linux_dsm_epyc7002/kernel/rcu/sync.c

227 lines
6.5 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0+
/*
* RCU-based infrastructure for lightweight reader-writer locking
*
* Copyright (c) 2015, Red Hat, Inc.
*
* Author: Oleg Nesterov <oleg@redhat.com>
*/
#include <linux/rcu_sync.h>
#include <linux/sched.h>
#ifdef CONFIG_PROVE_RCU
#define __INIT_HELD(func) .held = func,
#else
#define __INIT_HELD(func)
#endif
static const struct {
void (*sync)(void);
void (*call)(struct rcu_head *, void (*)(struct rcu_head *));
void (*wait)(void);
#ifdef CONFIG_PROVE_RCU
int (*held)(void);
#endif
} gp_ops[] = {
[RCU_SYNC] = {
.sync = synchronize_rcu,
.call = call_rcu,
.wait = rcu_barrier,
__INIT_HELD(rcu_read_lock_held)
},
[RCU_SCHED_SYNC] = {
.sync = synchronize_rcu,
.call = call_rcu,
.wait = rcu_barrier,
__INIT_HELD(rcu_read_lock_sched_held)
},
[RCU_BH_SYNC] = {
.sync = synchronize_rcu,
.call = call_rcu,
.wait = rcu_barrier,
__INIT_HELD(rcu_read_lock_bh_held)
},
};
enum { GP_IDLE = 0, GP_PENDING, GP_PASSED };
enum { CB_IDLE = 0, CB_PENDING, CB_REPLAY };
#define rss_lock gp_wait.lock
#ifdef CONFIG_PROVE_RCU
void rcu_sync_lockdep_assert(struct rcu_sync *rsp)
{
RCU_LOCKDEP_WARN(!gp_ops[rsp->gp_type].held(),
"suspicious rcu_sync_is_idle() usage");
}
EXPORT_SYMBOL_GPL(rcu_sync_lockdep_assert);
#endif
/**
* rcu_sync_init() - Initialize an rcu_sync structure
* @rsp: Pointer to rcu_sync structure to be initialized
* @type: Flavor of RCU with which to synchronize rcu_sync structure
*/
void rcu_sync_init(struct rcu_sync *rsp, enum rcu_sync_type type)
{
memset(rsp, 0, sizeof(*rsp));
init_waitqueue_head(&rsp->gp_wait);
rsp->gp_type = type;
}
locking, rcu, cgroup: Avoid synchronize_sched() in __cgroup_procs_write() The current percpu-rwsem read side is entirely free of serializing insns at the cost of having a synchronize_sched() in the write path. The latency of the synchronize_sched() is too high for cgroups. The commit 1ed1328792ff talks about the write path being a fairly cold path but this is not the case for Android which moves task to the foreground cgroup and back around binder IPC calls from foreground processes to background processes, so it is significantly hotter than human initiated operations. Switch cgroup_threadgroup_rwsem into the slow mode for now to avoid the problem, hopefully it should not be that slow after another commit: 80127a39681b ("locking/percpu-rwsem: Optimize readers and reduce global impact"). We could just add rcu_sync_enter() into cgroup_init() but we do not want another synchronize_sched() at boot time, so this patch adds the new helper which doesn't block but currently can only be called before the first use. Reported-by: John Stultz <john.stultz@linaro.org> Reported-by: Dmitry Shmidt <dimitrysh@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Colin Cross <ccross@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rom Lemarchand <romlem@google.com> Cc: Tejun Heo <tj@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Todd Kjos <tkjos@google.com> Link: http://lkml.kernel.org/r/20160811165413.GA22807@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-11 23:54:13 +07:00
/**
* rcu_sync_enter_start - Force readers onto slow path for multiple updates
* @rsp: Pointer to rcu_sync structure to use for synchronization
*
locking, rcu, cgroup: Avoid synchronize_sched() in __cgroup_procs_write() The current percpu-rwsem read side is entirely free of serializing insns at the cost of having a synchronize_sched() in the write path. The latency of the synchronize_sched() is too high for cgroups. The commit 1ed1328792ff talks about the write path being a fairly cold path but this is not the case for Android which moves task to the foreground cgroup and back around binder IPC calls from foreground processes to background processes, so it is significantly hotter than human initiated operations. Switch cgroup_threadgroup_rwsem into the slow mode for now to avoid the problem, hopefully it should not be that slow after another commit: 80127a39681b ("locking/percpu-rwsem: Optimize readers and reduce global impact"). We could just add rcu_sync_enter() into cgroup_init() but we do not want another synchronize_sched() at boot time, so this patch adds the new helper which doesn't block but currently can only be called before the first use. Reported-by: John Stultz <john.stultz@linaro.org> Reported-by: Dmitry Shmidt <dimitrysh@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Colin Cross <ccross@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rom Lemarchand <romlem@google.com> Cc: Tejun Heo <tj@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Todd Kjos <tkjos@google.com> Link: http://lkml.kernel.org/r/20160811165413.GA22807@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-11 23:54:13 +07:00
* Must be called after rcu_sync_init() and before first use.
*
* Ensures rcu_sync_is_idle() returns false and rcu_sync_{enter,exit}()
* pairs turn into NO-OPs.
*/
void rcu_sync_enter_start(struct rcu_sync *rsp)
{
rsp->gp_count++;
rsp->gp_state = GP_PASSED;
}
/**
* rcu_sync_enter() - Force readers onto slowpath
* @rsp: Pointer to rcu_sync structure to use for synchronization
*
* This function is used by updaters who need readers to make use of
* a slowpath during the update. After this function returns, all
* subsequent calls to rcu_sync_is_idle() will return false, which
* tells readers to stay off their fastpaths. A later call to
* rcu_sync_exit() re-enables reader slowpaths.
*
* When called in isolation, rcu_sync_enter() must wait for a grace
* period, however, closely spaced calls to rcu_sync_enter() can
* optimize away the grace-period wait via a state machine implemented
* by rcu_sync_enter(), rcu_sync_exit(), and rcu_sync_func().
*/
void rcu_sync_enter(struct rcu_sync *rsp)
{
bool need_wait, need_sync;
spin_lock_irq(&rsp->rss_lock);
need_wait = rsp->gp_count++;
need_sync = rsp->gp_state == GP_IDLE;
if (need_sync)
rsp->gp_state = GP_PENDING;
spin_unlock_irq(&rsp->rss_lock);
WARN_ON_ONCE(need_wait && need_sync);
if (need_sync) {
gp_ops[rsp->gp_type].sync();
rsp->gp_state = GP_PASSED;
wake_up_all(&rsp->gp_wait);
} else if (need_wait) {
wait_event(rsp->gp_wait, rsp->gp_state == GP_PASSED);
} else {
/*
* Possible when there's a pending CB from a rcu_sync_exit().
* Nobody has yet been allowed the 'fast' path and thus we can
* avoid doing any sync(). The callback will get 'dropped'.
*/
WARN_ON_ONCE(rsp->gp_state != GP_PASSED);
}
}
/**
* rcu_sync_func() - Callback function managing reader access to fastpath
* @rhp: Pointer to rcu_head in rcu_sync structure to use for synchronization
*
* This function is passed to one of the call_rcu() functions by
* rcu_sync_exit(), so that it is invoked after a grace period following the
* that invocation of rcu_sync_exit(). It takes action based on events that
* have taken place in the meantime, so that closely spaced rcu_sync_enter()
* and rcu_sync_exit() pairs need not wait for a grace period.
*
* If another rcu_sync_enter() is invoked before the grace period
* ended, reset state to allow the next rcu_sync_exit() to let the
* readers back onto their fastpaths (after a grace period). If both
* another rcu_sync_enter() and its matching rcu_sync_exit() are invoked
* before the grace period ended, re-invoke call_rcu() on behalf of that
* rcu_sync_exit(). Otherwise, set all state back to idle so that readers
* can again use their fastpaths.
*/
static void rcu_sync_func(struct rcu_head *rhp)
{
struct rcu_sync *rsp = container_of(rhp, struct rcu_sync, cb_head);
unsigned long flags;
WARN_ON_ONCE(rsp->gp_state != GP_PASSED);
WARN_ON_ONCE(rsp->cb_state == CB_IDLE);
spin_lock_irqsave(&rsp->rss_lock, flags);
if (rsp->gp_count) {
/*
* A new rcu_sync_begin() has happened; drop the callback.
*/
rsp->cb_state = CB_IDLE;
} else if (rsp->cb_state == CB_REPLAY) {
/*
* A new rcu_sync_exit() has happened; requeue the callback
* to catch a later GP.
*/
rsp->cb_state = CB_PENDING;
gp_ops[rsp->gp_type].call(&rsp->cb_head, rcu_sync_func);
} else {
/*
* We're at least a GP after rcu_sync_exit(); eveybody will now
* have observed the write side critical section. Let 'em rip!.
*/
rsp->cb_state = CB_IDLE;
rsp->gp_state = GP_IDLE;
}
spin_unlock_irqrestore(&rsp->rss_lock, flags);
}
/**
* rcu_sync_exit() - Allow readers back onto fast patch after grace period
* @rsp: Pointer to rcu_sync structure to use for synchronization
*
* This function is used by updaters who have completed, and can therefore
* now allow readers to make use of their fastpaths after a grace period
* has elapsed. After this grace period has completed, all subsequent
* calls to rcu_sync_is_idle() will return true, which tells readers that
* they can once again use their fastpaths.
*/
void rcu_sync_exit(struct rcu_sync *rsp)
{
spin_lock_irq(&rsp->rss_lock);
if (!--rsp->gp_count) {
if (rsp->cb_state == CB_IDLE) {
rsp->cb_state = CB_PENDING;
gp_ops[rsp->gp_type].call(&rsp->cb_head, rcu_sync_func);
} else if (rsp->cb_state == CB_PENDING) {
rsp->cb_state = CB_REPLAY;
}
}
spin_unlock_irq(&rsp->rss_lock);
}
/**
* rcu_sync_dtor() - Clean up an rcu_sync structure
* @rsp: Pointer to rcu_sync structure to be cleaned up
*/
void rcu_sync_dtor(struct rcu_sync *rsp)
{
int cb_state;
WARN_ON_ONCE(rsp->gp_count);
spin_lock_irq(&rsp->rss_lock);
if (rsp->cb_state == CB_REPLAY)
rsp->cb_state = CB_PENDING;
cb_state = rsp->cb_state;
spin_unlock_irq(&rsp->rss_lock);
if (cb_state != CB_IDLE) {
gp_ops[rsp->gp_type].wait();
WARN_ON_ONCE(rsp->cb_state != CB_IDLE);
}
}