X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;ds=sidebyside;f=io_u.c;h=2e9dac0db6b4aa0adae673a3b21d26173e48da3e;hb=8f933caebc879fa06bb8d2d25802dd6f27da8071;hp=40fd1968eb08b4a9e8d1bcd6108ab7b6f780d1fe;hpb=0a4957c60fb35db0973c15c4c2adbd3bac854ccd;p=fio.git

diff --git a/io_u.c b/io_u.c
index 40fd1968..2e9dac0d 100644
--- a/io_u.c
+++ b/io_u.c
@@ -231,11 +231,6 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u)
 	return 0;
 }
 
-static inline int is_power_of_2(unsigned int val)
-{
-	return (val != 0 && ((val & (val - 1)) == 0));
-}
-
 static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
 {
 	const int ddir = io_u->ddir;
@@ -307,6 +302,53 @@ static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
 	return DDIR_WRITE;
 }
 
+static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
+{
+	enum fio_ddir odir = ddir ^ 1;
+	struct timeval t;
+	long usec;
+
+	if (td->rate_pending_usleep[ddir] <= 0)
+		return ddir;
+
+	/*
+	 * We have too much pending sleep in this direction. See if we
+	 * should switch.
+	 */
+	if (td_rw(td)) {
+		/*
+		 * Other direction does not have too much pending, switch
+		 */
+		if (td->rate_pending_usleep[odir] < 100000)
+			return odir;
+
+		/*
+		 * Both directions have pending sleep. Sleep the minimum time
+		 * and deduct from both.
+		 */
+		if (td->rate_pending_usleep[ddir] <=
+			td->rate_pending_usleep[odir]) {
+			usec = td->rate_pending_usleep[ddir];
+		} else {
+			usec = td->rate_pending_usleep[odir];
+			ddir = odir;
+		}
+	} else
+		usec = td->rate_pending_usleep[ddir];
+
+	fio_gettime(&t, NULL);
+	usec_sleep(td, usec);
+	usec = utime_since_now(&t);
+
+	td->rate_pending_usleep[ddir] -= usec;
+
+	odir = ddir ^ 1;
+	if (td_rw(td) && __should_check_rate(td, odir))
+		td->rate_pending_usleep[odir] -= usec;
+	
+	return ddir;
+}
+
 /*
  * Return the data direction for the next io_u. If the job is a
  * mixed read/write workload, check the rwmix cycle and switch if
@@ -314,13 +356,29 @@ static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
  */
 static enum fio_ddir get_rw_ddir(struct thread_data *td)
 {
+	enum fio_ddir ddir;
+
+	/*
+	 * see if it's time to fsync
+	 */
+	if (td->o.fsync_blocks &&
+	   !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) &&
+	     td->io_issues[DDIR_WRITE] && should_fsync(td))
+		return DDIR_SYNC;
+
+	/*
+	 * see if it's time to fdatasync
+	 */
+	if (td->o.fdatasync_blocks &&
+	   !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) &&
+	     td->io_issues[DDIR_WRITE] && should_fsync(td))
+		return DDIR_DATASYNC;
+
 	if (td_rw(td)) {
 		/*
 		 * Check if it's time to seed a new data direction.
 		 */
 		if (td->io_issues[td->rwmix_ddir] >= td->rwmix_issues) {
-			enum fio_ddir ddir;
-
 			/*
 			 * Put a top limit on how many bytes we do for
 			 * one data direction, to avoid overflowing the
@@ -333,11 +391,14 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 
 			td->rwmix_ddir = ddir;
 		}
-		return td->rwmix_ddir;
+		ddir = td->rwmix_ddir;
 	} else if (td_read(td))
-		return DDIR_READ;
+		ddir = DDIR_READ;
 	else
-		return DDIR_WRITE;
+		ddir = DDIR_WRITE;
+
+	td->rwmix_ddir = rate_ddir(td, ddir);
+	return td->rwmix_ddir;
 }
 
 static void put_file_log(struct thread_data *td, struct fio_file *f)
@@ -362,6 +423,12 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 	td->cur_depth--;
 }
 
+void clear_io_u(struct thread_data *td, struct io_u *io_u)
+{
+	io_u->flags &= ~IO_U_F_FLIGHT;
+	put_io_u(td, io_u);
+}
+
 void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 {
 	struct io_u *__io_u = *io_u;
@@ -369,7 +436,7 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 	dprint(FD_IO, "requeue %p\n", __io_u);
 
 	__io_u->flags |= IO_U_F_FREE;
-	if ((__io_u->flags & IO_U_F_FLIGHT) && (__io_u->ddir != DDIR_SYNC))
+	if ((__io_u->flags & IO_U_F_FLIGHT) && !ddir_sync(__io_u->ddir))
 		td->io_issues[__io_u->ddir]--;
 
 	__io_u->flags &= ~IO_U_F_FLIGHT;
@@ -385,17 +452,13 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 	if (td->io_ops->flags & FIO_NOIO)
 		goto out;
 
+	io_u->ddir = get_rw_ddir(td);
+
 	/*
-	 * see if it's time to sync
+	 * fsync() or fdatasync(), we are done
 	 */
-	if (td->o.fsync_blocks &&
-	   !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td)) {
-		io_u->ddir = DDIR_SYNC;
+	if (ddir_sync(io_u->ddir))
 		goto out;
-	}
-
-	io_u->ddir = get_rw_ddir(td);
 
 	/*
 	 * See if it's time to switch to a new zone
@@ -611,7 +674,8 @@ static void io_u_mark_latency(struct thread_data *td, unsigned long usec)
 /*
  * Get next file to service by choosing one at random
  */
-static struct fio_file *get_next_file_rand(struct thread_data *td, enum fio_file_flags goodf,
+static struct fio_file *get_next_file_rand(struct thread_data *td,
+					   enum fio_file_flags goodf,
 					   enum fio_file_flags badf)
 {
 	struct fio_file *f;
@@ -821,7 +885,7 @@ struct io_u *get_io_u(struct thread_data *td)
 	f = io_u->file;
 	assert(fio_file_open(f));
 
-	if (io_u->ddir != DDIR_SYNC) {
+	if (!ddir_sync(io_u->ddir)) {
 		if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
 			dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
 			goto err_put;
@@ -885,7 +949,7 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 	assert(io_u->flags & IO_U_F_FLIGHT);
 	io_u->flags &= ~IO_U_F_FLIGHT;
 
-	if (io_u->ddir == DDIR_SYNC) {
+	if (ddir_sync(io_u->ddir)) {
 		td->last_was_sync = 1;
 		return;
 	}
@@ -902,16 +966,29 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 		td->this_io_bytes[idx] += bytes;
 
 		if (ramp_time_over(td)) {
+			unsigned long uninitialized_var(lusec);
+			unsigned long uninitialized_var(rusec);
+
 			if (!td->o.disable_clat || !td->o.disable_bw)
-				usec = utime_since(&io_u->issue_time,
+				lusec = utime_since(&io_u->issue_time,
+							&icd->time);
+			if (__should_check_rate(td, idx) ||
+			    __should_check_rate(td, idx ^ 1))
+				rusec = utime_since(&io_u->start_time,
 							&icd->time);
 
 			if (!td->o.disable_clat) {
 				add_clat_sample(td, idx, usec, bytes);
-				io_u_mark_latency(td, usec);
+				io_u_mark_latency(td, lusec);
 			}
 			if (!td->o.disable_bw)
 				add_bw_sample(td, idx, bytes, &icd->time);
+			if (__should_check_rate(td, idx)) {
+				td->rate_pending_usleep[idx] +=
+					(long) td->rate_usec_cycle[idx] - rusec;
+			}
+			if (__should_check_rate(td, idx ^ 1))
+				td->rate_pending_usleep[idx ^ 1] -= rusec;
 		}
 
 		if (td_write(td) && idx == DDIR_WRITE &&
@@ -930,6 +1007,17 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 		icd->error = io_u->error;
 		io_u_log_error(td, io_u);
 	}
+	if (td->o.continue_on_error && icd->error &&
+	    td_non_fatal_error(icd->error)) {
+		/*
+		 * If there is a non_fatal error, then add to the error count
+		 * and clear all the errors.
+		 */
+		update_error_count(td, icd->error);
+		td_clear_error(td);
+		icd->error = 0;
+		io_u->error = 0;
+	}
 }
 
 static void init_icd(struct thread_data *td, struct io_completion_data *icd,
@@ -961,7 +1049,8 @@ static void ios_completed(struct thread_data *td,
 /*
  * Complete a single io_u for the sync engines.
  */
-long io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
+int io_u_sync_complete(struct thread_data *td, struct io_u *io_u,
+		       unsigned long *bytes)
 {
 	struct io_completion_data icd;
 
@@ -969,17 +1058,24 @@ long io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
 	io_completed(td, io_u, &icd);
 	put_io_u(td, io_u);
 
-	if (!icd.error)
-		return icd.bytes_done[0] + icd.bytes_done[1];
+	if (icd.error) {
+		td_verror(td, icd.error, "io_u_sync_complete");
+		return -1;
+	}
 
-	td_verror(td, icd.error, "io_u_sync_complete");
-	return -1;
+	if (bytes) {
+		bytes[0] += icd.bytes_done[0];
+		bytes[1] += icd.bytes_done[1];
+	}
+
+	return 0;
 }
 
 /*
  * Called to complete min_events number of io for the async engines.
  */
-long io_u_queued_complete(struct thread_data *td, int min_evts)
+int io_u_queued_complete(struct thread_data *td, int min_evts,
+			 unsigned long *bytes)
 {
 	struct io_completion_data icd;
 	struct timespec *tvp = NULL;
@@ -1000,11 +1096,17 @@ long io_u_queued_complete(struct thread_data *td, int min_evts)
 
 	init_icd(td, &icd, ret);
 	ios_completed(td, &icd);
-	if (!icd.error)
-		return icd.bytes_done[0] + icd.bytes_done[1];
+	if (icd.error) {
+		td_verror(td, icd.error, "io_u_queued_complete");
+		return -1;
+	}
 
-	td_verror(td, icd.error, "io_u_queued_complete");
-	return -1;
+	if (bytes) {
+		bytes[0] += icd.bytes_done[0];
+		bytes[1] += icd.bytes_done[1];
+	}
+
+	return 0;
 }
 
 /*
@@ -1016,7 +1118,7 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
 		unsigned long slat_time;
 
 		slat_time = utime_since(&io_u->start_time, &io_u->issue_time);
-		add_slat_sample(td, io_u->ddir, io_u->xfer_buflen, slat_time);
+		add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen);
 	}
 }