[RFC PATCH-queue/locking/rfc 2/2] locking/mutex: Enable optimistic spinning of woken waiter

[Waiman Long]

This patch makes the waiter that sets the HANDOFF flag start spinning
instead of sleeping until the handoff is complete or the owner
sleeps. Otherwise, the handoff will cause the optimistic spinners to
abort spinning as the handed-off owner may not be running.

Signed-off-by: Waiman Long <Waima...@hpe.com>
---
kernel/locking/mutex.c | 104 ++++++++++++++++++++++++++++++++----------------
1 files changed, 70 insertions(+), 34 deletions(-)

diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
index 4a0e16e..064dd61 100644
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -380,22 +380,38 @@ static inline int mutex_can_spin_on_owner(struct mutex *lock)
*
* Returns true when the lock was taken, otherwise false, indicating
* that we need to jump to the slowpath and sleep.
+ *
+ * The waiter flag is set to true if the spinner is a waiter in the wait
+ * queue. The waiter-spinner will spin on the lock directly and concurrently
+ * with the spinner at the head of the OSQ, if present, until the owner is
+ * changed to itself.
*/
static bool mutex_optimistic_spin(struct mutex *lock,
- struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
+ struct ww_acquire_ctx *ww_ctx,
+ const bool use_ww_ctx, bool waiter)
{
struct task_struct *task = current;
+ bool acquired = false;

- if (!mutex_can_spin_on_owner(lock))
- goto done;
+ if (!waiter) {
+ /*
+ * The purpose of the mutex_can_spin_on_owner() function is
+ * to eliminate the overhead of osq_lock() and osq_unlock()
+ * in case spinning isn't possible. As a waiter-spinner
+ * is not going to take OSQ lock anyway, there is no need
+ * to call mutex_can_spin_on_owner().
+ */
+ if (!mutex_can_spin_on_owner(lock))
+ goto done;

- /*
- * In order to avoid a stampede of mutex spinners trying to
- * acquire the mutex all at once, the spinners need to take a
- * MCS (queued) lock first before spinning on the owner field.
- */
- if (!osq_lock(&lock->osq))
- goto done;
+ /*
+ * In order to avoid a stampede of mutex spinners trying to
+ * acquire the mutex all at once, the spinners need to take a
+ * MCS (queued) lock first before spinning on the owner field.
+ */
+ if (!osq_lock(&lock->osq))
+ goto done;
+ }

while (true) {
struct task_struct *owner;
@@ -421,23 +437,24 @@ static bool mutex_optimistic_spin(struct mutex *lock,
* release the lock or go to sleep.
*/
owner = __mutex_owner(lock);
- if (owner && !mutex_spin_on_owner(lock, owner))
- break;

- /* Try to acquire the mutex if it is unlocked. */
- if (__mutex_trylock(lock)) {
- osq_unlock(&lock->osq);
- return true;
+ if (owner == task)
+ goto gotlock;
+
+ if (owner) {
+ if (!mutex_spin_on_owner(lock, owner))
+ break;
+ /*
+ * For waiter-spinner, recheck the owner field
+ * as it may have been changed to itself.
+ */
+ if (waiter && (__mutex_owner(lock) == task))
+ goto gotlock;
}

- /*
- * When there's no owner, we might have preempted between the
- * owner acquiring the lock and setting the owner field. If
- * we're an RT task that will live-lock because we won't let
- * the owner complete.
- */
- if (!owner && (need_resched() || rt_task(task)))
- break;
+ /* Try to acquire the mutex if it is unlocked. */
+ if (__mutex_trylock(lock))
+ goto gotlock;

/*
* The cpu_relax() call is a compiler barrier which forces
@@ -446,16 +463,21 @@ static bool mutex_optimistic_spin(struct mutex *lock,
* values at the cost of a few extra spins.
*/
cpu_relax_lowlatency();
+ continue;
+gotlock:
+ acquired = true;
+ break;
}

- osq_unlock(&lock->osq);
+ if (!waiter)
+ osq_unlock(&lock->osq);
done:
/*
* If we fell out of the spin path because of need_resched(),
* reschedule now, before we try-lock the mutex. This avoids getting
* scheduled out right after we obtained the mutex.
*/
- if (need_resched()) {
+ if (!acquired && need_resched()) {
/*
* We _should_ have TASK_RUNNING here, but just in case
* we do not, make it so, otherwise we might get stuck.
@@ -464,11 +486,12 @@ done:
schedule_preempt_disabled();
}

- return false;
+ return acquired;
}
#else
static bool mutex_optimistic_spin(struct mutex *lock,
- struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
+ struct ww_acquire_ctx *ww_ctx,
+ const bool use_ww_ctx, bool waiter)
{
return false;
}
@@ -561,6 +584,7 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
unsigned long flags;
struct ww_mutex *ww;
int ret;
+ bool acquired;

if (use_ww_ctx) {
ww = container_of(lock, struct ww_mutex, base);
@@ -571,7 +595,8 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
preempt_disable();
mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);

- if (__mutex_trylock(lock) || mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx)) {
+ if (__mutex_trylock(lock) ||
+ mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, false)) {
/* got the lock, yay! */
lock_acquired(&lock->dep_map, ip);
if (use_ww_ctx)
@@ -603,7 +628,7 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,

lock_contended(&lock->dep_map, ip);

- for (;;) {
+ for (acquired = false; !acquired; ) {
/*
* got a signal? (This code gets eliminated in the
* TASK_UNINTERRUPTIBLE case.)
@@ -624,13 +649,24 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
/* didn't get the lock, go to sleep: */
spin_unlock_mutex(&lock->wait_lock, flags);
schedule_preempt_disabled();
- spin_lock_mutex(&lock->wait_lock, flags);

- if (__mutex_trylock(lock))
- break;
+ /*
+ * Both __mutex_trylock() and __mutex_waiter_is_first()
+ * can be done without the protection of wait_lock.
+ */
+ acquired = __mutex_trylock(lock);

- if (__mutex_waiter_is_first(lock, &waiter))
+ if (!acquired && __mutex_waiter_is_first(lock, &waiter)) {
__mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
+ /*
+ * Wait until the lock is handed off or the owner
+ * sleeps.
+ */
+ acquired = mutex_optimistic_spin(lock, ww_ctx,
+ use_ww_ctx, true);
+ }
+
+ spin_lock_mutex(&lock->wait_lock, flags);
}
__set_task_state(task, TASK_RUNNING);

--
1.7.1

[Peter Zijlstra]

On Fri, Aug 26, 2016 at 07:35:09PM -0400, Waiman Long wrote:

> @@ -624,13 +649,24 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
> /* didn't get the lock, go to sleep: */
> spin_unlock_mutex(&lock->wait_lock, flags);
> schedule_preempt_disabled();
>

> + /*
> + * Both __mutex_trylock() and __mutex_waiter_is_first()
> + * can be done without the protection of wait_lock.
> + */

True, but it took me a little while to figure out why
__mutex_waiter_is_first() is safe without the lock :-)

> + acquired = __mutex_trylock(lock);

>
> + if (!acquired && __mutex_waiter_is_first(lock, &waiter)) {
> __mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
> + /*
> + * Wait until the lock is handed off or the owner
> + * sleeps.
> + */
> + acquired = mutex_optimistic_spin(lock, ww_ctx,
> + use_ww_ctx, true);
> + }

That said; I think there's a few problems with this. Since we now poke
at the loop termination conditions outside of the wait_lock, it becomes
important where we do the task->state vs wakeup bits.

Specifically, since we still have state==RUNNING here, its possible
we'll fail to acquire the lock _and_ miss the wakeup from
mutex_unlock(). Leaving us stuck forever more.

Also, we should do the __mutex_trylock _after_ we set the handoff,
otherwise its possible we get the lock handed (miss the wakeup as per
the above) and fail to notice, again going back to sleep forever more.

> +
> + spin_lock_mutex(&lock->wait_lock, flags);
> }
> __set_task_state(task, TASK_RUNNING);

I'm thinking something like the below on top of yours will cure things..
have not tested yet...

---
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -394,10 +394,9 @@ static inline int mutex_can_spin_on_owne

*/
static bool mutex_optimistic_spin(struct mutex *lock,

struct ww_acquire_ctx *ww_ctx,
- const bool use_ww_ctx, bool waiter)
+ const bool use_ww_ctx, const bool waiter)

{
struct task_struct *task = current;

- bool acquired = false;

if (!waiter) {
/*
@@ -408,7 +407,7 @@ static bool mutex_optimistic_spin(struct
* to call mutex_can_spin_on_owner().
*/

if (!mutex_can_spin_on_owner(lock))
- goto done;

+ goto fail;

/*

* In order to avoid a stampede of mutex spinners trying to

@@ -416,10 +415,10 @@ static bool mutex_optimistic_spin(struct

* MCS (queued) lock first before spinning on the owner field.

*/

if (!osq_lock(&lock->osq))
- goto done;

+ goto fail;
}

- while (true) {
+ for (;;) {
struct task_struct *owner;

if (use_ww_ctx && ww_ctx->acquired > 0) {
@@ -435,7 +434,7 @@ static bool mutex_optimistic_spin(struct
* performed the optimistic spinning cannot be done.
*/
if (READ_ONCE(ww->ctx))
- break;
+ goto fail_unlock;
}

/*
@@ -443,23 +442,16 @@ static bool mutex_optimistic_spin(struct

* release the lock or go to sleep.
*/
owner = __mutex_owner(lock);
-

- if (owner == task)
- goto gotlock;
-
if (owner) {
+ if (waiter && owner == task)
+ goto gotlock_acquire;
+
if (!mutex_spin_on_owner(lock, owner))
- break;
- /*
- * For waiter-spinner, recheck the owner field
- * as it may have been changed to itself.
- */
- if (waiter && (__mutex_owner(lock) == task))
- goto gotlock;
+ goto fail_unlock;

}

/* Try to acquire the mutex if it is unlocked. */

- if (__mutex_trylock(lock, false))
+ if (__mutex_trylock(lock, waiter))
goto gotlock;

/*
@@ -469,21 +461,28 @@ static bool mutex_optimistic_spin(struct

* values at the cost of a few extra spins.
*/
cpu_relax_lowlatency();

- continue;
-gotlock:
- acquired = true;
- break;
}

+gotlock_acquire:
+ smp_mb(); /* ACQUIRE */
+gotlock:

+ if (!waiter)
+ osq_unlock(&lock->osq);

+
+ return true;
+
+
+fail_unlock:
if (!waiter)
osq_unlock(&lock->osq);
-done:
+
+fail:

/*
* If we fell out of the spin path because of need_resched(),
* reschedule now, before we try-lock the mutex. This avoids getting
* scheduled out right after we obtained the mutex.
*/

- if (!acquired && need_resched()) {
+ if (need_resched()) {

/*
* We _should_ have TASK_RUNNING here, but just in case
* we do not, make it so, otherwise we might get stuck.

@@ -492,12 +491,12 @@ static bool mutex_optimistic_spin(struct
schedule_preempt_disabled();
}

- return acquired;
+ return false;

}
#else
static bool mutex_optimistic_spin(struct mutex *lock,

struct ww_acquire_ctx *ww_ctx,
- const bool use_ww_ctx, bool waiter)
+ const bool use_ww_ctx, const bool waiter)
{
return false;
}
@@ -590,7 +589,6 @@ __mutex_lock_common(struct mutex *lock,

unsigned long flags;
struct ww_mutex *ww;
int ret;

- bool acquired;

if (use_ww_ctx) {
ww = container_of(lock, struct ww_mutex, base);

@@ -606,7 +604,7 @@ __mutex_lock_common(struct mutex *lock,
* possible to allow recursive lock attempts by accident.
*/
if (__mutex_trylock(lock, false) ||
- mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx)) {

+ mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, false)) {
/* got the lock, yay! */
lock_acquired(&lock->dep_map, ip);
if (use_ww_ctx)

@@ -638,7 +636,8 @@ __mutex_lock_common(struct mutex *lock,

lock_contended(&lock->dep_map, ip);

- for (acquired = false; !acquired; ) {
+ set_task_state(task, state);
+ for (;;) {

/*
* got a signal? (This code gets eliminated in the
* TASK_UNINTERRUPTIBLE case.)

@@ -654,30 +653,23 @@ __mutex_lock_common(struct mutex *lock,
goto err;
}

- __set_task_state(task, state);
-
- /* didn't get the lock, go to sleep: */
spin_unlock_mutex(&lock->wait_lock, flags);
schedule_preempt_disabled();

- /*
- * Both __mutex_trylock() and __mutex_waiter_is_first()
- * can be done without the protection of wait_lock.
- */
- acquired = __mutex_trylock(lock, true);
+ set_task_state(task, state);

- if (!acquired && __mutex_waiter_is_first(lock, &waiter)) {
+ if (__mutex_waiter_is_first(lock, &waiter)) {
__mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
- /*
- * Wait until the lock is handed off or the owner
- * sleeps.
- */
- acquired = mutex_optimistic_spin(lock, ww_ctx,
- use_ww_ctx, true);
+ if (mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, true))
+ break;
}

+ if (__mutex_trylock(lock, true))
+ break;
+
spin_lock_mutex(&lock->wait_lock, flags);

}
+ spin_lock_mutex(&lock->wait_lock, flags);
__set_task_state(task, TASK_RUNNING);

remove_waiter:
@@ -700,6 +692,7 @@ __mutex_lock_common(struct mutex *lock,
return 0;

err:
+ __set_task_state(task, TASK_RUNNING);
mutex_remove_waiter(lock, &waiter, task);
spin_unlock_mutex(&lock->wait_lock, flags);
debug_mutex_free_waiter(&waiter);

[tip-bot for Waiman Long]

Commit-ID: b341afb325eb390f707a82cbefd65cda887302ab
Gitweb: http://git.kernel.org/tip/b341afb325eb390f707a82cbefd65cda887302ab
Author: Waiman Long <Waima...@hpe.com>
AuthorDate: Fri, 26 Aug 2016 19:35:09 -0400
Committer: Ingo Molnar <mi...@kernel.org>
CommitDate: Tue, 25 Oct 2016 11:31:54 +0200

locking/mutex: Enable optimistic spinning of woken waiter

This patch makes the waiter that sets the HANDOFF flag start spinning
instead of sleeping until the handoff is complete or the owner
sleeps. Otherwise, the handoff will cause the optimistic spinners to
abort spinning as the handed-off owner may not be running.

Tested-by: Jason Low <jason...@hpe.com>
Signed-off-by: Waiman Long <Waima...@hpe.com>
Signed-off-by: Peter Zijlstra (Intel) <pet...@infradead.org>
Cc: Andrew Morton <ak...@linux-foundation.org>
Cc: Davidlohr Bueso <da...@stgolabs.net>
Cc: Ding Tianhong <dingti...@huawei.com>
Cc: Imre Deak <imre...@intel.com>
Cc: Linus Torvalds <torv...@linux-foundation.org>
Cc: Paul E. McKenney <pau...@linux.vnet.ibm.com>
Cc: Paul E. McKenney <pau...@us.ibm.com>
Cc: Peter Zijlstra <pet...@infradead.org>
Cc: Thomas Gleixner <tg...@linutronix.de>
Cc: Tim Chen <tim.c...@linux.intel.com>
Cc: Will Deacon <Will....@arm.com>
Link: http://lkml.kernel.org/r/1472254509-27508-2-git...@hpe.com
Signed-off-by: Ingo Molnar <mi...@kernel.org>
---
kernel/locking/mutex.c | 77 +++++++++++++++++++++++++++++++++++---------------
1 file changed, 54 insertions(+), 23 deletions(-)

diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
index 6c0d304..17a88e9 100644
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -416,24 +416,39 @@ static inline int mutex_can_spin_on_owner(struct mutex *lock)

*
* Returns true when the lock was taken, otherwise false, indicating
* that we need to jump to the slowpath and sleep.
+ *
+ * The waiter flag is set to true if the spinner is a waiter in the wait
+ * queue. The waiter-spinner will spin on the lock directly and concurrently
+ * with the spinner at the head of the OSQ, if present, until the owner is
+ * changed to itself.
*/
static bool mutex_optimistic_spin(struct mutex *lock,
- struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
+ struct ww_acquire_ctx *ww_ctx,

+ const bool use_ww_ctx, const bool waiter)

{
struct task_struct *task = current;

- if (!mutex_can_spin_on_owner(lock))
- goto done;
+ if (!waiter) {
+ /*
+ * The purpose of the mutex_can_spin_on_owner() function is
+ * to eliminate the overhead of osq_lock() and osq_unlock()
+ * in case spinning isn't possible. As a waiter-spinner
+ * is not going to take OSQ lock anyway, there is no need
+ * to call mutex_can_spin_on_owner().
+ */
+ if (!mutex_can_spin_on_owner(lock))

+ goto fail;

- /*
- * In order to avoid a stampede of mutex spinners trying to
- * acquire the mutex all at once, the spinners need to take a
- * MCS (queued) lock first before spinning on the owner field.
- */
- if (!osq_lock(&lock->osq))
- goto done;
+ /*
+ * In order to avoid a stampede of mutex spinners trying to
+ * acquire the mutex all at once, the spinners need to take a
+ * MCS (queued) lock first before spinning on the owner field.
+ */
+ if (!osq_lock(&lock->osq))

+ goto fail;
+ }

- while (true) {
+ for (;;) {
struct task_struct *owner;

if (use_ww_ctx && ww_ctx->acquired > 0) {

@@ -449,7 +464,7 @@ static bool mutex_optimistic_spin(struct mutex *lock,

* performed the optimistic spinning cannot be done.
*/
if (READ_ONCE(ww->ctx))
- break;
+ goto fail_unlock;
}

/*

@@ -457,15 +472,20 @@ static bool mutex_optimistic_spin(struct mutex *lock,

* release the lock or go to sleep.
*/
owner = __mutex_owner(lock);
- if (owner && !mutex_spin_on_owner(lock, owner))
- break;

+ if (owner) {

+ if (waiter && owner == task) {

+ smp_mb(); /* ACQUIRE */
+ break;
+ }

- /* Try to acquire the mutex if it is unlocked. */

- if (__mutex_trylock(lock, false)) {

- osq_unlock(&lock->osq);
- return true;

+ if (!mutex_spin_on_owner(lock, owner))
+ goto fail_unlock;

}

+ /* Try to acquire the mutex if it is unlocked. */

+ if (__mutex_trylock(lock, waiter))
+ break;
+

/*
* The cpu_relax() call is a compiler barrier which forces

* everything in this loop to be re-loaded. We don't need
@@ -475,8 +495,17 @@ static bool mutex_optimistic_spin(struct mutex *lock,
cpu_relax_lowlatency();
}

- osq_unlock(&lock->osq);
-done:

+ if (!waiter)
+ osq_unlock(&lock->osq);

+
+ return true;
+
+
+fail_unlock:

+ if (!waiter)
+ osq_unlock(&lock->osq);

+
+fail:

/*
* If we fell out of the spin path because of need_resched(),
* reschedule now, before we try-lock the mutex. This avoids getting

@@ -495,7 +524,8 @@ done:

}
#else
static bool mutex_optimistic_spin(struct mutex *lock,
- struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
+ struct ww_acquire_ctx *ww_ctx,

+ const bool use_ww_ctx, const bool waiter)
{
return false;
}
@@ -600,7 +630,7 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,

mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);

if (__mutex_trylock(lock, false) ||
- mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx)) {

+ mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, false)) {
/* got the lock, yay! */
lock_acquired(&lock->dep_map, ip);
if (use_ww_ctx)

@@ -669,7 +699,8 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
* state back to RUNNING and fall through the next schedule(),
* or we must see its unlock and acquire.
*/
- if (__mutex_trylock(lock, first))
+ if ((first && mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, true)) ||
+ __mutex_trylock(lock, first))
break;

spin_lock_mutex(&lock->wait_lock, flags);