/* * __do_six_trylock() - main trylock routine * * Returns 1 on success, 0 on failure * * In percpu reader mode, a failed trylock may cause a spurious trylock failure * for anoter thread taking the competing lock type, and we may havve to do a * wakeup: when a wakeup is required, we return -1 - wakeup_type.
*/ staticint __do_six_trylock(struct six_lock *lock, enum six_lock_type type, struct task_struct *task, booltry)
{ int ret;
u32 old;
/* * Percpu reader mode: * * The basic idea behind this algorithm is that you can implement a lock * between two threads without any atomics, just memory barriers: * * For two threads you'll need two variables, one variable for "thread a * has the lock" and another for "thread b has the lock". * * To take the lock, a thread sets its variable indicating that it holds * the lock, then issues a full memory barrier, then reads from the * other thread's variable to check if the other thread thinks it has * the lock. If we raced, we backoff and retry/sleep. * * Failure to take the lock may cause a spurious trylock failure in * another thread, because we temporarily set the lock to indicate that * we held it. This would be a problem for a thread in six_lock(), when * they are calling trylock after adding themself to the waitlist and * prior to sleeping. * * Therefore, if we fail to get the lock, and there were waiters of the * type we conflict with, we will have to issue a wakeup. * * Since we may be called under wait_lock (and by the wakeup code * itself), we return that the wakeup has to be done instead of doing it * here.
*/ if (type == SIX_LOCK_read && lock->readers) {
preempt_disable();
this_cpu_inc(*lock->readers); /* signal that we own lock */
smp_mb();
old = atomic_read(&lock->state);
ret = !(old & l[type].lock_fail);
if (!ret) {
smp_mb(); if (atomic_read(&lock->state) & SIX_LOCK_WAITING_write)
ret = -1 - SIX_LOCK_write;
}
} elseif (type == SIX_LOCK_write && lock->readers) { if (try)
atomic_add(SIX_LOCK_HELD_write, &lock->state);
/* * Make sure atomic_add happens before pcpu_read_count and * six_set_bitmask in slow path happens before pcpu_read_count. * * Paired with the smp_mb() in read lock fast path (per-cpu mode) * and the one before atomic_read in read unlock path.
*/
smp_mb();
ret = !pcpu_read_count(lock);
if (try && !ret) {
old = atomic_sub_return(SIX_LOCK_HELD_write, &lock->state); if (old & SIX_LOCK_WAITING_read)
ret = -1 - SIX_LOCK_read;
}
} else {
old = atomic_read(&lock->state); do {
ret = !(old & l[type].lock_fail); if (!ret || (type == SIX_LOCK_write && !try)) {
smp_mb(); break;
}
} while (!atomic_try_cmpxchg_acquire(&lock->state, &old, old + l[type].lock_val));
ret = __do_six_trylock(lock, lock_type, w->task, false); if (ret <= 0) goto unlock;
/* * Similar to percpu_rwsem_wake_function(), we need to guard * against the wakee noticing w->lock_acquired, returning, and * then exiting before we do the wakeup:
*/
task = get_task_struct(w->task);
__list_del(w->list.prev, w->list.next); /* * The release barrier here ensures the ordering of the * __list_del before setting w->lock_acquired; @w is on the * stack of the thread doing the waiting and will be reused * after it sees w->lock_acquired with no other locking: * pairs with smp_load_acquire() in six_lock_slowpath()
*/
smp_store_release(&w->lock_acquired, true);
wake_up_process(task);
put_task_struct(task);
}
ret = __do_six_trylock(lock, type, current, try); if (ret < 0)
__six_lock_wakeup(lock, -ret - 1);
return ret > 0;
}
/** * six_trylock_ip - attempt to take a six lock without blocking * @lock: lock to take * @type: SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write * @ip: ip parameter for lockdep/lockstat, i.e. _THIS_IP_ * * Return: true on success, false on failure.
*/ bool six_trylock_ip(struct six_lock *lock, enum six_lock_type type, unsignedlong ip)
{ if (!do_six_trylock(lock, type, true)) returnfalse;
if (type != SIX_LOCK_write)
six_acquire(&lock->dep_map, 1, type == SIX_LOCK_read, ip); returntrue;
}
EXPORT_SYMBOL_GPL(six_trylock_ip);
/** * six_relock_ip - attempt to re-take a lock that was held previously * @lock: lock to take * @type: SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write * @seq: lock sequence number obtained from six_lock_seq() while lock was * held previously * @ip: ip parameter for lockdep/lockstat, i.e. _THIS_IP_ * * Return: true on success, false on failure.
*/ bool six_relock_ip(struct six_lock *lock, enum six_lock_type type, unsigned seq, unsignedlong ip)
{ if (six_lock_seq(lock) != seq || !six_trylock_ip(lock, type, ip)) returnfalse;
if (six_lock_seq(lock) != seq) {
six_unlock_ip(lock, type, ip); returnfalse;
}
returntrue;
}
EXPORT_SYMBOL_GPL(six_relock_ip);
#ifdef CONFIG_BCACHEFS_SIX_OPTIMISTIC_SPIN
staticinlinebool six_owner_running(struct six_lock *lock)
{ /* * When there's no owner, we might have preempted between the owner * acquiring the lock and setting the owner field. If we're an RT task * that will live-lock because we won't let the owner complete.
*/
guard(rcu)(); struct task_struct *owner = READ_ONCE(lock->owner); return owner ? owner_on_cpu(owner) : !rt_or_dl_task(current);
}
while (!need_resched() && six_owner_running(lock)) { /* * Ensures that writes to the waitlist entry happen after we see * wait->lock_acquired: pairs with the smp_store_release in * __six_lock_wakeup
*/ if (smp_load_acquire(&wait->lock_acquired)) {
preempt_enable(); returntrue;
}
/* * The cpu_relax() call is a compiler barrier which forces * everything in this loop to be re-loaded. We don't need * memory barriers as we'll eventually observe the right * values at the cost of a few extra spins.
*/
cpu_relax();
}
raw_spin_lock(&lock->wait_lock);
six_set_bitmask(lock, SIX_LOCK_WAITING_read << type); /* * Retry taking the lock after taking waitlist lock, in case we raced * with an unlock:
*/
ret = __do_six_trylock(lock, type, current, false); if (ret <= 0) {
wait->start_time = local_clock();
if (unlikely(ret < 0)) {
__six_lock_wakeup(lock, -ret - 1);
ret = 0;
}
if (six_optimistic_spin(lock, wait, type)) goto out;
while (1) {
set_current_state(TASK_UNINTERRUPTIBLE);
/* * Ensures that writes to the waitlist entry happen after we see * wait->lock_acquired: pairs with the smp_store_release in * __six_lock_wakeup
*/ if (smp_load_acquire(&wait->lock_acquired)) break;
ret = should_sleep_fn ? should_sleep_fn(lock, p) : 0; if (unlikely(ret)) { bool acquired;
/* * If should_sleep_fn() returns an error, we are * required to return that error even if we already * acquired the lock - should_sleep_fn() might have * modified external state (e.g. when the deadlock cycle * detector in bcachefs issued a transaction restart)
*/
raw_spin_lock(&lock->wait_lock);
acquired = wait->lock_acquired; if (!acquired)
list_del(&wait->list);
raw_spin_unlock(&lock->wait_lock);
/** * six_lock_ip_waiter - take a lock, with full waitlist interface * @lock: lock to take * @type: SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write * @wait: pointer to wait object, which will be added to lock's waitlist * @should_sleep_fn: callback run after adding to waitlist, immediately prior * to scheduling * @p: passed through to @should_sleep_fn * @ip: ip parameter for lockdep/lockstat, i.e. _THIS_IP_ * * This is the most general six_lock() variant, with parameters to support full * cycle detection for deadlock avoidance. * * The code calling this function must implement tracking of held locks, and the * @wait object should be embedded into the struct that tracks held locks - * which must also be accessible in a thread-safe way. * * @should_sleep_fn should invoke the cycle detector; it should walk each * lock's waiters, and for each waiter recursively walk their held locks. * * When this function must block, @wait will be added to @lock's waitlist before * calling trylock, and before calling @should_sleep_fn, and @wait will not be * removed from the lock waitlist until the lock has been successfully acquired, * or we abort. * * @wait.start_time will be monotonically increasing for any given waitlist, and * thus may be used as a loop cursor. * * Return: 0 on success, or the return code from @should_sleep_fn on failure.
*/ int six_lock_ip_waiter(struct six_lock *lock, enum six_lock_type type, struct six_lock_waiter *wait,
six_lock_should_sleep_fn should_sleep_fn, void *p, unsignedlong ip)
{ int ret;
wait->start_time = 0;
if (type != SIX_LOCK_write)
six_acquire(&lock->dep_map, 0, type == SIX_LOCK_read, ip);
/** * six_lock_downgrade - convert an intent lock to a read lock * @lock: lock to dowgrade * * @lock will have read count incremented and intent count decremented
*/ void six_lock_downgrade(struct six_lock *lock)
{
six_lock_increment(lock, SIX_LOCK_read);
six_unlock_intent(lock);
}
EXPORT_SYMBOL_GPL(six_lock_downgrade);
/** * six_lock_tryupgrade - attempt to convert read lock to an intent lock * @lock: lock to upgrade * * On success, @lock will have intent count incremented and read count * decremented * * Return: true on success, false on failure
*/ bool six_lock_tryupgrade(struct six_lock *lock)
{
u32 old = atomic_read(&lock->state), new;
do { new = old;
if (new & SIX_LOCK_HELD_intent) returnfalse;
if (!lock->readers) {
EBUG_ON(!(new & SIX_LOCK_HELD_read)); new -= l[SIX_LOCK_read].lock_val;
}
new |= SIX_LOCK_HELD_intent;
} while (!atomic_try_cmpxchg_acquire(&lock->state, &old, new));
/** * six_trylock_convert - attempt to convert a held lock from one type to another * @lock: lock to upgrade * @from: SIX_LOCK_read or SIX_LOCK_intent * @to: SIX_LOCK_read or SIX_LOCK_intent * * On success, @lock will have intent count incremented and read count * decremented * * Return: true on success, false on failure
*/ bool six_trylock_convert(struct six_lock *lock, enum six_lock_type from, enum six_lock_type to)
{
EBUG_ON(to == SIX_LOCK_write || from == SIX_LOCK_write);
/** * six_lock_increment - increase held lock count on a lock that is already held * @lock: lock to increment * @type: SIX_LOCK_read or SIX_LOCK_intent * * @lock must already be held, with a lock type that is greater than or equal to * @type * * A corresponding six_unlock_type() call will be required for @lock to be fully * unlocked.
*/ void six_lock_increment(struct six_lock *lock, enum six_lock_type type)
{
six_acquire(&lock->dep_map, 0, type == SIX_LOCK_read, _RET_IP_);
/* XXX: assert already locked, and that we don't overflow: */
switch (type) { case SIX_LOCK_read: if (lock->readers) {
this_cpu_inc(*lock->readers);
} else {
EBUG_ON(!(atomic_read(&lock->state) &
(SIX_LOCK_HELD_read|
SIX_LOCK_HELD_intent)));
atomic_add(l[type].lock_val, &lock->state);
} break; case SIX_LOCK_write:
lock->write_lock_recurse++;
fallthrough; case SIX_LOCK_intent:
EBUG_ON(!(atomic_read(&lock->state) & SIX_LOCK_HELD_intent));
lock->intent_lock_recurse++; break;
}
}
EXPORT_SYMBOL_GPL(six_lock_increment);
/** * six_lock_wakeup_all - wake up all waiters on @lock * @lock: lock to wake up waiters for * * Wakeing up waiters will cause them to re-run should_sleep_fn, which may then * abort the lock operation. * * This function is never needed in a bug-free program; it's only useful in * debug code, e.g. to determine if a cycle detector is at fault.
*/ void six_lock_wakeup_all(struct six_lock *lock)
{
u32 state = atomic_read(&lock->state); struct six_lock_waiter *w;
/** * six_lock_counts - return held lock counts, for each lock type * @lock: lock to return counters for * * Return: the number of times a lock is held for read, intent and write.
*/ struct six_lock_count six_lock_counts(struct six_lock *lock)
{ struct six_lock_count ret;
/** * six_lock_readers_add - directly manipulate reader count of a lock * @lock: lock to add/subtract readers for * @nr: reader count to add/subtract * * When an upper layer is implementing lock reentrency, we may have both read * and intent locks on the same lock. * * When we need to take a write lock, the read locks will cause self-deadlock, * because six locks themselves do not track which read locks are held by the * current thread and which are held by a different thread - it does no * per-thread tracking of held locks. * * The upper layer that is tracking held locks may however, if trylock() has * failed, count up its own read locks, subtract them, take the write lock, and * then re-add them. * * As in any other situation when taking a write lock, @lock must be held for * intent one (or more) times, so @lock will never be left unlocked.
*/ void six_lock_readers_add(struct six_lock *lock, int nr)
{ if (lock->readers) {
this_cpu_add(*lock->readers, nr);
} else {
EBUG_ON((int) (atomic_read(&lock->state) & SIX_LOCK_HELD_read) + nr < 0); /* reader count starts at bit 0 */
atomic_add(nr, &lock->state);
}
}
EXPORT_SYMBOL_GPL(six_lock_readers_add);
/** * six_lock_exit - release resources held by a lock prior to freeing * @lock: lock to exit * * When a lock was initialized in percpu mode (SIX_OLCK_INIT_PCPU), this is * required to free the percpu read counts.
*/ void six_lock_exit(struct six_lock *lock)
{
WARN_ON(lock->readers && pcpu_read_count(lock));
WARN_ON(atomic_read(&lock->state) & SIX_LOCK_HELD_read);
/* * Don't assume that we have real percpu variables available in * userspace:
*/ #ifdef __KERNEL__ if (flags & SIX_LOCK_INIT_PCPU) { /* * We don't return an error here on memory allocation failure * since percpu is an optimization, and locks will work with the * same semantics in non-percpu mode: callers can check for * failure if they wish by checking lock->readers, but generally * will not want to treat it as an error.
*/
lock->readers = alloc_percpu_gfp(unsigned, gfp);
} #endif
}
EXPORT_SYMBOL_GPL(__six_lock_init);
Messung V0.5
¤ Dauer der Verarbeitung: 0.36 Sekunden
(vorverarbeitet)
¤
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.