linux/kernel/locking/osq_lock.c
Waiman Long f5bfdc8e39 locking/osq: Use optimized spinning loop for arm64
Arm64 has a more optimized spinning loop (atomic_cond_read_acquire)
using wfe for spinlock that can boost performance of sibling threads
by putting the current cpu to a wait state that is broken only when
the monitored variable changes or an external event happens.

OSQ has a more complicated spinning loop. Besides the lock value, it
also checks for need_resched() and vcpu_is_preempted(). The check for
need_resched() is not a problem as it is only set by the tick interrupt
handler. That will be detected by the spinning cpu right after iret.

The vcpu_is_preempted() check, however, is a problem as changes to the
preempt state of of previous node will not affect the wait state. For
ARM64, vcpu_is_preempted is not currently defined and so is a no-op.
Will has indicated that he is planning to para-virtualize wfe instead
of defining vcpu_is_preempted for PV support. So just add a comment in
arch/arm64/include/asm/spinlock.h to indicate that vcpu_is_preempted()
should not be defined as suggested.

On a 2-socket 56-core 224-thread ARM64 system, a kernel mutex locking
microbenchmark was run for 10s with and without the patch. The
performance numbers before patch were:

Running locktest with mutex [runtime = 10s, load = 1]
Threads = 224, Min/Mean/Max = 316/123,143/2,121,269
Threads = 224, Total Rate = 2,757 kop/s; Percpu Rate = 12 kop/s

After patch, the numbers were:

Running locktest with mutex [runtime = 10s, load = 1]
Threads = 224, Min/Mean/Max = 334/147,836/1,304,787
Threads = 224, Total Rate = 3,311 kop/s; Percpu Rate = 15 kop/s

So there was about 20% performance improvement.

Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lkml.kernel.org/r/20200113150735.21956-1-longman@redhat.com
2020-01-17 10:19:30 +01:00

229 lines
5.7 KiB
C

// SPDX-License-Identifier: GPL-2.0
#include <linux/percpu.h>
#include <linux/sched.h>
#include <linux/osq_lock.h>
/*
* An MCS like lock especially tailored for optimistic spinning for sleeping
* lock implementations (mutex, rwsem, etc).
*
* Using a single mcs node per CPU is safe because sleeping locks should not be
* called from interrupt context and we have preemption disabled while
* spinning.
*/
static DEFINE_PER_CPU_SHARED_ALIGNED(struct optimistic_spin_node, osq_node);
/*
* We use the value 0 to represent "no CPU", thus the encoded value
* will be the CPU number incremented by 1.
*/
static inline int encode_cpu(int cpu_nr)
{
return cpu_nr + 1;
}
static inline int node_cpu(struct optimistic_spin_node *node)
{
return node->cpu - 1;
}
static inline struct optimistic_spin_node *decode_cpu(int encoded_cpu_val)
{
int cpu_nr = encoded_cpu_val - 1;
return per_cpu_ptr(&osq_node, cpu_nr);
}
/*
* 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 optimistic_spin_node *
osq_wait_next(struct optimistic_spin_queue *lock,
struct optimistic_spin_node *node,
struct optimistic_spin_node *prev)
{
struct optimistic_spin_node *next = NULL;
int curr = encode_cpu(smp_processor_id());
int old;
/*
* If there is a prev node in queue, then the 'old' value will be
* the prev node's CPU #, else it's set to OSQ_UNLOCKED_VAL since if
* we're currently last in queue, then the queue will then become empty.
*/
old = prev ? prev->cpu : OSQ_UNLOCKED_VAL;
for (;;) {
if (atomic_read(&lock->tail) == curr &&
atomic_cmpxchg_acquire(&lock->tail, curr, old) == curr) {
/*
* 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;
}
cpu_relax();
}
return next;
}
bool osq_lock(struct optimistic_spin_queue *lock)
{
struct optimistic_spin_node *node = this_cpu_ptr(&osq_node);
struct optimistic_spin_node *prev, *next;
int curr = encode_cpu(smp_processor_id());
int old;
node->locked = 0;
node->next = NULL;
node->cpu = curr;
/*
* We need both ACQUIRE (pairs with corresponding RELEASE in
* unlock() uncontended, or fastpath) and RELEASE (to publish
* the node fields we just initialised) semantics when updating
* the lock tail.
*/
old = atomic_xchg(&lock->tail, curr);
if (old == OSQ_UNLOCKED_VAL)
return true;
prev = decode_cpu(old);
node->prev = prev;
/*
* osq_lock() unqueue
*
* node->prev = prev osq_wait_next()
* WMB MB
* prev->next = node next->prev = prev // unqueue-C
*
* Here 'node->prev' and 'next->prev' are the same variable and we need
* to ensure these stores happen in-order to avoid corrupting the list.
*/
smp_wmb();
WRITE_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.
*/
/*
* Wait to acquire the lock or cancelation. Note that need_resched()
* will come with an IPI, which will wake smp_cond_load_relaxed() if it
* is implemented with a monitor-wait. vcpu_is_preempted() relies on
* polling, be careful.
*/
if (smp_cond_load_relaxed(&node->locked, VAL || need_resched() ||
vcpu_is_preempted(node_cpu(node->prev))))
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;
cpu_relax();
/*
* 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 = READ_ONCE(node->prev);
}
/*
* Step - B -- stabilize @next
*
* Similar to unlock(), wait for @node->next or move @lock from @node
* back to @prev.
*/
next = osq_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.
*/
WRITE_ONCE(next->prev, prev);
WRITE_ONCE(prev->next, next);
return false;
}
void osq_unlock(struct optimistic_spin_queue *lock)
{
struct optimistic_spin_node *node, *next;
int curr = encode_cpu(smp_processor_id());
/*
* Fast path for the uncontended case.
*/
if (likely(atomic_cmpxchg_release(&lock->tail, curr,
OSQ_UNLOCKED_VAL) == curr))
return;
/*
* Second most likely case.
*/
node = this_cpu_ptr(&osq_node);
next = xchg(&node->next, NULL);
if (next) {
WRITE_ONCE(next->locked, 1);
return;
}
next = osq_wait_next(lock, node, NULL);
if (next)
WRITE_ONCE(next->locked, 1);
}