From: Peter Zijlstra <peterz@infradead.org>
To: Jason Low <jason.low2@hp.com>
Cc: Ingo Molnar <mingo@redhat.com>,
Paul McKenney <paulmck@linux.vnet.ibm.com>,
Waiman Long <Waiman.Long@hp.com>,
Linus Torvalds <torvalds@linux-foundation.org>,
Thomas Gleixner <tglx@linutronix.de>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
Rik van Riel <riel@redhat.com>,
Andrew Morton <akpm@linux-foundation.org>,
Davidlohr Bueso <davidlohr@hp.com>,
"H. Peter Anvin" <hpa@zytor.com>,
Andi Kleen <andi@firstfloor.org>,
"Chandramouleeswaran, Aswin" <aswin@hp.com>,
"Norton, Scott J" <scott.norton@hp.com>,
chegu_vinod@hp.com
Subject: Re: [RFC][PATCH v2 5/5] mutex: Give spinners a chance to spin_on_owner if need_resched() triggered while queued
Date: Mon, 3 Feb 2014 20:25:25 +0100 [thread overview]
Message-ID: <20140203192525.GN8874@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <1391452760.7498.26.camel@j-VirtualBox>
On Mon, Feb 03, 2014 at 10:39:20AM -0800, Jason Low wrote:
> > To avoid the xchg on every loop.
>
> Ah yes, we want to use xchg() on &node->next.
>
> Since the cmpxchg() is now in a loop in the unlock function, an
> additional (*lock == node) check before the cmpxchg() would also be nice
> to avoid spinning on cmpxchg() there too.
Right, I have the below; you can find the patches this depends upon
here:
http://programming.kicks-ass.net/sekrit/patches.tar.bz2
---
Subject: locking, mutex: Cancelable MCS lock for adaptive spinning
From: Peter Zijlstra <peterz@infradead.org>
Date: Wed, 29 Jan 2014 12:51:42 +0100
Since we want a task waiting for a mutex_lock() to go to sleep and
reschedule on need_resched() we must be able to abort the
mcs_spin_lock() around the adaptive spin.
Therefore implement a cancelable mcs lock.
XXX: anybody got a better name than m_spinlock?
Cc: paulmck@linux.vnet.ibm.com
Cc: Waiman.Long@hp.com
Cc: torvalds@linux-foundation.org
Cc: tglx@linutronix.de
Cc: riel@redhat.com
Cc: akpm@linux-foundation.org
Cc: davidlohr@hp.com
Cc: hpa@zytor.com
Cc: andi@firstfloor.org
Cc: aswin@hp.com
Cc: scott.norton@hp.com
Cc: chegu_vinod@hp.com
Cc: mingo@redhat.com
Cc: Jason Low <jason.low2@hp.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-7jr68p4f447w2e0ck7y1yl06@git.kernel.org
---
include/linux/mutex.h | 4 -
kernel/locking/Makefile | 2
kernel/locking/mcs_spinlock.c | 156 ++++++++++++++++++++++++++++++++++++++++++
kernel/locking/mcs_spinlock.h | 18 ++++
kernel/locking/mutex.c | 10 +-
5 files changed, 183 insertions(+), 7 deletions(-)
--- a/include/linux/mutex.h
+++ b/include/linux/mutex.h
@@ -46,7 +46,7 @@
* - detects multi-task circular deadlocks and prints out all affected
* locks and tasks (and only those tasks)
*/
-struct mcs_spinlock;
+struct m_spinlock;
struct mutex {
/* 1: unlocked, 0: locked, negative: locked, possible waiters */
atomic_t count;
@@ -56,7 +56,7 @@ struct mutex {
struct task_struct *owner;
#endif
#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
- struct mcs_spinlock *mcs_lock; /* Spinner MCS lock */
+ struct m_spinlock *m_lock; /* Spinner MCS lock */
#endif
#ifdef CONFIG_DEBUG_MUTEXES
const char *name;
--- a/kernel/locking/Makefile
+++ b/kernel/locking/Makefile
@@ -1,5 +1,5 @@
-obj-y += mutex.o semaphore.o rwsem.o lglock.o
+obj-y += mutex.o semaphore.o rwsem.o lglock.o mcs_spinlock.o
ifdef CONFIG_FUNCTION_TRACER
CFLAGS_REMOVE_lockdep.o = -pg
--- /dev/null
+++ b/kernel/locking/mcs_spinlock.c
@@ -0,0 +1,156 @@
+
+#include <linux/percpu.h>
+#include <linux/mutex.h>
+#include <linux/sched.h>
+#include "mcs_spinlock.h"
+
+#ifdef CONFIG_SMP
+
+/*
+ * Using a single mcs node per CPU is safe because mutex_lock() should not be
+ * called from interrupt context and we have preemption disabled over the mcs
+ * lock usage.
+ */
+static DEFINE_PER_CPU_SHARED_ALIGNED(struct m_spinlock, m_node);
+
+/*
+ * Get a stable @node->next pointer, either for unlock() or unqueue() purposes.
+ * Can return NULL in case we were the last queued and we updated @lock instead.
+ */
+static inline struct m_spinlock *
+m_spin_wait_next(struct m_spinlock **lock, struct m_spinlock *node,
+ struct m_spinlock *prev)
+{
+ struct m_spinlock *next = NULL;
+
+ for (;;) {
+ if (*lock == node && cmpxchg(lock, node, prev) == node) {
+ /*
+ * We were the last queued, we moved @lock back. @prev
+ * will now observe @lock and will complete its
+ * unlock()/unqueue().
+ */
+ break;
+ }
+
+ /*
+ * We must xchg() the @node->next value, because if we were to
+ * leave it in, a concurrent unlock()/unqueue() from
+ * @node->next might complete Step-A and think its @prev is
+ * still valid.
+ *
+ * If the concurrent unlock()/unqueue() wins the race, we'll
+ * wait for either @lock to point to us, through its Step-B, or
+ * wait for a new @node->next from its Step-C.
+ */
+ if (node->next) {
+ next = xchg(&node->next, NULL);
+ if (next)
+ break;
+ }
+
+ arch_mutex_cpu_relax();
+ }
+
+ return next;
+}
+
+bool m_spin_lock(struct m_spinlock **lock)
+{
+ struct m_spinlock *node = this_cpu_ptr(&m_node);
+ struct m_spinlock *prev, *next;
+
+ node->locked = 0;
+ node->next = NULL;
+
+ node->prev = prev = xchg(lock, node);
+ if (likely(prev == NULL))
+ return true;
+
+ ACCESS_ONCE(prev->next) = node;
+
+ /*
+ * Normally @prev is untouchable after the above store; because at that
+ * moment unlock can proceed and wipe the node element from stack.
+ *
+ * However, since our nodes are static per-cpu storage, we're
+ * guaranteed their existence -- this allows us to apply
+ * cmpxchg in an attempt to undo our queueing.
+ */
+
+ while (!smp_load_acquire(&node->locked)) {
+ if (need_resched())
+ goto unqueue;
+ arch_mutex_cpu_relax();
+ }
+ return true;
+
+unqueue:
+ /*
+ * Step - A -- stabilize @prev
+ *
+ * Undo our @prev->next assignment; this will make @prev's
+ * unlock()/unqueue() wait for a next pointer since @lock points to us
+ * (or later).
+ */
+
+ for (;;) {
+ if (prev->next == node &&
+ cmpxchg(&prev->next, node, NULL) == node)
+ break;
+
+ /*
+ * We can only fail the cmpxchg() racing against an unlock(),
+ * in which case we should observe @node->locked becomming
+ * true.
+ */
+ if (smp_load_acquire(&node->locked))
+ return true;
+
+ /*
+ * Or we race against a concurrent unqueue()'s step-B, in which
+ * case its step-C will write us a new @node->prev pointer.
+ */
+ prev = ACCESS_ONCE(node->prev);
+ }
+
+ /*
+ * Step - B -- stabilize @next
+ *
+ * Similar to unlock(), wait for @node->next or move @lock from @node
+ * back to @prev.
+ */
+
+ next = m_spin_wait_next(lock, node, prev);
+ if (!next)
+ return false;
+
+ /*
+ * Step - C -- unlink
+ *
+ * @prev is stable because its still waiting for a new @prev->next
+ * pointer, @next is stable because our @node->next pointer is NULL and
+ * it will wait in Step-A.
+ */
+
+ ACCESS_ONCE(next->prev) = prev;
+ ACCESS_ONCE(prev->next) = next;
+
+ return false;
+}
+
+void m_spin_unlock(struct m_spinlock **lock)
+{
+ struct m_spinlock *node = this_cpu_ptr(&m_node);
+ struct m_spinlock *next;
+
+ if (likely(cmpxchg(lock, node, NULL) == node))
+ return;
+
+ next = m_spin_wait_next(lock, node, NULL);
+ if (next)
+ ACCESS_ONCE(next->locked) = 1;
+}
+
+#endif
+
--- a/kernel/locking/mcs_spinlock.h
+++ b/kernel/locking/mcs_spinlock.h
@@ -109,4 +109,22 @@ void mcs_spin_unlock(struct mcs_spinlock
arch_mcs_spin_unlock_contended(&next->locked);
}
+/*
+ * Cancellable version of the MCS lock above.
+ *
+ * This version can fail acquisition and unqueue a spinner; it assumes no
+ * nesting.
+ *
+ * Intended for adaptive spinning of sleeping locks:
+ * mutex_lock()/rwsem_down_{read,write}() etc.
+ */
+
+struct m_spinlock {
+ struct m_spinlock *next, *prev;
+ int locked; /* 1 if lock acquired */
+};
+
+extern bool m_spin_lock(struct m_spinlock **lock);
+extern void m_spin_unlock(struct m_spinlock **lock);
+
#endif /* __LINUX_MCS_SPINLOCK_H */
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -53,7 +53,7 @@ __mutex_init(struct mutex *lock, const c
INIT_LIST_HEAD(&lock->wait_list);
mutex_clear_owner(lock);
#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
- lock->mcs_lock = NULL;
+ lock->m_lock = NULL;
#endif
debug_mutex_init(lock, name, key);
@@ -403,7 +403,9 @@ __mutex_lock_common(struct mutex *lock,
if (!mutex_can_spin_on_owner(lock))
goto slowpath;
- mcs_spin_lock(&lock->mcs_lock);
+ if (!m_spin_lock(&lock->m_lock))
+ goto slowpath;
+
for (;;) {
struct task_struct *owner;
@@ -442,7 +444,7 @@ __mutex_lock_common(struct mutex *lock,
}
mutex_set_owner(lock);
- mcs_spin_unlock(&lock->mcs_lock);
+ m_spin_unlock(&lock->m_lock);
preempt_enable();
return 0;
}
@@ -464,7 +466,7 @@ __mutex_lock_common(struct mutex *lock,
*/
arch_mutex_cpu_relax();
}
- mcs_spin_unlock(&lock->mcs_lock);
+ m_spin_unlock(&lock->m_lock);
slowpath:
#endif
spin_lock_mutex(&lock->wait_lock, flags);
next prev parent reply other threads:[~2014-02-03 19:25 UTC|newest]
Thread overview: 51+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-01-28 19:13 [PATCH v2 0/5] mutex: Mutex scalability patches Jason Low
2014-01-28 19:13 ` [PATCH v2 1/5] mutex: In mutex_can_spin_on_owner(), return false if task need_resched() Jason Low
2014-01-28 20:20 ` Paul E. McKenney
2014-01-28 22:01 ` Jason Low
2014-01-28 21:09 ` Davidlohr Bueso
2014-03-11 12:41 ` [tip:core/locking] locking/mutexes: Return false if task need_resched() in mutex_can_spin_on_owner() tip-bot for Jason Low
2014-01-28 19:13 ` [PATCH v2 2/5] mutex: Modify the way optimistic spinners are queued Jason Low
2014-01-28 20:23 ` Paul E. McKenney
2014-01-28 20:24 ` Paul E. McKenney
2014-01-28 21:17 ` Davidlohr Bueso
2014-01-28 22:10 ` Jason Low
2014-02-02 21:58 ` Paul E. McKenney
2014-03-11 12:41 ` [tip:core/locking] locking/mutexes: " tip-bot for Jason Low
2014-03-11 15:24 ` Jason Low
2014-03-11 15:33 ` Peter Zijlstra
2014-01-28 19:13 ` [PATCH v2 3/5] mutex: Unlock the mutex without the wait_lock Jason Low
2014-03-11 12:41 ` [tip:core/locking] locking/mutexes: " tip-bot for Jason Low
2014-03-12 12:24 ` Peter Zijlstra
2014-03-12 18:44 ` Jason Low
2014-03-13 7:28 ` [tip:core/locking] locking/mutex: Fix debug checks tip-bot for Peter Zijlstra
2014-01-28 19:13 ` [RFC][PATCH v2 4/5] mutex: Disable preemtion between modifying lock->owner and locking/unlocking mutex Jason Low
2014-01-28 20:54 ` Peter Zijlstra
2014-01-28 22:17 ` Jason Low
2014-01-28 19:13 ` [RFC][PATCH v2 5/5] mutex: Give spinners a chance to spin_on_owner if need_resched() triggered while queued Jason Low
2014-01-28 21:07 ` Peter Zijlstra
2014-01-28 22:51 ` Jason Low
2014-01-29 11:51 ` Peter Zijlstra
2014-01-31 3:29 ` Jason Low
2014-01-31 14:09 ` Peter Zijlstra
2014-01-31 20:01 ` Jason Low
2014-01-31 20:08 ` Peter Zijlstra
2014-02-02 21:01 ` Jason Low
2014-02-02 21:12 ` Peter Zijlstra
2014-02-03 18:39 ` Jason Low
2014-02-03 19:25 ` Peter Zijlstra [this message]
2014-02-03 20:55 ` Jason Low
2014-02-03 21:06 ` Peter Zijlstra
2014-02-03 21:56 ` Jason Low
2014-02-04 7:13 ` Jason Low
2014-02-02 22:02 ` Paul E. McKenney
2014-02-02 20:02 ` Peter Zijlstra
2014-02-05 21:44 ` Waiman Long
2014-02-06 14:04 ` Peter Zijlstra
2014-02-06 18:45 ` Waiman Long
2014-02-06 20:10 ` Norton, Scott J
2014-02-10 17:01 ` Peter Zijlstra
2014-02-06 17:44 ` Jason Low
2014-02-06 18:37 ` Waiman Long
2014-01-28 21:08 ` [PATCH v2 0/5] mutex: Mutex scalability patches Davidlohr Bueso
2014-01-28 23:11 ` Jason Low
2014-02-06 14:52 [RFC][PATCH v2 5/5] mutex: Give spinners a chance to spin_on_owner if need_resched() triggered while queued Daniel J Blueman
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=20140203192525.GN8874@twins.programming.kicks-ass.net \
--to=peterz@infradead.org \
--cc=Waiman.Long@hp.com \
--cc=akpm@linux-foundation.org \
--cc=andi@firstfloor.org \
--cc=aswin@hp.com \
--cc=chegu_vinod@hp.com \
--cc=davidlohr@hp.com \
--cc=hpa@zytor.com \
--cc=jason.low2@hp.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=paulmck@linux.vnet.ibm.com \
--cc=riel@redhat.com \
--cc=scott.norton@hp.com \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).