Index: head/share/man/man9/taskqueue.9 =================================================================== --- head/share/man/man9/taskqueue.9 (revision 276664) +++ head/share/man/man9/taskqueue.9 (revision 276665) @@ -1,502 +1,508 @@ .\" -*- nroff -*- .\" .\" Copyright (c) 2000 Doug Rabson .\" .\" All rights reserved. .\" .\" This program is free software. .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions and the following disclaimer. .\" 2. Redistributions in binary form must reproduce the above copyright .\" notice, this list of conditions and the following disclaimer in the .\" documentation and/or other materials provided with the distribution. .\" .\" THIS SOFTWARE IS PROVIDED BY THE DEVELOPERS ``AS IS'' AND ANY EXPRESS OR .\" IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES .\" OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. .\" IN NO EVENT SHALL THE DEVELOPERS BE LIABLE FOR ANY DIRECT, INDIRECT, .\" INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT .\" NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, .\" DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY .\" THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT .\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF .\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. .\" .\" $FreeBSD$ .\" -.Dd May 24, 2014 +.Dd January 4, 2015 .Dt TASKQUEUE 9 .Os .Sh NAME .Nm taskqueue .Nd asynchronous task execution .Sh SYNOPSIS .In sys/param.h .In sys/kernel.h .In sys/malloc.h .In sys/queue.h .In sys/taskqueue.h .Bd -literal typedef void (*task_fn_t)(void *context, int pending); typedef void (*taskqueue_enqueue_fn)(void *context); struct task { STAILQ_ENTRY(task) ta_link; /* link for queue */ u_short ta_pending; /* count times queued */ u_short ta_priority; /* priority of task in queue */ task_fn_t ta_func; /* task handler */ void *ta_context; /* argument for handler */ }; enum taskqueue_callback_type { TASKQUEUE_CALLBACK_TYPE_INIT, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN, }; typedef void (*taskqueue_callback_fn)(void *context); struct timeout_task; .Ed .Ft struct taskqueue * .Fn taskqueue_create "const char *name" "int mflags" "taskqueue_enqueue_fn enqueue" "void *context" .Ft struct taskqueue * .Fn taskqueue_create_fast "const char *name" "int mflags" "taskqueue_enqueue_fn enqueue" "void *context" .Ft int .Fn taskqueue_start_threads "struct taskqueue **tqp" "int count" "int pri" "const char *name" "..." .Ft int .Fo taskqueue_start_threads_pinned .Fa "struct taskqueue **tqp" "int count" "int pri" "int cpu_id" .Fa "const char *name" "..." .Fc .Ft void .Fn taskqueue_set_callback "struct taskqueue *queue" "enum taskqueue_callback_type cb_type" "taskqueue_callback_fn callback" "void *context" .Ft void .Fn taskqueue_free "struct taskqueue *queue" .Ft int .Fn taskqueue_enqueue "struct taskqueue *queue" "struct task *task" .Ft int .Fn taskqueue_enqueue_fast "struct taskqueue *queue" "struct task *task" .Ft int .Fn taskqueue_enqueue_timeout "struct taskqueue *queue" "struct timeout_task *timeout_task" "int ticks" .Ft int .Fn taskqueue_cancel "struct taskqueue *queue" "struct task *task" "u_int *pendp" .Ft int .Fn taskqueue_cancel_timeout "struct taskqueue *queue" "struct timeout_task *timeout_task" "u_int *pendp" .Ft void .Fn taskqueue_drain "struct taskqueue *queue" "struct task *task" .Ft void .Fn taskqueue_drain_timeout "struct taskqueue *queue" "struct timeout_task *timeout_task" .Ft void .Fn taskqueue_drain_all "struct taskqueue *queue" .Ft void .Fn taskqueue_block "struct taskqueue *queue" .Ft void .Fn taskqueue_unblock "struct taskqueue *queue" .Ft int .Fn taskqueue_member "struct taskqueue *queue" "struct thread *td" .Ft void .Fn taskqueue_run "struct taskqueue *queue" .Fn TASK_INIT "struct task *task" "int priority" "task_fn_t func" "void *context" .Fn TASK_INITIALIZER "int priority" "task_fn_t func" "void *context" .Fn TASKQUEUE_DECLARE "name" .Fn TASKQUEUE_DEFINE "name" "taskqueue_enqueue_fn enqueue" "void *context" "init" .Fn TASKQUEUE_FAST_DEFINE "name" "taskqueue_enqueue_fn enqueue" "void *context" "init" .Fn TASKQUEUE_DEFINE_THREAD "name" .Fn TASKQUEUE_FAST_DEFINE_THREAD "name" .Fn TIMEOUT_TASK_INIT "struct taskqueue *queue" "struct timeout_task *timeout_task" "int priority" "task_fn_t func" "void *context" .Sh DESCRIPTION These functions provide a simple interface for asynchronous execution of code. .Pp The function .Fn taskqueue_create is used to create new queues. The arguments to .Fn taskqueue_create include a name that should be unique, a set of .Xr malloc 9 flags that specify whether the call to .Fn malloc is allowed to sleep, a function that is called from .Fn taskqueue_enqueue when a task is added to the queue, and a pointer to the memory location where the identity of the thread that services the queue is recorded. .\" XXX The rest of the sentence gets lots in relation to the first part. The function called from .Fn taskqueue_enqueue must arrange for the queue to be processed (for instance by scheduling a software interrupt or waking a kernel thread). The memory location where the thread identity is recorded is used to signal the service thread(s) to terminate--when this value is set to zero and the thread is signaled it will terminate. If the queue is intended for use in fast interrupt handlers .Fn taskqueue_create_fast should be used in place of .Fn taskqueue_create . .Pp The function .Fn taskqueue_free should be used to free the memory used by the queue. Any tasks that are on the queue will be executed at this time after which the thread servicing the queue will be signaled that it should exit. .Pp Once a taskqueue has been created, its threads should be started using .Fn taskqueue_start_threads or .Fn taskqueue_start_threads_pinned . .Fn taskqueue_start_threads_pinned takes a .Va cpu_id argument which will cause the threads which are started for the taskqueue to be pinned to run on the given CPU. Callbacks may optionally be registered using .Fn taskqueue_set_callback . Currently, callbacks may be registered for the following purposes: .Bl -tag -width TASKQUEUE_CALLBACK_TYPE_SHUTDOWN .It Dv TASKQUEUE_CALLBACK_TYPE_INIT This callback is called by every thread in the taskqueue, before it executes any tasks. This callback must be set before the taskqueue's threads are started. .It Dv TASKQUEUE_CALLBACK_TYPE_SHUTDOWN This callback is called by every thread in the taskqueue, after it executes its last task. This callback will always be called before the taskqueue structure is reclaimed. .El .Pp To add a task to the list of tasks queued on a taskqueue, call .Fn taskqueue_enqueue with pointers to the queue and task. If the task's .Va ta_pending field is non-zero, then it is simply incremented to reflect the number of times the task was enqueued, up to a cap of USHRT_MAX. Otherwise, the task is added to the list before the first task which has a lower .Va ta_priority value or at the end of the list if no tasks have a lower priority. Enqueueing a task does not perform any memory allocation which makes it suitable for calling from an interrupt handler. This function will return .Er EPIPE if the queue is being freed. .Pp The function .Fn taskqueue_enqueue_fast should be used in place of .Fn taskqueue_enqueue when the enqueuing must happen from a fast interrupt handler. This method uses spin locks to avoid the possibility of sleeping in the fast interrupt context. .Pp When a task is executed, first it is removed from the queue, the value of .Va ta_pending is recorded and then the field is zeroed. The function .Va ta_func from the task structure is called with the value of the field .Va ta_context as its first argument and the value of .Va ta_pending as its second argument. After the function .Va ta_func returns, .Xr wakeup 9 is called on the task pointer passed to .Fn taskqueue_enqueue . .Pp The .Fn taskqueue_enqueue_timeout is used to schedule the enqueue after the specified amount of .Va ticks . Only non-fast task queues can be used for .Va timeout_task scheduling. If the .Va ticks argument is negative, the already scheduled enqueueing is not re-scheduled. Otherwise, the task is scheduled for enqueueing in the future, after the absolute value of .Va ticks is passed. .Pp The .Fn taskqueue_cancel function is used to cancel a task. The .Va ta_pending count is cleared, and the old value returned in the reference parameter .Fa pendp , if it is .Pf non- Dv NULL . If the task is currently running, .Dv EBUSY is returned, otherwise 0. To implement a blocking .Fn taskqueue_cancel that waits for a running task to finish, it could look like: .Bd -literal -offset indent while (taskqueue_cancel(tq, task, NULL) != 0) taskqueue_drain(tq, task); .Ed .Pp Note that, as with .Fn taskqueue_drain , the caller is responsible for ensuring that the task is not re-enqueued after being canceled. .Pp Similarly, the .Fn taskqueue_cancel_timeout function is used to cancel the scheduled task execution. .Pp The .Fn taskqueue_drain function is used to wait for the task to finish, and the .Fn taskqueue_drain_timeout function is used to wait for the scheduled task to finish. There is no guarantee that the task will not be enqueued after call to .Fn taskqueue_drain . If the caller wants to put the task into a known state, then before calling .Fn taskqueue_drain the caller should use out-of-band means to ensure that the task would not be enqueued. For example, if the task is enqueued by an interrupt filter, then the interrupt could be disabled. .Pp The .Fn taskqueue_drain_all function is used to wait for all pending and running tasks that are enqueued on the taskqueue to finish. -The caller must arrange that the tasks are not re-enqueued. -Note that +Tasks posted to the taskqueue after .Fn taskqueue_drain_all -currently does not handle tasks with delayed enqueueing. +begins processing, +including pending enqueues scheduled by a previous call to +.Fn taskqueue_enqueue_timeout , +do not extend the wait time of +.Fn taskqueue_drain_all +and may complete after +.Fn taskqueue_drain_all +returns. .Pp The .Fn taskqueue_block function blocks the taskqueue. It prevents any enqueued but not running tasks from being executed. Future calls to .Fn taskqueue_enqueue will enqueue tasks, but the tasks will not be run until .Fn taskqueue_unblock is called. Please note that .Fn taskqueue_block does not wait for any currently running tasks to finish. Thus, the .Fn taskqueue_block does not provide a guarantee that .Fn taskqueue_run is not running after .Fn taskqueue_block returns, but it does provide a guarantee that .Fn taskqueue_run will not be called again until .Fn taskqueue_unblock is called. If the caller requires a guarantee that .Fn taskqueue_run is not running, then this must be arranged by the caller. Note that if .Fn taskqueue_drain is called on a task that is enqueued on a taskqueue that is blocked by .Fn taskqueue_block , then .Fn taskqueue_drain can not return until the taskqueue is unblocked. This can result in a deadlock if the thread blocked in .Fn taskqueue_drain is the thread that is supposed to call .Fn taskqueue_unblock . Thus, use of .Fn taskqueue_drain after .Fn taskqueue_block is discouraged, because the state of the task can not be known in advance. The same caveat applies to .Fn taskqueue_drain_all . .Pp The .Fn taskqueue_unblock function unblocks the previously blocked taskqueue. All enqueued tasks can be run after this call. .Pp The .Fn taskqueue_member function returns .No 1 if the given thread .Fa td is part of the given taskqueue .Fa queue and .No 0 otherwise. .Pp The .Fn taskqueue_run function will run all pending tasks in the specified .Fa queue . Normally this function is only used internally. .Pp A convenience macro, .Fn TASK_INIT "task" "priority" "func" "context" is provided to initialise a .Va task structure. The .Fn TASK_INITIALIZER macro generates an initializer for a task structure. A macro .Fn TIMEOUT_TASK_INIT "queue" "timeout_task" "priority" "func" "context" initializes the .Va timeout_task structure. The values of .Va priority , .Va func , and .Va context are simply copied into the task structure fields and the .Va ta_pending field is cleared. .Pp Five macros .Fn TASKQUEUE_DECLARE "name" , .Fn TASKQUEUE_DEFINE "name" "enqueue" "context" "init" , .Fn TASKQUEUE_FAST_DEFINE "name" "enqueue" "context" "init" , and .Fn TASKQUEUE_DEFINE_THREAD "name" .Fn TASKQUEUE_FAST_DEFINE_THREAD "name" are used to declare a reference to a global queue, to define the implementation of the queue, and declare a queue that uses its own thread. The .Fn TASKQUEUE_DEFINE macro arranges to call .Fn taskqueue_create with the values of its .Va name , .Va enqueue and .Va context arguments during system initialisation. After calling .Fn taskqueue_create , the .Va init argument to the macro is executed as a C statement, allowing any further initialisation to be performed (such as registering an interrupt handler etc.) .Pp The .Fn TASKQUEUE_DEFINE_THREAD macro defines a new taskqueue with its own kernel thread to serve tasks. The variable .Vt struct taskqueue *taskqueue_name is used to enqueue tasks onto the queue. .Pp .Fn TASKQUEUE_FAST_DEFINE and .Fn TASKQUEUE_FAST_DEFINE_THREAD act just like .Fn TASKQUEUE_DEFINE and .Fn TASKQUEUE_DEFINE_THREAD respectively but taskqueue is created with .Fn taskqueue_create_fast . .Ss Predefined Task Queues The system provides four global taskqueues, .Va taskqueue_fast , .Va taskqueue_swi , .Va taskqueue_swi_giant , and .Va taskqueue_thread . The .Va taskqueue_fast queue is for swi handlers dispatched from fast interrupt handlers, where sleep mutexes cannot be used. The swi taskqueues are run via a software interrupt mechanism. The .Va taskqueue_swi queue runs without the protection of the .Va Giant kernel lock, and the .Va taskqueue_swi_giant queue runs with the protection of the .Va Giant kernel lock. The thread taskqueue .Va taskqueue_thread runs in a kernel thread context, and tasks run from this thread do not run under the .Va Giant kernel lock. If the caller wants to run under .Va Giant , he should explicitly acquire and release .Va Giant in his taskqueue handler routine. .Pp To use these queues, call .Fn taskqueue_enqueue with the value of the global taskqueue variable for the queue you wish to use .Va ( taskqueue_swi , .Va taskqueue_swi_giant , or .Va taskqueue_thread ) . Use .Fn taskqueue_enqueue_fast for the global taskqueue variable .Va taskqueue_fast . .Pp The software interrupt queues can be used, for instance, for implementing interrupt handlers which must perform a significant amount of processing in the handler. The hardware interrupt handler would perform minimal processing of the interrupt and then enqueue a task to finish the work. This reduces to a minimum the amount of time spent with interrupts disabled. .Pp The thread queue can be used, for instance, by interrupt level routines that need to call kernel functions that do things that can only be done from a thread context. (e.g., call malloc with the M_WAITOK flag.) .Pp Note that tasks queued on shared taskqueues such as .Va taskqueue_swi may be delayed an indeterminate amount of time before execution. If queueing delays cannot be tolerated then a private taskqueue should be created with a dedicated processing thread. .Sh SEE ALSO .Xr ithread 9 , .Xr kthread 9 , .Xr swi 9 .Sh HISTORY This interface first appeared in .Fx 5.0 . There is a similar facility called work_queue in the Linux kernel. .Sh AUTHORS This manual page was written by .An Doug Rabson . Index: head/sys/kern/subr_taskqueue.c =================================================================== --- head/sys/kern/subr_taskqueue.c (revision 276664) +++ head/sys/kern/subr_taskqueue.c (revision 276665) @@ -1,721 +1,793 @@ /*- * Copyright (c) 2000 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_TASKQUEUE, "taskqueue", "Task Queues"); static void *taskqueue_giant_ih; static void *taskqueue_ih; static void taskqueue_fast_enqueue(void *); static void taskqueue_swi_enqueue(void *); static void taskqueue_swi_giant_enqueue(void *); struct taskqueue_busy { struct task *tb_running; TAILQ_ENTRY(taskqueue_busy) tb_link; }; +struct task * const TB_DRAIN_WAITER = (struct task *)0x1; + struct taskqueue { STAILQ_HEAD(, task) tq_queue; taskqueue_enqueue_fn tq_enqueue; void *tq_context; TAILQ_HEAD(, taskqueue_busy) tq_active; struct mtx tq_mutex; struct thread **tq_threads; int tq_tcount; int tq_spin; int tq_flags; int tq_callouts; taskqueue_callback_fn tq_callbacks[TASKQUEUE_NUM_CALLBACKS]; void *tq_cb_contexts[TASKQUEUE_NUM_CALLBACKS]; }; #define TQ_FLAGS_ACTIVE (1 << 0) #define TQ_FLAGS_BLOCKED (1 << 1) #define TQ_FLAGS_UNLOCKED_ENQUEUE (1 << 2) #define DT_CALLOUT_ARMED (1 << 0) #define TQ_LOCK(tq) \ do { \ if ((tq)->tq_spin) \ mtx_lock_spin(&(tq)->tq_mutex); \ else \ mtx_lock(&(tq)->tq_mutex); \ } while (0) #define TQ_ASSERT_LOCKED(tq) mtx_assert(&(tq)->tq_mutex, MA_OWNED) #define TQ_UNLOCK(tq) \ do { \ if ((tq)->tq_spin) \ mtx_unlock_spin(&(tq)->tq_mutex); \ else \ mtx_unlock(&(tq)->tq_mutex); \ } while (0) #define TQ_ASSERT_UNLOCKED(tq) mtx_assert(&(tq)->tq_mutex, MA_NOTOWNED) void _timeout_task_init(struct taskqueue *queue, struct timeout_task *timeout_task, int priority, task_fn_t func, void *context) { TASK_INIT(&timeout_task->t, priority, func, context); callout_init_mtx(&timeout_task->c, &queue->tq_mutex, CALLOUT_RETURNUNLOCKED); timeout_task->q = queue; timeout_task->f = 0; } static __inline int TQ_SLEEP(struct taskqueue *tq, void *p, struct mtx *m, int pri, const char *wm, int t) { if (tq->tq_spin) return (msleep_spin(p, m, wm, t)); return (msleep(p, m, pri, wm, t)); } static struct taskqueue * _taskqueue_create(const char *name __unused, int mflags, taskqueue_enqueue_fn enqueue, void *context, int mtxflags, const char *mtxname) { struct taskqueue *queue; queue = malloc(sizeof(struct taskqueue), M_TASKQUEUE, mflags | M_ZERO); if (!queue) return NULL; STAILQ_INIT(&queue->tq_queue); TAILQ_INIT(&queue->tq_active); queue->tq_enqueue = enqueue; queue->tq_context = context; queue->tq_spin = (mtxflags & MTX_SPIN) != 0; queue->tq_flags |= TQ_FLAGS_ACTIVE; if (enqueue == taskqueue_fast_enqueue || enqueue == taskqueue_swi_enqueue || enqueue == taskqueue_swi_giant_enqueue || enqueue == taskqueue_thread_enqueue) queue->tq_flags |= TQ_FLAGS_UNLOCKED_ENQUEUE; mtx_init(&queue->tq_mutex, mtxname, NULL, mtxflags); return queue; } struct taskqueue * taskqueue_create(const char *name, int mflags, taskqueue_enqueue_fn enqueue, void *context) { return _taskqueue_create(name, mflags, enqueue, context, MTX_DEF, "taskqueue"); } void taskqueue_set_callback(struct taskqueue *queue, enum taskqueue_callback_type cb_type, taskqueue_callback_fn callback, void *context) { KASSERT(((cb_type >= TASKQUEUE_CALLBACK_TYPE_MIN) && (cb_type <= TASKQUEUE_CALLBACK_TYPE_MAX)), ("Callback type %d not valid, must be %d-%d", cb_type, TASKQUEUE_CALLBACK_TYPE_MIN, TASKQUEUE_CALLBACK_TYPE_MAX)); KASSERT((queue->tq_callbacks[cb_type] == NULL), ("Re-initialization of taskqueue callback?")); queue->tq_callbacks[cb_type] = callback; queue->tq_cb_contexts[cb_type] = context; } /* * Signal a taskqueue thread to terminate. */ static void taskqueue_terminate(struct thread **pp, struct taskqueue *tq) { while (tq->tq_tcount > 0 || tq->tq_callouts > 0) { wakeup(tq); TQ_SLEEP(tq, pp, &tq->tq_mutex, PWAIT, "taskqueue_destroy", 0); } } void taskqueue_free(struct taskqueue *queue) { TQ_LOCK(queue); queue->tq_flags &= ~TQ_FLAGS_ACTIVE; taskqueue_terminate(queue->tq_threads, queue); KASSERT(TAILQ_EMPTY(&queue->tq_active), ("Tasks still running?")); KASSERT(queue->tq_callouts == 0, ("Armed timeout tasks")); mtx_destroy(&queue->tq_mutex); free(queue->tq_threads, M_TASKQUEUE); free(queue, M_TASKQUEUE); } static int taskqueue_enqueue_locked(struct taskqueue *queue, struct task *task) { struct task *ins; struct task *prev; /* * Count multiple enqueues. */ if (task->ta_pending) { if (task->ta_pending < USHRT_MAX) task->ta_pending++; TQ_UNLOCK(queue); return (0); } /* * Optimise the case when all tasks have the same priority. */ prev = STAILQ_LAST(&queue->tq_queue, task, ta_link); if (!prev || prev->ta_priority >= task->ta_priority) { STAILQ_INSERT_TAIL(&queue->tq_queue, task, ta_link); } else { prev = NULL; for (ins = STAILQ_FIRST(&queue->tq_queue); ins; prev = ins, ins = STAILQ_NEXT(ins, ta_link)) if (ins->ta_priority < task->ta_priority) break; if (prev) STAILQ_INSERT_AFTER(&queue->tq_queue, prev, task, ta_link); else STAILQ_INSERT_HEAD(&queue->tq_queue, task, ta_link); } task->ta_pending = 1; if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) != 0) TQ_UNLOCK(queue); if ((queue->tq_flags & TQ_FLAGS_BLOCKED) == 0) queue->tq_enqueue(queue->tq_context); if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) == 0) TQ_UNLOCK(queue); /* Return with lock released. */ return (0); } + int taskqueue_enqueue(struct taskqueue *queue, struct task *task) { int res; TQ_LOCK(queue); res = taskqueue_enqueue_locked(queue, task); /* The lock is released inside. */ return (res); } static void taskqueue_timeout_func(void *arg) { struct taskqueue *queue; struct timeout_task *timeout_task; timeout_task = arg; queue = timeout_task->q; KASSERT((timeout_task->f & DT_CALLOUT_ARMED) != 0, ("Stray timeout")); timeout_task->f &= ~DT_CALLOUT_ARMED; queue->tq_callouts--; taskqueue_enqueue_locked(timeout_task->q, &timeout_task->t); /* The lock is released inside. */ } int taskqueue_enqueue_timeout(struct taskqueue *queue, struct timeout_task *timeout_task, int ticks) { int res; TQ_LOCK(queue); KASSERT(timeout_task->q == NULL || timeout_task->q == queue, ("Migrated queue")); KASSERT(!queue->tq_spin, ("Timeout for spin-queue")); timeout_task->q = queue; res = timeout_task->t.ta_pending; if (ticks == 0) { taskqueue_enqueue_locked(queue, &timeout_task->t); /* The lock is released inside. */ } else { if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) { res++; } else { queue->tq_callouts++; timeout_task->f |= DT_CALLOUT_ARMED; if (ticks < 0) ticks = -ticks; /* Ignore overflow. */ } if (ticks > 0) { callout_reset(&timeout_task->c, ticks, taskqueue_timeout_func, timeout_task); } TQ_UNLOCK(queue); } return (res); } static void -taskqueue_drain_running(struct taskqueue *queue) +taskqueue_task_nop_fn(void *context, int pending) { +} - while (!TAILQ_EMPTY(&queue->tq_active)) - TQ_SLEEP(queue, &queue->tq_active, &queue->tq_mutex, - PWAIT, "-", 0); +/* + * Block until all currently queued tasks in this taskqueue + * have begun execution. Tasks queued during execution of + * this function are ignored. + */ +static void +taskqueue_drain_tq_queue(struct taskqueue *queue) +{ + struct task t_barrier; + + if (STAILQ_EMPTY(&queue->tq_queue)) + return; + + /* + * Enqueue our barrier with the lowest possible priority + * so we are inserted after all current tasks. + */ + TASK_INIT(&t_barrier, 0, taskqueue_task_nop_fn, &t_barrier); + taskqueue_enqueue_locked(queue, &t_barrier); + + /* + * Raise the barrier's priority so newly queued tasks cannot + * pass it. + */ + t_barrier.ta_priority = USHRT_MAX; + + /* + * Once the barrier has executed, all previously queued tasks + * have completed or are currently executing. + */ + while (t_barrier.ta_pending != 0) + TQ_SLEEP(queue, &t_barrier, &queue->tq_mutex, PWAIT, "-", 0); } +/* + * Block until all currently executing tasks for this taskqueue + * complete. Tasks that begin execution during the execution + * of this function are ignored. + */ +static void +taskqueue_drain_tq_active(struct taskqueue *queue) +{ + struct taskqueue_busy tb_marker, *tb_first; + + if (TAILQ_EMPTY(&queue->tq_active)) + return; + + /* Block taskq_terminate().*/ + queue->tq_callouts++; + + /* + * Wait for all currently executing taskqueue threads + * to go idle. + */ + tb_marker.tb_running = TB_DRAIN_WAITER; + TAILQ_INSERT_TAIL(&queue->tq_active, &tb_marker, tb_link); + while (TAILQ_FIRST(&queue->tq_active) != &tb_marker) + TQ_SLEEP(queue, &tb_marker, &queue->tq_mutex, PWAIT, "-", 0); + TAILQ_REMOVE(&queue->tq_active, &tb_marker, tb_link); + + /* + * Wakeup any other drain waiter that happened to queue up + * without any intervening active thread. + */ + tb_first = TAILQ_FIRST(&queue->tq_active); + if (tb_first != NULL && tb_first->tb_running == TB_DRAIN_WAITER) + wakeup(tb_first); + + /* Release taskqueue_terminate(). */ + queue->tq_callouts--; + if ((queue->tq_flags & TQ_FLAGS_ACTIVE) == 0) + wakeup_one(queue->tq_threads); +} + void taskqueue_block(struct taskqueue *queue) { TQ_LOCK(queue); queue->tq_flags |= TQ_FLAGS_BLOCKED; TQ_UNLOCK(queue); } void taskqueue_unblock(struct taskqueue *queue) { TQ_LOCK(queue); queue->tq_flags &= ~TQ_FLAGS_BLOCKED; if (!STAILQ_EMPTY(&queue->tq_queue)) queue->tq_enqueue(queue->tq_context); TQ_UNLOCK(queue); } static void taskqueue_run_locked(struct taskqueue *queue) { struct taskqueue_busy tb; + struct taskqueue_busy *tb_first; struct task *task; int pending; TQ_ASSERT_LOCKED(queue); tb.tb_running = NULL; - TAILQ_INSERT_TAIL(&queue->tq_active, &tb, tb_link); while (STAILQ_FIRST(&queue->tq_queue)) { + TAILQ_INSERT_TAIL(&queue->tq_active, &tb, tb_link); + /* * Carefully remove the first task from the queue and * zero its pending count. */ task = STAILQ_FIRST(&queue->tq_queue); STAILQ_REMOVE_HEAD(&queue->tq_queue, ta_link); pending = task->ta_pending; task->ta_pending = 0; tb.tb_running = task; TQ_UNLOCK(queue); task->ta_func(task->ta_context, pending); TQ_LOCK(queue); tb.tb_running = NULL; wakeup(task); + + TAILQ_REMOVE(&queue->tq_active, &tb, tb_link); + tb_first = TAILQ_FIRST(&queue->tq_active); + if (tb_first != NULL && + tb_first->tb_running == TB_DRAIN_WAITER) + wakeup(tb_first); } - TAILQ_REMOVE(&queue->tq_active, &tb, tb_link); - if (TAILQ_EMPTY(&queue->tq_active)) - wakeup(&queue->tq_active); } void taskqueue_run(struct taskqueue *queue) { TQ_LOCK(queue); taskqueue_run_locked(queue); TQ_UNLOCK(queue); } static int task_is_running(struct taskqueue *queue, struct task *task) { struct taskqueue_busy *tb; TQ_ASSERT_LOCKED(queue); TAILQ_FOREACH(tb, &queue->tq_active, tb_link) { if (tb->tb_running == task) return (1); } return (0); } static int taskqueue_cancel_locked(struct taskqueue *queue, struct task *task, u_int *pendp) { if (task->ta_pending > 0) STAILQ_REMOVE(&queue->tq_queue, task, task, ta_link); if (pendp != NULL) *pendp = task->ta_pending; task->ta_pending = 0; return (task_is_running(queue, task) ? EBUSY : 0); } int taskqueue_cancel(struct taskqueue *queue, struct task *task, u_int *pendp) { int error; TQ_LOCK(queue); error = taskqueue_cancel_locked(queue, task, pendp); TQ_UNLOCK(queue); return (error); } int taskqueue_cancel_timeout(struct taskqueue *queue, struct timeout_task *timeout_task, u_int *pendp) { u_int pending, pending1; int error; TQ_LOCK(queue); pending = !!callout_stop(&timeout_task->c); error = taskqueue_cancel_locked(queue, &timeout_task->t, &pending1); if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) { timeout_task->f &= ~DT_CALLOUT_ARMED; queue->tq_callouts--; } TQ_UNLOCK(queue); if (pendp != NULL) *pendp = pending + pending1; return (error); } void taskqueue_drain(struct taskqueue *queue, struct task *task) { if (!queue->tq_spin) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__); TQ_LOCK(queue); while (task->ta_pending != 0 || task_is_running(queue, task)) TQ_SLEEP(queue, task, &queue->tq_mutex, PWAIT, "-", 0); TQ_UNLOCK(queue); } void taskqueue_drain_all(struct taskqueue *queue) { - struct task *task; if (!queue->tq_spin) WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__); TQ_LOCK(queue); - task = STAILQ_LAST(&queue->tq_queue, task, ta_link); - if (task != NULL) - while (task->ta_pending != 0) - TQ_SLEEP(queue, task, &queue->tq_mutex, PWAIT, "-", 0); - taskqueue_drain_running(queue); - KASSERT(STAILQ_EMPTY(&queue->tq_queue), - ("taskqueue queue is not empty after draining")); + taskqueue_drain_tq_queue(queue); + taskqueue_drain_tq_active(queue); TQ_UNLOCK(queue); } void taskqueue_drain_timeout(struct taskqueue *queue, struct timeout_task *timeout_task) { callout_drain(&timeout_task->c); taskqueue_drain(queue, &timeout_task->t); } static void taskqueue_swi_enqueue(void *context) { swi_sched(taskqueue_ih, 0); } static void taskqueue_swi_run(void *dummy) { taskqueue_run(taskqueue_swi); } static void taskqueue_swi_giant_enqueue(void *context) { swi_sched(taskqueue_giant_ih, 0); } static void taskqueue_swi_giant_run(void *dummy) { taskqueue_run(taskqueue_swi_giant); } static int _taskqueue_start_threads(struct taskqueue **tqp, int count, int pri, cpuset_t *mask, const char *ktname) { struct thread *td; struct taskqueue *tq; int i, error; if (count <= 0) return (EINVAL); tq = *tqp; tq->tq_threads = malloc(sizeof(struct thread *) * count, M_TASKQUEUE, M_NOWAIT | M_ZERO); if (tq->tq_threads == NULL) { printf("%s: no memory for %s threads\n", __func__, ktname); return (ENOMEM); } for (i = 0; i < count; i++) { if (count == 1) error = kthread_add(taskqueue_thread_loop, tqp, NULL, &tq->tq_threads[i], RFSTOPPED, 0, "%s", ktname); else error = kthread_add(taskqueue_thread_loop, tqp, NULL, &tq->tq_threads[i], RFSTOPPED, 0, "%s_%d", ktname, i); if (error) { /* should be ok to continue, taskqueue_free will dtrt */ printf("%s: kthread_add(%s): error %d", __func__, ktname, error); tq->tq_threads[i] = NULL; /* paranoid */ } else tq->tq_tcount++; } for (i = 0; i < count; i++) { if (tq->tq_threads[i] == NULL) continue; td = tq->tq_threads[i]; if (mask) { error = cpuset_setthread(td->td_tid, mask); /* * Failing to pin is rarely an actual fatal error; * it'll just affect performance. */ if (error) printf("%s: curthread=%llu: can't pin; " "error=%d\n", __func__, (unsigned long long) td->td_tid, error); } thread_lock(td); sched_prio(td, pri); sched_add(td, SRQ_BORING); thread_unlock(td); } return (0); } int taskqueue_start_threads(struct taskqueue **tqp, int count, int pri, const char *name, ...) { char ktname[MAXCOMLEN + 1]; va_list ap; va_start(ap, name); vsnprintf(ktname, sizeof(ktname), name, ap); va_end(ap); return (_taskqueue_start_threads(tqp, count, pri, NULL, ktname)); } int taskqueue_start_threads_pinned(struct taskqueue **tqp, int count, int pri, int cpu_id, const char *name, ...) { char ktname[MAXCOMLEN + 1]; va_list ap; cpuset_t mask; va_start(ap, name); vsnprintf(ktname, sizeof(ktname), name, ap); va_end(ap); /* * In case someone passes in NOCPU, just fall back to the * default behaviour of "don't pin". */ if (cpu_id != NOCPU) { CPU_ZERO(&mask); CPU_SET(cpu_id, &mask); } return (_taskqueue_start_threads(tqp, count, pri, cpu_id == NOCPU ? NULL : &mask, ktname)); } static inline void taskqueue_run_callback(struct taskqueue *tq, enum taskqueue_callback_type cb_type) { taskqueue_callback_fn tq_callback; TQ_ASSERT_UNLOCKED(tq); tq_callback = tq->tq_callbacks[cb_type]; if (tq_callback != NULL) tq_callback(tq->tq_cb_contexts[cb_type]); } void taskqueue_thread_loop(void *arg) { struct taskqueue **tqp, *tq; tqp = arg; tq = *tqp; taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_INIT); TQ_LOCK(tq); while ((tq->tq_flags & TQ_FLAGS_ACTIVE) != 0) { taskqueue_run_locked(tq); /* * Because taskqueue_run() can drop tq_mutex, we need to * check if the TQ_FLAGS_ACTIVE flag wasn't removed in the * meantime, which means we missed a wakeup. */ if ((tq->tq_flags & TQ_FLAGS_ACTIVE) == 0) break; TQ_SLEEP(tq, tq, &tq->tq_mutex, 0, "-", 0); } taskqueue_run_locked(tq); /* * This thread is on its way out, so just drop the lock temporarily * in order to call the shutdown callback. This allows the callback * to look at the taskqueue, even just before it dies. */ TQ_UNLOCK(tq); taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN); TQ_LOCK(tq); /* rendezvous with thread that asked us to terminate */ tq->tq_tcount--; wakeup_one(tq->tq_threads); TQ_UNLOCK(tq); kthread_exit(); } void taskqueue_thread_enqueue(void *context) { struct taskqueue **tqp, *tq; tqp = context; tq = *tqp; wakeup_one(tq); } TASKQUEUE_DEFINE(swi, taskqueue_swi_enqueue, NULL, swi_add(NULL, "task queue", taskqueue_swi_run, NULL, SWI_TQ, INTR_MPSAFE, &taskqueue_ih)); TASKQUEUE_DEFINE(swi_giant, taskqueue_swi_giant_enqueue, NULL, swi_add(NULL, "Giant taskq", taskqueue_swi_giant_run, NULL, SWI_TQ_GIANT, 0, &taskqueue_giant_ih)); TASKQUEUE_DEFINE_THREAD(thread); struct taskqueue * taskqueue_create_fast(const char *name, int mflags, taskqueue_enqueue_fn enqueue, void *context) { return _taskqueue_create(name, mflags, enqueue, context, MTX_SPIN, "fast_taskqueue"); } /* NB: for backwards compatibility */ int taskqueue_enqueue_fast(struct taskqueue *queue, struct task *task) { return taskqueue_enqueue(queue, task); } static void *taskqueue_fast_ih; static void taskqueue_fast_enqueue(void *context) { swi_sched(taskqueue_fast_ih, 0); } static void taskqueue_fast_run(void *dummy) { taskqueue_run(taskqueue_fast); } TASKQUEUE_FAST_DEFINE(fast, taskqueue_fast_enqueue, NULL, swi_add(NULL, "fast taskq", taskqueue_fast_run, NULL, SWI_TQ_FAST, INTR_MPSAFE, &taskqueue_fast_ih)); int taskqueue_member(struct taskqueue *queue, struct thread *td) { int i, j, ret = 0; for (i = 0, j = 0; ; i++) { if (queue->tq_threads[i] == NULL) continue; if (queue->tq_threads[i] == td) { ret = 1; break; } if (++j >= queue->tq_tcount) break; } return (ret); }