X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=592ea8bfaa05cc17b4a8a3e23c1cd54b5aebea91;hp=ce0a0098299294279ea4bf16dfd88b7079d9466e;hb=90b7a96d9573a14f20470af2aa0dd4e60611a477;hpb=f3afa57e36550288340f1b6c694f354ae72654b9

diff --git a/backend.c b/backend.c
index ce0a0098..592ea8bf 100644
--- a/backend.c
+++ b/backend.c
@@ -34,7 +34,9 @@
 #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"
@@ -48,6 +50,8 @@
 #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;
@@ -57,16 +61,18 @@ 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];
 
 int groupid = 0;
 unsigned int thread_number = 0;
-unsigned int nr_process = 0;
-unsigned int nr_thread = 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 *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
@@ -107,6 +113,14 @@ 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;
@@ -204,7 +218,7 @@ 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;
 
@@ -337,17 +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.
@@ -381,8 +395,9 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
  * 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;
@@ -410,6 +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);
@@ -425,18 +441,53 @@ static void do_verify(struct thread_data *td)
 		if (flow_threshold_exceeded(td))
 			continue;
 
-		io_u = __get_io_u(td);
-		if (!io_u)
-			break;
+		if (!td->o.experimental_verify) {
+			io_u = __get_io_u(td);
+			if (!io_u)
+				break;
 
-		if (get_next_verify(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;
+			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)
@@ -444,6 +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:
@@ -476,7 +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;
 			}
@@ -495,7 +548,7 @@ sync_done:
 			break;
 		}
 
-		if (break_on_this_error(td, io_u->ddir, &ret))
+		if (break_on_this_error(td, ddir, &ret))
 			break;
 
 		/*
@@ -519,7 +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;
 				}
@@ -561,9 +614,12 @@ static int io_bytes_exceeded(struct thread_data *td)
 /*
  * 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;
 
@@ -576,13 +632,12 @@ static void do_io(struct thread_data *td)
 		(!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
 		td->o.time_based) {
 		struct timeval comp_time;
-		unsigned long bytes_done[DDIR_RWDIR_CNT] = { 0, 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);
@@ -717,7 +772,7 @@ sync_done:
 
 		if (ret < 0)
 			break;
-		if (!ddir_rw_sum(bytes_done))
+		if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO))
 			continue;
 
 		if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) {
@@ -758,7 +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;
 		}
@@ -780,6 +835,8 @@ sync_done:
 	 */
 	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)
@@ -791,6 +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));
 	}
 
@@ -802,6 +863,7 @@ static int init_io_u(struct thread_data *td)
 	struct io_u *io_u;
 	unsigned int max_bs, min_write;
 	int cl_align, i, max_units;
+	int data_xfer = 1;
 	char *p;
 
 	max_units = td->o.iodepth;
@@ -811,6 +873,9 @@ static int init_io_u(struct thread_data *td)
 	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;
 
@@ -823,7 +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 ||
@@ -851,7 +916,7 @@ 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);
 
@@ -869,6 +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;
 	}
 
@@ -942,8 +1017,19 @@ static int keep_running(struct thread_data *td)
 		return 1;
 	}
 
-	if (ddir_rw_sum(td->io_bytes) < 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;
 }
@@ -972,17 +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);
@@ -990,6 +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;
 
@@ -1009,16 +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;
 	}
@@ -1027,18 +1122,64 @@ 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
 	 * do this first.
@@ -1049,11 +1190,12 @@ 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;
 		}
@@ -1063,15 +1205,15 @@ static void *thread_main(void *data)
 		goto err;
 
 	errno = 0;
-	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))
@@ -1080,21 +1222,22 @@ 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));
@@ -1115,7 +1258,7 @@ static void *thread_main(void *data)
 
 		prune_io_piece_log(td);
 
-		do_io(td);
+		verify_bytes = do_io(td);
 
 		clear_state = 1;
 
@@ -1144,7 +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);
 
@@ -1161,6 +1304,8 @@ static void *thread_main(void *data)
 	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) {
 		if (td->o.bw_log_file) {
@@ -1214,12 +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");
 	}
@@ -1326,7 +1471,7 @@ 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;
@@ -1373,14 +1518,21 @@ 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();
 
+	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)
@@ -1435,6 +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) {
@@ -1580,25 +1735,18 @@ static void run_threads(void)
 
 		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)
@@ -1606,6 +1754,13 @@ 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;
@@ -1669,9 +1824,9 @@ 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_TRIM], 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(FIO_MUTEX_LOCKED);