All of lore.kernel.org
 help / color / mirror / Atom feed
From: Alexey Dobriyan <adobriyan@gmail.com>
To: axboe@kernel.dk
Cc: fio@vger.kernel.org, damien.lemoal@wdc.com,
	Alexey Dobriyan <adobriyan@gmail.com>
Subject: [PATCH 3/9] zbd: don't lock zones outside working area
Date: Tue,  5 May 2020 20:56:29 +0300	[thread overview]
Message-ID: <20200505175634.2517-3-adobriyan@gmail.com> (raw)
In-Reply-To: <20200505175634.2517-1-adobriyan@gmail.com>

Currently threads lock each other zones even if their working areas as
defined by [f->file_offset, f->file_offset + f->io_size) don't intersect.
This leads to unnecessary quiescing.

Patch clamps every zone to [->min_zone, ->max_zone) when doing search
for opened zone and more importantly adds an assert so that any
unnecessary zone locking becomes very visible.

Signed-off-by: Alexey Dobriyan (SK hynix) <adobriyan@gmail.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 file.h |  3 +++
 zbd.c  | 40 ++++++++++++++++++++++++++--------------
 2 files changed, 29 insertions(+), 14 deletions(-)

diff --git a/file.h b/file.h
index ae0e6fc8..375bbfd3 100644
--- a/file.h
+++ b/file.h
@@ -104,6 +104,9 @@ struct fio_file {
 	 * Zoned block device information. See also zonemode=zbd.
 	 */
 	struct zoned_block_device_info *zbd_info;
+	/* zonemode=zbd working area */
+	uint32_t min_zone;	/* inclusive */
+	uint32_t max_zone;	/* exclusive */
 
 	/*
 	 * Track last end and last start of IO for a given data direction
diff --git a/zbd.c b/zbd.c
index 8dc3c397..76771f2e 100644
--- a/zbd.c
+++ b/zbd.c
@@ -156,8 +156,14 @@ static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
 		z->wp + required > z->start + f->zbd_info->zone_size;
 }
 
-static void zone_lock(struct thread_data *td, struct fio_zone_info *z)
+static void zone_lock(struct thread_data *td, struct fio_file *f, struct fio_zone_info *z)
 {
+	struct zoned_block_device_info *zbd = f->zbd_info;
+	uint32_t nz = z - zbd->zone_info;
+
+	/* A thread should never lock zones outside its working area. */
+	assert(f->min_zone <= nz && nz < f->max_zone);
+
 	/*
 	 * Lock the io_u target zone. The zone will be unlocked if io_u offset
 	 * is changed or when io_u completes and zbd_put_io() executed.
@@ -289,6 +295,9 @@ static bool zbd_verify_sizes(void)
 					 (unsigned long long) new_end - f->file_offset);
 				f->io_size = new_end - f->file_offset;
 			}
+
+			f->min_zone = zbd_zone_idx(f, f->file_offset);
+			f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size);
 		}
 	}
 
@@ -730,7 +739,7 @@ static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
 
 		if (!zbd_zone_swr(z))
 			continue;
-		zone_lock(td, z);
+		zone_lock(td, f, z);
 		if (all_zones) {
 			unsigned int i;
 
@@ -854,14 +863,12 @@ static void zbd_init_swd(struct fio_file *f)
 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
 {
 	struct fio_zone_info *zb, *ze;
-	uint32_t zone_idx_e;
 
 	if (!f->zbd_info)
 		return;
 
-	zb = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)];
-	zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size);
-	ze = &f->zbd_info->zone_info[zone_idx_e];
+	zb = &f->zbd_info->zone_info[f->min_zone];
+	ze = &f->zbd_info->zone_info[f->max_zone];
 	zbd_init_swd(f);
 	/*
 	 * If data verification is enabled reset the affected zones before
@@ -953,7 +960,7 @@ static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
 						      struct io_u *io_u)
 {
 	const uint32_t min_bs = td->o.min_bs[io_u->ddir];
-	const struct fio_file *f = io_u->file;
+	struct fio_file *f = io_u->file;
 	struct fio_zone_info *z;
 	unsigned int open_zone_idx = -1;
 	uint32_t zone_idx, new_zone_idx;
@@ -970,6 +977,10 @@ static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
 	} else {
 		zone_idx = zbd_zone_idx(f, io_u->offset);
 	}
+	if (zone_idx < f->min_zone)
+		zone_idx = f->min_zone;
+	else if (zone_idx >= f->max_zone)
+		zone_idx = f->max_zone - 1;
 	dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
 	       __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
 
@@ -984,7 +995,7 @@ static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
 
 		z = &f->zbd_info->zone_info[zone_idx];
 
-		zone_lock(td, z);
+		zone_lock(td, f, z);
 		pthread_mutex_lock(&f->zbd_info->mutex);
 		if (td->o.max_open_zones == 0)
 			goto examine_zone;
@@ -1010,8 +1021,7 @@ static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
 			if (tmp_idx >= f->zbd_info->num_open_zones)
 				tmp_idx = 0;
 			tmpz = f->zbd_info->open_zones[tmp_idx];
-
-			if (is_valid_offset(f, f->zbd_info->zone_info[tmpz].start)) {
+			if (f->min_zone <= tmpz && tmpz < f->max_zone) {
 				open_zone_idx = tmp_idx;
 				goto found_candidate_zone;
 			}
@@ -1056,11 +1066,11 @@ examine_zone:
 		z++;
 		if (!is_valid_offset(f, z->start)) {
 			/* Wrap-around. */
-			zone_idx = zbd_zone_idx(f, f->file_offset);
+			zone_idx = f->min_zone;
 			z = &f->zbd_info->zone_info[zone_idx];
 		}
 		assert(is_valid_offset(f, z->start));
-		zone_lock(td, z);
+		zone_lock(td, f, z);
 		if (z->open)
 			continue;
 		if (zbd_open_zone(td, io_u, zone_idx))
@@ -1073,12 +1083,14 @@ examine_zone:
 	pthread_mutex_lock(&f->zbd_info->mutex);
 	for (i = 0; i < f->zbd_info->num_open_zones; i++) {
 		zone_idx = f->zbd_info->open_zones[i];
+		if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
+			continue;
 		pthread_mutex_unlock(&f->zbd_info->mutex);
 		pthread_mutex_unlock(&z->mutex);
 
 		z = &f->zbd_info->zone_info[zone_idx];
 
-		zone_lock(td, z);
+		zone_lock(td, f, z);
 		if (z->wp + min_bs <= (z+1)->start)
 			goto out;
 		pthread_mutex_lock(&f->zbd_info->mutex);
@@ -1404,7 +1416,7 @@ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
 
 	zbd_check_swd(f);
 
-	zone_lock(td, zb);
+	zone_lock(td, f, zb);
 
 	switch (io_u->ddir) {
 	case DDIR_READ:
-- 
2.26.2



  parent reply	other threads:[~2020-05-05 17:56 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-05-05 17:56 [PATCH 1/9] verify: decouple seed generation from buffer fill Alexey Dobriyan
2020-05-05 17:56 ` [PATCH 2/9] zbd: bump ZBD_MAX_OPEN_ZONES Alexey Dobriyan
2020-05-05 17:56 ` Alexey Dobriyan [this message]
2020-05-05 17:56 ` [PATCH 4/9] zbd: introduce per job maximum open zones limit Alexey Dobriyan
2020-05-11  1:09   ` Damien Le Moal
2020-05-05 17:56 ` [PATCH 5/9] zbd: make zbd_info->mutex non-recursive Alexey Dobriyan
2020-05-11  1:11   ` Damien Le Moal
2020-05-05 17:56 ` [PATCH 6/9] zbd: consolidate zone mutex initialisation Alexey Dobriyan
2020-05-11  1:16   ` Damien Le Moal
2020-05-21 21:44     ` Alexey Dobriyan
2020-05-21 22:02       ` Damien Le Moal
2020-05-21 22:27         ` Alexey Dobriyan
2020-05-05 17:56 ` [PATCH 7/9] fio: parse "io_size=1%" Alexey Dobriyan
2020-05-11  1:20   ` Damien Le Moal
2020-05-21 22:13     ` Alexey Dobriyan
2020-05-05 17:56 ` [PATCH 8/9] zbd: support verification Alexey Dobriyan
2020-05-05 17:59   ` Alexey Dobriyan
2020-05-11  2:01     ` Damien Le Moal
2020-05-11  1:59   ` Damien Le Moal
2020-05-11 16:00     ` Alexey Dobriyan
2020-05-11  0:56 ` [PATCH 1/9] verify: decouple seed generation from buffer fill Damien Le Moal

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200505175634.2517-3-adobriyan@gmail.com \
    --to=adobriyan@gmail.com \
    --cc=axboe@kernel.dk \
    --cc=damien.lemoal@wdc.com \
    --cc=fio@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.