Return-Path: Received: from mail-qt0-f194.google.com ([209.85.216.194]:44457 "EHLO mail-qt0-f194.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728327AbeIQSbz (ORCPT ); Mon, 17 Sep 2018 14:31:55 -0400 Received: by mail-qt0-f194.google.com with SMTP id k38-v6so15106596qtk.11 for ; Mon, 17 Sep 2018 06:04:38 -0700 (PDT) Received: from leira.trondhjem.org.localdomain ([66.187.232.65]) by smtp.gmail.com with ESMTPSA id q1-v6sm10499607qkl.31.2018.09.17.06.04.37 for (version=TLS1_2 cipher=ECDHE-RSA-CHACHA20-POLY1305 bits=256/256); Mon, 17 Sep 2018 06:04:37 -0700 (PDT) From: Trond Myklebust To: linux-nfs@vger.kernel.org Subject: [PATCH v3 36/44] SUNRPC: Fix priority queue fairness Date: Mon, 17 Sep 2018 09:03:27 -0400 Message-Id: <20180917130335.112832-37-trond.myklebust@hammerspace.com> In-Reply-To: <20180917130335.112832-36-trond.myklebust@hammerspace.com> References: <20180917130335.112832-1-trond.myklebust@hammerspace.com> <20180917130335.112832-2-trond.myklebust@hammerspace.com> <20180917130335.112832-3-trond.myklebust@hammerspace.com> <20180917130335.112832-4-trond.myklebust@hammerspace.com> <20180917130335.112832-5-trond.myklebust@hammerspace.com> <20180917130335.112832-6-trond.myklebust@hammerspace.com> <20180917130335.112832-7-trond.myklebust@hammerspace.com> <20180917130335.112832-8-trond.myklebust@hammerspace.com> <20180917130335.112832-9-trond.myklebust@hammerspace.com> <20180917130335.112832-10-trond.myklebust@hammerspace.com> <20180917130335.112832-11-trond.myklebust@hammerspace.com> <20180917130335.112832-12-trond.myklebust@hammerspace.com> <20180917130335.112832-13-trond.myklebust@hammerspace.com> <20180917130335.112832-14-trond.myklebust@hammerspace.com> <20180917130335.112832-15-trond.myklebust@hammerspace.com> <20180917130335.112832-16-trond.myklebust@hammerspace.com> <20180917130335.112832-17-trond.myklebust@hammerspace.com> <20180917130335.112832-18-trond.myklebust@hammerspace.com> <20180917130335.112832-19-trond.myklebust@hammerspace.com> <20180917130335.112832-20-trond.myklebust@hammerspace.com> <20180917130335.112832-21-trond.myklebust@hammerspace.com> <20180917130335.112832-22-trond.myklebust@hammerspace.com> <20180917130335.112832-23-trond.myklebust@hammerspace.com> <20180917130335.112832-24-trond.myklebust@hammerspace.com> <20180917130335.112832-25-trond.myklebust@hammerspace.com> <20180917130335.112832-26-trond.myklebust@hammerspace.com> <20180917130335.112832-27-trond.myklebust@hammerspace.com> <20180917130335.112832-28-trond.myklebust@hammerspace.com> <20180917130335.112832-29-trond.myklebust@hammerspace.com> <20180917130335.112832-30-trond.myklebust@hammerspace.com> <20180917130335.112832-31-trond.myklebust@hammerspace.com> <20180917130335.112832-32-trond.myklebust@hammerspace.com> <20180917130335.112832-33-trond.myklebust@hammerspace.com> <20180917130335.112832-34-trond.myklebust@hammerspace.com> <20180917130335.112832-35-trond.myklebust@hammerspace.com> <20180917130335.112832-36-trond.myklebust@hammerspace.com> MIME-Version: 1.0 Sender: linux-nfs-owner@vger.kernel.org List-ID: Fix up the priority queue to not batch by owner, but by queue, so that we allow '1 << priority' elements to be dequeued before switching to the next priority queue. The owner field is still used to wake up requests in round robin order by owner to avoid single processes hogging the RPC layer by loading the queues. Signed-off-by: Trond Myklebust --- include/linux/sunrpc/sched.h | 2 - net/sunrpc/sched.c | 109 +++++++++++++++++------------------ 2 files changed, 54 insertions(+), 57 deletions(-) diff --git a/include/linux/sunrpc/sched.h b/include/linux/sunrpc/sched.h index 8840a420cf4c..7b540c066594 100644 --- a/include/linux/sunrpc/sched.h +++ b/include/linux/sunrpc/sched.h @@ -189,7 +189,6 @@ struct rpc_timer { struct rpc_wait_queue { spinlock_t lock; struct list_head tasks[RPC_NR_PRIORITY]; /* task queue for each priority level */ - pid_t owner; /* process id of last task serviced */ unsigned char maxpriority; /* maximum priority (0 if queue is not a priority queue) */ unsigned char priority; /* current priority */ unsigned char nr; /* # tasks remaining for cookie */ @@ -205,7 +204,6 @@ struct rpc_wait_queue { * from a single cookie. The aim is to improve * performance of NFS operations such as read/write. */ -#define RPC_BATCH_COUNT 16 #define RPC_IS_PRIORITY(q) ((q)->maxpriority > 0) /* diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c index 9a8ec012b449..57ca5bead1cb 100644 --- a/net/sunrpc/sched.c +++ b/net/sunrpc/sched.c @@ -99,64 +99,78 @@ __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task) list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list); } -static void rpc_rotate_queue_owner(struct rpc_wait_queue *queue) -{ - struct list_head *q = &queue->tasks[queue->priority]; - struct rpc_task *task; - - if (!list_empty(q)) { - task = list_first_entry(q, struct rpc_task, u.tk_wait.list); - if (task->tk_owner == queue->owner) - list_move_tail(&task->u.tk_wait.list, q); - } -} - static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority) { if (queue->priority != priority) { - /* Fairness: rotate the list when changing priority */ - rpc_rotate_queue_owner(queue); queue->priority = priority; + queue->nr = 1U << priority; } } -static void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid) -{ - queue->owner = pid; - queue->nr = RPC_BATCH_COUNT; -} - static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue) { rpc_set_waitqueue_priority(queue, queue->maxpriority); - rpc_set_waitqueue_owner(queue, 0); } /* - * Add new request to a priority queue. + * Add a request to a queue list */ -static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, - struct rpc_task *task, - unsigned char queue_priority) +static void +__rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task) { - struct list_head *q; struct rpc_task *t; - INIT_LIST_HEAD(&task->u.tk_wait.links); - if (unlikely(queue_priority > queue->maxpriority)) - queue_priority = queue->maxpriority; - if (queue_priority > queue->priority) - rpc_set_waitqueue_priority(queue, queue_priority); - q = &queue->tasks[queue_priority]; list_for_each_entry(t, q, u.tk_wait.list) { if (t->tk_owner == task->tk_owner) { - list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links); + list_add_tail(&task->u.tk_wait.links, + &t->u.tk_wait.links); + /* Cache the queue head in task->u.tk_wait.list */ + task->u.tk_wait.list.next = q; + task->u.tk_wait.list.prev = NULL; return; } } + INIT_LIST_HEAD(&task->u.tk_wait.links); list_add_tail(&task->u.tk_wait.list, q); } +/* + * Remove request from a queue list + */ +static void +__rpc_list_dequeue_task(struct rpc_task *task) +{ + struct list_head *q; + struct rpc_task *t; + + if (task->u.tk_wait.list.prev == NULL) { + list_del(&task->u.tk_wait.links); + return; + } + if (!list_empty(&task->u.tk_wait.links)) { + t = list_first_entry(&task->u.tk_wait.links, + struct rpc_task, + u.tk_wait.links); + /* Assume __rpc_list_enqueue_task() cached the queue head */ + q = t->u.tk_wait.list.next; + list_add_tail(&t->u.tk_wait.list, q); + list_del(&task->u.tk_wait.links); + } + list_del(&task->u.tk_wait.list); +} + +/* + * Add new request to a priority queue. + */ +static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, + struct rpc_task *task, + unsigned char queue_priority) +{ + if (unlikely(queue_priority > queue->maxpriority)) + queue_priority = queue->maxpriority; + __rpc_list_enqueue_task(&queue->tasks[queue_priority], task); +} + /* * Add new request to wait queue. * @@ -194,13 +208,7 @@ static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, */ static void __rpc_remove_wait_queue_priority(struct rpc_task *task) { - struct rpc_task *t; - - if (!list_empty(&task->u.tk_wait.links)) { - t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list); - list_move(&t->u.tk_wait.list, &task->u.tk_wait.list); - list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links); - } + __rpc_list_dequeue_task(task); } /* @@ -212,7 +220,8 @@ static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_tas __rpc_disable_timer(queue, task); if (RPC_IS_PRIORITY(queue)) __rpc_remove_wait_queue_priority(task); - list_del(&task->u.tk_wait.list); + else + list_del(&task->u.tk_wait.list); queue->qlen--; dprintk("RPC: %5u removed from queue %p \"%s\"\n", task->tk_pid, queue, rpc_qname(queue)); @@ -545,17 +554,9 @@ static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *q * Service a batch of tasks from a single owner. */ q = &queue->tasks[queue->priority]; - if (!list_empty(q)) { - task = list_entry(q->next, struct rpc_task, u.tk_wait.list); - if (queue->owner == task->tk_owner) { - if (--queue->nr) - goto out; - list_move_tail(&task->u.tk_wait.list, q); - } - /* - * Check if we need to switch queues. - */ - goto new_owner; + if (!list_empty(q) && --queue->nr) { + task = list_first_entry(q, struct rpc_task, u.tk_wait.list); + goto out; } /* @@ -567,7 +568,7 @@ static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *q else q = q - 1; if (!list_empty(q)) { - task = list_entry(q->next, struct rpc_task, u.tk_wait.list); + task = list_first_entry(q, struct rpc_task, u.tk_wait.list); goto new_queue; } } while (q != &queue->tasks[queue->priority]); @@ -577,8 +578,6 @@ static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *q new_queue: rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); -new_owner: - rpc_set_waitqueue_owner(queue, task->tk_owner); out: return task; } -- 2.17.1