[PATCH-tip v3 07/11] locking/qspinlock_stat: Introduce a generic lockevent counting APIs

From: Waiman Long
Date: Thu Feb 28 2019 - 14:10:57 EST

The percpu event counts used by qspinlock code can be useful for
other locking code as well. So a new set of lockevent_* counting APIs
is introduced with the lock event names extracted out into the new
lock_events_list.h header file for easier addition in the future.

The existing qstat_inc() calls are replaced by either lockevent_inc() or
lockevent_cond_inc() calls.

The qstat_hop() call is renamed to lockevent_pv_hop(). The "reset_counters"
debugfs file is also renamed to ".reset_counts".

Signed-off-by: Waiman Long <longman@xxxxxxxxxx>
Acked-by: Davidlohr Bueso <dbueso@xxxxxxx>
---
kernel/locking/lock_events.h | 55 ++++++++++++
kernel/locking/lock_events_list.h | 50 +++++++++++
kernel/locking/qspinlock.c | 8 +-
kernel/locking/qspinlock_paravirt.h | 19 +++--
kernel/locking/qspinlock_stat.h | 163 ++++++++++++++----------------------
5 files changed, 181 insertions(+), 114 deletions(-)
create mode 100644 kernel/locking/lock_events.h
create mode 100644 kernel/locking/lock_events_list.h

diff --git a/kernel/locking/lock_events.h b/kernel/locking/lock_events.h
new file mode 100644
index 0000000..4009e07
--- /dev/null
+++ b/kernel/locking/lock_events.h
@@ -0,0 +1,55 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * Authors: Waiman Long <longman@xxxxxxxxxx>
+ */
+
+enum lock_events {
+
+#include "lock_events_list.h"
+
+ lockevent_num, /* Total number of lock event counts */
+ LOCKEVENT_reset_cnts = lockevent_num,
+};
+
+#ifdef CONFIG_QUEUED_LOCK_STAT
+/*
+ * Per-cpu counters
+ */
+DECLARE_PER_CPU(unsigned long, lockevents[lockevent_num]);
+
+/*
+ * Increment the PV qspinlock statistical counters
+ */
+static inline void __lockevent_inc(enum lock_events event, bool cond)
+{
+ if (cond)
+ __this_cpu_inc(lockevents[event]);
+}
+
+#define lockevent_inc(ev) __lockevent_inc(LOCKEVENT_ ##ev, true)
+#define lockevent_cond_inc(ev, c) __lockevent_inc(LOCKEVENT_ ##ev, c)
+
+static inline void __lockevent_add(enum lock_events event, int inc)
+{
+ __this_cpu_add(lockevents[event], inc);
+}
+
+#define lockevent_add(ev, c) __lockevent_add(LOCKEVENT_ ##ev, c)
+
+#else /* CONFIG_QUEUED_LOCK_STAT */
+
+#define lockevent_inc(ev)
+#define lockevent_add(ev, c)
+#define lockevent_cond_inc(ev, c)
+
+#endif /* CONFIG_QUEUED_LOCK_STAT */
diff --git a/kernel/locking/lock_events_list.h b/kernel/locking/lock_events_list.h
new file mode 100644
index 0000000..8b4d2e1
--- /dev/null
+++ b/kernel/locking/lock_events_list.h
@@ -0,0 +1,50 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * Authors: Waiman Long <longman@xxxxxxxxxx>
+ */
+
+#ifndef LOCK_EVENT
+#define LOCK_EVENT(name) LOCKEVENT_ ## name,
+#endif
+
+#ifdef CONFIG_QUEUED_SPINLOCKS
+#ifdef CONFIG_PARAVIRT_SPINLOCKS
+/*
+ * Locking events for PV qspinlock.
+ */
+LOCK_EVENT(pv_hash_hops) /* Average # of hops per hashing operation */
+LOCK_EVENT(pv_kick_unlock) /* # of vCPU kicks issued at unlock time */
+LOCK_EVENT(pv_kick_wake) /* # of vCPU kicks for pv_latency_wake */
+LOCK_EVENT(pv_latency_kick) /* Average latency (ns) of vCPU kick */
+LOCK_EVENT(pv_latency_wake) /* Average latency (ns) of kick-to-wakeup */
+LOCK_EVENT(pv_lock_stealing) /* # of lock stealing operations */
+LOCK_EVENT(pv_spurious_wakeup) /* # of spurious wakeups in non-head vCPUs */
+LOCK_EVENT(pv_wait_again) /* # of wait's after queue head vCPU kick */
+LOCK_EVENT(pv_wait_early) /* # of early vCPU wait's */
+LOCK_EVENT(pv_wait_head) /* # of vCPU wait's at the queue head */
+LOCK_EVENT(pv_wait_node) /* # of vCPU wait's at non-head queue node */
+#endif /* CONFIG_PARAVIRT_SPINLOCKS */
+
+/*
+ * Locking events for qspinlock
+ *
+ * Subtracting lock_use_node[234] from lock_slowpath will give you
+ * lock_use_node1.
+ */
+LOCK_EVENT(lock_pending) /* # of locking ops via pending code */
+LOCK_EVENT(lock_slowpath) /* # of locking ops via MCS lock queue */
+LOCK_EVENT(lock_use_node2) /* # of locking ops that use 2nd percpu node */
+LOCK_EVENT(lock_use_node3) /* # of locking ops that use 3rd percpu node */
+LOCK_EVENT(lock_use_node4) /* # of locking ops that use 4th percpu node */
+LOCK_EVENT(lock_no_node) /* # of locking ops w/o using percpu node */
+#endif /* CONFIG_QUEUED_SPINLOCKS */
diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
index 21ee51b..fe0dc3c 100644
--- a/kernel/locking/qspinlock.c
+++ b/kernel/locking/qspinlock.c
@@ -398,7 +398,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
* 0,1,0 -> 0,0,1
*/
clear_pending_set_locked(lock);
- qstat_inc(qstat_lock_pending, true);
+ lockevent_inc(lock_pending);
return;

/*
@@ -406,7 +406,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
* queuing.
*/
queue:
- qstat_inc(qstat_lock_slowpath, true);
+ lockevent_inc(lock_slowpath);
pv_queue:
node = this_cpu_ptr(&qnodes[0].mcs);
idx = node->count++;
@@ -422,7 +422,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
* simple enough.
*/
if (unlikely(idx >= MAX_NODES)) {
- qstat_inc(qstat_lock_no_node, true);
+ lockevent_inc(lock_no_node);
while (!queued_spin_trylock(lock))
cpu_relax();
goto release;
@@ -433,7 +433,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
/*
* Keep counts of non-zero index values:
*/
- qstat_inc(qstat_lock_use_node2 + idx - 1, idx);
+ lockevent_cond_inc(lock_use_node2 + idx - 1, idx);

/*
* Ensure that we increment the head node->count before initialising
diff --git a/kernel/locking/qspinlock_paravirt.h b/kernel/locking/qspinlock_paravirt.h
index 8f36c27..89bab07 100644
--- a/kernel/locking/qspinlock_paravirt.h
+++ b/kernel/locking/qspinlock_paravirt.h
@@ -89,7 +89,7 @@ static inline bool pv_hybrid_queued_unfair_trylock(struct qspinlock *lock)

if (!(val & _Q_LOCKED_PENDING_MASK) &&
(cmpxchg_acquire(&lock->locked, 0, _Q_LOCKED_VAL) == 0)) {
- qstat_inc(qstat_pv_lock_stealing, true);
+ lockevent_inc(pv_lock_stealing);
return true;
}
if (!(val & _Q_TAIL_MASK) || (val & _Q_PENDING_MASK))
@@ -219,7 +219,7 @@ static struct qspinlock **pv_hash(struct qspinlock *lock, struct pv_node *node)
hopcnt++;
if (!cmpxchg(&he->lock, NULL, lock)) {
WRITE_ONCE(he->node, node);
- qstat_hop(hopcnt);
+ lockevent_pv_hop(hopcnt);
return &he->lock;
}
}
@@ -320,8 +320,8 @@ static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
smp_store_mb(pn->state, vcpu_halted);

if (!READ_ONCE(node->locked)) {
- qstat_inc(qstat_pv_wait_node, true);
- qstat_inc(qstat_pv_wait_early, wait_early);
+ lockevent_inc(pv_wait_node);
+ lockevent_cond_inc(pv_wait_early, wait_early);
pv_wait(&pn->state, vcpu_halted);
}

@@ -339,7 +339,8 @@ static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
* So it is better to spin for a while in the hope that the
* MCS lock will be released soon.
*/
- qstat_inc(qstat_pv_spurious_wakeup, !READ_ONCE(node->locked));
+ lockevent_cond_inc(pv_spurious_wakeup,
+ !READ_ONCE(node->locked));
}

/*
@@ -416,7 +417,7 @@ static void pv_kick_node(struct qspinlock *lock, struct mcs_spinlock *node)
/*
* Tracking # of slowpath locking operations
*/
- qstat_inc(qstat_lock_slowpath, true);
+ lockevent_inc(lock_slowpath);

for (;; waitcnt++) {
/*
@@ -464,8 +465,8 @@ static void pv_kick_node(struct qspinlock *lock, struct mcs_spinlock *node)
}
}
WRITE_ONCE(pn->state, vcpu_hashed);
- qstat_inc(qstat_pv_wait_head, true);
- qstat_inc(qstat_pv_wait_again, waitcnt);
+ lockevent_inc(pv_wait_head);
+ lockevent_cond_inc(pv_wait_again, waitcnt);
pv_wait(&lock->locked, _Q_SLOW_VAL);

/*
@@ -528,7 +529,7 @@ static void pv_kick_node(struct qspinlock *lock, struct mcs_spinlock *node)
* vCPU is harmless other than the additional latency in completing
* the unlock.
*/
- qstat_inc(qstat_pv_kick_unlock, true);
+ lockevent_inc(pv_kick_unlock);
pv_kick(node->cpu);
}

diff --git a/kernel/locking/qspinlock_stat.h b/kernel/locking/qspinlock_stat.h
index d73f853..1db5b37 100644
--- a/kernel/locking/qspinlock_stat.h
+++ b/kernel/locking/qspinlock_stat.h
@@ -38,8 +38,8 @@
* Subtracting lock_use_node[234] from lock_slowpath will give you
* lock_use_node1.
*
- * Writing to the "reset_counters" file will reset all the above counter
- * values.
+ * Writing to the special ".reset_counts" file will reset all the above
+ * counter values.
*
* These statistical counters are implemented as per-cpu variables which are
* summed and computed whenever the corresponding debugfs files are read. This
@@ -48,27 +48,7 @@
*
* There may be slight difference between pv_kick_wake and pv_kick_unlock.
*/
-enum qlock_stats {
- qstat_pv_hash_hops,
- qstat_pv_kick_unlock,
- qstat_pv_kick_wake,
- qstat_pv_latency_kick,
- qstat_pv_latency_wake,
- qstat_pv_lock_stealing,
- qstat_pv_spurious_wakeup,
- qstat_pv_wait_again,
- qstat_pv_wait_early,
- qstat_pv_wait_head,
- qstat_pv_wait_node,
- qstat_lock_pending,
- qstat_lock_slowpath,
- qstat_lock_use_node2,
- qstat_lock_use_node3,
- qstat_lock_use_node4,
- qstat_lock_no_node,
- qstat_num, /* Total number of statistical counters */
- qstat_reset_cnts = qstat_num,
-};
+#include "lock_events.h"

#ifdef CONFIG_QUEUED_LOCK_STAT
/*
@@ -79,99 +59,91 @@ enum qlock_stats {
#include <linux/sched/clock.h>
#include <linux/fs.h>

-static const char * const qstat_names[qstat_num + 1] = {
- [qstat_pv_hash_hops] = "pv_hash_hops",
- [qstat_pv_kick_unlock] = "pv_kick_unlock",
- [qstat_pv_kick_wake] = "pv_kick_wake",
- [qstat_pv_spurious_wakeup] = "pv_spurious_wakeup",
- [qstat_pv_latency_kick] = "pv_latency_kick",
- [qstat_pv_latency_wake] = "pv_latency_wake",
- [qstat_pv_lock_stealing] = "pv_lock_stealing",
- [qstat_pv_wait_again] = "pv_wait_again",
- [qstat_pv_wait_early] = "pv_wait_early",
- [qstat_pv_wait_head] = "pv_wait_head",
- [qstat_pv_wait_node] = "pv_wait_node",
- [qstat_lock_pending] = "lock_pending",
- [qstat_lock_slowpath] = "lock_slowpath",
- [qstat_lock_use_node2] = "lock_use_node2",
- [qstat_lock_use_node3] = "lock_use_node3",
- [qstat_lock_use_node4] = "lock_use_node4",
- [qstat_lock_no_node] = "lock_no_node",
- [qstat_reset_cnts] = "reset_counters",
+#define EVENT_COUNT(ev) lockevents[LOCKEVENT_ ## ev]
+
+#undef LOCK_EVENT
+#define LOCK_EVENT(name) [LOCKEVENT_ ## name] = #name,
+
+static const char * const lockevent_names[lockevent_num + 1] = {
+
+#include "lock_events_list.h"
+
+ [LOCKEVENT_reset_cnts] = ".reset_counts",
};

/*
* Per-cpu counters
*/
-static DEFINE_PER_CPU(unsigned long, qstats[qstat_num]);
+DEFINE_PER_CPU(unsigned long, lockevents[lockevent_num]);
static DEFINE_PER_CPU(u64, pv_kick_time);

/*
* Function to read and return the qlock statistical counter values
*
* The following counters are handled specially:
- * 1. qstat_pv_latency_kick
+ * 1. pv_latency_kick
* Average kick latency (ns) = pv_latency_kick/pv_kick_unlock
- * 2. qstat_pv_latency_wake
+ * 2. pv_latency_wake
* Average wake latency (ns) = pv_latency_wake/pv_kick_wake
- * 3. qstat_pv_hash_hops
+ * 3. pv_hash_hops
* Average hops/hash = pv_hash_hops/pv_kick_unlock
*/
-static ssize_t qstat_read(struct file *file, char __user *user_buf,
- size_t count, loff_t *ppos)
+static ssize_t lockevent_read(struct file *file, char __user *user_buf,
+ size_t count, loff_t *ppos)
{
char buf[64];
- int cpu, counter, len;
- u64 stat = 0, kicks = 0;
+ int cpu, id, len;
+ u64 sum = 0, kicks = 0;

/*
* Get the counter ID stored in file->f_inode->i_private
*/
- counter = (long)file_inode(file)->i_private;
+ id = (long)file_inode(file)->i_private;

- if (counter >= qstat_num)
+ if (id >= lockevent_num)
return -EBADF;

for_each_possible_cpu(cpu) {
- stat += per_cpu(qstats[counter], cpu);
+ sum += per_cpu(lockevents[id], cpu);
/*
- * Need to sum additional counter for some of them
+ * Need to sum additional counters for some of them
*/
- switch (counter) {
+ switch (id) {

- case qstat_pv_latency_kick:
- case qstat_pv_hash_hops:
- kicks += per_cpu(qstats[qstat_pv_kick_unlock], cpu);
+ case LOCKEVENT_pv_latency_kick:
+ case LOCKEVENT_pv_hash_hops:
+ kicks += per_cpu(EVENT_COUNT(pv_kick_unlock), cpu);
break;

- case qstat_pv_latency_wake:
- kicks += per_cpu(qstats[qstat_pv_kick_wake], cpu);
+ case LOCKEVENT_pv_latency_wake:
+ kicks += per_cpu(EVENT_COUNT(pv_kick_wake), cpu);
break;
}
}

- if (counter == qstat_pv_hash_hops) {
+ if (id == LOCKEVENT_pv_hash_hops) {
u64 frac = 0;

if (kicks) {
- frac = 100ULL * do_div(stat, kicks);
+ frac = 100ULL * do_div(sum, kicks);
frac = DIV_ROUND_CLOSEST_ULL(frac, kicks);
}

/*
* Return a X.XX decimal number
*/
- len = snprintf(buf, sizeof(buf) - 1, "%llu.%02llu\n", stat, frac);
+ len = snprintf(buf, sizeof(buf) - 1, "%llu.%02llu\n",
+ sum, frac);
} else {
/*
* Round to the nearest ns
*/
- if ((counter == qstat_pv_latency_kick) ||
- (counter == qstat_pv_latency_wake)) {
+ if ((id == LOCKEVENT_pv_latency_kick) ||
+ (id == LOCKEVENT_pv_latency_wake)) {
if (kicks)
- stat = DIV_ROUND_CLOSEST_ULL(stat, kicks);
+ sum = DIV_ROUND_CLOSEST_ULL(sum, kicks);
}
- len = snprintf(buf, sizeof(buf) - 1, "%llu\n", stat);
+ len = snprintf(buf, sizeof(buf) - 1, "%llu\n", sum);
}

return simple_read_from_buffer(user_buf, count, ppos, buf, len);
@@ -180,11 +152,9 @@ static ssize_t qstat_read(struct file *file, char __user *user_buf,
/*
* Function to handle write request
*
- * When counter = reset_cnts, reset all the counter values.
- * Since the counter updates aren't atomic, the resetting is done twice
- * to make sure that the counters are very likely to be all cleared.
+ * When id = .reset_cnts, reset all the counter values.
*/
-static ssize_t qstat_write(struct file *file, const char __user *user_buf,
+static ssize_t lockevent_write(struct file *file, const char __user *user_buf,
size_t count, loff_t *ppos)
{
int cpu;
@@ -192,14 +162,14 @@ static ssize_t qstat_write(struct file *file, const char __user *user_buf,
/*
* Get the counter ID stored in file->f_inode->i_private
*/
- if ((long)file_inode(file)->i_private != qstat_reset_cnts)
+ if ((long)file_inode(file)->i_private != LOCKEVENT_reset_cnts)
return count;

for_each_possible_cpu(cpu) {
int i;
- unsigned long *ptr = per_cpu_ptr(qstats, cpu);
+ unsigned long *ptr = per_cpu_ptr(lockevents, cpu);

- for (i = 0 ; i < qstat_num; i++)
+ for (i = 0 ; i < lockevent_num; i++)
WRITE_ONCE(ptr[i], 0);
}
return count;
@@ -208,9 +178,9 @@ static ssize_t qstat_write(struct file *file, const char __user *user_buf,
/*
* Debugfs data structures
*/
-static const struct file_operations fops_qstat = {
- .read = qstat_read,
- .write = qstat_write,
+static const struct file_operations fops_lockevent = {
+ .read = lockevent_read,
+ .write = lockevent_write,
.llseek = default_llseek,
};

@@ -219,10 +189,10 @@ static ssize_t qstat_write(struct file *file, const char __user *user_buf,
*/
static int __init init_qspinlock_stat(void)
{
- struct dentry *d_qstat = debugfs_create_dir("qlockstat", NULL);
+ struct dentry *d_counts = debugfs_create_dir("qlockstat", NULL);
int i;

- if (!d_qstat)
+ if (!d_counts)
goto out;

/*
@@ -232,18 +202,19 @@ static int __init init_qspinlock_stat(void)
* root is allowed to do the read/write to limit impact to system
* performance.
*/
- for (i = 0; i < qstat_num; i++)
- if (!debugfs_create_file(qstat_names[i], 0400, d_qstat,
- (void *)(long)i, &fops_qstat))
+ for (i = 0; i < lockevent_num; i++)
+ if (!debugfs_create_file(lockevent_names[i], 0400, d_counts,
+ (void *)(long)i, &fops_lockevent))
goto fail_undo;

- if (!debugfs_create_file(qstat_names[qstat_reset_cnts], 0200, d_qstat,
- (void *)(long)qstat_reset_cnts, &fops_qstat))
+ if (!debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
+ d_counts, (void *)(long)LOCKEVENT_reset_cnts,
+ &fops_lockevent))
goto fail_undo;

return 0;
fail_undo:
- debugfs_remove_recursive(d_qstat);
+ debugfs_remove_recursive(d_counts);
out:
pr_warn("Could not create 'qlockstat' debugfs entries\n");
return -ENOMEM;
@@ -251,20 +222,11 @@ static int __init init_qspinlock_stat(void)
fs_initcall(init_qspinlock_stat);

/*
- * Increment the PV qspinlock statistical counters
- */
-static inline void qstat_inc(enum qlock_stats stat, bool cond)
-{
- if (cond)
- this_cpu_inc(qstats[stat]);
-}
-
-/*
* PV hash hop count
*/
-static inline void qstat_hop(int hopcnt)
+static inline void lockevent_pv_hop(int hopcnt)
{
- this_cpu_add(qstats[qstat_pv_hash_hops], hopcnt);
+ this_cpu_add(EVENT_COUNT(pv_hash_hops), hopcnt);
}

/*
@@ -276,7 +238,7 @@ static inline void __pv_kick(int cpu)

per_cpu(pv_kick_time, cpu) = start;
pv_kick(cpu);
- this_cpu_add(qstats[qstat_pv_latency_kick], sched_clock() - start);
+ this_cpu_add(EVENT_COUNT(pv_latency_kick), sched_clock() - start);
}

/*
@@ -289,9 +251,9 @@ static inline void __pv_wait(u8 *ptr, u8 val)
*pkick_time = 0;
pv_wait(ptr, val);
if (*pkick_time) {
- this_cpu_add(qstats[qstat_pv_latency_wake],
+ this_cpu_add(EVENT_COUNT(pv_latency_wake),
sched_clock() - *pkick_time);
- qstat_inc(qstat_pv_kick_wake, true);
+ lockevent_inc(pv_kick_wake);
}
}

@@ -300,7 +262,6 @@ static inline void __pv_wait(u8 *ptr, u8 val)

#else /* CONFIG_QUEUED_LOCK_STAT */

-static inline void qstat_inc(enum qlock_stats stat, bool cond) { }
-static inline void qstat_hop(int hopcnt) { }
+static inline void lockevent_pv_hop(int hopcnt) { }

#endif /* CONFIG_QUEUED_LOCK_STAT */
--
1.8.3.1