linux/lib/percpu-refcount.c
Tejun Heo a4244454df percpu-refcount: use RCU-sched insted of normal RCU
percpu-refcount was incorrectly using preempt_disable/enable() for RCU
critical sections against call_rcu().  6a24474da8 ("percpu-refcount:
consistently use plain (non-sched) RCU") fixed it by converting the
preepmtion operations with rcu_read_[un]lock() citing that there isn't
any advantage in using sched-RCU over using the usual one; however,
rcu_read_[un]lock() for the preemptible RCU implementation -
CONFIG_TREE_PREEMPT_RCU, chosen when CONFIG_PREEMPT - are slightly
more expensive than preempt_disable/enable().

In a contrived microbench which repeats the followings,

 - percpu_ref_get()
 - copy 32 bytes of data into percpu buffer
 - percpu_put_get()
 - copy 32 bytes of data into percpu buffer

rcu_read_[un]lock() used in percpu_ref_get/put() makes it go slower by
about 15% when compared to using sched-RCU.

As the RCU critical sections are extremely short, using sched-RCU
shouldn't have any latency implications.  Convert to RCU-sched.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Kent Overstreet <koverstreet@google.com>
Acked-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Rusty Russell <rusty@rustcorp.com.au>
2013-06-16 16:12:26 -07:00

159 lines
5.5 KiB
C

#define pr_fmt(fmt) "%s: " fmt "\n", __func__
#include <linux/kernel.h>
#include <linux/percpu-refcount.h>
/*
* Initially, a percpu refcount is just a set of percpu counters. Initially, we
* don't try to detect the ref hitting 0 - which means that get/put can just
* increment or decrement the local counter. Note that the counter on a
* particular cpu can (and will) wrap - this is fine, when we go to shutdown the
* percpu counters will all sum to the correct value
*
* (More precisely: because moduler arithmatic is commutative the sum of all the
* pcpu_count vars will be equal to what it would have been if all the gets and
* puts were done to a single integer, even if some of the percpu integers
* overflow or underflow).
*
* The real trick to implementing percpu refcounts is shutdown. We can't detect
* the ref hitting 0 on every put - this would require global synchronization
* and defeat the whole purpose of using percpu refs.
*
* What we do is require the user to keep track of the initial refcount; we know
* the ref can't hit 0 before the user drops the initial ref, so as long as we
* convert to non percpu mode before the initial ref is dropped everything
* works.
*
* Converting to non percpu mode is done with some RCUish stuff in
* percpu_ref_kill. Additionally, we need a bias value so that the atomic_t
* can't hit 0 before we've added up all the percpu refs.
*/
#define PCPU_COUNT_BIAS (1U << 31)
/**
* percpu_ref_init - initialize a percpu refcount
* @ref: percpu_ref to initialize
* @release: function which will be called when refcount hits 0
*
* Initializes the refcount in single atomic counter mode with a refcount of 1;
* analagous to atomic_set(ref, 1).
*
* Note that @release must not sleep - it may potentially be called from RCU
* callback context by percpu_ref_kill().
*/
int percpu_ref_init(struct percpu_ref *ref, percpu_ref_func_t *release)
{
atomic_set(&ref->count, 1 + PCPU_COUNT_BIAS);
ref->pcpu_count = alloc_percpu(unsigned);
if (!ref->pcpu_count)
return -ENOMEM;
ref->release = release;
return 0;
}
/**
* percpu_ref_cancel_init - cancel percpu_ref_init()
* @ref: percpu_ref to cancel init for
*
* Once a percpu_ref is initialized, its destruction is initiated by
* percpu_ref_kill() and completes asynchronously, which can be painful to
* do when destroying a half-constructed object in init failure path.
*
* This function destroys @ref without invoking @ref->release and the
* memory area containing it can be freed immediately on return. To
* prevent accidental misuse, it's required that @ref has finished
* percpu_ref_init(), whether successful or not, but never used.
*
* The weird name and usage restriction are to prevent people from using
* this function by mistake for normal shutdown instead of
* percpu_ref_kill().
*/
void percpu_ref_cancel_init(struct percpu_ref *ref)
{
unsigned __percpu *pcpu_count = ref->pcpu_count;
int cpu;
WARN_ON_ONCE(atomic_read(&ref->count) != 1 + PCPU_COUNT_BIAS);
if (pcpu_count) {
for_each_possible_cpu(cpu)
WARN_ON_ONCE(*per_cpu_ptr(pcpu_count, cpu));
free_percpu(ref->pcpu_count);
}
}
static void percpu_ref_kill_rcu(struct rcu_head *rcu)
{
struct percpu_ref *ref = container_of(rcu, struct percpu_ref, rcu);
unsigned __percpu *pcpu_count = ref->pcpu_count;
unsigned count = 0;
int cpu;
/* Mask out PCPU_REF_DEAD */
pcpu_count = (unsigned __percpu *)
(((unsigned long) pcpu_count) & ~PCPU_STATUS_MASK);
for_each_possible_cpu(cpu)
count += *per_cpu_ptr(pcpu_count, cpu);
free_percpu(pcpu_count);
pr_debug("global %i pcpu %i", atomic_read(&ref->count), (int) count);
/*
* It's crucial that we sum the percpu counters _before_ adding the sum
* to &ref->count; since gets could be happening on one cpu while puts
* happen on another, adding a single cpu's count could cause
* @ref->count to hit 0 before we've got a consistent value - but the
* sum of all the counts will be consistent and correct.
*
* Subtracting the bias value then has to happen _after_ adding count to
* &ref->count; we need the bias value to prevent &ref->count from
* reaching 0 before we add the percpu counts. But doing it at the same
* time is equivalent and saves us atomic operations:
*/
atomic_add((int) count - PCPU_COUNT_BIAS, &ref->count);
/* @ref is viewed as dead on all CPUs, send out kill confirmation */
if (ref->confirm_kill)
ref->confirm_kill(ref);
/*
* Now we're in single atomic_t mode with a consistent refcount, so it's
* safe to drop our initial ref:
*/
percpu_ref_put(ref);
}
/**
* percpu_ref_kill_and_confirm - drop the initial ref and schedule confirmation
* @ref: percpu_ref to kill
* @confirm_kill: optional confirmation callback
*
* Equivalent to percpu_ref_kill() but also schedules kill confirmation if
* @confirm_kill is not NULL. @confirm_kill, which may not block, will be
* called after @ref is seen as dead from all CPUs - all further
* invocations of percpu_ref_tryget() will fail. See percpu_ref_tryget()
* for more details.
*
* Due to the way percpu_ref is implemented, @confirm_kill will be called
* after at least one full RCU grace period has passed but this is an
* implementation detail and callers must not depend on it.
*/
void percpu_ref_kill_and_confirm(struct percpu_ref *ref,
percpu_ref_func_t *confirm_kill)
{
WARN_ONCE(REF_STATUS(ref->pcpu_count) == PCPU_REF_DEAD,
"percpu_ref_kill() called more than once!\n");
ref->pcpu_count = (unsigned __percpu *)
(((unsigned long) ref->pcpu_count)|PCPU_REF_DEAD);
ref->confirm_kill = confirm_kill;
call_rcu_sched(&ref->rcu, percpu_ref_kill_rcu);
}