X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=iolog.c;h=e674171e5195f71b74b6c03514a33d597ad4312e;hp=cfcab2712d1026e20a71c27dcc2428a1a76923f2;hb=949ae6dce35c81348cfb15f77c9d0a6785fe4053;hpb=e75cc8f3a296fe5910eee9b7f25cc8cb524b4518

diff --git a/iolog.c b/iolog.c
index cfcab271..e674171e 100644
--- a/iolog.c
+++ b/iolog.c
@@ -18,7 +18,6 @@
 #include "verify.h"
 #include "trim.h"
 #include "filelock.h"
-#include "lib/tp.h"
 
 static const char iolog_ver2[] = "fio version 2 iolog";
 
@@ -28,7 +27,7 @@ void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
 	td->total_io_size += ipo->len;
 }
 
-void log_io_u(struct thread_data *td, struct io_u *io_u)
+void log_io_u(const struct thread_data *td, const struct io_u *io_u)
 {
 	if (!td->o.write_iolog_file)
 		return;
@@ -60,14 +59,22 @@ void log_file(struct thread_data *td, struct fio_file *f,
 
 static void iolog_delay(struct thread_data *td, unsigned long delay)
 {
-	unsigned long usec = utime_since_now(&td->last_issue);
-	unsigned long this_delay;
+	uint64_t usec = utime_since_now(&td->last_issue);
+	uint64_t this_delay;
+	struct timeval tv;
 
+	if (delay < td->time_offset) {
+		td->time_offset = 0;
+		return;
+	}
+
+	delay -= td->time_offset;
 	if (delay < usec)
 		return;
 
 	delay -= usec;
 
+	fio_gettime(&tv, NULL);
 	while (delay && !td->terminate) {
 		this_delay = delay;
 		if (this_delay > 500000)
@@ -76,6 +83,12 @@ static void iolog_delay(struct thread_data *td, unsigned long delay)
 		usec_sleep(td, this_delay);
 		delay -= this_delay;
 	}
+
+	usec = utime_since_now(&tv);
+	if (usec > delay)
+		td->time_offset = usec - delay;
+	else
+		td->time_offset = 0;
 }
 
 static int ipo_special(struct thread_data *td, struct io_piece *ipo)
@@ -173,7 +186,7 @@ void prune_io_piece_log(struct thread_data *td)
 	}
 
 	while (!flist_empty(&td->io_hist_list)) {
-		ipo = flist_entry(&td->io_hist_list, struct io_piece, list);
+		ipo = flist_first_entry(&td->io_hist_list, struct io_piece, list);
 		flist_del(&ipo->list);
 		remove_trim_entry(td, ipo);
 		td->io_hist_len--;
@@ -236,6 +249,7 @@ restart:
 	p = &td->io_hist_tree.rb_node;
 	parent = NULL;
 	while (*p) {
+		int overlap = 0;
 		parent = *p;
 
 		__ipo = rb_entry(parent, struct io_piece, rb_node);
@@ -243,11 +257,18 @@ restart:
 			p = &(*p)->rb_left;
 		else if (ipo->file > __ipo->file)
 			p = &(*p)->rb_right;
-		else if (ipo->offset < __ipo->offset)
+		else if (ipo->offset < __ipo->offset) {
 			p = &(*p)->rb_left;
-		else if (ipo->offset > __ipo->offset)
+			overlap = ipo->offset + ipo->len > __ipo->offset;
+		}
+		else if (ipo->offset > __ipo->offset) {
 			p = &(*p)->rb_right;
-		else {
+			overlap = __ipo->offset + __ipo->len > ipo->offset;
+		}
+		else
+			overlap = 1;
+
+		if (overlap) {
 			dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu",
 				__ipo->offset, __ipo->len,
 				ipo->offset, ipo->len);
@@ -269,6 +290,18 @@ void unlog_io_piece(struct thread_data *td, struct io_u *io_u)
 {
 	struct io_piece *ipo = io_u->ipo;
 
+	if (td->ts.nr_block_infos) {
+		uint32_t *info = io_u_block_info(td, io_u);
+		if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) {
+			if (io_u->ddir == DDIR_TRIM)
+				*info = BLOCK_INFO_SET_STATE(*info,
+						BLOCK_STATE_TRIM_FAILURE);
+			else if (io_u->ddir == DDIR_WRITE)
+				*info = BLOCK_INFO_SET_STATE(*info,
+						BLOCK_STATE_WRITE_FAILURE);
+		}
+	}
+
 	if (!ipo)
 		return;
 
@@ -282,7 +315,7 @@ void unlog_io_piece(struct thread_data *td, struct io_u *io_u)
 	td->io_hist_len--;
 }
 
-void trim_io_piece(struct thread_data *td, struct io_u *io_u)
+void trim_io_piece(struct thread_data *td, const struct io_u *io_u)
 {
 	struct io_piece *ipo = io_u->ipo;
 
@@ -560,7 +593,7 @@ void setup_log(struct io_log **log, struct log_params *p,
 
 	if (l->log_gz && !p->td)
 		l->log_gz = 0;
-	else if (l->log_gz) {
+	else if (l->log_gz || l->log_gz_store) {
 		pthread_mutex_init(&l->chunk_lock, NULL);
 		p->td->flags |= TD_F_COMPRESS_LOG;
 	}
@@ -638,7 +671,12 @@ static void flush_samples(FILE *f, void *samples, uint64_t sample_size)
 #ifdef CONFIG_ZLIB
 
 struct iolog_flush_data {
-	struct tp_work work;
+	struct workqueue_work work;
+	pthread_mutex_t lock;
+	pthread_cond_t cv;
+	int wait;
+	volatile int done;
+	volatile int refs;
 	struct io_log *log;
 	void *samples;
 	uint64_t nr_samples;
@@ -854,10 +892,12 @@ int iolog_file_inflate(const char *file)
 	if (ret < 0) {
 		perror("fread");
 		fclose(f);
+		free(buf);
 		return 1;
 	} else if (ret != 1) {
 		log_err("fio: short read on reading log\n");
 		fclose(f);
+		free(buf);
 		return 1;
 	}
 
@@ -871,18 +911,18 @@ int iolog_file_inflate(const char *file)
 	 */
 	total = ic.len;
 	do {
-		size_t ret;
+		size_t iret;
 
-		ret = inflate_chunk(&ic,  1, stdout, &stream, &iter);
-		total -= ret;
+		iret = inflate_chunk(&ic,  1, stdout, &stream, &iter);
+		total -= iret;
 		if (!total)
 			break;
 		if (iter.err)
 			break;
 
 		ic.seq++;
-		ic.len -= ret;
-		ic.buf += ret;
+		ic.len -= iret;
+		ic.buf += iret;
 	} while (1);
 
 	if (iter.seq) {
@@ -909,12 +949,15 @@ int iolog_file_inflate(const char *file)
 
 #endif
 
-void flush_log(struct io_log *log)
+void flush_log(struct io_log *log, int do_append)
 {
 	void *buf;
 	FILE *f;
 
-	f = fopen(log->filename, "w");
+	if (!do_append)
+		f = fopen(log->filename, "w");
+	else
+		f = fopen(log->filename, "a");
 	if (!f) {
 		perror("fopen log");
 		return;
@@ -932,7 +975,7 @@ void flush_log(struct io_log *log)
 
 static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 {
-	if (td->tp_data)
+	if (td->flags & TD_F_COMPRESS_LOG)
 		iolog_flush(log, 1);
 
 	if (trylock) {
@@ -944,7 +987,7 @@ static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 	if (td->client_type == FIO_CLIENT_TYPE_GUI)
 		fio_send_iolog(td, log, log->filename);
 	else
-		flush_log(log);
+		flush_log(log, !td->o.per_job_logs);
 
 	fio_unlock_file(log->filename);
 	free_log(log);
@@ -953,12 +996,26 @@ static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 
 #ifdef CONFIG_ZLIB
 
+static void drop_data_unlock(struct iolog_flush_data *data)
+{
+	int refs;
+
+	refs = --data->refs;
+	pthread_mutex_unlock(&data->lock);
+
+	if (!refs) {
+		free(data);
+		pthread_mutex_destroy(&data->lock);
+		pthread_cond_destroy(&data->cv);
+	}
+}
+
 /*
  * Invoked from our compress helper thread, when logging would have exceeded
  * the specified memory limitation. Compresses the previously stored
  * entries.
  */
-static int gz_work(struct tp_work *work)
+static int gz_work(struct submit_worker *sw, struct workqueue_work *work)
 {
 	struct iolog_flush_data *data;
 	struct iolog_compress *c;
@@ -1039,12 +1096,14 @@ static int gz_work(struct tp_work *work)
 
 	ret = 0;
 done:
-	if (work->wait) {
-		work->done = 1;
-		pthread_cond_signal(&work->cv);
+	if (data->wait) {
+		pthread_mutex_lock(&data->lock);
+		data->done = 1;
+		pthread_cond_signal(&data->cv);
+
+		drop_data_unlock(data);
 	} else
 		free(data);
-
 	return ret;
 err:
 	while (!flist_empty(&list)) {
@@ -1056,17 +1115,54 @@ err:
 	goto done;
 }
 
+static int gz_init_worker(struct submit_worker *sw)
+{
+	struct thread_data *td = sw->wq->td;
+
+	if (!fio_option_is_set(&td->o, log_gz_cpumask))
+		return 0;
+
+	if (fio_setaffinity(gettid(), td->o.log_gz_cpumask) == -1) {
+		log_err("gz: failed to set CPU affinity\n");
+		return 1;
+	}
+
+	return 0;
+}
+
+static struct workqueue_ops log_compress_wq_ops = {
+	.fn		= gz_work,
+	.init_worker_fn	= gz_init_worker,
+	.nice	= 1,
+};
+
+int iolog_compress_init(struct thread_data *td)
+{
+	if (!(td->flags & TD_F_COMPRESS_LOG))
+		return 0;
+
+	workqueue_init(td, &td->log_compress_wq, &log_compress_wq_ops, 1);
+	return 0;
+}
+
+void iolog_compress_exit(struct thread_data *td)
+{
+	if (!(td->flags & TD_F_COMPRESS_LOG))
+		return;
+
+	workqueue_exit(&td->log_compress_wq);
+}
+
 /*
- * Queue work item to compress the existing log entries. We copy the
- * samples, and reset the log sample count to 0 (so the logging will
- * continue to use the memory associated with the log). If called with
- * wait == 1, will not return until the log compression has completed.
+ * Queue work item to compress the existing log entries. We reset the
+ * current log to a small size, and reference the existing log in the
+ * data that we queue for compression. Once compression has been done,
+ * this old log is freed. If called with wait == 1, will not return until
+ * the log compression has completed.
  */
 int iolog_flush(struct io_log *log, int wait)
 {
-	struct tp_data *tdat = log->td->tp_data;
 	struct iolog_flush_data *data;
-	size_t sample_size;
 
 	data = malloc(sizeof(*data));
 	if (!data)
@@ -1074,34 +1170,29 @@ int iolog_flush(struct io_log *log, int wait)
 
 	data->log = log;
 
-	sample_size = log->nr_samples * log_entry_sz(log);
-	data->samples = malloc(sample_size);
-	if (!data->samples) {
-		free(data);
-		return 1;
-	}
-
-	memcpy(data->samples, log->log, sample_size);
+	data->samples = log->log;
 	data->nr_samples = log->nr_samples;
-	data->work.fn = gz_work;
-	log->nr_samples = 0;
 
-	if (wait) {
-		pthread_mutex_init(&data->work.lock, NULL);
-		pthread_cond_init(&data->work.cv, NULL);
-		data->work.wait = 1;
-	} else
-		data->work.wait = 0;
+	log->nr_samples = 0;
+	log->max_samples = 128;
+	log->log = malloc(log->max_samples * log_entry_sz(log));
+
+	data->wait = wait;
+	if (data->wait) {
+		pthread_mutex_init(&data->lock, NULL);
+		pthread_cond_init(&data->cv, NULL);
+		data->done = 0;
+		data->refs = 2;
+	}
 
-	data->work.prio = 1;
-	tp_queue_work(tdat, &data->work);
+	workqueue_enqueue(&log->td->log_compress_wq, &data->work);
 
 	if (wait) {
-		pthread_mutex_lock(&data->work.lock);
-		while (!data->work.done)
-			pthread_cond_wait(&data->work.cv, &data->work.lock);
-		pthread_mutex_unlock(&data->work.lock);
-		free(data);
+		pthread_mutex_lock(&data->lock);
+		while (!data->done)
+			pthread_cond_wait(&data->cv, &data->lock);
+
+		drop_data_unlock(data);
 	}
 
 	return 0;
@@ -1114,56 +1205,48 @@ int iolog_flush(struct io_log *log, int wait)
 	return 1;
 }
 
+int iolog_compress_init(struct thread_data *td)
+{
+	return 0;
+}
+
+void iolog_compress_exit(struct thread_data *td)
+{
+}
+
 #endif
 
-static int write_iops_log(struct thread_data *td, int try)
+static int __write_log(struct thread_data *td, struct io_log *log, int try)
 {
-	struct io_log *log = td->iops_log;
+	if (log)
+		return finish_log(td, log, try);
 
-	if (!log)
-		return 0;
+	return 0;
+}
 
-	return finish_log(td, log, try);
+static int write_iops_log(struct thread_data *td, int try)
+{
+	return __write_log(td, td->iops_log, try);
 }
 
 static int write_slat_log(struct thread_data *td, int try)
 {
-	struct io_log *log = td->slat_log;
-
-	if (!log)
-		return 0;
-
-	return finish_log(td, log, try);
+	return __write_log(td, td->slat_log, try);
 }
 
 static int write_clat_log(struct thread_data *td, int try)
 {
-	struct io_log *log = td->clat_log;
-
-	if (!log)
-		return 0;
-
-	return finish_log(td, log, try);
+	return __write_log(td, td->clat_log, try);
 }
 
 static int write_lat_log(struct thread_data *td, int try)
 {
-	struct io_log *log = td->lat_log;
-
-	if (!log)
-		return 0;
-
-	return finish_log(td, log, try);
+	return __write_log(td, td->lat_log, try);
 }
 
 static int write_bandw_log(struct thread_data *td, int try)
 {
-	struct io_log *log = td->bw_log;
-
-	if (!log)
-		return 0;
-
-	return finish_log(td, log, try);
+	return __write_log(td, td->bw_log, try);
 }
 
 enum {