6.1-stable review patch. If anyone has any objections, please let me know.
------------------
From: Nam Cao namcao@linutronix.de
[ Upstream commit 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd ]
The ready event list of an epoll object is protected by read-write semaphore:
- The consumer (waiter) acquires the write lock and takes items. - the producer (waker) takes the read lock and adds items.
The point of this design is enabling epoll to scale well with large number of producers, as multiple producers can hold the read lock at the same time.
Unfortunately, this implementation may cause scheduling priority inversion problem. Suppose the consumer has higher scheduling priority than the producer. The consumer needs to acquire the write lock, but may be blocked by the producer holding the read lock. Since read-write semaphore does not support priority-boosting for the readers (even with CONFIG_PREEMPT_RT=y), we have a case of priority inversion: a higher priority consumer is blocked by a lower priority producer. This problem was reported in [1].
Furthermore, this could also cause stall problem, as described in [2].
Fix this problem by replacing rwlock with spinlock.
This reduces the event bandwidth, as the producers now have to contend with each other for the spinlock. According to the benchmark from https://github.com/rouming/test-tools/blob/master/stress-epoll.c:
On 12 x86 CPUs: Before After Diff threads events/ms events/ms 8 7162 4956 -31% 16 8733 5383 -38% 32 7968 5572 -30% 64 10652 5739 -46% 128 11236 5931 -47%
On 4 riscv CPUs: Before After Diff threads events/ms events/ms 8 2958 2833 -4% 16 3323 3097 -7% 32 3451 3240 -6% 64 3554 3178 -11% 128 3601 3235 -10%
Although the numbers look bad, it should be noted that this benchmark creates multiple threads who do nothing except constantly generating new epoll events, thus contention on the spinlock is high. For real workload, the event rate is likely much lower, and the performance drop is not as bad.
Using another benchmark (perf bench epoll wait) where spinlock contention is lower, improvement is even observed on x86:
On 12 x86 CPUs: Before: Averaged 110279 operations/sec (+- 1.09%), total secs = 8 After: Averaged 114577 operations/sec (+- 2.25%), total secs = 8
On 4 riscv CPUs: Before: Averaged 175767 operations/sec (+- 0.62%), total secs = 8 After: Averaged 167396 operations/sec (+- 0.23%), total secs = 8
In conclusion, no one is likely to be upset over this change. After all, spinlock was used originally for years, and the commit which converted to rwlock didn't mention a real workload, just that the benchmark numbers are nice.
This patch is not exactly the revert of commit a218cc491420 ("epoll: use rwlock in order to reduce ep_poll_callback() contention"), because git revert conflicts in some places which are not obvious on the resolution. This patch is intended to be backported, therefore go with the obvious approach:
- Replace rwlock_t with spinlock_t one to one
- Delete list_add_tail_lockless() and chain_epi_lockless(). These were introduced to allow producers to concurrently add items to the list. But now that spinlock no longer allows producers to touch the event list concurrently, these two functions are not necessary anymore.
Fixes: a218cc491420 ("epoll: use rwlock in order to reduce ep_poll_callback() contention") Signed-off-by: Nam Cao namcao@linutronix.de Link: https://lore.kernel.org/ec92458ea357ec503c737ead0f10b2c6e4c37d47.1752581388.... Tested-by: K Prateek Nayak kprateek.nayak@amd.com Cc: stable@vger.kernel.org Reported-by: Frederic Weisbecker frederic@kernel.org Closes: https://lore.kernel.org/linux-rt-users/20210825132754.GA895675@lothringen/ [1] Reported-by: Valentin Schneider vschneid@redhat.com Closes: https://lore.kernel.org/linux-rt-users/xhsmhttqvnall.mognet@vschneid.remote.... [2] Signed-off-by: Christian Brauner brauner@kernel.org Signed-off-by: Florian Bezdeka florian.bezdeka@siemens.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- fs/eventpoll.c | 139 ++++++++++----------------------------------------------- 1 file changed, 26 insertions(+), 113 deletions(-)
--- a/fs/eventpoll.c +++ b/fs/eventpoll.c @@ -45,10 +45,10 @@ * * 1) epmutex (mutex) * 2) ep->mtx (mutex) - * 3) ep->lock (rwlock) + * 3) ep->lock (spinlock) * * The acquire order is the one listed above, from 1 to 3. - * We need a rwlock (ep->lock) because we manipulate objects + * We need a spinlock (ep->lock) because we manipulate objects * from inside the poll callback, that might be triggered from * a wake_up() that in turn might be called from IRQ context. * So we can't sleep inside the poll callback and hence we need @@ -193,7 +193,7 @@ struct eventpoll { struct list_head rdllist;
/* Lock which protects rdllist and ovflist */ - rwlock_t lock; + spinlock_t lock;
/* RB tree root used to store monitored fd structs */ struct rb_root_cached rbr; @@ -621,10 +621,10 @@ static void ep_start_scan(struct eventpo * in a lockless way. */ lockdep_assert_irqs_enabled(); - write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock); list_splice_init(&ep->rdllist, txlist); WRITE_ONCE(ep->ovflist, NULL); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); }
static void ep_done_scan(struct eventpoll *ep, @@ -632,7 +632,7 @@ static void ep_done_scan(struct eventpol { struct epitem *epi, *nepi;
- write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock); /* * During the time we spent inside the "sproc" callback, some * other events might have been queued by the poll callback. @@ -673,7 +673,7 @@ static void ep_done_scan(struct eventpol wake_up(&ep->wq); }
- write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); }
static void epi_rcu_free(struct rcu_head *head) @@ -719,10 +719,10 @@ static int ep_remove(struct eventpoll *e
rb_erase_cached(&epi->rbn, &ep->rbr);
- write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock); if (ep_is_linked(epi)) list_del_init(&epi->rdllink); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock);
wakeup_source_unregister(ep_wakeup_source(epi)); /* @@ -986,7 +986,7 @@ static int ep_alloc(struct eventpoll **p goto free_uid;
mutex_init(&ep->mtx); - rwlock_init(&ep->lock); + spin_lock_init(&ep->lock); init_waitqueue_head(&ep->wq); init_waitqueue_head(&ep->poll_wait); INIT_LIST_HEAD(&ep->rdllist); @@ -1077,99 +1077,9 @@ struct file *get_epoll_tfile_raw_ptr(str #endif /* CONFIG_KCMP */
/* - * Adds a new entry to the tail of the list in a lockless way, i.e. - * multiple CPUs are allowed to call this function concurrently. - * - * Beware: it is necessary to prevent any other modifications of the - * existing list until all changes are completed, in other words - * concurrent list_add_tail_lockless() calls should be protected - * with a read lock, where write lock acts as a barrier which - * makes sure all list_add_tail_lockless() calls are fully - * completed. - * - * Also an element can be locklessly added to the list only in one - * direction i.e. either to the tail or to the head, otherwise - * concurrent access will corrupt the list. - * - * Return: %false if element has been already added to the list, %true - * otherwise. - */ -static inline bool list_add_tail_lockless(struct list_head *new, - struct list_head *head) -{ - struct list_head *prev; - - /* - * This is simple 'new->next = head' operation, but cmpxchg() - * is used in order to detect that same element has been just - * added to the list from another CPU: the winner observes - * new->next == new. - */ - if (!try_cmpxchg(&new->next, &new, head)) - return false; - - /* - * Initially ->next of a new element must be updated with the head - * (we are inserting to the tail) and only then pointers are atomically - * exchanged. XCHG guarantees memory ordering, thus ->next should be - * updated before pointers are actually swapped and pointers are - * swapped before prev->next is updated. - */ - - prev = xchg(&head->prev, new); - - /* - * It is safe to modify prev->next and new->prev, because a new element - * is added only to the tail and new->next is updated before XCHG. - */ - - prev->next = new; - new->prev = prev; - - return true; -} - -/* - * Chains a new epi entry to the tail of the ep->ovflist in a lockless way, - * i.e. multiple CPUs are allowed to call this function concurrently. - * - * Return: %false if epi element has been already chained, %true otherwise. - */ -static inline bool chain_epi_lockless(struct epitem *epi) -{ - struct eventpoll *ep = epi->ep; - - /* Fast preliminary check */ - if (epi->next != EP_UNACTIVE_PTR) - return false; - - /* Check that the same epi has not been just chained from another CPU */ - if (cmpxchg(&epi->next, EP_UNACTIVE_PTR, NULL) != EP_UNACTIVE_PTR) - return false; - - /* Atomically exchange tail */ - epi->next = xchg(&ep->ovflist, epi); - - return true; -} - -/* * This is the callback that is passed to the wait queue wakeup * mechanism. It is called by the stored file descriptors when they * have events to report. - * - * This callback takes a read lock in order not to contend with concurrent - * events from another file descriptor, thus all modifications to ->rdllist - * or ->ovflist are lockless. Read lock is paired with the write lock from - * ep_scan_ready_list(), which stops all list modifications and guarantees - * that lists state is seen correctly. - * - * Another thing worth to mention is that ep_poll_callback() can be called - * concurrently for the same @epi from different CPUs if poll table was inited - * with several wait queues entries. Plural wakeup from different CPUs of a - * single wait queue is serialized by wq.lock, but the case when multiple wait - * queues are used should be detected accordingly. This is detected using - * cmpxchg() operation. */ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) { @@ -1180,7 +1090,7 @@ static int ep_poll_callback(wait_queue_e unsigned long flags; int ewake = 0;
- read_lock_irqsave(&ep->lock, flags); + spin_lock_irqsave(&ep->lock, flags);
ep_set_busy_poll_napi_id(epi);
@@ -1209,12 +1119,15 @@ static int ep_poll_callback(wait_queue_e * chained in ep->ovflist and requeued later on. */ if (READ_ONCE(ep->ovflist) != EP_UNACTIVE_PTR) { - if (chain_epi_lockless(epi)) + if (epi->next == EP_UNACTIVE_PTR) { + epi->next = READ_ONCE(ep->ovflist); + WRITE_ONCE(ep->ovflist, epi); ep_pm_stay_awake_rcu(epi); + } } else if (!ep_is_linked(epi)) { /* In the usual case, add event to ready list. */ - if (list_add_tail_lockless(&epi->rdllink, &ep->rdllist)) - ep_pm_stay_awake_rcu(epi); + list_add_tail(&epi->rdllink, &ep->rdllist); + ep_pm_stay_awake_rcu(epi); }
/* @@ -1247,7 +1160,7 @@ static int ep_poll_callback(wait_queue_e pwake++;
out_unlock: - read_unlock_irqrestore(&ep->lock, flags); + spin_unlock_irqrestore(&ep->lock, flags);
/* We have to call this outside the lock */ if (pwake) @@ -1576,7 +1489,7 @@ static int ep_insert(struct eventpoll *e }
/* We have to drop the new item inside our item list to keep track of it */ - write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock);
/* record NAPI ID of new item if present */ ep_set_busy_poll_napi_id(epi); @@ -1593,7 +1506,7 @@ static int ep_insert(struct eventpoll *e pwake++; }
- write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock);
/* We have to call this outside the lock */ if (pwake) @@ -1657,7 +1570,7 @@ static int ep_modify(struct eventpoll *e * list, push it inside. */ if (ep_item_poll(epi, &pt, 1)) { - write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock); if (!ep_is_linked(epi)) { list_add_tail(&epi->rdllink, &ep->rdllist); ep_pm_stay_awake(epi); @@ -1668,7 +1581,7 @@ static int ep_modify(struct eventpoll *e if (waitqueue_active(&ep->poll_wait)) pwake++; } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); }
/* We have to call this outside the lock */ @@ -1901,7 +1814,7 @@ static int ep_poll(struct eventpoll *ep, init_wait(&wait); wait.func = ep_autoremove_wake_function;
- write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock); /* * Barrierless variant, waitqueue_active() is called under * the same lock on wakeup ep_poll_callback() side, so it @@ -1920,7 +1833,7 @@ static int ep_poll(struct eventpoll *ep, if (!eavail) __add_wait_queue_exclusive(&ep->wq, &wait);
- write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock);
if (!eavail) timed_out = !schedule_hrtimeout_range(to, slack, @@ -1935,7 +1848,7 @@ static int ep_poll(struct eventpoll *ep, eavail = 1;
if (!list_empty_careful(&wait.entry)) { - write_lock_irq(&ep->lock); + spin_lock_irq(&ep->lock); /* * If the thread timed out and is not on the wait queue, * it means that the thread was woken up after its @@ -1946,7 +1859,7 @@ static int ep_poll(struct eventpoll *ep, if (timed_out) eavail = list_empty(&wait.entry); __remove_wait_queue(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } } }