);
+	result = get_sd_event_locked();
 	pthread_mutex_unlock(&mutex);
-	return events ? events->event : NULL;
+
+	/* release the faked thread environment if needed */
+	if (current_thread == &lt) {
+		/*
+		 * Releasing it is needed because there is no way to guess
+		 * when it has to be released really. But here is where it is
+		 * hazardous: if the caller modifies the eventloop when it
+		 * is waiting, there is no way to make the change effective.
+		 * A workaround to achieve that goal is for the caller to
+		 * require the event loop a second time after having modified it.
+		 */
+		NOTICE("Requiring sd_event loop out of binder callbacks is hazardous!");
+		if (verbose_wants(Log_Level_Info))
+			sig_monitor_dumpstack();
+		evloop_release();
+		current_thread = NULL;
+	}
+
+	return result;
 }
 
+#if defined(REMOVE_SYSTEMD_EVENT)
+/**
+ * Gets the fdev_epoll item.
+ * @return a fdev_epoll or NULL in case of error
+ */
+struct fdev_epoll *jobs_get_fdev_epoll()
+{
+	struct fdev_epoll *result;
+
+	pthread_mutex_lock(&mutex);
+	result = get_fdevepoll();
+	pthread_mutex_unlock(&mutex);
+
+	return result;
+}
+#endif
+
 /**
  * Enter the jobs processing loop.
  * @param allowed_count Maximum count of thread for jobs including this one
@@ -700,7 +932,7 @@ struct sd_event *jobs_get_sd_event()
  * @param start         The start routine to activate (can't be NULL)
  * @return 0 in case of success or -1 in case of error.
  */
-int jobs_start(int allowed_count, int start_count, int waiter_count, void (*start)(int signum))
+int jobs_start(int allowed_count, int start_count, int waiter_count, void (*start)(int signum, void* arg), void *arg)
 {
 	int rc, launched;
 	struct thread me;
@@ -715,24 +947,24 @@ int jobs_start(int allowed_count, int start_count, int waiter_count, void (*star
 	pthread_mutex_lock(&mutex);
 
 	/* check whether already running */
-	if (current || allowed) {
+	if (current_thread || allowed) {
 		ERROR("thread already started");
 		errno = EINVAL;
 		goto error;
 	}
 
-	/* start */
-	if (sig_monitor_init() < 0) {
-		ERROR("failed to initialise signal handlers");
-		goto error;
-	}
-
 	/* records the allowed count */
 	allowed = allowed_count;
 	started = 0;
-	waiting = 0;
+	running = 0;
 	remains = waiter_count;
 
+#if HAS_WATCHDOG
+	/* set the watchdog */
+	if (sd_watchdog_enabled(0, NULL))
+		sd_event_set_watchdog(get_sd_event_locked(), 1);
+#endif
+
 	/* start at least one thread */
 	launched = 0;
 	while ((launched + 1) < start_count) {
@@ -744,7 +976,7 @@ int jobs_start(int allowed_count, int start_count, int waiter_count, void (*star
 	}
 
 	/* queue the start job */
-	job = job_create(NULL, 0, (job_cb_t)start, NULL);
+	job = job_create(NULL, 0, start, arg);
 	if (!job) {
 		ERROR("out of memory");
 		errno = ENOMEM;
@@ -754,7 +986,9 @@ int jobs_start(int allowed_count, int start_count, int waiter_count, void (*star
 	remains--;
 
 	/* run until end */
+	running++;
 	thread_run(&me);
+	running--;
 	rc = 0;
 error:
 	pthread_mutex_unlock(&mutex);
@@ -822,7 +1056,7 @@ void jobs_terminate()
 		head = job->next;
 
 		/* search if job is stacked for current */
-		t = current;
+		t = current_thread;
 		while (t && t->job != job)
 			t = t->upper;
 		if (t) {