[PATCH 2/4] GFS2: Automatically adjust glock min hold time

From: Steven Whitehouse
Date: Fri Jul 22 2011 - 05:40:29 EST


From: Bob Peterson <rpeterso@xxxxxxxxxx>

This patch is a performance improvement for GFS2 in a clustered
environment. It makes the glock hold time self-adjusting.

Signed-off-by: Bob Peterson <rpeterso@xxxxxxxxxx>
Signed-off-by: Steven Whitehouse <swhiteho@xxxxxxxxxx>

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index 1c1336e..88e8a23 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -409,6 +409,10 @@ static void state_change(struct gfs2_glock *gl, unsigned int new_state)
if (held1 && held2 && list_empty(&gl->gl_holders))
clear_bit(GLF_QUEUED, &gl->gl_flags);

+ if (new_state != gl->gl_target)
+ /* shorten our minimum hold time */
+ gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
+ GL_GLOCK_MIN_HOLD);
gl->gl_state = new_state;
gl->gl_tchange = jiffies;
}
@@ -668,7 +672,7 @@ static void glock_work_func(struct work_struct *work)
gl->gl_demote_state != LM_ST_EXCLUSIVE) {
unsigned long holdtime, now = jiffies;

- holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
+ holdtime = gl->gl_tchange + gl->gl_hold_time;
if (time_before(now, holdtime))
delay = holdtime - now;

@@ -679,9 +683,14 @@ static void glock_work_func(struct work_struct *work)
}
run_queue(gl, 0);
spin_unlock(&gl->gl_spin);
- if (!delay ||
- queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
+ if (!delay)
gfs2_glock_put(gl);
+ else {
+ if (gl->gl_name.ln_type != LM_TYPE_INODE)
+ delay = 0;
+ if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
+ gfs2_glock_put(gl);
+ }
if (drop_ref)
gfs2_glock_put(gl);
}
@@ -743,6 +752,7 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
gl->gl_tchange = jiffies;
gl->gl_object = NULL;
gl->gl_sbd = sdp;
+ gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
INIT_WORK(&gl->gl_delete, delete_work_func);

@@ -855,8 +865,15 @@ static int gfs2_glock_demote_wait(void *word)

static void wait_on_holder(struct gfs2_holder *gh)
{
+ unsigned long time1 = jiffies;
+
might_sleep();
wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
+ if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
+ /* Lengthen the minimum hold time. */
+ gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
+ GL_GLOCK_HOLD_INCR,
+ GL_GLOCK_MAX_HOLD);
}

static void wait_on_demote(struct gfs2_glock *gl)
@@ -1093,8 +1110,9 @@ void gfs2_glock_dq(struct gfs2_holder *gh)

gfs2_glock_hold(gl);
if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
- !test_bit(GLF_DEMOTE, &gl->gl_flags))
- delay = gl->gl_ops->go_min_hold_time;
+ !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
+ gl->gl_name.ln_type == LM_TYPE_INODE)
+ delay = gl->gl_hold_time;
if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
gfs2_glock_put(gl);
}
@@ -1273,12 +1291,13 @@ void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
unsigned long now = jiffies;

gfs2_glock_hold(gl);
- holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
- if (test_bit(GLF_QUEUED, &gl->gl_flags)) {
+ holdtime = gl->gl_tchange + gl->gl_hold_time;
+ if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
+ gl->gl_name.ln_type == LM_TYPE_INODE) {
if (time_before(now, holdtime))
delay = holdtime - now;
if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
- delay = gl->gl_ops->go_min_hold_time;
+ delay = gl->gl_hold_time;
}

spin_lock(&gl->gl_spin);
@@ -1667,7 +1686,7 @@ static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
dtime *= 1000000/HZ; /* demote time in uSec */
if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
dtime = 0;
- gfs2_print_dbg(seq, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d\n",
+ gfs2_print_dbg(seq, "G: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
state2str(gl->gl_state),
gl->gl_name.ln_type,
(unsigned long long)gl->gl_name.ln_number,
@@ -1676,7 +1695,7 @@ static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
state2str(gl->gl_demote_state), dtime,
atomic_read(&gl->gl_ail_count),
atomic_read(&gl->gl_revokes),
- atomic_read(&gl->gl_ref));
+ atomic_read(&gl->gl_ref), gl->gl_hold_time);

list_for_each_entry(gh, &gl->gl_holders, gh_list) {
error = dump_holder(seq, gh);
diff --git a/fs/gfs2/glock.h b/fs/gfs2/glock.h
index 6b2f757..6670711 100644
--- a/fs/gfs2/glock.h
+++ b/fs/gfs2/glock.h
@@ -113,6 +113,12 @@ enum {

#define GLR_TRYFAILED 13

+#define GL_GLOCK_MAX_HOLD (long)(HZ / 5)
+#define GL_GLOCK_DFT_HOLD (long)(HZ / 5)
+#define GL_GLOCK_MIN_HOLD (long)(10)
+#define GL_GLOCK_HOLD_INCR (long)(HZ / 20)
+#define GL_GLOCK_HOLD_DECR (long)(HZ / 40)
+
struct lm_lockops {
const char *lm_proto_name;
int (*lm_mount) (struct gfs2_sbd *sdp, const char *fsname);
diff --git a/fs/gfs2/glops.c b/fs/gfs2/glops.c
index 95788ae..a4222c3 100644
--- a/fs/gfs2/glops.c
+++ b/fs/gfs2/glops.c
@@ -551,7 +551,6 @@ const struct gfs2_glock_operations gfs2_inode_glops = {
.go_lock = inode_go_lock,
.go_dump = inode_go_dump,
.go_type = LM_TYPE_INODE,
- .go_min_hold_time = HZ / 5,
.go_flags = GLOF_ASPACE,
};

@@ -562,7 +561,6 @@ const struct gfs2_glock_operations gfs2_rgrp_glops = {
.go_unlock = rgrp_go_unlock,
.go_dump = gfs2_rgrp_dump,
.go_type = LM_TYPE_RGRP,
- .go_min_hold_time = HZ / 5,
.go_flags = GLOF_ASPACE,
};

diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index 24cd55f..892ac37 100644
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -163,7 +163,6 @@ struct gfs2_glock_operations {
int (*go_dump)(struct seq_file *seq, const struct gfs2_glock *gl);
void (*go_callback) (struct gfs2_glock *gl);
const int go_type;
- const unsigned long go_min_hold_time;
const unsigned long go_flags;
#define GLOF_ASPACE 1
};
@@ -221,6 +220,7 @@ struct gfs2_glock {

unsigned int gl_hash;
unsigned long gl_demote_time; /* time of first demote request */
+ long gl_hold_time;
struct list_head gl_holders;

const struct gfs2_glock_operations *gl_ops;
--
1.7.4

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/