X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=iolog.c;h=460d7a2e51af8ada55d8b324547d6fb1dc12b74b;hp=975ce6f7a481c438025992f4b828b6d69378b290;hb=ae626d4ead6416adf464cf209cdf3e8b85d58190;hpb=9973b0f961a57c19f885ffca05f86ae6ef85f8c7

diff --git a/iolog.c b/iolog.c
index 975ce6f7..460d7a2e 100644
--- a/iolog.c
+++ b/iolog.c
@@ -19,6 +19,8 @@
 #include "trim.h"
 #include "filelock.h"
 #include "smalloc.h"
+#include "blktrace.h"
+#include "pshared.h"
 
 static int iolog_flush(struct io_log *log);
 
@@ -64,7 +66,7 @@ static void iolog_delay(struct thread_data *td, unsigned long delay)
 {
 	uint64_t usec = utime_since_now(&td->last_issue);
 	uint64_t this_delay;
-	struct timeval tv;
+	struct timespec ts;
 
 	if (delay < td->time_offset) {
 		td->time_offset = 0;
@@ -77,7 +79,7 @@ static void iolog_delay(struct thread_data *td, unsigned long delay)
 
 	delay -= usec;
 
-	fio_gettime(&tv, NULL);
+	fio_gettime(&ts, NULL);
 	while (delay && !td->terminate) {
 		this_delay = delay;
 		if (this_delay > 500000)
@@ -87,7 +89,7 @@ static void iolog_delay(struct thread_data *td, unsigned long delay)
 		delay -= this_delay;
 	}
 
-	usec = utime_since_now(&tv);
+	usec = utime_since_now(&ts);
 	if (usec > delay)
 		td->time_offset = usec - delay;
 	else
@@ -109,6 +111,11 @@ static int ipo_special(struct thread_data *td, struct io_piece *ipo)
 
 	switch (ipo->file_action) {
 	case FIO_LOG_OPEN_FILE:
+		if (td->o.replay_redirect && fio_file_open(f)) {
+			dprint(FD_FILE, "iolog: ignoring re-open of file %s\n",
+					f->file_name);
+			break;
+		}
 		ret = td_io_open_file(td, f);
 		if (!ret)
 			break;
@@ -178,7 +185,7 @@ int read_iolog_get(struct thread_data *td, struct io_u *io_u)
 void prune_io_piece_log(struct thread_data *td)
 {
 	struct io_piece *ipo;
-	struct rb_node *n;
+	struct fio_rb_node *n;
 
 	while ((n = rb_first(&td->io_hist_tree)) != NULL) {
 		ipo = rb_entry(n, struct io_piece, rb_node);
@@ -202,7 +209,7 @@ void prune_io_piece_log(struct thread_data *td)
  */
 void log_io_piece(struct thread_data *td, struct io_u *io_u)
 {
-	struct rb_node **p, *parent;
+	struct fio_rb_node **p, *parent;
 	struct io_piece *ipo, *__ipo;
 
 	ipo = malloc(sizeof(struct io_piece));
@@ -221,21 +228,16 @@ void log_io_piece(struct thread_data *td, struct io_u *io_u)
 	}
 
 	/*
-	 * We don't need to sort the entries, if:
-	 *
-	 *	Sequential writes, or
-	 *	Random writes that lay out the file as it goes along
+	 * We don't need to sort the entries if we only performed sequential
+	 * writes. In this case, just reading back data in the order we wrote
+	 * it out is the faster but still safe.
 	 *
-	 * For both these cases, just reading back data in the order we
-	 * wrote it out is the fastest.
-	 *
-	 * One exception is if we don't have a random map AND we are doing
-	 * verifies, in that case we need to check for duplicate blocks and
-	 * drop the old one, which we rely on the rb insert/lookup for
-	 * handling.
+	 * One exception is if we don't have a random map in which case we need
+	 * to check for duplicate blocks and drop the old one, which we rely on
+	 * the rb insert/lookup for handling.
 	 */
-	if (((!td->o.verifysort) || !td_random(td) || !td->o.overwrite) &&
-	      (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
+	if (((!td->o.verifysort) || !td_random(td)) &&
+	      file_randommap(td, ipo->file)) {
 		INIT_FLIST_HEAD(&ipo->list);
 		flist_add_tail(&ipo->list, &td->io_hist_list);
 		ipo->flags |= IP_F_ONLIST;
@@ -272,13 +274,14 @@ restart:
 			overlap = 1;
 
 		if (overlap) {
-			dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu",
+			dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu\n",
 				__ipo->offset, __ipo->len,
 				ipo->offset, ipo->len);
 			td->io_hist_len--;
 			rb_erase(parent, &td->io_hist_tree);
 			remove_trim_entry(td, __ipo);
-			free(__ipo);
+			if (!(__ipo->flags & IP_F_IN_FLIGHT))
+				free(__ipo);
 			goto restart;
 		}
 	}
@@ -346,7 +349,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 	unsigned long long offset;
 	unsigned int bytes;
 	int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */
-	char *fname, *act;
+	char *rfname, *fname, *act;
 	char *str, *p;
 	enum fio_ddir rw;
 
@@ -357,7 +360,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 	 * for doing verifications.
 	 */
 	str = malloc(4096);
-	fname = malloc(256+16);
+	rfname = fname = malloc(256+16);
 	act = malloc(256+16);
 
 	reads = writes = waits = 0;
@@ -365,8 +368,12 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 		struct io_piece *ipo;
 		int r;
 
-		r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
+		r = sscanf(p, "%256s %256s %llu %u", rfname, act, &offset,
 									&bytes);
+
+		if (td->o.replay_redirect)
+			fname = td->o.replay_redirect;
+
 		if (r == 4) {
 			/*
 			 * Check action first
@@ -392,8 +399,14 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 		} else if (r == 2) {
 			rw = DDIR_INVAL;
 			if (!strcmp(act, "add")) {
-				fileno = add_file(td, fname, 0, 1);
-				file_action = FIO_LOG_ADD_FILE;
+				if (td->o.replay_redirect &&
+				    get_fileno(td, fname) != -1) {
+					dprint(FD_FILE, "iolog: ignoring"
+						" re-add of file %s\n", fname);
+				} else {
+					fileno = add_file(td, fname, 0, 1);
+					file_action = FIO_LOG_ADD_FILE;
+				}
 				continue;
 			} else if (!strcmp(act, "open")) {
 				fileno = get_fileno(td, fname);
@@ -407,7 +420,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 				continue;
 			}
 		} else {
-			log_err("bad iolog2: %s", p);
+			log_err("bad iolog2: %s\n", p);
 			continue;
 		}
 
@@ -421,6 +434,8 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 				continue;
 			writes++;
 		} else if (rw == DDIR_WAIT) {
+			if (td->o.no_stall)
+				continue;
 			waits++;
 		} else if (rw == DDIR_INVAL) {
 		} else if (!ddir_sync(rw)) {
@@ -437,7 +452,12 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 		if (rw == DDIR_WAIT) {
 			ipo->delay = offset;
 		} else {
-			ipo->offset = offset;
+			if (td->o.replay_scale)
+				ipo->offset = offset / td->o.replay_scale;
+			else
+				ipo->offset = offset;
+			ipo_bytes_align(td->o.replay_align, ipo);
+
 			ipo->len = bytes;
 			if (rw != DDIR_INVAL && bytes > td->o.max_bs[rw])
 				td->o.max_bs[rw] = bytes;
@@ -451,7 +471,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 
 	free(str);
 	free(act);
-	free(fname);
+	free(rfname);
 
 	if (writes && read_only) {
 		log_err("fio: <%s> skips replay of %d writes due to"
@@ -576,6 +596,9 @@ void setup_log(struct io_log **log, struct log_params *p,
 	       const char *filename)
 {
 	struct io_log *l;
+	int i;
+	struct io_u_plat_entry *entry;
+	struct flist_head *list;
 
 	l = scalloc(1, sizeof(*l));
 	INIT_FLIST_HEAD(&l->io_logs);
@@ -589,6 +612,16 @@ void setup_log(struct io_log **log, struct log_params *p,
 	l->filename = strdup(filename);
 	l->td = p->td;
 
+	/* Initialize histogram lists for each r/w direction,
+	 * with initial io_u_plat of all zeros:
+	 */
+	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+		list = &l->hist_window[i].list;
+		INIT_FLIST_HEAD(list);
+		entry = calloc(1, sizeof(struct io_u_plat_entry));
+		flist_add(&entry->list, list);
+	}
+
 	if (l->td && l->td->o.io_submit_mode != IO_MODE_OFFLOAD) {
 		struct io_logs *p;
 
@@ -607,6 +640,7 @@ void setup_log(struct io_log **log, struct log_params *p,
 		l->log_gz = 0;
 	else if (l->log_gz || l->log_gz_store) {
 		mutex_init_pshared(&l->chunk_lock);
+		mutex_init_pshared(&l->deferred_free_lock);
 		p->td->flags |= TD_F_COMPRESS_LOG;
 	}
 
@@ -661,24 +695,32 @@ void free_log(struct io_log *log)
 	sfree(log);
 }
 
-static inline unsigned long hist_sum(int j, int stride, unsigned int *io_u_plat)
+uint64_t hist_sum(int j, int stride, uint64_t *io_u_plat,
+		uint64_t *io_u_plat_last)
 {
-	unsigned long sum;
+	uint64_t sum;
 	int k;
 
-	for (k = sum = 0; k < stride; k++)
-		sum += io_u_plat[j + k];
+	if (io_u_plat_last) {
+		for (k = sum = 0; k < stride; k++)
+			sum += io_u_plat[j + k] - io_u_plat_last[j + k];
+	} else {
+		for (k = sum = 0; k < stride; k++)
+			sum += io_u_plat[j + k];
+	}
 
 	return sum;
 }
 
-void flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
-			uint64_t sample_size)
+static void flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
+			       uint64_t sample_size)
 {
 	struct io_sample *s;
 	int log_offset;
 	uint64_t i, j, nr_samples;
-	unsigned int *io_u_plat;
+	struct io_u_plat_entry *entry, *entry_before;
+	uint64_t *io_u_plat;
+	uint64_t *io_u_plat_before;
 
 	int stride = 1 << hist_coarseness;
 	
@@ -692,15 +734,25 @@ void flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
 
 	for (i = 0; i < nr_samples; i++) {
 		s = __get_sample(samples, log_offset, i);
-		io_u_plat = (unsigned int *) (uintptr_t) s->val;
-		fprintf(f, "%lu, %u, %u, ", (unsigned long)s->time,
-		        io_sample_ddir(s), s->bs);
+
+		entry = s->data.plat_entry;
+		io_u_plat = entry->io_u_plat;
+
+		entry_before = flist_first_entry(&entry->list, struct io_u_plat_entry, list);
+		io_u_plat_before = entry_before->io_u_plat;
+
+		fprintf(f, "%lu, %u, %u, ", (unsigned long) s->time,
+						io_sample_ddir(s), s->bs);
 		for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
-			fprintf(f, "%lu, ", hist_sum(j, stride, io_u_plat));
+			fprintf(f, "%llu, ", (unsigned long long)
+			        hist_sum(j, stride, io_u_plat, io_u_plat_before));
 		}
-		fprintf(f, "%lu\n", (unsigned long) 
-		        hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat));
-		free(io_u_plat);
+		fprintf(f, "%llu\n", (unsigned long long)
+		        hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat,
+					io_u_plat_before));
+
+		flist_del(&entry_before->list);
+		free(entry_before);
 	}
 }
 
@@ -722,16 +774,16 @@ void flush_samples(FILE *f, void *samples, uint64_t sample_size)
 		s = __get_sample(samples, log_offset, i);
 
 		if (!log_offset) {
-			fprintf(f, "%lu, %lu, %u, %u\n",
+			fprintf(f, "%lu, %" PRId64 ", %u, %u\n",
 					(unsigned long) s->time,
-					(unsigned long) s->val,
+					s->data.val,
 					io_sample_ddir(s), s->bs);
 		} else {
 			struct io_sample_offset *so = (void *) s;
 
-			fprintf(f, "%lu, %lu, %u, %u, %llu\n",
+			fprintf(f, "%lu, %" PRId64 ", %u, %u, %llu\n",
 					(unsigned long) s->time,
-					(unsigned long) s->val,
+					s->data.val,
 					io_sample_ddir(s), s->bs,
 					(unsigned long long) so->offset);
 		}
@@ -1034,11 +1086,11 @@ void flush_log(struct io_log *log, bool do_append)
 		cur_log = flist_first_entry(&log->io_logs, struct io_logs, list);
 		flist_del_init(&cur_log->list);
 		
-		if (log == log->td->clat_hist_log)
+		if (log->td && log == log->td->clat_hist_log)
 			flush_hist_samples(f, log->hist_coarseness, cur_log->log,
-			                   cur_log->nr_samples * log_entry_sz(log));
+			                   log_sample_sz(log, cur_log));
 		else
-			flush_samples(f, cur_log->log, cur_log->nr_samples * log_entry_sz(log));
+			flush_samples(f, cur_log->log, log_sample_sz(log, cur_log));
 		
 		sfree(cur_log);
 	}
@@ -1090,6 +1142,38 @@ size_t log_chunk_sizes(struct io_log *log)
 
 #ifdef CONFIG_ZLIB
 
+static void iolog_put_deferred(struct io_log *log, void *ptr)
+{
+	if (!ptr)
+		return;
+
+	pthread_mutex_lock(&log->deferred_free_lock);
+	if (log->deferred < IOLOG_MAX_DEFER) {
+		log->deferred_items[log->deferred] = ptr;
+		log->deferred++;
+	} else if (!fio_did_warn(FIO_WARN_IOLOG_DROP))
+		log_err("fio: had to drop log entry free\n");
+	pthread_mutex_unlock(&log->deferred_free_lock);
+}
+
+static void iolog_free_deferred(struct io_log *log)
+{
+	int i;
+
+	if (!log->deferred)
+		return;
+
+	pthread_mutex_lock(&log->deferred_free_lock);
+
+	for (i = 0; i < log->deferred; i++) {
+		free(log->deferred_items[i]);
+		log->deferred_items[i] = NULL;
+	}
+
+	log->deferred = 0;
+	pthread_mutex_unlock(&log->deferred_free_lock);
+}
+
 static int gz_work(struct iolog_flush_data *data)
 {
 	struct iolog_compress *c = NULL;
@@ -1122,7 +1206,8 @@ static int gz_work(struct iolog_flush_data *data)
 				data->log->filename);
 	do {
 		if (c)
-			dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, c->len);
+			dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq,
+				(unsigned long) c->len);
 		c = get_new_chunk(seq);
 		stream.avail_out = GZ_CHUNK;
 		stream.next_out = c->buf;
@@ -1159,7 +1244,7 @@ static int gz_work(struct iolog_flush_data *data)
 	total -= c->len;
 	c->len = GZ_CHUNK - stream.avail_out;
 	total += c->len;
-	dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, c->len);
+	dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, (unsigned long) c->len);
 
 	if (ret != Z_STREAM_END) {
 		do {
@@ -1170,7 +1255,8 @@ static int gz_work(struct iolog_flush_data *data)
 			c->len = GZ_CHUNK - stream.avail_out;
 			total += c->len;
 			flist_add_tail(&c->list, &list);
-			dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, c->len);
+			dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq,
+				(unsigned long) c->len);
 		} while (ret != Z_STREAM_END);
 	}
 
@@ -1180,7 +1266,7 @@ static int gz_work(struct iolog_flush_data *data)
 	if (ret != Z_OK)
 		log_err("fio: deflateEnd %d\n", ret);
 
-	free(data->samples);
+	iolog_put_deferred(data->log, data->samples);
 
 	if (!flist_empty(&list)) {
 		pthread_mutex_lock(&data->log->chunk_lock);
@@ -1191,7 +1277,7 @@ static int gz_work(struct iolog_flush_data *data)
 	ret = 0;
 done:
 	if (data->free)
-		free(data);
+		sfree(data);
 	return ret;
 err:
 	while (!flist_empty(&list)) {
@@ -1292,7 +1378,7 @@ int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log)
 {
 	struct iolog_flush_data *data;
 
-	data = malloc(sizeof(*data));
+	data = smalloc(sizeof(*data));
 	if (!data)
 		return 1;
 
@@ -1306,6 +1392,9 @@ int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log)
 	cur_log->log = NULL;
 
 	workqueue_enqueue(&log->td->log_compress_wq, &data->work);
+
+	iolog_free_deferred(log);
+
 	return 0;
 }
 #else