X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=592ea8bfaa05cc17b4a8a3e23c1cd54b5aebea91;hp=821e978a98b0152b2e4c099c5af6eb6d8e0b1abc;hb=90b7a96d9573a14f20470af2aa0dd4e60611a477;hpb=2e1df07d1ea30e0304cc65370f3ed161a6f22cd4

diff --git a/backend.c b/backend.c
index 821e978a..592ea8bf 100644
--- a/backend.c
+++ b/backend.c
@@ -30,10 +30,13 @@
 #include <locale.h>
 #include <assert.h>
 #include <time.h>
+#include <inttypes.h>
 #include <sys/stat.h>
 #include <sys/wait.h>
 #include <sys/ipc.h>
+#ifndef FIO_NO_HAVE_SHM_H
 #include <sys/shm.h>
+#endif
 #include <sys/mman.h>
 
 #include "fio.h"
@@ -47,19 +50,32 @@
 #include "lib/rand.h"
 #include "memalign.h"
 #include "server.h"
+#include "lib/getrusage.h"
+#include "idletime.h"
 
 static pthread_t disk_util_thread;
+static struct fio_mutex *disk_thread_mutex;
 static struct fio_mutex *startup_mutex;
 static struct fio_mutex *writeout_mutex;
 static struct flist_head *cgroup_list;
 static char *cgroup_mnt;
 static int exit_value;
 static volatile int fio_abort;
+static unsigned int nr_process = 0;
+static unsigned int nr_thread = 0;
+
+struct io_log *agg_io_log[DDIR_RWDIR_CNT];
 
-struct io_log *agg_io_log[2];
+int groupid = 0;
+unsigned int thread_number = 0;
+unsigned int stat_number = 0;
+int shm_id = 0;
+int temp_stall_ts;
+unsigned long done_secs = 0;
+volatile int disk_util_exit = 0;
 
 #define PAGE_ALIGN(buf)	\
-	(char *) (((unsigned long) (buf) + page_mask) & ~page_mask)
+	(char *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
 
 #define JOB_START_TIMEOUT	(5 * 1000)
 
@@ -78,6 +94,11 @@ static void sig_int(int sig)
 	}
 }
 
+static void sig_show_status(int sig)
+{
+	show_running_run_stats();
+}
+
 static void set_sig_handlers(void)
 {
 	struct sigaction act;
@@ -92,6 +113,19 @@ static void set_sig_handlers(void)
 	act.sa_flags = SA_RESTART;
 	sigaction(SIGTERM, &act, NULL);
 
+/* Windows uses SIGBREAK as a quit signal from other applications */
+#ifdef WIN32
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_int;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGBREAK, &act, NULL);
+#endif
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_show_status;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGUSR1, &act, NULL);
+
 	if (is_backend) {
 		memset(&act, 0, sizeof(act));
 		act.sa_handler = sig_int;
@@ -184,14 +218,16 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
 }
 
 static int check_min_rate(struct thread_data *td, struct timeval *now,
-			  unsigned long *bytes_done)
+			  uint64_t *bytes_done)
 {
 	int ret = 0;
 
-	if (bytes_done[0])
-		ret |= __check_min_rate(td, now, 0);
-	if (bytes_done[1])
-		ret |= __check_min_rate(td, now, 1);
+	if (bytes_done[DDIR_READ])
+		ret |= __check_min_rate(td, now, DDIR_READ);
+	if (bytes_done[DDIR_WRITE])
+		ret |= __check_min_rate(td, now, DDIR_WRITE);
+	if (bytes_done[DDIR_TRIM])
+		ret |= __check_min_rate(td, now, DDIR_TRIM);
 
 	return ret;
 }
@@ -285,6 +321,7 @@ requeue:
 
 	return 0;
 }
+
 static inline void __update_tv_cache(struct thread_data *td)
 {
 	fio_gettime(&td->tv_cache, NULL);
@@ -314,17 +351,17 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 	int ret = *retptr;
 
 	if (ret < 0 || td->error) {
-		int err;
+		int err = td->error;
+		enum error_type_bit eb;
 
 		if (ret < 0)
 			err = -ret;
-		else
-			err = td->error;
 
-		if (!(td->o.continue_on_error & td_error_type(ddir, err)))
+		eb = td_error_type(ddir, err);
+		if (!(td->o.continue_on_error & (1 << eb)))
 			return 1;
 
-		if (td_non_fatal_error(err)) {
+		if (td_non_fatal_error(td, eb, err)) {
 		        /*
 		         * Continue with the I/Os in case of
 			 * a non fatal error.
@@ -354,14 +391,13 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 	return 0;
 }
 
-
-
 /*
  * The main verify engine. Runs over the writes we previously submitted,
  * reads the blocks back in, and checks the crc/md5 of the data.
  */
-static void do_verify(struct thread_data *td)
+static void do_verify(struct thread_data *td, uint64_t verify_bytes)
 {
+	uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
 	struct fio_file *f;
 	struct io_u *io_u;
 	int ret, min_events;
@@ -389,6 +425,7 @@ static void do_verify(struct thread_data *td)
 
 	io_u = NULL;
 	while (!td->terminate) {
+		enum fio_ddir ddir;
 		int ret2, full;
 
 		update_tv_cache(td);
@@ -401,18 +438,56 @@ static void do_verify(struct thread_data *td)
 			}
 		}
 
-		io_u = __get_io_u(td);
-		if (!io_u)
-			break;
+		if (flow_threshold_exceeded(td))
+			continue;
 
-		if (get_next_verify(td, io_u)) {
-			put_io_u(td, io_u);
-			break;
-		}
+		if (!td->o.experimental_verify) {
+			io_u = __get_io_u(td);
+			if (!io_u)
+				break;
 
-		if (td_io_prep(td, io_u)) {
-			put_io_u(td, io_u);
-			break;
+			if (get_next_verify(td, io_u)) {
+				put_io_u(td, io_u);
+				break;
+			}
+
+			if (td_io_prep(td, io_u)) {
+				put_io_u(td, io_u);
+				break;
+			}
+		} else {
+			if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes)
+				break;
+
+			while ((io_u = get_io_u(td)) != NULL) {
+				/*
+				 * We are only interested in the places where
+				 * we wrote or trimmed IOs. Turn those into
+				 * reads for verification purposes.
+				 */
+				if (io_u->ddir == DDIR_READ) {
+					/*
+					 * Pretend we issued it for rwmix
+					 * accounting
+					 */
+					td->io_issues[DDIR_READ]++;
+					put_io_u(td, io_u);
+					continue;
+				} else if (io_u->ddir == DDIR_TRIM) {
+					io_u->ddir = DDIR_READ;
+					io_u->flags |= IO_U_F_TRIMMED;
+					break;
+				} else if (io_u->ddir == DDIR_WRITE) {
+					io_u->ddir = DDIR_READ;
+					break;
+				} else {
+					put_io_u(td, io_u);
+					continue;
+				}
+			}
+
+			if (!io_u)
+				break;
 		}
 
 		if (td->o.verify_async)
@@ -420,6 +495,8 @@ static void do_verify(struct thread_data *td)
 		else
 			io_u->end_io = verify_io_u;
 
+		ddir = io_u->ddir;
+
 		ret = td_io_queue(td, io_u);
 		switch (ret) {
 		case FIO_Q_COMPLETED:
@@ -452,7 +529,7 @@ static void do_verify(struct thread_data *td)
 				requeue_io_u(td, &io_u);
 			} else {
 sync_done:
-				ret = io_u_sync_complete(td, io_u, NULL);
+				ret = io_u_sync_complete(td, io_u, bytes_done);
 				if (ret < 0)
 					break;
 			}
@@ -471,7 +548,7 @@ sync_done:
 			break;
 		}
 
-		if (break_on_this_error(td, io_u->ddir, &ret))
+		if (break_on_this_error(td, ddir, &ret))
 			break;
 
 		/*
@@ -483,7 +560,10 @@ sync_done:
 		if (full || !td->o.iodepth_batch_complete) {
 			min_events = min(td->o.iodepth_batch_complete,
 					 td->cur_depth);
-			if (full && !min_events && td->o.iodepth_batch_complete != 0)
+			/*
+			 * if the queue is full, we MUST reap at least 1 event
+			 */
+			if (full && !min_events)
 				min_events = 1;
 
 			do {
@@ -492,7 +572,7 @@ sync_done:
 				 * and do the verification on them through
 				 * the callback handler
 				 */
-				if (io_u_queued_complete(td, min_events, NULL) < 0) {
+				if (io_u_queued_complete(td, min_events, bytes_done) < 0) {
 					ret = -1;
 					break;
 				}
@@ -515,12 +595,31 @@ sync_done:
 	dprint(FD_VERIFY, "exiting loop\n");
 }
 
+static int io_bytes_exceeded(struct thread_data *td)
+{
+	unsigned long long bytes;
+
+	if (td_rw(td))
+		bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE];
+	else if (td_write(td))
+		bytes = td->this_io_bytes[DDIR_WRITE];
+	else if (td_read(td))
+		bytes = td->this_io_bytes[DDIR_READ];
+	else
+		bytes = td->this_io_bytes[DDIR_TRIM];
+
+	return bytes >= td->o.size;
+}
+
 /*
  * Main IO worker function. It retrieves io_u's to process and queues
  * and reaps them, checking for rate and errors along the way.
+ *
+ * Returns number of bytes written and trimmed.
  */
-static void do_io(struct thread_data *td)
+static uint64_t do_io(struct thread_data *td)
 {
+	uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
 	unsigned int i;
 	int ret = 0;
 
@@ -529,17 +628,16 @@ static void do_io(struct thread_data *td)
 	else
 		td_set_runstate(td, TD_RUNNING);
 
-	while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
-		(!flist_empty(&td->trim_list)) ||
-	        ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) {
+	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
+		(!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
+		td->o.time_based) {
 		struct timeval comp_time;
-		unsigned long bytes_done[2] = { 0, 0 };
 		int min_evts = 0;
 		struct io_u *io_u;
 		int ret2, full;
 		enum fio_ddir ddir;
 
-		if (td->terminate)
+		if (td->terminate || td->done)
 			break;
 
 		update_tv_cache(td);
@@ -552,6 +650,9 @@ static void do_io(struct thread_data *td)
 			}
 		}
 
+		if (flow_threshold_exceeded(td))
+			continue;
+
 		io_u = get_io_u(td);
 		if (!io_u)
 			break;
@@ -559,11 +660,12 @@ static void do_io(struct thread_data *td)
 		ddir = io_u->ddir;
 
 		/*
-		 * Add verification end_io handler, if asked to verify
-		 * a previously written file.
+		 * Add verification end_io handler if:
+		 *	- Asked to verify (!td_rw(td))
+		 *	- Or the io_u is from our verify list (mixed write/ver)
 		 */
 		if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ &&
-		    !td_rw(td)) {
+		    ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
 			if (td->o.verify_async)
 				io_u->end_io = verify_io_u_async;
 			else
@@ -606,8 +708,9 @@ static void do_io(struct thread_data *td)
 				requeue_io_u(td, &io_u);
 			} else {
 sync_done:
-				if (__should_check_rate(td, 0) ||
-				    __should_check_rate(td, 1))
+				if (__should_check_rate(td, DDIR_READ) ||
+				    __should_check_rate(td, DDIR_WRITE) ||
+				    __should_check_rate(td, DDIR_TRIM))
 					fio_gettime(&comp_time, NULL);
 
 				ret = io_u_sync_complete(td, io_u, bytes_done);
@@ -648,11 +751,15 @@ sync_done:
 		if (full || !td->o.iodepth_batch_complete) {
 			min_evts = min(td->o.iodepth_batch_complete,
 					td->cur_depth);
-			if (full && !min_evts && td->o.iodepth_batch_complete != 0)
+			/*
+			 * if the queue is full, we MUST reap at least 1 event
+			 */
+			if (full && !min_evts)
 				min_evts = 1;
 
-			if (__should_check_rate(td, 0) ||
-			    __should_check_rate(td, 1))
+			if (__should_check_rate(td, DDIR_READ) ||
+			    __should_check_rate(td, DDIR_WRITE) ||
+			    __should_check_rate(td, DDIR_TRIM))
 				fio_gettime(&comp_time, NULL);
 
 			do {
@@ -665,7 +772,7 @@ sync_done:
 
 		if (ret < 0)
 			break;
-		if (!(bytes_done[0] + bytes_done[1]))
+		if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO))
 			continue;
 
 		if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) {
@@ -680,7 +787,7 @@ sync_done:
 		if (td->o.thinktime) {
 			unsigned long long b;
 
-			b = td->io_blocks[0] + td->io_blocks[1];
+			b = ddir_rw_sum(td->io_blocks);
 			if (!(b % td->o.thinktime_blocks)) {
 				int left;
 
@@ -706,7 +813,7 @@ sync_done:
 
 		i = td->cur_depth;
 		if (i) {
-			ret = io_u_queued_complete(td, i, NULL);
+			ret = io_u_queued_complete(td, i, bytes_done);
 			if (td->o.fill_device && td->error == ENOSPC)
 				td->error = 0;
 		}
@@ -726,8 +833,10 @@ sync_done:
 	/*
 	 * stop job if we failed doing any IO
 	 */
-	if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0)
+	if (!ddir_rw_sum(td->this_io_bytes))
 		td->done = 1;
+
+	return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
 }
 
 static void cleanup_io_u(struct thread_data *td)
@@ -739,6 +848,10 @@ static void cleanup_io_u(struct thread_data *td)
 		io_u = flist_entry(entry, struct io_u, list);
 
 		flist_del(&io_u->list);
+
+		if (td->io_ops->io_u_free)
+			td->io_ops->io_u_free(td, io_u);
+
 		fio_memfree(io_u, sizeof(*io_u));
 	}
 
@@ -748,15 +861,21 @@ static void cleanup_io_u(struct thread_data *td)
 static int init_io_u(struct thread_data *td)
 {
 	struct io_u *io_u;
-	unsigned int max_bs;
+	unsigned int max_bs, min_write;
 	int cl_align, i, max_units;
+	int data_xfer = 1;
 	char *p;
 
 	max_units = td->o.iodepth;
 	max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
+	max_bs = max(td->o.max_bs[DDIR_TRIM], max_bs);
+	min_write = td->o.min_bs[DDIR_WRITE];
 	td->orig_buffer_size = (unsigned long long) max_bs
 					* (unsigned long long) max_units;
 
+	if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td)))
+		data_xfer = 0;
+
 	if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) {
 		unsigned long bs;
 
@@ -769,7 +888,7 @@ static int init_io_u(struct thread_data *td)
 		return 1;
 	}
 
-	if (allocate_io_mem(td))
+	if (data_xfer && allocate_io_mem(td))
 		return 1;
 
 	if (td->o.odirect || td->o.mem_align ||
@@ -797,12 +916,12 @@ static int init_io_u(struct thread_data *td)
 		INIT_FLIST_HEAD(&io_u->list);
 		dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i);
 
-		if (!(td->io_ops->flags & FIO_NOIO)) {
+		if (data_xfer) {
 			io_u->buf = p;
 			dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf);
 
 			if (td_write(td))
-				io_u_fill_buffer(td, io_u, max_bs);
+				io_u_fill_buffer(td, io_u, min_write, max_bs);
 			if (td_write(td) && td->o.verify_pattern_bytes) {
 				/*
 				 * Fill the buffer with the pattern if we are
@@ -815,6 +934,16 @@ static int init_io_u(struct thread_data *td)
 		io_u->index = i;
 		io_u->flags = IO_U_F_FREE;
 		flist_add(&io_u->list, &td->io_u_freelist);
+
+		if (td->io_ops->io_u_init) {
+			int ret = td->io_ops->io_u_init(td, io_u);
+
+			if (ret) {
+				log_err("fio: failed to init engine data: %d\n", ret);
+				return 1;
+			}
+		}
+
 		p += max_bs;
 	}
 
@@ -879,8 +1008,6 @@ static int switch_ioscheduler(struct thread_data *td)
 
 static int keep_running(struct thread_data *td)
 {
-	unsigned long long io_done;
-
 	if (td->done)
 		return 0;
 	if (td->o.time_based)
@@ -890,10 +1017,19 @@ static int keep_running(struct thread_data *td)
 		return 1;
 	}
 
-	io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]
-			+ td->io_skip_bytes;
-	if (io_done < td->o.size)
+	if (ddir_rw_sum(td->io_bytes) < td->o.size) {
+		uint64_t diff;
+
+		/*
+		 * If the difference is less than the minimum IO size, we
+		 * are done.
+		 */
+		diff = td->o.size - ddir_rw_sum(td->io_bytes);
+		if (diff < td->o.rw_min_bs)
+			return 0;
+
 		return 1;
+	}
 
 	return 0;
 }
@@ -922,17 +1058,24 @@ static void *thread_main(void *data)
 {
 	unsigned long long elapsed;
 	struct thread_data *td = data;
+	struct thread_options *o = &td->o;
 	pthread_condattr_t attr;
 	int clear_state;
+	int ret;
 
-	if (!td->o.use_thread) {
+	if (!o->use_thread) {
 		setsid();
 		td->pid = getpid();
 	} else
 		td->pid = gettid();
 
+	fio_local_clock_init(td->o.use_thread);
+
 	dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
 
+	if (is_backend)
+		fio_server_send_start(td);
+
 	INIT_FLIST_HEAD(&td->io_u_freelist);
 	INIT_FLIST_HEAD(&td->io_u_busylist);
 	INIT_FLIST_HEAD(&td->io_u_requeues);
@@ -940,6 +1083,7 @@ static void *thread_main(void *data)
 	INIT_FLIST_HEAD(&td->io_hist_list);
 	INIT_FLIST_HEAD(&td->verify_list);
 	INIT_FLIST_HEAD(&td->trim_list);
+	INIT_FLIST_HEAD(&td->next_rand_list);
 	pthread_mutex_init(&td->io_u_lock, NULL);
 	td->io_hist_tree = RB_ROOT;
 
@@ -959,16 +1103,17 @@ static void *thread_main(void *data)
 	 * eating a file descriptor
 	 */
 	fio_mutex_remove(td->mutex);
+	td->mutex = NULL;
 
 	/*
 	 * A new gid requires privilege, so we need to do this before setting
 	 * the uid.
 	 */
-	if (td->o.gid != -1U && setgid(td->o.gid)) {
+	if (o->gid != -1U && setgid(o->gid)) {
 		td_verror(td, errno, "setgid");
 		goto err;
 	}
-	if (td->o.uid != -1U && setuid(td->o.uid)) {
+	if (o->uid != -1U && setuid(o->uid)) {
 		td_verror(td, errno, "setuid");
 		goto err;
 	}
@@ -977,17 +1122,63 @@ static void *thread_main(void *data)
 	 * If we have a gettimeofday() thread, make sure we exclude that
 	 * thread from this job
 	 */
-	if (td->o.gtod_cpu)
-		fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu);
+	if (o->gtod_cpu)
+		fio_cpu_clear(&o->cpumask, o->gtod_cpu);
 
 	/*
 	 * Set affinity first, in case it has an impact on the memory
 	 * allocations.
 	 */
-	if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) {
-		td_verror(td, errno, "cpu_set_affinity");
-		goto err;
+	if (o->cpumask_set) {
+		ret = fio_setaffinity(td->pid, o->cpumask);
+		if (ret == -1) {
+			td_verror(td, errno, "cpu_set_affinity");
+			goto err;
+		}
+	}
+
+#ifdef CONFIG_LIBNUMA
+	/* numa node setup */
+	if (td->o.numa_cpumask_set || td->o.numa_memmask_set) {
+		int ret;
+
+		if (numa_available() < 0) {
+			td_verror(td, errno, "Does not support NUMA API\n");
+			goto err;
+		}
+
+		if (td->o.numa_cpumask_set) {
+			ret = numa_run_on_node_mask(td->o.numa_cpunodesmask);
+			if (ret == -1) {
+				td_verror(td, errno, \
+					"numa_run_on_node_mask failed\n");
+				goto err;
+			}
+		}
+
+		if (td->o.numa_memmask_set) {
+
+			switch (td->o.numa_mem_mode) {
+			case MPOL_INTERLEAVE:
+				numa_set_interleave_mask(td->o.numa_memnodesmask);
+				break;
+			case MPOL_BIND:
+				numa_set_membind(td->o.numa_memnodesmask);
+				break;
+			case MPOL_LOCAL:
+				numa_set_localalloc();
+				break;
+			case MPOL_PREFERRED:
+				numa_set_preferred(td->o.numa_mem_prefer_node);
+				break;
+			case MPOL_DEFAULT:
+			default:
+				break;
+			}
+
+		}
 	}
+#endif
 
 	/*
 	 * May alter parameters that init_io_u() will use, so we need to
@@ -999,28 +1190,30 @@ static void *thread_main(void *data)
 	if (init_io_u(td))
 		goto err;
 
-	if (td->o.verify_async && verify_async_init(td))
+	if (o->verify_async && verify_async_init(td))
 		goto err;
 
-	if (td->ioprio_set) {
-		if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
+	if (o->ioprio) {
+		ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
+		if (ret == -1) {
 			td_verror(td, errno, "ioprio_set");
 			goto err;
 		}
 	}
 
-	if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt))
+	if (td->o.cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt))
 		goto err;
 
-	if (nice(td->o.nice) == -1) {
+	errno = 0;
+	if (nice(o->nice) == -1 && errno != 0) {
 		td_verror(td, errno, "nice");
 		goto err;
 	}
 
-	if (td->o.ioscheduler && switch_ioscheduler(td))
+	if (o->ioscheduler && switch_ioscheduler(td))
 		goto err;
 
-	if (!td->o.create_serialize && setup_files(td))
+	if (!o->create_serialize && setup_files(td))
 		goto err;
 
 	if (td_io_init(td))
@@ -1029,30 +1222,34 @@ static void *thread_main(void *data)
 	if (init_random_map(td))
 		goto err;
 
-	if (td->o.exec_prerun) {
-		if (exec_string(td->o.exec_prerun))
-			goto err;
-	}
+	if (o->exec_prerun && exec_string(o->exec_prerun))
+		goto err;
 
-	if (td->o.pre_read) {
+	if (o->pre_read) {
 		if (pre_read_files(td) < 0)
 			goto err;
 	}
 
-	fio_gettime(&td->epoch, NULL);
-	getrusage(RUSAGE_SELF, &td->ru_start);
+	fio_verify_init(td);
 
+	fio_gettime(&td->epoch, NULL);
+	fio_getrusage(&td->ru_start);
 	clear_state = 0;
 	while (keep_running(td)) {
+		uint64_t verify_bytes;
+
 		fio_gettime(&td->start, NULL);
 		memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
 		memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));
 		memcpy(&td->tv_cache, &td->start, sizeof(td->start));
 
-		if (td->o.ratemin[0] || td->o.ratemin[1]) {
-		        memcpy(&td->lastrate[0], &td->bw_sample_time,
+		if (td->o.ratemin[DDIR_READ] || td->o.ratemin[DDIR_WRITE] ||
+				td->o.ratemin[DDIR_TRIM]) {
+		        memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
+						sizeof(td->bw_sample_time));
+		        memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
 						sizeof(td->bw_sample_time));
-		        memcpy(&td->lastrate[1], &td->bw_sample_time,
+		        memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
 						sizeof(td->bw_sample_time));
 		}
 
@@ -1061,7 +1258,7 @@ static void *thread_main(void *data)
 
 		prune_io_piece_log(td);
 
-		do_io(td);
+		verify_bytes = do_io(td);
 
 		clear_state = 1;
 
@@ -1073,6 +1270,10 @@ static void *thread_main(void *data)
 			elapsed = utime_since_now(&td->start);
 			td->ts.runtime[DDIR_WRITE] += elapsed;
 		}
+		if (td_trim(td) && td->io_bytes[DDIR_TRIM]) {
+			elapsed = utime_since_now(&td->start);
+			td->ts.runtime[DDIR_TRIM] += elapsed;
+		}
 
 		if (td->error || td->terminate)
 			break;
@@ -1086,7 +1287,7 @@ static void *thread_main(void *data)
 
 		fio_gettime(&td->start, NULL);
 
-		do_verify(td);
+		do_verify(td, verify_bytes);
 
 		td->ts.runtime[DDIR_READ] += utime_since_now(&td->start);
 
@@ -1095,11 +1296,15 @@ static void *thread_main(void *data)
 	}
 
 	update_rusage_stat(td);
-	td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000;
-	td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000;
+	td->ts.runtime[DDIR_READ] = (td->ts.runtime[DDIR_READ] + 999) / 1000;
+	td->ts.runtime[DDIR_WRITE] = (td->ts.runtime[DDIR_WRITE] + 999) / 1000;
+	td->ts.runtime[DDIR_TRIM] = (td->ts.runtime[DDIR_TRIM] + 999) / 1000;
 	td->ts.total_run_time = mtime_since_now(&td->epoch);
-	td->ts.io_bytes[0] = td->io_bytes[0];
-	td->ts.io_bytes[1] = td->io_bytes[1];
+	td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ];
+	td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE];
+	td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
+
+	fio_unpin_memory(td);
 
 	fio_mutex_down(writeout_mutex);
 	if (td->bw_log) {
@@ -1154,12 +1359,12 @@ err:
 		verify_async_exit(td);
 
 	close_and_free_files(td);
-	close_ioengine(td);
 	cleanup_io_u(td);
+	close_ioengine(td);
 	cgroup_shutdown(td, &cgroup_mnt);
 
-	if (td->o.cpumask_set) {
-		int ret = fio_cpuset_exit(&td->o.cpumask);
+	if (o->cpumask_set) {
+		int ret = fio_cpuset_exit(&o->cpumask);
 
 		td_verror(td, ret, "fio_cpuset_exit");
 	}
@@ -1171,7 +1376,7 @@ err:
 		write_iolog_close(td);
 
 	td_set_runstate(td, TD_EXITED);
-	return (void *) (unsigned long) td->error;
+	return (void *) (uintptr_t) td->error;
 }
 
 
@@ -1202,7 +1407,7 @@ static int fork_main(int shmid, int offset)
 	td = data + offset * sizeof(struct thread_data);
 	ret = thread_main(td);
 	shmdt(data);
-	return (int) (unsigned long) ret;
+	return (int) (uintptr_t) ret;
 }
 
 /*
@@ -1257,6 +1462,7 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
 			if (errno == ECHILD) {
 				log_err("fio: pid=%d disappeared %d\n",
 						(int) td->pid, td->runstate);
+				td->sig = ECHILD;
 				td_set_runstate(td, TD_REAPED);
 				goto reaped;
 			}
@@ -1265,9 +1471,10 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
 			if (WIFSIGNALED(status)) {
 				int sig = WTERMSIG(status);
 
-				if (sig != SIGTERM)
+				if (sig != SIGTERM && sig != SIGUSR2)
 					log_err("fio: pid=%d, got signal=%d\n",
 							(int) td->pid, sig);
+				td->sig = sig;
 				td_set_runstate(td, TD_REAPED);
 				goto reaped;
 			}
@@ -1287,8 +1494,8 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
 		continue;
 reaped:
 		(*nr_running)--;
-		(*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]);
-		(*t_rate) -= (td->o.rate[0] + td->o.rate[1]);
+		(*m_rate) -= ddir_rw_sum(td->o.ratemin);
+		(*t_rate) -= ddir_rw_sum(td->o.rate);
 		if (!td->pid)
 			pending--;
 
@@ -1302,8 +1509,6 @@ reaped:
 		fio_terminate_threads(TERMINATE_ALL);
 }
 
-
-
 /*
  * Main function for kicking off and reaping jobs, as needed.
  */
@@ -1313,15 +1518,22 @@ static void run_threads(void)
 	unsigned long spent;
 	unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
 
-	if (fio_pin_memory())
-		return;
-
 	if (fio_gtod_offload && fio_start_gtod_thread())
 		return;
+	
+	fio_idle_prof_init();
 
 	set_sig_handlers();
 
-	if (!terse_output) {
+	nr_thread = nr_process = 0;
+	for_each_td(td, i) {
+		if (td->o.use_thread)
+			nr_thread++;
+		else
+			nr_process++;
+	}
+
+	if (output_format == FIO_OUTPUT_NORMAL) {
 		log_info("Starting ");
 		if (nr_thread)
 			log_info("%d thread%s", nr_thread,
@@ -1375,6 +1587,9 @@ static void run_threads(void)
 		}
 	}
 
+	/* start idle threads before io threads start to run */
+	fio_idle_prof_start();
+
 	set_genesis_time();
 
 	while (todo) {
@@ -1488,7 +1703,8 @@ static void run_threads(void)
 		}
 
 		if (left) {
-			log_err("fio: %d jobs failed to start\n", left);
+			log_err("fio: %d job%s failed to start\n", left,
+					left > 1 ? "s" : "");
 			for (i = 0; i < this_jobs; i++) {
 				td = map[i];
 				if (!td)
@@ -1511,49 +1727,57 @@ static void run_threads(void)
 				td_set_runstate(td, TD_RUNNING);
 			nr_running++;
 			nr_started--;
-			m_rate += td->o.ratemin[0] + td->o.ratemin[1];
-			t_rate += td->o.rate[0] + td->o.rate[1];
+			m_rate += ddir_rw_sum(td->o.ratemin);
+			t_rate += ddir_rw_sum(td->o.rate);
 			todo--;
 			fio_mutex_up(td->mutex);
 		}
 
 		reap_threads(&nr_running, &t_rate, &m_rate);
 
-		if (todo) {
-			if (is_backend)
-				fio_server_idle_loop();
-			else
-				usleep(100000);
-		}
+		if (todo)
+			usleep(100000);
 	}
 
 	while (nr_running) {
 		reap_threads(&nr_running, &t_rate, &m_rate);
-
-		if (is_backend)
-			fio_server_idle_loop();
-		else
-			usleep(10000);
+		usleep(10000);
 	}
 
+	fio_idle_prof_stop();
+
 	update_io_ticks();
-	fio_unpin_memory();
+}
+
+void wait_for_disk_thread_exit(void)
+{
+	fio_mutex_down(disk_thread_mutex);
+}
+
+static void free_disk_util(void)
+{
+	disk_util_start_exit();
+	wait_for_disk_thread_exit();
+	disk_util_prune_entries();
 }
 
 static void *disk_thread_main(void *data)
 {
+	int ret = 0;
+
 	fio_mutex_up(startup_mutex);
 
-	while (threads) {
+	while (threads && !ret) {
 		usleep(DISK_UTIL_MSEC * 1000);
 		if (!threads)
 			break;
-		update_io_ticks();
+		ret = update_io_ticks();
 
 		if (!is_backend)
 			print_thread_status();
 	}
 
+	fio_mutex_up(disk_thread_mutex);
 	return NULL;
 }
 
@@ -1561,14 +1785,20 @@ static int create_disk_util_thread(void)
 {
 	int ret;
 
+	setup_disk_util();
+
+	disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
+
 	ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL);
 	if (ret) {
+		fio_mutex_remove(disk_thread_mutex);
 		log_err("Can't create disk util thread: %s\n", strerror(ret));
 		return 1;
 	}
 
 	ret = pthread_detach(disk_util_thread);
 	if (ret) {
+		fio_mutex_remove(disk_thread_mutex);
 		log_err("Can't detatch disk util thread: %s\n", strerror(ret));
 		return 1;
 	}
@@ -1579,7 +1809,6 @@ static int create_disk_util_thread(void)
 	return 0;
 }
 
-
 int fio_backend(void)
 {
 	struct thread_data *td;
@@ -1595,14 +1824,15 @@ int fio_backend(void)
 		return 0;
 
 	if (write_bw_log) {
-		setup_log(&agg_io_log[DDIR_READ], 0);
-		setup_log(&agg_io_log[DDIR_WRITE], 0);
+		setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW);
+		setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW);
+		setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW);
 	}
 
-	startup_mutex = fio_mutex_init(0);
+	startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
 	if (startup_mutex == NULL)
 		return 1;
-	writeout_mutex = fio_mutex_init(1);
+	writeout_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);
 	if (writeout_mutex == NULL)
 		return 1;
 
@@ -1620,19 +1850,21 @@ int fio_backend(void)
 			__finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log");
 			__finish_log(agg_io_log[DDIR_WRITE],
 					"agg-write_bw.log");
+			__finish_log(agg_io_log[DDIR_TRIM],
+					"agg-write_bw.log");
 		}
 	}
 
 	for_each_td(td, i)
 		fio_options_free(td);
 
+	free_disk_util();
 	cgroup_kill(cgroup_list);
 	sfree(cgroup_list);
 	sfree(cgroup_mnt);
 
 	fio_mutex_remove(startup_mutex);
 	fio_mutex_remove(writeout_mutex);
+	fio_mutex_remove(disk_thread_mutex);
 	return exit_value;
 }
-
-