October 2025 - Linux-stable-mirror

FAILED: patch "[PATCH] eventpoll: Replace rwlock with spinlock" failed to apply to 5.10-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.10-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.10.y git checkout FETCH_HEAD git cherry-pick -x 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101619-cesarean-cresting-01cb@gregkh' --subject-prefix 'PATCH 5.10.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd Mon Sep 17 00:00:00 2001 From: Nam Cao <namcao(a)linutronix.de> Date: Tue, 15 Jul 2025 14:46:34 +0200 Subject: [PATCH] eventpoll: Replace rwlock with spinlock 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(a)linutronix.de> Link: https://lore.kernel.org/ec92458ea357ec503c737ead0f10b2c6e4c37d47.1752581388… Tested-by: K Prateek Nayak <kprateek.nayak(a)amd.com> Cc: stable(a)vger.kernel.org Reported-by: Frederic Weisbecker <frederic(a)kernel.org> Closes: https://lore.kernel.org/linux-rt-users/20210825132754.GA895675@lothringen/ [1] Reported-by: Valentin Schneider <vschneid(a)redhat.com> Closes: https://lore.kernel.org/linux-rt-users/xhsmhttqvnall.mognet@vschneid.remote… [2] Signed-off-by: Christian Brauner <brauner(a)kernel.org> diff --git a/fs/eventpoll.c b/fs/eventpoll.c index b22d6f819f78..ee7c4b683ec3 100644 --- a/fs/eventpoll.c +++ b/fs/eventpoll.c @@ -46,10 +46,10 @@ * * 1) epnested_mutex (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 @@ -195,7 +195,7 @@ 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; @@ -741,10 +741,10 @@ static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist) * 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, @@ -752,7 +752,7 @@ static void ep_done_scan(struct eventpoll *ep, { 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. @@ -793,7 +793,7 @@ static void ep_done_scan(struct eventpoll *ep, wake_up(&ep->wq); } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } static void ep_get(struct eventpoll *ep) @@ -868,10 +868,10 @@ static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force) 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)); /* @@ -1152,7 +1152,7 @@ static int ep_alloc(struct eventpoll **pep) return -ENOMEM; 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); @@ -1239,100 +1239,10 @@ struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, } #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_start/done_scan(), 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) { @@ -1343,7 +1253,7 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v 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); @@ -1372,12 +1282,15 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v * 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); } /* @@ -1410,7 +1323,7 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v 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) @@ -1745,7 +1658,7 @@ static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, } /* 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); @@ -1762,7 +1675,7 @@ static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, pwake++; } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); /* We have to call this outside the lock */ if (pwake) @@ -1826,7 +1739,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, * 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); @@ -1837,7 +1750,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, 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 */ @@ -2089,7 +2002,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, 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 @@ -2108,7 +2021,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, if (!eavail) __add_wait_queue_exclusive(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); if (!eavail) timed_out = !ep_schedule_timeout(to) || @@ -2124,7 +2037,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, 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 @@ -2135,7 +2048,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, if (timed_out) eavail = list_empty(&wait.entry); __remove_wait_queue(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } } }

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] eventpoll: Replace rwlock with spinlock" failed to apply to 5.15-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.15-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.15.y git checkout FETCH_HEAD git cherry-pick -x 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101617-gaming-preplan-1aaf@gregkh' --subject-prefix 'PATCH 5.15.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd Mon Sep 17 00:00:00 2001 From: Nam Cao <namcao(a)linutronix.de> Date: Tue, 15 Jul 2025 14:46:34 +0200 Subject: [PATCH] eventpoll: Replace rwlock with spinlock 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(a)linutronix.de> Link: https://lore.kernel.org/ec92458ea357ec503c737ead0f10b2c6e4c37d47.1752581388… Tested-by: K Prateek Nayak <kprateek.nayak(a)amd.com> Cc: stable(a)vger.kernel.org Reported-by: Frederic Weisbecker <frederic(a)kernel.org> Closes: https://lore.kernel.org/linux-rt-users/20210825132754.GA895675@lothringen/ [1] Reported-by: Valentin Schneider <vschneid(a)redhat.com> Closes: https://lore.kernel.org/linux-rt-users/xhsmhttqvnall.mognet@vschneid.remote… [2] Signed-off-by: Christian Brauner <brauner(a)kernel.org> diff --git a/fs/eventpoll.c b/fs/eventpoll.c index b22d6f819f78..ee7c4b683ec3 100644 --- a/fs/eventpoll.c +++ b/fs/eventpoll.c @@ -46,10 +46,10 @@ * * 1) epnested_mutex (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 @@ -195,7 +195,7 @@ 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; @@ -741,10 +741,10 @@ static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist) * 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, @@ -752,7 +752,7 @@ static void ep_done_scan(struct eventpoll *ep, { 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. @@ -793,7 +793,7 @@ static void ep_done_scan(struct eventpoll *ep, wake_up(&ep->wq); } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } static void ep_get(struct eventpoll *ep) @@ -868,10 +868,10 @@ static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force) 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)); /* @@ -1152,7 +1152,7 @@ static int ep_alloc(struct eventpoll **pep) return -ENOMEM; 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); @@ -1239,100 +1239,10 @@ struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, } #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_start/done_scan(), 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) { @@ -1343,7 +1253,7 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v 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); @@ -1372,12 +1282,15 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v * 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); } /* @@ -1410,7 +1323,7 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v 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) @@ -1745,7 +1658,7 @@ static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, } /* 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); @@ -1762,7 +1675,7 @@ static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, pwake++; } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); /* We have to call this outside the lock */ if (pwake) @@ -1826,7 +1739,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, * 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); @@ -1837,7 +1750,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, 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 */ @@ -2089,7 +2002,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, 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 @@ -2108,7 +2021,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, if (!eavail) __add_wait_queue_exclusive(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); if (!eavail) timed_out = !ep_schedule_timeout(to) || @@ -2124,7 +2037,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, 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 @@ -2135,7 +2048,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, if (timed_out) eavail = list_empty(&wait.entry); __remove_wait_queue(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } } }

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] eventpoll: Replace rwlock with spinlock" failed to apply to 6.1-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.1-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.1.y git checkout FETCH_HEAD git cherry-pick -x 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101616-causal-numerator-0c1e@gregkh' --subject-prefix 'PATCH 6.1.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 0c43094f8cc9d3d99d835c0ac9c4fe1ccc62babd Mon Sep 17 00:00:00 2001 From: Nam Cao <namcao(a)linutronix.de> Date: Tue, 15 Jul 2025 14:46:34 +0200 Subject: [PATCH] eventpoll: Replace rwlock with spinlock 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(a)linutronix.de> Link: https://lore.kernel.org/ec92458ea357ec503c737ead0f10b2c6e4c37d47.1752581388… Tested-by: K Prateek Nayak <kprateek.nayak(a)amd.com> Cc: stable(a)vger.kernel.org Reported-by: Frederic Weisbecker <frederic(a)kernel.org> Closes: https://lore.kernel.org/linux-rt-users/20210825132754.GA895675@lothringen/ [1] Reported-by: Valentin Schneider <vschneid(a)redhat.com> Closes: https://lore.kernel.org/linux-rt-users/xhsmhttqvnall.mognet@vschneid.remote… [2] Signed-off-by: Christian Brauner <brauner(a)kernel.org> diff --git a/fs/eventpoll.c b/fs/eventpoll.c index b22d6f819f78..ee7c4b683ec3 100644 --- a/fs/eventpoll.c +++ b/fs/eventpoll.c @@ -46,10 +46,10 @@ * * 1) epnested_mutex (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 @@ -195,7 +195,7 @@ 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; @@ -741,10 +741,10 @@ static void ep_start_scan(struct eventpoll *ep, struct list_head *txlist) * 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, @@ -752,7 +752,7 @@ static void ep_done_scan(struct eventpoll *ep, { 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. @@ -793,7 +793,7 @@ static void ep_done_scan(struct eventpoll *ep, wake_up(&ep->wq); } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } static void ep_get(struct eventpoll *ep) @@ -868,10 +868,10 @@ static bool __ep_remove(struct eventpoll *ep, struct epitem *epi, bool force) 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)); /* @@ -1152,7 +1152,7 @@ static int ep_alloc(struct eventpoll **pep) return -ENOMEM; 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); @@ -1239,100 +1239,10 @@ struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, } #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_start/done_scan(), 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) { @@ -1343,7 +1253,7 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v 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); @@ -1372,12 +1282,15 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v * 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); } /* @@ -1410,7 +1323,7 @@ static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, v 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) @@ -1745,7 +1658,7 @@ static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, } /* 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); @@ -1762,7 +1675,7 @@ static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, pwake++; } - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); /* We have to call this outside the lock */ if (pwake) @@ -1826,7 +1739,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, * 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); @@ -1837,7 +1750,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, 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 */ @@ -2089,7 +2002,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, 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 @@ -2108,7 +2021,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, if (!eavail) __add_wait_queue_exclusive(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); if (!eavail) timed_out = !ep_schedule_timeout(to) || @@ -2124,7 +2037,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, 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 @@ -2135,7 +2048,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, if (timed_out) eavail = list_empty(&wait.entry); __remove_wait_queue(&ep->wq, &wait); - write_unlock_irq(&ep->lock); + spin_unlock_irq(&ep->lock); } } }

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] cpufreq: CPPC: Avoid using CPUFREQ_ETERNAL as transition" failed to apply to 5.4-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.4-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.4.y git checkout FETCH_HEAD git cherry-pick -x f965d111e68f4a993cc44d487d416e3d954eea11 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101656-peso-landfill-d499@gregkh' --subject-prefix 'PATCH 5.4.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From f965d111e68f4a993cc44d487d416e3d954eea11 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" <rafael.j.wysocki(a)intel.com> Date: Fri, 26 Sep 2025 12:19:41 +0200 Subject: [PATCH] cpufreq: CPPC: Avoid using CPUFREQ_ETERNAL as transition delay If cppc_get_transition_latency() returns CPUFREQ_ETERNAL to indicate a failure to retrieve the transition latency value from the platform firmware, the CPPC cpufreq driver will use that value (converted to microseconds) as the policy transition delay, but it is way too large for any practical use. Address this by making the driver use the cpufreq's default transition latency value (in microseconds) as the transition delay if CPUFREQ_ETERNAL is returned by cppc_get_transition_latency(). Fixes: d4f3388afd48 ("cpufreq / CPPC: Set platform specific transition_delay_us") Cc: 5.19+ <stable(a)vger.kernel.org> # 5.19 Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki(a)intel.com> Reviewed-by: Mario Limonciello (AMD) <superm1(a)kernel.org> Reviewed-by: Jie Zhan <zhanjie9(a)hisilicon.com> Acked-by: Viresh Kumar <viresh.kumar(a)linaro.org> Reviewed-by: Qais Yousef <qyousef(a)layalina.io> diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c index 12de0ac7bbaf..b71946937c52 100644 --- a/drivers/cpufreq/cppc_cpufreq.c +++ b/drivers/cpufreq/cppc_cpufreq.c @@ -308,6 +308,16 @@ static int cppc_verify_policy(struct cpufreq_policy_data *policy) return 0; } +static unsigned int __cppc_cpufreq_get_transition_delay_us(unsigned int cpu) +{ + unsigned int transition_latency_ns = cppc_get_transition_latency(cpu); + + if (transition_latency_ns == CPUFREQ_ETERNAL) + return CPUFREQ_DEFAULT_TRANSITION_LATENCY_NS / NSEC_PER_USEC; + + return transition_latency_ns / NSEC_PER_USEC; +} + /* * The PCC subspace describes the rate at which platform can accept commands * on the shared PCC channel (including READs which do not count towards freq @@ -330,12 +340,12 @@ static unsigned int cppc_cpufreq_get_transition_delay_us(unsigned int cpu) return 10000; } } - return cppc_get_transition_latency(cpu) / NSEC_PER_USEC; + return __cppc_cpufreq_get_transition_delay_us(cpu); } #else static unsigned int cppc_cpufreq_get_transition_delay_us(unsigned int cpu) { - return cppc_get_transition_latency(cpu) / NSEC_PER_USEC; + return __cppc_cpufreq_get_transition_delay_us(cpu); } #endif

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] cpufreq: CPPC: Avoid using CPUFREQ_ETERNAL as transition" failed to apply to 5.10-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.10-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.10.y git checkout FETCH_HEAD git cherry-pick -x f965d111e68f4a993cc44d487d416e3d954eea11 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101655-hardhead-variably-b7de@gregkh' --subject-prefix 'PATCH 5.10.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From f965d111e68f4a993cc44d487d416e3d954eea11 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" <rafael.j.wysocki(a)intel.com> Date: Fri, 26 Sep 2025 12:19:41 +0200 Subject: [PATCH] cpufreq: CPPC: Avoid using CPUFREQ_ETERNAL as transition delay If cppc_get_transition_latency() returns CPUFREQ_ETERNAL to indicate a failure to retrieve the transition latency value from the platform firmware, the CPPC cpufreq driver will use that value (converted to microseconds) as the policy transition delay, but it is way too large for any practical use. Address this by making the driver use the cpufreq's default transition latency value (in microseconds) as the transition delay if CPUFREQ_ETERNAL is returned by cppc_get_transition_latency(). Fixes: d4f3388afd48 ("cpufreq / CPPC: Set platform specific transition_delay_us") Cc: 5.19+ <stable(a)vger.kernel.org> # 5.19 Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki(a)intel.com> Reviewed-by: Mario Limonciello (AMD) <superm1(a)kernel.org> Reviewed-by: Jie Zhan <zhanjie9(a)hisilicon.com> Acked-by: Viresh Kumar <viresh.kumar(a)linaro.org> Reviewed-by: Qais Yousef <qyousef(a)layalina.io> diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c index 12de0ac7bbaf..b71946937c52 100644 --- a/drivers/cpufreq/cppc_cpufreq.c +++ b/drivers/cpufreq/cppc_cpufreq.c @@ -308,6 +308,16 @@ static int cppc_verify_policy(struct cpufreq_policy_data *policy) return 0; } +static unsigned int __cppc_cpufreq_get_transition_delay_us(unsigned int cpu) +{ + unsigned int transition_latency_ns = cppc_get_transition_latency(cpu); + + if (transition_latency_ns == CPUFREQ_ETERNAL) + return CPUFREQ_DEFAULT_TRANSITION_LATENCY_NS / NSEC_PER_USEC; + + return transition_latency_ns / NSEC_PER_USEC; +} + /* * The PCC subspace describes the rate at which platform can accept commands * on the shared PCC channel (including READs which do not count towards freq @@ -330,12 +340,12 @@ static unsigned int cppc_cpufreq_get_transition_delay_us(unsigned int cpu) return 10000; } } - return cppc_get_transition_latency(cpu) / NSEC_PER_USEC; + return __cppc_cpufreq_get_transition_delay_us(cpu); } #else static unsigned int cppc_cpufreq_get_transition_delay_us(unsigned int cpu) { - return cppc_get_transition_latency(cpu) / NSEC_PER_USEC; + return __cppc_cpufreq_get_transition_delay_us(cpu); } #endif

2 weeks, 3 days

1
0
0 0

[PATCH v3] drm/sched: Fix potential double free in drm_sched_job_add_resv_dependencies

by Tvrtko Ursulin

When adding dependencies with drm_sched_job_add_dependency(), that function consumes the fence reference both on success and failure, so in the latter case the dma_fence_put() on the error path (xarray failed to expand) is a double free. Interestingly this bug appears to have been present ever since commit ebd5f74255b9 ("drm/sched: Add dependency tracking"), since the code back then looked like this: drm_sched_job_add_implicit_dependencies(): ... for (i = 0; i < fence_count; i++) { ret = drm_sched_job_add_dependency(job, fences[i]); if (ret) break; } for (; i < fence_count; i++) dma_fence_put(fences[i]); Which means for the failing 'i' the dma_fence_put was already a double free. Possibly there were no users at that time, or the test cases were insufficient to hit it. The bug was then only noticed and fixed after commit 9c2ba265352a ("drm/scheduler: use new iterator in drm_sched_job_add_implicit_dependencies v2") landed, with its fixup of commit 4eaf02d6076c ("drm/scheduler: fix drm_sched_job_add_implicit_dependencies"). At that point it was a slightly different flavour of a double free, which commit 963d0b356935 ("drm/scheduler: fix drm_sched_job_add_implicit_dependencies harder") noticed and attempted to fix. But it only moved the double free from happening inside the drm_sched_job_add_dependency(), when releasing the reference not yet obtained, to the caller, when releasing the reference already released by the former in the failure case. As such it is not easy to identify the right target for the fixes tag so lets keep it simple and just continue the chain. While fixing we also improve the comment and explain the reason for taking the reference and not dropping it. Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin(a)igalia.com> Fixes: 963d0b356935 ("drm/scheduler: fix drm_sched_job_add_implicit_dependencies harder") Reported-by: Dan Carpenter <dan.carpenter(a)linaro.org> Closes: https://lore.kernel.org/dri-devel/aNFbXq8OeYl3QSdm@stanley.mountain/ Cc: Christian König <christian.koenig(a)amd.com> Cc: Rob Clark <robdclark(a)chromium.org> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: Matthew Brost <matthew.brost(a)intel.com> Cc: Danilo Krummrich <dakr(a)kernel.org> Cc: Philipp Stanner <phasta(a)kernel.org> Cc: "Christian König" <ckoenig.leichtzumerken(a)gmail.com> Cc: dri-devel(a)lists.freedesktop.org Cc: <stable(a)vger.kernel.org> # v5.16+ --- v2: * Re-arrange commit text so discussion around sentences starting with capital letters in all cases can be avoided. * Keep double return for now. * Improved comment instead of dropping it. v3: * Commit SHA formatting in the commit message. --- drivers/gpu/drm/scheduler/sched_main.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c index 46119aacb809..c39f0245e3a9 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -965,13 +965,14 @@ int drm_sched_job_add_resv_dependencies(struct drm_sched_job *job, dma_resv_assert_held(resv); dma_resv_for_each_fence(&cursor, resv, usage, fence) { - /* Make sure to grab an additional ref on the added fence */ - dma_fence_get(fence); - ret = drm_sched_job_add_dependency(job, fence); - if (ret) { - dma_fence_put(fence); + /* + * As drm_sched_job_add_dependency always consumes the fence + * reference (even when it fails), and dma_resv_for_each_fence + * is not obtaining one, we need to grab one before calling. + */ + ret = drm_sched_job_add_dependency(job, dma_fence_get(fence)); + if (ret) return ret; - } } return 0; } -- 2.48.0

2 weeks, 3 days

2
1
0 0

FAILED: patch "[PATCH] arm64: kprobes: call set_memory_rox() for kprobe page" failed to apply to 6.1-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.1-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.1.y git checkout FETCH_HEAD git cherry-pick -x 195a1b7d8388c0ec2969a39324feb8bebf9bb907 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101634-riveter-crust-f6b3@gregkh' --subject-prefix 'PATCH 6.1.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 195a1b7d8388c0ec2969a39324feb8bebf9bb907 Mon Sep 17 00:00:00 2001 From: Yang Shi <yang(a)os.amperecomputing.com> Date: Thu, 18 Sep 2025 09:23:49 -0700 Subject: [PATCH] arm64: kprobes: call set_memory_rox() for kprobe page The kprobe page is allocated by execmem allocator with ROX permission. It needs to call set_memory_rox() to set proper permission for the direct map too. It was missed. Fixes: 10d5e97c1bf8 ("arm64: use PAGE_KERNEL_ROX directly in alloc_insn_page") Cc: <stable(a)vger.kernel.org> Signed-off-by: Yang Shi <yang(a)os.amperecomputing.com> Reviewed-by: Catalin Marinas <catalin.marinas(a)arm.com> Signed-off-by: Will Deacon <will(a)kernel.org> diff --git a/arch/arm64/kernel/probes/kprobes.c b/arch/arm64/kernel/probes/kprobes.c index 0c5d408afd95..8ab6104a4883 100644 --- a/arch/arm64/kernel/probes/kprobes.c +++ b/arch/arm64/kernel/probes/kprobes.c @@ -10,6 +10,7 @@ #define pr_fmt(fmt) "kprobes: " fmt +#include <linux/execmem.h> #include <linux/extable.h> #include <linux/kasan.h> #include <linux/kernel.h> @@ -41,6 +42,17 @@ DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); static void __kprobes post_kprobe_handler(struct kprobe *, struct kprobe_ctlblk *, struct pt_regs *); +void *alloc_insn_page(void) +{ + void *addr; + + addr = execmem_alloc(EXECMEM_KPROBES, PAGE_SIZE); + if (!addr) + return NULL; + set_memory_rox((unsigned long)addr, 1); + return addr; +} + static void __kprobes arch_prepare_ss_slot(struct kprobe *p) { kprobe_opcode_t *addr = p->ainsn.xol_insn;

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] arm64: kprobes: call set_memory_rox() for kprobe page" failed to apply to 5.15-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.15-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.15.y git checkout FETCH_HEAD git cherry-pick -x 195a1b7d8388c0ec2969a39324feb8bebf9bb907 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101608-fastball-thaw-dc90@gregkh' --subject-prefix 'PATCH 5.15.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 195a1b7d8388c0ec2969a39324feb8bebf9bb907 Mon Sep 17 00:00:00 2001 From: Yang Shi <yang(a)os.amperecomputing.com> Date: Thu, 18 Sep 2025 09:23:49 -0700 Subject: [PATCH] arm64: kprobes: call set_memory_rox() for kprobe page The kprobe page is allocated by execmem allocator with ROX permission. It needs to call set_memory_rox() to set proper permission for the direct map too. It was missed. Fixes: 10d5e97c1bf8 ("arm64: use PAGE_KERNEL_ROX directly in alloc_insn_page") Cc: <stable(a)vger.kernel.org> Signed-off-by: Yang Shi <yang(a)os.amperecomputing.com> Reviewed-by: Catalin Marinas <catalin.marinas(a)arm.com> Signed-off-by: Will Deacon <will(a)kernel.org> diff --git a/arch/arm64/kernel/probes/kprobes.c b/arch/arm64/kernel/probes/kprobes.c index 0c5d408afd95..8ab6104a4883 100644 --- a/arch/arm64/kernel/probes/kprobes.c +++ b/arch/arm64/kernel/probes/kprobes.c @@ -10,6 +10,7 @@ #define pr_fmt(fmt) "kprobes: " fmt +#include <linux/execmem.h> #include <linux/extable.h> #include <linux/kasan.h> #include <linux/kernel.h> @@ -41,6 +42,17 @@ DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); static void __kprobes post_kprobe_handler(struct kprobe *, struct kprobe_ctlblk *, struct pt_regs *); +void *alloc_insn_page(void) +{ + void *addr; + + addr = execmem_alloc(EXECMEM_KPROBES, PAGE_SIZE); + if (!addr) + return NULL; + set_memory_rox((unsigned long)addr, 1); + return addr; +} + static void __kprobes arch_prepare_ss_slot(struct kprobe *p) { kprobe_opcode_t *addr = p->ainsn.xol_insn;

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] bus: mhi: ep: Fix chained transfer handling in read path" failed to apply to 6.1-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.1-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.1.y git checkout FETCH_HEAD git cherry-pick -x f5225a34bd8f9f64eec37f6ae1461289aaa3eb86 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101623-error-attain-05f5@gregkh' --subject-prefix 'PATCH 6.1.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From f5225a34bd8f9f64eec37f6ae1461289aaa3eb86 Mon Sep 17 00:00:00 2001 From: Sumit Kumar <sumit.kumar(a)oss.qualcomm.com> Date: Wed, 10 Sep 2025 18:11:09 +0530 Subject: [PATCH] bus: mhi: ep: Fix chained transfer handling in read path The mhi_ep_read_channel function incorrectly assumes the End of Transfer (EOT) bit is present for each packet in a chained transactions, causing it to advance mhi_chan->rd_offset beyond wr_offset during host-to-device transfers when EOT has not yet arrived. This leads to access of unmapped host memory, causing IOMMU faults and processing of stale TREs. Modify the loop condition to ensure mhi_queue is not empty, allowing the function to process only valid TREs up to the current write pointer to prevent premature reads and ensure safe traversal of chained TREs. Due to this change, buf_left needs to be removed from the while loop condition to avoid exiting prematurely before reading the ring completely, and also remove write_offset since it will always be zero because the new cache buffer is allocated every time. Fixes: 5301258899773 ("bus: mhi: ep: Add support for reading from the host") Co-developed-by: Akhil Vinod <akhil.vinod(a)oss.qualcomm.com> Signed-off-by: Akhil Vinod <akhil.vinod(a)oss.qualcomm.com> Signed-off-by: Sumit Kumar <sumit.kumar(a)oss.qualcomm.com> [mani: reworded description slightly] Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam(a)oss.qualcomm.com> Reviewed-by: Krishna Chaitanya Chundru <krishna.chundru(a)oss.qualcomm.com> Cc: stable(a)vger.kernel.org Link: https://patch.msgid.link/20250910-final_chained-v3-1-ec77c9d88ace@oss.qualc… diff --git a/drivers/bus/mhi/ep/main.c b/drivers/bus/mhi/ep/main.c index b3eafcf2a2c5..cdea24e92919 100644 --- a/drivers/bus/mhi/ep/main.c +++ b/drivers/bus/mhi/ep/main.c @@ -403,17 +403,13 @@ static int mhi_ep_read_channel(struct mhi_ep_cntrl *mhi_cntrl, { struct mhi_ep_chan *mhi_chan = &mhi_cntrl->mhi_chan[ring->ch_id]; struct device *dev = &mhi_cntrl->mhi_dev->dev; - size_t tr_len, read_offset, write_offset; + size_t tr_len, read_offset; struct mhi_ep_buf_info buf_info = {}; u32 len = MHI_EP_DEFAULT_MTU; struct mhi_ring_element *el; - bool tr_done = false; void *buf_addr; - u32 buf_left; int ret; - buf_left = len; - do { /* Don't process the transfer ring if the channel is not in RUNNING state */ if (mhi_chan->state != MHI_CH_STATE_RUNNING) { @@ -426,24 +422,23 @@ static int mhi_ep_read_channel(struct mhi_ep_cntrl *mhi_cntrl, /* Check if there is data pending to be read from previous read operation */ if (mhi_chan->tre_bytes_left) { dev_dbg(dev, "TRE bytes remaining: %u\n", mhi_chan->tre_bytes_left); - tr_len = min(buf_left, mhi_chan->tre_bytes_left); + tr_len = min(len, mhi_chan->tre_bytes_left); } else { mhi_chan->tre_loc = MHI_TRE_DATA_GET_PTR(el); mhi_chan->tre_size = MHI_TRE_DATA_GET_LEN(el); mhi_chan->tre_bytes_left = mhi_chan->tre_size; - tr_len = min(buf_left, mhi_chan->tre_size); + tr_len = min(len, mhi_chan->tre_size); } read_offset = mhi_chan->tre_size - mhi_chan->tre_bytes_left; - write_offset = len - buf_left; buf_addr = kmem_cache_zalloc(mhi_cntrl->tre_buf_cache, GFP_KERNEL); if (!buf_addr) return -ENOMEM; buf_info.host_addr = mhi_chan->tre_loc + read_offset; - buf_info.dev_addr = buf_addr + write_offset; + buf_info.dev_addr = buf_addr; buf_info.size = tr_len; buf_info.cb = mhi_ep_read_completion; buf_info.cb_buf = buf_addr; @@ -459,16 +454,12 @@ static int mhi_ep_read_channel(struct mhi_ep_cntrl *mhi_cntrl, goto err_free_buf_addr; } - buf_left -= tr_len; mhi_chan->tre_bytes_left -= tr_len; - if (!mhi_chan->tre_bytes_left) { - if (MHI_TRE_DATA_GET_IEOT(el)) - tr_done = true; - + if (!mhi_chan->tre_bytes_left) mhi_chan->rd_offset = (mhi_chan->rd_offset + 1) % ring->ring_size; - } - } while (buf_left && !tr_done); + /* Read until the some buffer is left or the ring becomes not empty */ + } while (!mhi_ep_queue_is_empty(mhi_chan->mhi_dev, DMA_TO_DEVICE)); return 0; @@ -502,15 +493,11 @@ static int mhi_ep_process_ch_ring(struct mhi_ep_ring *ring) mhi_chan->xfer_cb(mhi_chan->mhi_dev, &result); } else { /* UL channel */ - do { - ret = mhi_ep_read_channel(mhi_cntrl, ring); - if (ret < 0) { - dev_err(&mhi_chan->mhi_dev->dev, "Failed to read channel\n"); - return ret; - } - - /* Read until the ring becomes empty */ - } while (!mhi_ep_queue_is_empty(mhi_chan->mhi_dev, DMA_TO_DEVICE)); + ret = mhi_ep_read_channel(mhi_cntrl, ring); + if (ret < 0) { + dev_err(&mhi_chan->mhi_dev->dev, "Failed to read channel\n"); + return ret; + } } return 0;

2 weeks, 3 days

1
0
0 0

FAILED: patch "[PATCH] rseq: Protect event mask against membarrier IPI" failed to apply to 6.1-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.1-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.1.y git checkout FETCH_HEAD git cherry-pick -x 6eb350a2233100a283f882c023e5ad426d0ed63b # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2025101555-untitled-sighing-27f5@gregkh' --subject-prefix 'PATCH 6.1.y' HEAD^.. Possible dependencies: thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 6eb350a2233100a283f882c023e5ad426d0ed63b Mon Sep 17 00:00:00 2001 From: Thomas Gleixner <tglx(a)linutronix.de> Date: Wed, 13 Aug 2025 17:02:30 +0200 Subject: [PATCH] rseq: Protect event mask against membarrier IPI rseq_need_restart() reads and clears task::rseq_event_mask with preemption disabled to guard against the scheduler. But membarrier() uses an IPI and sets the PREEMPT bit in the event mask from the IPI, which leaves that RMW operation unprotected. Use guard(irq) if CONFIG_MEMBARRIER is enabled to fix that. Fixes: 2a36ab717e8f ("rseq/membarrier: Add MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ") Signed-off-by: Thomas Gleixner <tglx(a)linutronix.de> Reviewed-by: Boqun Feng <boqun.feng(a)gmail.com> Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers(a)efficios.com> Cc: stable(a)vger.kernel.org diff --git a/include/linux/rseq.h b/include/linux/rseq.h index bc8af3eb5598..1fbeb61babeb 100644 --- a/include/linux/rseq.h +++ b/include/linux/rseq.h @@ -7,6 +7,12 @@ #include <linux/preempt.h> #include <linux/sched.h> +#ifdef CONFIG_MEMBARRIER +# define RSEQ_EVENT_GUARD irq +#else +# define RSEQ_EVENT_GUARD preempt +#endif + /* * Map the event mask on the user-space ABI enum rseq_cs_flags * for direct mask checks. @@ -41,9 +47,8 @@ static inline void rseq_handle_notify_resume(struct ksignal *ksig, static inline void rseq_signal_deliver(struct ksignal *ksig, struct pt_regs *regs) { - preempt_disable(); - __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask); - preempt_enable(); + scoped_guard(RSEQ_EVENT_GUARD) + __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask); rseq_handle_notify_resume(ksig, regs); } diff --git a/kernel/rseq.c b/kernel/rseq.c index b7a1ec327e81..2452b7366b00 100644 --- a/kernel/rseq.c +++ b/kernel/rseq.c @@ -342,12 +342,12 @@ static int rseq_need_restart(struct task_struct *t, u32 cs_flags) /* * Load and clear event mask atomically with respect to - * scheduler preemption. + * scheduler preemption and membarrier IPIs. */ - preempt_disable(); - event_mask = t->rseq_event_mask; - t->rseq_event_mask = 0; - preempt_enable(); + scoped_guard(RSEQ_EVENT_GUARD) { + event_mask = t->rseq_event_mask; + t->rseq_event_mask = 0; + } return !!event_mask; }

2 weeks, 3 days

2
1
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror October 2025