2012-12-18 00:01:36 +00:00
|
|
|
#include <linux/atomic.h>
|
2012-12-18 00:01:32 +00:00
|
|
|
#include <linux/rwsem.h>
|
|
|
|
#include <linux/percpu.h>
|
|
|
|
#include <linux/wait.h>
|
2012-12-18 00:01:38 +00:00
|
|
|
#include <linux/lockdep.h>
|
2012-12-18 00:01:32 +00:00
|
|
|
#include <linux/percpu-rwsem.h>
|
|
|
|
#include <linux/rcupdate.h>
|
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/errno.h>
|
|
|
|
|
2012-12-18 00:01:38 +00:00
|
|
|
int __percpu_init_rwsem(struct percpu_rw_semaphore *brw,
|
|
|
|
const char *name, struct lock_class_key *rwsem_key)
|
2012-12-18 00:01:32 +00:00
|
|
|
{
|
|
|
|
brw->fast_read_ctr = alloc_percpu(int);
|
|
|
|
if (unlikely(!brw->fast_read_ctr))
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2012-12-18 00:01:38 +00:00
|
|
|
/* ->rw_sem represents the whole percpu_rw_semaphore for lockdep */
|
|
|
|
__init_rwsem(&brw->rw_sem, name, rwsem_key);
|
2015-08-21 17:42:57 +00:00
|
|
|
rcu_sync_init(&brw->rss, RCU_SCHED_SYNC);
|
2012-12-18 00:01:32 +00:00
|
|
|
atomic_set(&brw->slow_read_ctr, 0);
|
|
|
|
init_waitqueue_head(&brw->write_waitq);
|
|
|
|
return 0;
|
|
|
|
}
|
2015-09-01 03:21:59 +00:00
|
|
|
EXPORT_SYMBOL_GPL(__percpu_init_rwsem);
|
2012-12-18 00:01:32 +00:00
|
|
|
|
|
|
|
void percpu_free_rwsem(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
2015-08-21 17:42:55 +00:00
|
|
|
/*
|
|
|
|
* XXX: temporary kludge. The error path in alloc_super()
|
|
|
|
* assumes that percpu_free_rwsem() is safe after kzalloc().
|
|
|
|
*/
|
|
|
|
if (!brw->fast_read_ctr)
|
|
|
|
return;
|
|
|
|
|
2015-08-21 17:42:57 +00:00
|
|
|
rcu_sync_dtor(&brw->rss);
|
2012-12-18 00:01:32 +00:00
|
|
|
free_percpu(brw->fast_read_ctr);
|
|
|
|
brw->fast_read_ctr = NULL; /* catch use after free bugs */
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2015-08-21 17:43:00 +00:00
|
|
|
* This is the fast-path for down_read/up_read. If it succeeds we rely
|
|
|
|
* on the barriers provided by rcu_sync_enter/exit; see the comments in
|
|
|
|
* percpu_down_write() and percpu_up_write().
|
2012-12-18 00:01:32 +00:00
|
|
|
*
|
|
|
|
* If this helper fails the callers rely on the normal rw_semaphore and
|
|
|
|
* atomic_dec_and_test(), so in this case we have the necessary barriers.
|
|
|
|
*/
|
|
|
|
static bool update_fast_ctr(struct percpu_rw_semaphore *brw, unsigned int val)
|
|
|
|
{
|
2015-08-21 17:42:57 +00:00
|
|
|
bool success;
|
2012-12-18 00:01:32 +00:00
|
|
|
|
|
|
|
preempt_disable();
|
2015-08-21 17:42:57 +00:00
|
|
|
success = rcu_sync_is_idle(&brw->rss);
|
|
|
|
if (likely(success))
|
2012-12-18 00:01:32 +00:00
|
|
|
__this_cpu_add(*brw->fast_read_ctr, val);
|
|
|
|
preempt_enable();
|
|
|
|
|
|
|
|
return success;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Like the normal down_read() this is not recursive, the writer can
|
|
|
|
* come after the first percpu_down_read() and create the deadlock.
|
2012-12-18 00:01:38 +00:00
|
|
|
*
|
|
|
|
* Note: returns with lock_is_held(brw->rw_sem) == T for lockdep,
|
|
|
|
* percpu_up_read() does rwsem_release(). This pairs with the usage
|
|
|
|
* of ->rw_sem in percpu_down/up_write().
|
2012-12-18 00:01:32 +00:00
|
|
|
*/
|
|
|
|
void percpu_down_read(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
2012-12-18 00:01:38 +00:00
|
|
|
might_sleep();
|
2015-08-21 17:43:03 +00:00
|
|
|
rwsem_acquire_read(&brw->rw_sem.dep_map, 0, 0, _RET_IP_);
|
|
|
|
|
|
|
|
if (likely(update_fast_ctr(brw, +1)))
|
2012-12-18 00:01:32 +00:00
|
|
|
return;
|
|
|
|
|
2015-08-21 17:43:03 +00:00
|
|
|
/* Avoid rwsem_acquire_read() and rwsem_release() */
|
|
|
|
__down_read(&brw->rw_sem);
|
2012-12-18 00:01:32 +00:00
|
|
|
atomic_inc(&brw->slow_read_ctr);
|
2012-12-18 00:01:38 +00:00
|
|
|
__up_read(&brw->rw_sem);
|
2012-12-18 00:01:32 +00:00
|
|
|
}
|
2015-09-01 03:21:59 +00:00
|
|
|
EXPORT_SYMBOL_GPL(percpu_down_read);
|
2012-12-18 00:01:32 +00:00
|
|
|
|
2015-07-21 15:45:57 +00:00
|
|
|
int percpu_down_read_trylock(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
|
|
|
if (unlikely(!update_fast_ctr(brw, +1))) {
|
|
|
|
if (!__down_read_trylock(&brw->rw_sem))
|
|
|
|
return 0;
|
|
|
|
atomic_inc(&brw->slow_read_ctr);
|
|
|
|
__up_read(&brw->rw_sem);
|
|
|
|
}
|
|
|
|
|
|
|
|
rwsem_acquire_read(&brw->rw_sem.dep_map, 0, 1, _RET_IP_);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2012-12-18 00:01:32 +00:00
|
|
|
void percpu_up_read(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
2012-12-18 00:01:38 +00:00
|
|
|
rwsem_release(&brw->rw_sem.dep_map, 1, _RET_IP_);
|
|
|
|
|
2012-12-18 00:01:32 +00:00
|
|
|
if (likely(update_fast_ctr(brw, -1)))
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* false-positive is possible but harmless */
|
|
|
|
if (atomic_dec_and_test(&brw->slow_read_ctr))
|
|
|
|
wake_up_all(&brw->write_waitq);
|
|
|
|
}
|
2015-09-01 03:21:59 +00:00
|
|
|
EXPORT_SYMBOL_GPL(percpu_up_read);
|
2012-12-18 00:01:32 +00:00
|
|
|
|
|
|
|
static int clear_fast_ctr(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
|
|
|
unsigned int sum = 0;
|
|
|
|
int cpu;
|
|
|
|
|
|
|
|
for_each_possible_cpu(cpu) {
|
|
|
|
sum += per_cpu(*brw->fast_read_ctr, cpu);
|
|
|
|
per_cpu(*brw->fast_read_ctr, cpu) = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return sum;
|
|
|
|
}
|
|
|
|
|
|
|
|
void percpu_down_write(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
|
|
|
/*
|
2015-08-21 17:43:00 +00:00
|
|
|
* Make rcu_sync_is_idle() == F and thus disable the fast-path in
|
|
|
|
* percpu_down_read() and percpu_up_read(), and wait for gp pass.
|
2012-12-18 00:01:32 +00:00
|
|
|
*
|
2015-08-21 17:43:00 +00:00
|
|
|
* The latter synchronises us with the preceding readers which used
|
|
|
|
* the fast-past, so we can not miss the result of __this_cpu_add()
|
|
|
|
* or anything else inside their criticial sections.
|
2012-12-18 00:01:32 +00:00
|
|
|
*/
|
2015-08-21 17:42:57 +00:00
|
|
|
rcu_sync_enter(&brw->rss);
|
2012-12-18 00:01:32 +00:00
|
|
|
|
2012-12-18 00:01:36 +00:00
|
|
|
/* exclude other writers, and block the new readers completely */
|
|
|
|
down_write(&brw->rw_sem);
|
|
|
|
|
2012-12-18 00:01:32 +00:00
|
|
|
/* nobody can use fast_read_ctr, move its sum into slow_read_ctr */
|
|
|
|
atomic_add(clear_fast_ctr(brw), &brw->slow_read_ctr);
|
|
|
|
|
|
|
|
/* wait for all readers to complete their percpu_up_read() */
|
|
|
|
wait_event(brw->write_waitq, !atomic_read(&brw->slow_read_ctr));
|
|
|
|
}
|
2015-09-01 03:21:59 +00:00
|
|
|
EXPORT_SYMBOL_GPL(percpu_down_write);
|
2012-12-18 00:01:32 +00:00
|
|
|
|
|
|
|
void percpu_up_write(struct percpu_rw_semaphore *brw)
|
|
|
|
{
|
2012-12-18 00:01:36 +00:00
|
|
|
/* release the lock, but the readers can't use the fast-path */
|
2012-12-18 00:01:32 +00:00
|
|
|
up_write(&brw->rw_sem);
|
|
|
|
/*
|
2015-08-21 17:43:00 +00:00
|
|
|
* Enable the fast-path in percpu_down_read() and percpu_up_read()
|
|
|
|
* but only after another gp pass; this adds the necessary barrier
|
|
|
|
* to ensure the reader can't miss the changes done by us.
|
2012-12-18 00:01:32 +00:00
|
|
|
*/
|
2015-08-21 17:42:57 +00:00
|
|
|
rcu_sync_exit(&brw->rss);
|
2012-12-18 00:01:32 +00:00
|
|
|
}
|
2015-09-01 03:21:59 +00:00
|
|
|
EXPORT_SYMBOL_GPL(percpu_up_write);
|