X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=022122a2d9914ca78a6406075e41e3c58b58242a;hp=3a1df299fd80f920ee97d8f1ecb79b525726a838;hb=62232d0618e3f5aa4805b4be34e872895ba1beb2;hpb=73cfb8a7da365c37253662477edb54ecc19575df

diff --git a/backend.c b/backend.c
index 3a1df299..022122a2 100644
--- a/backend.c
+++ b/backend.c
@@ -34,10 +34,12 @@
 #include <sys/stat.h>
 #include <sys/wait.h>
 #include <sys/ipc.h>
-#include <sys/shm.h>
 #include <sys/mman.h>
 
 #include "fio.h"
+#ifndef FIO_NO_HAVE_SHM_H
+#include <sys/shm.h>
+#endif
 #include "hash.h"
 #include "smalloc.h"
 #include "verify.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;
@@ -64,6 +68,7 @@ struct io_log *agg_io_log[DDIR_RWDIR_CNT];
 
 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;
@@ -108,6 +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;
@@ -205,7 +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;
 
@@ -309,6 +322,21 @@ requeue:
 	return 0;
 }
 
+static int fio_file_fsync(struct thread_data *td, struct fio_file *f)
+{
+	int ret;
+
+	if (fio_file_open(f))
+		return fio_io_sync(td, f);
+
+	if (td_io_open_file(td, f))
+		return 1;
+
+	ret = fio_io_sync(td, f);
+	td_io_close_file(td, f);
+	return ret;
+}
+
 static inline void __update_tv_cache(struct thread_data *td)
 {
 	fio_gettime(&td->tv_cache, NULL);
@@ -378,12 +406,22 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 	return 0;
 }
 
+static void check_update_rusage(struct thread_data *td)
+{
+	if (td->update_rusage) {
+		td->update_rusage = 0;
+		update_rusage_stat(td);
+		fio_mutex_up(td->rusage_sem);
+	}
+}
+
 /*
  * 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;
@@ -404,6 +442,8 @@ static void do_verify(struct thread_data *td)
 			break;
 	}
 
+	check_update_rusage(td);
+
 	if (td->error)
 		return;
 
@@ -411,9 +451,11 @@ 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);
+		check_update_rusage(td);
 
 		if (runtime_exceeded(td, &td->tv_cache)) {
 			__update_tv_cache(td);
@@ -426,18 +468,53 @@ 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)
@@ -445,6 +522,8 @@ 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:
@@ -477,7 +556,7 @@ 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;
 			}
@@ -496,7 +575,7 @@ sync_done:
 			break;
 		}
 
-		if (break_on_this_error(td, io_u->ddir, &ret))
+		if (break_on_this_error(td, ddir, &ret))
 			break;
 
 		/*
@@ -520,7 +599,7 @@ 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;
 				}
@@ -530,6 +609,8 @@ sync_done:
 			break;
 	}
 
+	check_update_rusage(td);
+
 	if (!td->error) {
 		min_events = td->cur_depth;
 
@@ -562,9 +643,12 @@ 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;
 
@@ -577,13 +661,14 @@ 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)
+		check_update_rusage(td);
+
+		if (td->terminate || td->done)
 			break;
 
 		update_tv_cache(td);
@@ -718,7 +803,7 @@ 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)) {
@@ -747,6 +832,8 @@ sync_done:
 		}
 	}
 
+	check_update_rusage(td);
+
 	if (td->trim_entries)
 		log_err("fio: %d trim entries leaked?\n", td->trim_entries);
 
@@ -759,7 +846,7 @@ 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;
 		}
@@ -768,9 +855,11 @@ sync_done:
 			td_set_runstate(td, TD_FSYNCING);
 
 			for_each_file(td, f, i) {
-				if (!fio_file_open(f))
+				if (!fio_file_fsync(td, f))
 					continue;
-				fio_io_sync(td, f);
+
+				log_err("fio: end_fsync failed for file %s\n",
+								f->file_name);
 			}
 		}
 	} else
@@ -781,6 +870,8 @@ 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)
@@ -792,6 +883,10 @@ 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));
 	}
 
@@ -807,8 +902,7 @@ static int init_io_u(struct thread_data *td)
 	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);
+	max_bs = td_max_bs(td);
 	min_write = td->o.min_bs[DDIR_WRITE];
 	td->orig_buffer_size = (unsigned long long) max_bs
 					* (unsigned long long) max_units;
@@ -874,6 +968,16 @@ 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;
 	}
 
@@ -947,8 +1051,19 @@ static int keep_running(struct thread_data *td)
 		return 1;
 	}
 
-	if (ddir_rw_sum(td->io_bytes) < td->o.size)
+	if (td->o.size != -1ULL && 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_max_bs(td))
+			return 0;
+
 		return 1;
+	}
 
 	return 0;
 }
@@ -988,6 +1103,8 @@ static void *thread_main(void *data)
 	} 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)
@@ -1000,6 +1117,7 @@ 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;
 
@@ -1053,8 +1171,48 @@ static void *thread_main(void *data)
 		}
 	}
 
-	if (fio_pin_memory(td))
-		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
@@ -1109,10 +1267,11 @@ static void *thread_main(void *data)
 	fio_verify_init(td);
 
 	fio_gettime(&td->epoch, NULL);
-	getrusage(RUSAGE_SELF, &td->ru_start);
-
+	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));
@@ -1133,7 +1292,7 @@ static void *thread_main(void *data)
 
 		prune_io_piece_log(td);
 
-		do_io(td);
+		verify_bytes = do_io(td);
 
 		clear_state = 1;
 
@@ -1162,7 +1321,7 @@ 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);
 
@@ -1234,8 +1393,8 @@ 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 (o->cpumask_set) {
@@ -1250,6 +1409,9 @@ err:
 	if (td->o.write_iolog_file)
 		write_iolog_close(td);
 
+	fio_mutex_remove(td->rusage_sem);
+	td->rusage_sem = NULL;
+
 	td_set_runstate(td, TD_EXITED);
 	return (void *) (uintptr_t) td->error;
 }
@@ -1346,7 +1508,7 @@ 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;
@@ -1395,6 +1557,8 @@ static void run_threads(void)
 
 	if (fio_gtod_offload && fio_start_gtod_thread())
 		return;
+	
+	fio_idle_prof_init();
 
 	set_sig_handlers();
 
@@ -1460,6 +1624,9 @@ static void run_threads(void)
 		}
 	}
 
+	/* start idle threads before io threads start to run */
+	fio_idle_prof_start();
+
 	set_genesis_time();
 
 	while (todo) {
@@ -1498,6 +1665,9 @@ static void run_threads(void)
 
 			init_disk_util(td);
 
+			td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
+			td->update_rusage = 0;
+
 			/*
 			 * Set state to created. Thread will transition
 			 * to TD_INITIALIZED when it's done setting up.
@@ -1614,6 +1784,8 @@ static void run_threads(void)
 		usleep(10000);
 	}
 
+	fio_idle_prof_stop();
+
 	update_io_ticks();
 }