From 13652c4b7e2dd74e8b206c8f724f064fe45e2c8a Mon Sep 17 00:00:00 2001 From: Mathieu Desnoyers Date: Sat, 18 Oct 2014 16:38:32 +0200 Subject: [PATCH] Implement waitqueue and workqueue APIs Signed-off-by: Mathieu Desnoyers --- Makefile.am | 3 +- urcu-qsbr.c | 2 +- urcu.c | 2 +- urcu/static/wfcqueue.h | 19 +- urcu-wait.h => urcu/waitqueue-lifo.h | 92 ++++++- urcu/workqueue-fifo.h | 345 +++++++++++++++++++++++++++ 6 files changed, 445 insertions(+), 18 deletions(-) rename urcu-wait.h => urcu/waitqueue-lifo.h (68%) create mode 100644 urcu/workqueue-fifo.h diff --git a/Makefile.am b/Makefile.am index 752510d..ab62bbf 100644 --- a/Makefile.am +++ b/Makefile.am @@ -21,13 +21,14 @@ nobase_dist_include_HEADERS = urcu/compiler.h urcu/hlist.h urcu/list.h \ urcu/ref.h urcu/cds.h urcu/urcu_ref.h urcu/urcu-futex.h \ urcu/uatomic_arch.h urcu/rculfhash.h urcu/wfcqueue.h \ urcu/lfstack.h urcu/syscall-compat.h \ + urcu/waitqueue-lifo.h urcu/workqueue-fifo.h \ $(top_srcdir)/urcu/map/*.h \ $(top_srcdir)/urcu/static/*.h \ urcu/rand-compat.h \ urcu/tls-compat.h nobase_nodist_include_HEADERS = urcu/arch.h urcu/uatomic.h urcu/config.h -dist_noinst_HEADERS = urcu-die.h urcu-wait.h +dist_noinst_HEADERS = urcu-die.h EXTRA_DIST = $(top_srcdir)/urcu/arch/*.h $(top_srcdir)/urcu/uatomic/*.h \ gpl-2.0.txt lgpl-2.1.txt lgpl-relicensing.txt \ diff --git a/urcu-qsbr.c b/urcu-qsbr.c index 71e7a39..d7178f4 100644 --- a/urcu-qsbr.c +++ b/urcu-qsbr.c @@ -41,9 +41,9 @@ #include "urcu/static/urcu-qsbr.h" #include "urcu-pointer.h" #include "urcu/tls-compat.h" +#include "urcu/waitqueue-lifo.h" #include "urcu-die.h" -#include "urcu-wait.h" /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */ #undef _LGPL_SOURCE diff --git a/urcu.c b/urcu.c index ae3490f..0241d3b 100644 --- a/urcu.c +++ b/urcu.c @@ -41,9 +41,9 @@ #include "urcu/static/urcu.h" #include "urcu-pointer.h" #include "urcu/tls-compat.h" +#include "urcu/waitqueue-lifo.h" #include "urcu-die.h" -#include "urcu-wait.h" /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */ #undef _LGPL_SOURCE diff --git a/urcu/static/wfcqueue.h b/urcu/static/wfcqueue.h index 48b2625..b2ad7ab 100644 --- a/urcu/static/wfcqueue.h +++ b/urcu/static/wfcqueue.h @@ -367,7 +367,8 @@ static inline struct cds_wfcq_node * ___cds_wfcq_dequeue_with_state(cds_wfcq_head_ptr_t u_head, struct cds_wfcq_tail *tail, int *state, - int blocking) + int blocking, + int safe) { struct __cds_wfcq_head *head = u_head._h; struct cds_wfcq_node *node, *next; @@ -385,6 +386,8 @@ ___cds_wfcq_dequeue_with_state(cds_wfcq_head_ptr_t u_head, } if ((next = CMM_LOAD_SHARED(node->next)) == NULL) { + struct cds_wfcq_node *ret_node; + /* * @node is probably the only node in the queue. * Try to move the tail to &q->head. @@ -400,7 +403,15 @@ ___cds_wfcq_dequeue_with_state(cds_wfcq_head_ptr_t u_head, * content. */ _cds_wfcq_node_init(&head->node); - if (uatomic_cmpxchg(&tail->p, node, &head->node) == node) { + if (safe) { + ret_node = uatomic_cmpxchg(&tail->p, node, + &head->node); + } else { + ret_node = tail->p; + if (ret_node == node) + tail->p = &head->node; + } + if (ret_node == node) { if (state) *state |= CDS_WFCQ_STATE_LAST; return node; @@ -440,7 +451,7 @@ static inline struct cds_wfcq_node * ___cds_wfcq_dequeue_with_state_blocking(cds_wfcq_head_ptr_t head, struct cds_wfcq_tail *tail, int *state) { - return ___cds_wfcq_dequeue_with_state(head, tail, state, 1); + return ___cds_wfcq_dequeue_with_state(head, tail, state, 1, 1); } /* @@ -466,7 +477,7 @@ static inline struct cds_wfcq_node * ___cds_wfcq_dequeue_with_state_nonblocking(cds_wfcq_head_ptr_t head, struct cds_wfcq_tail *tail, int *state) { - return ___cds_wfcq_dequeue_with_state(head, tail, state, 0); + return ___cds_wfcq_dequeue_with_state(head, tail, state, 0, 1); } /* diff --git a/urcu-wait.h b/urcu/waitqueue-lifo.h similarity index 68% rename from urcu-wait.h rename to urcu/waitqueue-lifo.h index d00842a..4d800f0 100644 --- a/urcu-wait.h +++ b/urcu/waitqueue-lifo.h @@ -1,12 +1,12 @@ -#ifndef _URCU_WAIT_H -#define _URCU_WAIT_H +#ifndef _URCU_WAITQUEUE_LIFO_H +#define _URCU_WAITQUEUE_LIFO_H /* - * urcu-wait.h + * urcu/waitqueue-lifo.h * - * Userspace RCU library wait/wakeup management + * Userspace RCU library - wait queue scheme with LIFO semantic * - * Copyright (c) 2012 Mathieu Desnoyers + * Copyright (c) 2012-2014 Mathieu Desnoyers * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public @@ -25,6 +25,7 @@ #include #include +#include /* * Number of busy-loop attempts before waiting on futex for grace period @@ -56,11 +57,11 @@ struct urcu_wait_node { struct urcu_wait_node name struct urcu_wait_queue { - struct cds_wfs_stack stack; + struct __cds_wfs_stack stack; }; #define URCU_WAIT_QUEUE_HEAD_INIT(name) \ - { .stack.head = CDS_WFS_END, .stack.lock = PTHREAD_MUTEX_INITIALIZER } + { .stack.head = CDS_WFS_END, } #define DECLARE_URCU_WAIT_QUEUE(name) \ struct urcu_wait_queue name @@ -68,6 +69,12 @@ struct urcu_wait_queue { #define DEFINE_URCU_WAIT_QUEUE(name) \ struct urcu_wait_queue name = URCU_WAIT_QUEUE_HEAD_INIT(name) +static inline +void urcu_wait_queue_init(struct urcu_wait_queue *queue) +{ + __cds_wfs_init(&queue->stack); +} + struct urcu_waiters { struct cds_wfs_head *head; }; @@ -102,6 +109,13 @@ void urcu_wait_set_state(struct urcu_wait_node *node, node->state = state; } +static inline +void urcu_wait_or_state(struct urcu_wait_node *node, + enum urcu_wait_state state) +{ + uatomic_or(&node->state, state); +} + static inline void urcu_wait_node_init(struct urcu_wait_node *node, enum urcu_wait_state state) @@ -120,8 +134,13 @@ static inline void urcu_adaptative_wake_up(struct urcu_wait_node *wait) { cmm_smp_mb(); - assert(uatomic_read(&wait->state) == URCU_WAIT_WAITING); - uatomic_set(&wait->state, URCU_WAIT_WAKEUP); + /* + * "or" of WAKEUP flag rather than "set" is useful for multiple + * concurrent wakeup sources. Note that "WAIT_TEARDOWN" becomes + * useless when we use multiple wakeup sources: lifetime of the + * "value" should then be handled by the caller. + */ + uatomic_or(&wait->state, URCU_WAIT_WAKEUP); if (!(uatomic_read(&wait->state) & URCU_WAIT_RUNNING)) futex_noasync(&wait->state, FUTEX_WAKE, 1, NULL, NULL, 0); /* Allow teardown of struct urcu_wait memory. */ @@ -165,21 +184,72 @@ skip_futex_wait: assert(uatomic_read(&wait->state) & URCU_WAIT_TEARDOWN); } +/* + * Need mutual exclusion against other wakeup and move waiters + * operations. It is provided by the caller. + */ +static inline +int urcu_dequeue_wake_single(struct urcu_wait_queue *queue) +{ + struct cds_wfs_node *node; + struct urcu_wait_node *wait_node; + int wakeup_done = 0; + + node = __cds_wfs_pop_blocking(&queue->stack); + if (!node) + return -ENOENT; + wait_node = caa_container_of(node, struct urcu_wait_node, node); + CMM_STORE_SHARED(wait_node->node.next, NULL); + /* Don't wake already running threads */ + if (!(wait_node->state & URCU_WAIT_RUNNING)) { + urcu_adaptative_wake_up(wait_node); + wakeup_done = 1; + } + return wakeup_done; +} + +/* + * Need mutual exclusion against other wakeup and move waiters + * operations. It is provided by the caller. + */ +static inline +int urcu_dequeue_wake_n(struct urcu_wait_queue *queue, int n) +{ + int nr_wakeup = 0; + + for (;;) { + int ret; + + ret = urcu_dequeue_wake_single(queue); + if (ret < 0) + return nr_wakeup; + else if (ret > 0) + nr_wakeup++; + else + break; + } + return nr_wakeup; +} + static inline -void urcu_wake_all_waiters(struct urcu_waiters *waiters) +int urcu_wake_all_waiters(struct urcu_waiters *waiters) { struct cds_wfs_node *iter, *iter_n; + int nr_wakeup = 0; /* Wake all waiters in our stack head */ cds_wfs_for_each_blocking_safe(waiters->head, iter, iter_n) { struct urcu_wait_node *wait_node = caa_container_of(iter, struct urcu_wait_node, node); + CMM_STORE_SHARED(wait_node->node.next, NULL); /* Don't wake already running threads */ if (wait_node->state & URCU_WAIT_RUNNING) continue; urcu_adaptative_wake_up(wait_node); + nr_wakeup++; } + return nr_wakeup; } -#endif /* _URCU_WAIT_H */ +#endif /* _URCU_WAITQUEUE_LIFO_H */ diff --git a/urcu/workqueue-fifo.h b/urcu/workqueue-fifo.h new file mode 100644 index 0000000..65d3a2e --- /dev/null +++ b/urcu/workqueue-fifo.h @@ -0,0 +1,345 @@ +#ifndef _URCU_WORKQUEUE_FIFO_H +#define _URCU_WORKQUEUE_FIFO_H + +/* + * urcu/workqueue-fifo.h + * + * Userspace RCU library - work queue scheme with FIFO semantic + * + * Copyright (c) 2014 Mathieu Desnoyers + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include +#include +#include +#include +#include + +/* + * We use RCU to steal work from siblings. Therefore, one of RCU flavors + * need to be included before this header. All worker that participate + * in stealing (initialized with the URCU_WORKER_STEAL flag) need to be + * registered RCU readers threads. + */ + +struct urcu_work { + struct cds_wfcq_node node; +}; + +struct urcu_workqueue { + /* FIFO work queue */ + struct __cds_wfcq_head head; + struct cds_wfcq_tail tail; + + /* Associated wait queue for LIFO wait/wakeup */ + struct urcu_wait_queue waitqueue; + + /* RCU linked list head of siblings for work stealing. */ + struct cds_list_head sibling_head; + pthread_mutex_t sibling_lock; /* Protect sibling list updates */ +}; + +struct urcu_worker { + struct cds_wfcq_head head; + struct cds_wfcq_tail tail; + + struct urcu_wait_node wait_node; + /* RCU linked list node of siblings for work stealing. */ + struct cds_list_head sibling_node; + int flags; /* enum urcu_worker_flags */ +}; + +enum urcu_worker_flags { + URCU_WORKER_STEAL = (1 << 0), +}; + +static inline +void urcu_workqueue_init(struct urcu_workqueue *queue) +{ + __cds_wfcq_init(&queue->head, &queue->tail); + urcu_wait_queue_init(&queue->waitqueue); + CDS_INIT_LIST_HEAD(&queue->sibling_head); +} + +static inline +void urcu_queue_work(struct urcu_workqueue *queue, struct urcu_work *work) +{ + bool was_empty; + + cds_wfcq_node_init(&work->node); + + /* Enqueue work. */ + was_empty = !cds_wfcq_enqueue(&queue->head, &queue->tail, + &work->node); + /* + * If workqueue was previously empty, wakeup one worker thread. + * It will eventually grab the entire content of the work-queue + * (therefore grabbing a "work batch"). After having grabbed the + * work batch, while that thread is running and taking care of + * that work batch, when we enqueue more work, we will wake + * another thread (if there is one waiting), which will + * eventually grab the new batch, and so on. This scheme ensures + * that contiguous batch of work are handled by the same thread + * (for locality), and also ensures that we scale work to many + * worker threads when threads are busy enough to still be + * running when work is enqueued. + */ + if (was_empty) + (void) urcu_dequeue_wake_single(&queue->waitqueue); +} + +static inline +void urcu_workqueue_wakeup_all(struct urcu_workqueue *queue) +{ + struct urcu_waiters waiters; + + urcu_move_waiters(&waiters, &queue->waitqueue); + (void) urcu_wake_all_waiters(&waiters); +} + +static inline +void urcu_worker_init(struct urcu_worker *worker, int flags) +{ + cds_wfcq_init(&worker->head, &worker->tail); + worker->flags = flags; + urcu_wait_node_init(&worker->wait_node, URCU_WAIT_RUNNING); +} + +static inline +void urcu_worker_register(struct urcu_workqueue *queue, + struct urcu_worker *worker) +{ + if (worker->flags & URCU_WORKER_STEAL) { + pthread_mutex_lock(&queue->sibling_lock); + cds_list_add_rcu(&worker->sibling_node, &queue->sibling_head); + pthread_mutex_unlock(&queue->sibling_lock); + } +} + +static inline +void urcu_worker_unregister(struct urcu_workqueue *queue, + struct urcu_worker *worker) +{ + enum cds_wfcq_ret wfcq_ret; + + if (worker->flags & URCU_WORKER_STEAL) { + pthread_mutex_lock(&queue->sibling_lock); + cds_list_del_rcu(&worker->sibling_node); + pthread_mutex_unlock(&queue->sibling_lock); + + /* + * Wait for grace period before freeing or reusing + * "worker" because used by RCU linked list. + */ + synchronize_rcu(); + } + + /* + * Put any local work we still have back into the workqueue. + */ + wfcq_ret = __cds_wfcq_splice_blocking(&queue->head, + &queue->tail, + &worker->head, + &worker->tail); + if (wfcq_ret != CDS_WFCQ_RET_SRC_EMPTY + && wfcq_ret == CDS_WFCQ_RET_DEST_EMPTY) { + /* + * Wakeup worker thread if we have put work back into + * workqueue that was previously empty. + */ + (void) urcu_dequeue_wake_single(&queue->waitqueue); + } +} + +/* + * Try stealing work from siblings when we have nothing to do. + */ +static inline +void ___urcu_steal_work(struct urcu_worker *worker, + struct urcu_worker *sibling) +{ + cds_wfcq_dequeue_lock(&sibling->head, &sibling->tail); + (void) __cds_wfcq_splice_blocking(&worker->head, + &worker->tail, + &sibling->head, + &sibling->tail); + cds_wfcq_dequeue_unlock(&sibling->head, &sibling->tail); +} + +static inline +int __urcu_steal_work(struct urcu_workqueue *queue, + struct urcu_worker *worker) +{ + struct urcu_worker *sibling_prev, *sibling_next; + struct cds_list_head *sibling_node; + + if (!(worker->flags & URCU_WORKER_STEAL)) + return 0; + + rcu_read_lock(); + + sibling_node = rcu_dereference(worker->sibling_node.next); + if (sibling_node == &queue->sibling_head) + sibling_node = rcu_dereference(sibling_node->next); + sibling_next = caa_container_of(sibling_node, struct urcu_worker, + sibling_node); + if (sibling_next != worker) + ___urcu_steal_work(worker, sibling_next); + + sibling_node = rcu_dereference(worker->sibling_node.prev); + if (sibling_node == &queue->sibling_head) + sibling_node = rcu_dereference(sibling_node->prev); + sibling_prev = caa_container_of(sibling_node, struct urcu_worker, + sibling_node); + if (sibling_prev != worker && sibling_prev != sibling_next) + ___urcu_steal_work(worker, sibling_prev); + + rcu_read_unlock(); + + return !cds_wfcq_empty(&worker->head, &worker->tail); +} + +static inline +void ___urcu_wakeup_sibling(struct urcu_worker *sibling) +{ + urcu_adaptative_wake_up(&sibling->wait_node); +} + +static inline +void __urcu_wakeup_siblings(struct urcu_workqueue *queue, + struct urcu_worker *worker) +{ + struct urcu_worker *sibling_prev, *sibling_next; + struct cds_list_head *sibling_node; + + if (!(worker->flags & URCU_WORKER_STEAL)) + return; + + /* Only wakeup siblings if we have work in our own queue. */ + if (cds_wfcq_empty(&worker->head, &worker->tail)) + return; + + rcu_read_lock(); + + sibling_node = rcu_dereference(worker->sibling_node.next); + if (sibling_node == &queue->sibling_head) + sibling_node = rcu_dereference(sibling_node->next); + sibling_next = caa_container_of(sibling_node, struct urcu_worker, + sibling_node); + if (sibling_next != worker) + ___urcu_wakeup_sibling(sibling_next); + + sibling_node = rcu_dereference(worker->sibling_node.prev); + if (sibling_node == &queue->sibling_head) + sibling_node = rcu_dereference(sibling_node->prev); + sibling_prev = caa_container_of(sibling_node, struct urcu_worker, + sibling_node); + if (sibling_prev != worker && sibling_prev != sibling_next) + ___urcu_wakeup_sibling(sibling_prev); + + rcu_read_unlock(); +} + +static inline +void urcu_accept_work(struct urcu_workqueue *queue, + struct urcu_worker *worker, + int blocking) +{ + enum cds_wfcq_ret wfcq_ret; + + wfcq_ret = __cds_wfcq_splice_blocking(&worker->head, + &worker->tail, + &queue->head, + &queue->tail); + /* Don't wait if we have work to do. */ + if (wfcq_ret != CDS_WFCQ_RET_SRC_EMPTY + || !cds_wfcq_empty(&worker->head, + &worker->tail)) + goto do_work; + /* Try to steal work from sibling instead of blocking */ + if (__urcu_steal_work(queue, worker)) + goto do_work; + if (!blocking) + return; + urcu_wait_set_state(&worker->wait_node, + URCU_WAIT_WAITING); + if (!CMM_LOAD_SHARED(worker->wait_node.node.next)) { + int was_empty; + + /* + * NULL next pointer. We are therefore not in + * the queue. + */ + cds_wfs_node_init(&worker->wait_node.node); + was_empty = !urcu_wait_add(&queue->waitqueue, + &worker->wait_node); + /* + * If the wait queue was empty, it means we are the + * first thread to be put back into an otherwise empty + * wait queue. Re-check if work queue is empty after + * adding ourself to wait queue, so we can wakeup the + * top of wait queue since new work have appeared, and + * work enqueuer may not have seen that it needed to do + * a wake up. + */ + if (was_empty && !cds_wfcq_empty(&queue->head, + &queue->tail)) + (void) urcu_dequeue_wake_single(&queue->waitqueue); + } else { + /* + * Non-NULL next pointer. We are therefore in + * the queue, or the dispatcher just removed us + * from it (after we read the next pointer), and + * is therefore awakening us. The state will + * therefore have been changed from WAITING to + * some other state, which will let the busy + * wait pass through. + */ + } + urcu_adaptative_busy_wait(&worker->wait_node); + return; + +do_work: + /* + * We will be busy handling the work batch, awaken siblings so + * they can steal from us. + */ + __urcu_wakeup_siblings(queue, worker); +} + +static inline +struct urcu_work *urcu_dequeue_work(struct urcu_worker *worker) +{ + struct cds_wfcq_node *node; + + /* + * If we are registered for work stealing, we need to dequeue + * safely against siblings. + */ + if (worker->flags & URCU_WORKER_STEAL) + node = cds_wfcq_dequeue_blocking(&worker->head, + &worker->tail); + else + node = ___cds_wfcq_dequeue_with_state(&worker->head, + &worker->tail, NULL, 1, 0); + if (!node) + return NULL; + return caa_container_of(node, struct urcu_work, node); +} + +#endif /* _URCU_WORKQUEUE_FIFO_H */ -- 2.34.1