mirror_zfs/module/splat/splat-taskq.c

1461 lines
40 KiB
C
Raw Normal View History

/*****************************************************************************\
* Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
* Copyright (C) 2007 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
* Written by Brian Behlendorf <behlendorf1@llnl.gov>.
* UCRL-CODE-235197
*
* This file is part of the SPL, Solaris Porting Layer.
* For details, see <http://zfsonlinux.org/>.
*
* The SPL is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the
* Free Software Foundation; either version 2 of the License, or (at your
* option) any later version.
*
* The SPL 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 General Public License
* for more details.
*
* You should have received a copy of the GNU General Public License along
* with the SPL. If not, see <http://www.gnu.org/licenses/>.
*****************************************************************************
* Solaris Porting LAyer Tests (SPLAT) Task Queue Tests.
\*****************************************************************************/
#include <sys/taskq.h>
#include <sys/random.h>
#include <sys/kmem.h>
#include "splat-internal.h"
#define SPLAT_TASKQ_NAME "taskq"
#define SPLAT_TASKQ_DESC "Kernel Task Queue Tests"
#define SPLAT_TASKQ_TEST1_ID 0x0201
#define SPLAT_TASKQ_TEST1_NAME "single"
#define SPLAT_TASKQ_TEST1_DESC "Single task queue, single task"
#define SPLAT_TASKQ_TEST2_ID 0x0202
#define SPLAT_TASKQ_TEST2_NAME "multiple"
#define SPLAT_TASKQ_TEST2_DESC "Multiple task queues, multiple tasks"
#define SPLAT_TASKQ_TEST3_ID 0x0203
2009-01-06 02:08:03 +03:00
#define SPLAT_TASKQ_TEST3_NAME "system"
#define SPLAT_TASKQ_TEST3_DESC "System task queue, multiple tasks"
#define SPLAT_TASKQ_TEST4_ID 0x0204
#define SPLAT_TASKQ_TEST4_NAME "wait"
#define SPLAT_TASKQ_TEST4_DESC "Multiple task waiting"
#define SPLAT_TASKQ_TEST5_ID 0x0205
#define SPLAT_TASKQ_TEST5_NAME "order"
#define SPLAT_TASKQ_TEST5_DESC "Correct task ordering"
#define SPLAT_TASKQ_TEST6_ID 0x0206
#define SPLAT_TASKQ_TEST6_NAME "front"
#define SPLAT_TASKQ_TEST6_DESC "Correct ordering with TQ_FRONT flag"
#define SPLAT_TASKQ_TEST7_ID 0x0207
#define SPLAT_TASKQ_TEST7_NAME "recurse"
#define SPLAT_TASKQ_TEST7_DESC "Single task queue, recursive dispatch"
#define SPLAT_TASKQ_TEST8_ID 0x0208
#define SPLAT_TASKQ_TEST8_NAME "contention"
#define SPLAT_TASKQ_TEST8_DESC "1 queue, 100 threads, 131072 tasks"
#define SPLAT_TASKQ_TEST9_ID 0x0209
#define SPLAT_TASKQ_TEST9_NAME "delay"
#define SPLAT_TASKQ_TEST9_DESC "Delayed task execution"
#define SPLAT_TASKQ_TEST10_ID 0x020a
#define SPLAT_TASKQ_TEST10_NAME "cancel"
#define SPLAT_TASKQ_TEST10_DESC "Cancel task execution"
#define SPLAT_TASKQ_ORDER_MAX 8
#define SPLAT_TASKQ_DEPTH_MAX 16
typedef struct splat_taskq_arg {
int flag;
int id;
atomic_t *count;
int order[SPLAT_TASKQ_ORDER_MAX];
unsigned int depth;
unsigned long expire;
taskq_t *tq;
taskq_ent_t *tqe;
spinlock_t lock;
struct file *file;
const char *name;
} splat_taskq_arg_t;
typedef struct splat_taskq_id {
int id;
splat_taskq_arg_t *arg;
} splat_taskq_id_t;
/*
* Create a taskq, queue a task, wait until task completes, ensure
* task ran properly, cleanup taskq.
*/
static void
2009-01-06 02:08:03 +03:00
splat_taskq_test13_func(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
ASSERT(tq_arg);
splat_vprint(tq_arg->file, SPLAT_TASKQ_TEST1_NAME,
"Taskq '%s' function '%s' setting flag\n",
2009-01-06 02:08:03 +03:00
tq_arg->name, sym2str(splat_taskq_test13_func));
tq_arg->flag = 1;
}
static int
splat_taskq_test1_impl(struct file *file, void *arg, boolean_t prealloc)
{
taskq_t *tq;
taskqid_t id;
splat_taskq_arg_t tq_arg;
taskq_ent_t tqe;
taskq_init_ent(&tqe);
splat_vprint(file, SPLAT_TASKQ_TEST1_NAME,
"Taskq '%s' creating (%s dispatch)\n",
SPLAT_TASKQ_TEST1_NAME,
prealloc ? "prealloc" : "dynamic");
if ((tq = taskq_create(SPLAT_TASKQ_TEST1_NAME, 1, maxclsyspri,
50, INT_MAX, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST1_NAME,
"Taskq '%s' create failed\n",
SPLAT_TASKQ_TEST1_NAME);
return -EINVAL;
}
tq_arg.flag = 0;
tq_arg.id = 0;
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST1_NAME;
splat_vprint(file, SPLAT_TASKQ_TEST1_NAME,
"Taskq '%s' function '%s' dispatching\n",
2009-01-06 02:08:03 +03:00
tq_arg.name, sym2str(splat_taskq_test13_func));
if (prealloc) {
taskq_dispatch_ent(tq, splat_taskq_test13_func,
&tq_arg, TQ_SLEEP, &tqe);
id = tqe.tqent_id;
} else {
id = taskq_dispatch(tq, splat_taskq_test13_func,
&tq_arg, TQ_SLEEP);
}
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST1_NAME,
"Taskq '%s' function '%s' dispatch failed\n",
tq_arg.name, sym2str(splat_taskq_test13_func));
taskq_destroy(tq);
return -EINVAL;
}
splat_vprint(file, SPLAT_TASKQ_TEST1_NAME, "Taskq '%s' waiting\n",
tq_arg.name);
taskq_wait(tq);
splat_vprint(file, SPLAT_TASKQ_TEST1_NAME, "Taskq '%s' destroying\n",
tq_arg.name);
taskq_destroy(tq);
return (tq_arg.flag) ? 0 : -EINVAL;
}
static int
splat_taskq_test1(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test1_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test1_impl(file, arg, B_TRUE);
return rc;
}
/*
* Create multiple taskq's, each with multiple tasks, wait until
* all tasks complete, ensure all tasks ran properly and in the
* correct order. Run order must be the same as the order submitted
* because we only have 1 thread per taskq. Finally cleanup the taskq.
*/
static void
splat_taskq_test2_func1(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
ASSERT(tq_arg);
splat_vprint(tq_arg->file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' function '%s' flag = %d = %d * 2\n",
tq_arg->name, tq_arg->id,
sym2str(splat_taskq_test2_func1),
tq_arg->flag * 2, tq_arg->flag);
tq_arg->flag *= 2;
}
static void
splat_taskq_test2_func2(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
ASSERT(tq_arg);
splat_vprint(tq_arg->file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' function '%s' flag = %d = %d + 1\n",
tq_arg->name, tq_arg->id,
sym2str(splat_taskq_test2_func2),
tq_arg->flag + 1, tq_arg->flag);
tq_arg->flag += 1;
}
#define TEST2_TASKQS 8
#define TEST2_THREADS_PER_TASKQ 1
static int
splat_taskq_test2_impl(struct file *file, void *arg, boolean_t prealloc) {
taskq_t *tq[TEST2_TASKQS] = { NULL };
taskqid_t id;
splat_taskq_arg_t tq_args[TEST2_TASKQS];
taskq_ent_t *func1_tqes = NULL;
taskq_ent_t *func2_tqes = NULL;
int i, rc = 0;
func1_tqes = kmalloc(sizeof(*func1_tqes) * TEST2_TASKQS, GFP_KERNEL);
if (func1_tqes == NULL) {
rc = -ENOMEM;
goto out;
}
func2_tqes = kmalloc(sizeof(*func2_tqes) * TEST2_TASKQS, GFP_KERNEL);
if (func2_tqes == NULL) {
rc = -ENOMEM;
goto out;
}
for (i = 0; i < TEST2_TASKQS; i++) {
taskq_init_ent(&func1_tqes[i]);
taskq_init_ent(&func2_tqes[i]);
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' creating (%s dispatch)\n",
SPLAT_TASKQ_TEST2_NAME, i,
prealloc ? "prealloc" : "dynamic");
if ((tq[i] = taskq_create(SPLAT_TASKQ_TEST2_NAME,
TEST2_THREADS_PER_TASKQ,
maxclsyspri, 50, INT_MAX,
TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' create failed\n",
SPLAT_TASKQ_TEST2_NAME, i);
rc = -EINVAL;
break;
}
tq_args[i].flag = i;
tq_args[i].id = i;
tq_args[i].file = file;
tq_args[i].name = SPLAT_TASKQ_TEST2_NAME;
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' function '%s' dispatching\n",
tq_args[i].name, tq_args[i].id,
sym2str(splat_taskq_test2_func1));
if (prealloc) {
taskq_dispatch_ent(tq[i], splat_taskq_test2_func1,
&tq_args[i], TQ_SLEEP, &func1_tqes[i]);
id = func1_tqes[i].tqent_id;
} else {
id = taskq_dispatch(tq[i], splat_taskq_test2_func1,
&tq_args[i], TQ_SLEEP);
}
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' function '%s' dispatch "
"failed\n", tq_args[i].name, tq_args[i].id,
sym2str(splat_taskq_test2_func1));
rc = -EINVAL;
break;
}
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' function '%s' dispatching\n",
tq_args[i].name, tq_args[i].id,
sym2str(splat_taskq_test2_func2));
if (prealloc) {
taskq_dispatch_ent(tq[i], splat_taskq_test2_func2,
&tq_args[i], TQ_SLEEP, &func2_tqes[i]);
id = func2_tqes[i].tqent_id;
} else {
id = taskq_dispatch(tq[i], splat_taskq_test2_func2,
&tq_args[i], TQ_SLEEP);
}
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME, "Taskq "
"'%s/%d' function '%s' dispatch failed\n",
tq_args[i].name, tq_args[i].id,
sym2str(splat_taskq_test2_func2));
rc = -EINVAL;
break;
}
}
/* When rc is set we're effectively just doing cleanup here, so
* ignore new errors in that case. They just cause noise. */
for (i = 0; i < TEST2_TASKQS; i++) {
if (tq[i] != NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' waiting\n",
tq_args[i].name, tq_args[i].id);
taskq_wait(tq[i]);
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d; destroying\n",
tq_args[i].name, tq_args[i].id);
taskq_destroy(tq[i]);
if (!rc && tq_args[i].flag != ((i * 2) + 1)) {
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' processed tasks "
"out of order; %d != %d\n",
tq_args[i].name, tq_args[i].id,
tq_args[i].flag, i * 2 + 1);
rc = -EINVAL;
} else {
splat_vprint(file, SPLAT_TASKQ_TEST2_NAME,
"Taskq '%s/%d' processed tasks "
"in the correct order; %d == %d\n",
tq_args[i].name, tq_args[i].id,
tq_args[i].flag, i * 2 + 1);
}
}
}
out:
if (func1_tqes)
kfree(func1_tqes);
if (func2_tqes)
kfree(func2_tqes);
return rc;
}
static int
splat_taskq_test2(struct file *file, void *arg) {
int rc;
rc = splat_taskq_test2_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test2_impl(file, arg, B_TRUE);
return rc;
}
/*
* Use the global system task queue with a single task, wait until task
* completes, ensure task ran properly.
2009-01-06 02:08:03 +03:00
*/
static int
splat_taskq_test3_impl(struct file *file, void *arg, boolean_t prealloc)
2009-01-06 02:08:03 +03:00
{
taskqid_t id;
splat_taskq_arg_t tq_arg;
taskq_ent_t tqe;
taskq_init_ent(&tqe);
2009-01-06 02:08:03 +03:00
tq_arg.flag = 0;
tq_arg.id = 0;
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST3_NAME;
splat_vprint(file, SPLAT_TASKQ_TEST3_NAME,
"Taskq '%s' function '%s' %s dispatch\n",
tq_arg.name, sym2str(splat_taskq_test13_func),
prealloc ? "prealloc" : "dynamic");
if (prealloc) {
taskq_dispatch_ent(system_taskq, splat_taskq_test13_func,
&tq_arg, TQ_SLEEP, &tqe);
id = tqe.tqent_id;
} else {
id = taskq_dispatch(system_taskq, splat_taskq_test13_func,
&tq_arg, TQ_SLEEP);
}
if (id == 0) {
2009-01-06 02:08:03 +03:00
splat_vprint(file, SPLAT_TASKQ_TEST3_NAME,
"Taskq '%s' function '%s' dispatch failed\n",
tq_arg.name, sym2str(splat_taskq_test13_func));
return -EINVAL;
}
splat_vprint(file, SPLAT_TASKQ_TEST3_NAME, "Taskq '%s' waiting\n",
tq_arg.name);
taskq_wait(system_taskq);
return (tq_arg.flag) ? 0 : -EINVAL;
}
static int
splat_taskq_test3(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test3_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test3_impl(file, arg, B_TRUE);
return rc;
}
/*
* Create a taskq and dispatch a large number of tasks to the queue.
* Then use taskq_wait() to block until all the tasks complete, then
* cross check that all the tasks ran by checking the shared atomic
* counter which is incremented in the task function.
*
* First we try with a large 'maxalloc' value, then we try with a small one.
* We should not drop tasks when TQ_SLEEP is used in taskq_dispatch(), even
* if the number of pending tasks is above maxalloc.
*/
static void
splat_taskq_test4_func(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
ASSERT(tq_arg);
atomic_inc(tq_arg->count);
}
static int
splat_taskq_test4_common(struct file *file, void *arg, int minalloc,
int maxalloc, int nr_tasks, boolean_t prealloc)
{
taskq_t *tq;
taskqid_t id;
splat_taskq_arg_t tq_arg;
taskq_ent_t *tqes;
atomic_t count;
int i, j, rc = 0;
tqes = kmalloc(sizeof(*tqes) * nr_tasks, GFP_KERNEL);
if (tqes == NULL)
return -ENOMEM;
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME,
"Taskq '%s' creating (%s dispatch) (%d/%d/%d)\n",
SPLAT_TASKQ_TEST4_NAME,
prealloc ? "prealloc" : "dynamic",
minalloc, maxalloc, nr_tasks);
if ((tq = taskq_create(SPLAT_TASKQ_TEST4_NAME, 1, maxclsyspri,
minalloc, maxalloc, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME,
"Taskq '%s' create failed\n",
SPLAT_TASKQ_TEST4_NAME);
rc = -EINVAL;
goto out_free;
}
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST4_NAME;
tq_arg.count = &count;
for (i = 1; i <= nr_tasks; i *= 2) {
atomic_set(tq_arg.count, 0);
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME,
"Taskq '%s' function '%s' dispatched %d times\n",
tq_arg.name, sym2str(splat_taskq_test4_func), i);
for (j = 0; j < i; j++) {
taskq_init_ent(&tqes[j]);
if (prealloc) {
taskq_dispatch_ent(tq, splat_taskq_test4_func,
&tq_arg, TQ_SLEEP, &tqes[j]);
id = tqes[j].tqent_id;
} else {
id = taskq_dispatch(tq, splat_taskq_test4_func,
&tq_arg, TQ_SLEEP);
}
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME,
"Taskq '%s' function '%s' dispatch "
"%d failed\n", tq_arg.name,
sym2str(splat_taskq_test4_func), j);
rc = -EINVAL;
goto out;
}
}
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME, "Taskq '%s' "
"waiting for %d dispatches\n", tq_arg.name, i);
taskq_wait(tq);
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME, "Taskq '%s' "
"%d/%d dispatches finished\n", tq_arg.name,
atomic_read(&count), i);
if (atomic_read(&count) != i) {
rc = -ERANGE;
goto out;
}
}
out:
splat_vprint(file, SPLAT_TASKQ_TEST4_NAME, "Taskq '%s' destroying\n",
tq_arg.name);
taskq_destroy(tq);
out_free:
kfree(tqes);
return rc;
}
static int
splat_taskq_test4_impl(struct file *file, void *arg, boolean_t prealloc)
{
int rc;
rc = splat_taskq_test4_common(file, arg, 50, INT_MAX, 1024, prealloc);
if (rc)
return rc;
rc = splat_taskq_test4_common(file, arg, 1, 1, 32, prealloc);
return rc;
}
static int
splat_taskq_test4(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test4_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test4_impl(file, arg, B_TRUE);
return rc;
}
/*
* Create a taskq and dispatch a specific sequence of tasks carefully
* crafted to validate the order in which tasks are processed. When
* there are multiple worker threads each thread will process the
* next pending task as soon as it completes its current task. This
* means that tasks do not strictly complete in order in which they
* were dispatched (increasing task id). This is fine but we need to
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
* verify that taskq_wait_all() blocks until the passed task id and all
* lower task ids complete. We do this by dispatching the following
* specific sequence of tasks each of which block for N time units.
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
* We then use taskq_wait_all() to unblock at specific task id and
* verify the only the expected task ids have completed and in the
* correct order. The two cases of interest are:
*
* 1) Task ids larger than the waited for task id can run and
* complete as long as there is an available worker thread.
* 2) All task ids lower than the waited one must complete before
* unblocking even if the waited task id itself has completed.
*
* The following table shows each task id and how they will be
* scheduled. Each rows represent one time unit and each column
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
* one of the three worker threads. The places taskq_wait_all()
* must unblock for a specific id are identified as well as the
* task ids which must have completed and their order.
*
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
* +-----+ <--- taskq_wait_all(tq, 8) unblocks
* | | Required Completion Order: 1,2,4,5,3,8,6,7
* +-----+ |
* | | |
* | | +-----+
* | | | 8 |
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
* | | +-----+ <--- taskq_wait_all(tq, 3) unblocks
* | | 7 | | Required Completion Order: 1,2,4,5,3
* | +-----+ |
* | 6 | | |
* +-----+ | |
* | | 5 | |
* | +-----+ |
* | 4 | | |
* +-----+ | |
* | 1 | 2 | 3 |
* +-----+-----+-----+
*
*/
static void
splat_taskq_test5_func(void *arg)
{
splat_taskq_id_t *tq_id = (splat_taskq_id_t *)arg;
splat_taskq_arg_t *tq_arg = tq_id->arg;
int factor;
/* Delays determined by above table */
switch (tq_id->id) {
default: factor = 0; break;
case 1: case 8: factor = 1; break;
case 2: case 4: case 5: factor = 2; break;
case 6: case 7: factor = 4; break;
case 3: factor = 5; break;
}
msleep(factor * 100);
splat_vprint(tq_arg->file, tq_arg->name,
"Taskqid %d complete for taskq '%s'\n",
tq_id->id, tq_arg->name);
spin_lock(&tq_arg->lock);
tq_arg->order[tq_arg->flag] = tq_id->id;
tq_arg->flag++;
spin_unlock(&tq_arg->lock);
}
static int
splat_taskq_test_order(splat_taskq_arg_t *tq_arg, int *order)
{
int i, j;
for (i = 0; i < SPLAT_TASKQ_ORDER_MAX; i++) {
if (tq_arg->order[i] != order[i]) {
splat_vprint(tq_arg->file, tq_arg->name,
"Taskq '%s' incorrect completion "
"order\n", tq_arg->name);
splat_vprint(tq_arg->file, tq_arg->name,
"%s", "Expected { ");
for (j = 0; j < SPLAT_TASKQ_ORDER_MAX; j++)
splat_print(tq_arg->file, "%d ", order[j]);
splat_print(tq_arg->file, "%s", "}\n");
splat_vprint(tq_arg->file, tq_arg->name,
"%s", "Got { ");
for (j = 0; j < SPLAT_TASKQ_ORDER_MAX; j++)
splat_print(tq_arg->file, "%d ",
tq_arg->order[j]);
splat_print(tq_arg->file, "%s", "}\n");
return -EILSEQ;
}
}
splat_vprint(tq_arg->file, tq_arg->name,
"Taskq '%s' validated correct completion order\n",
tq_arg->name);
return 0;
}
static int
splat_taskq_test5_impl(struct file *file, void *arg, boolean_t prealloc)
{
taskq_t *tq;
taskqid_t id;
splat_taskq_id_t tq_id[SPLAT_TASKQ_ORDER_MAX];
splat_taskq_arg_t tq_arg;
int order1[SPLAT_TASKQ_ORDER_MAX] = { 1,2,4,5,3,0,0,0 };
int order2[SPLAT_TASKQ_ORDER_MAX] = { 1,2,4,5,3,8,6,7 };
taskq_ent_t *tqes;
int i, rc = 0;
tqes = kmem_alloc(sizeof(*tqes) * SPLAT_TASKQ_ORDER_MAX, KM_SLEEP);
memset(tqes, 0, sizeof(*tqes) * SPLAT_TASKQ_ORDER_MAX);
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME,
"Taskq '%s' creating (%s dispatch)\n",
SPLAT_TASKQ_TEST5_NAME,
prealloc ? "prealloc" : "dynamic");
if ((tq = taskq_create(SPLAT_TASKQ_TEST5_NAME, 3, maxclsyspri,
50, INT_MAX, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME,
"Taskq '%s' create failed\n",
SPLAT_TASKQ_TEST5_NAME);
return -EINVAL;
}
tq_arg.flag = 0;
memset(&tq_arg.order, 0, sizeof(int) * SPLAT_TASKQ_ORDER_MAX);
spin_lock_init(&tq_arg.lock);
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST5_NAME;
for (i = 0; i < SPLAT_TASKQ_ORDER_MAX; i++) {
taskq_init_ent(&tqes[i]);
tq_id[i].id = i + 1;
tq_id[i].arg = &tq_arg;
if (prealloc) {
taskq_dispatch_ent(tq, splat_taskq_test5_func,
&tq_id[i], TQ_SLEEP, &tqes[i]);
id = tqes[i].tqent_id;
} else {
id = taskq_dispatch(tq, splat_taskq_test5_func,
&tq_id[i], TQ_SLEEP);
}
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME,
"Taskq '%s' function '%s' dispatch failed\n",
tq_arg.name, sym2str(splat_taskq_test5_func));
rc = -EINVAL;
goto out;
}
if (tq_id[i].id != id) {
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME,
"Taskq '%s' expected taskqid %d got %d\n",
tq_arg.name, (int)tq_id[i].id, (int)id);
rc = -EINVAL;
goto out;
}
}
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME, "Taskq '%s' "
"waiting for taskqid %d completion\n", tq_arg.name, 3);
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
taskq_wait_all(tq, 3);
if ((rc = splat_taskq_test_order(&tq_arg, order1)))
goto out;
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME, "Taskq '%s' "
"waiting for taskqid %d completion\n", tq_arg.name, 8);
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
taskq_wait_all(tq, 8);
rc = splat_taskq_test_order(&tq_arg, order2);
out:
splat_vprint(file, SPLAT_TASKQ_TEST5_NAME,
"Taskq '%s' destroying\n", tq_arg.name);
taskq_destroy(tq);
kmem_free(tqes, sizeof(*tqes) * SPLAT_TASKQ_ORDER_MAX);
return rc;
}
static int
splat_taskq_test5(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test5_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test5_impl(file, arg, B_TRUE);
return rc;
}
/*
* Create a single task queue with three threads. Dispatch 8 tasks,
* setting TQ_FRONT on only the last three. Sleep after
* dispatching tasks 1-3 to ensure they will run and hold the threads
* busy while we dispatch the remaining tasks. Verify that tasks 6-8
* run before task 4-5.
*
* The following table shows each task id and how they will be
* scheduled. Each rows represent one time unit and each column
* one of the three worker threads.
*
* NB: The Horizontal Line is the LAST Time unit consumed by the Task,
* and must be included in the factor calculation.
* T
* 17-> +-----+
* 16 | T6 |
* 15-> +-----+ |
* 14 | T6 | |
* 13-> | | 5 +-----+
* 12 | | | T6 |
* 11-> | +-----| |
* 10 | 4 | T6 | |
* 9-> +-----+ | 8 |
* 8 | T5 | | |
* 7-> | | 7 +-----+
* 6 | | | T7 |
* 5-> | +-----+ |
* 4 | 6 | T5 | |
* 3-> +-----+ | |
* 2 | T3 | | |
* 1 | 1 | 2 | 3 |
* 0 +-----+-----+-----+
*
*/
static void
splat_taskq_test6_func(void *arg)
{
/* Delays determined by above table */
static const int factor[SPLAT_TASKQ_ORDER_MAX+1] = {0,3,5,7,6,6,5,6,6};
splat_taskq_id_t *tq_id = (splat_taskq_id_t *)arg;
splat_taskq_arg_t *tq_arg = tq_id->arg;
splat_vprint(tq_arg->file, tq_arg->name,
"Taskqid %d starting for taskq '%s'\n",
tq_id->id, tq_arg->name);
if (tq_id->id < SPLAT_TASKQ_ORDER_MAX+1) {
msleep(factor[tq_id->id] * 50);
}
spin_lock(&tq_arg->lock);
tq_arg->order[tq_arg->flag] = tq_id->id;
tq_arg->flag++;
splat_vprint(tq_arg->file, tq_arg->name,
"Taskqid %d complete for taskq '%s'\n",
tq_id->id, tq_arg->name);
spin_unlock(&tq_arg->lock);
}
static int
splat_taskq_test6_impl(struct file *file, void *arg, boolean_t prealloc)
{
taskq_t *tq;
taskqid_t id;
splat_taskq_id_t tq_id[SPLAT_TASKQ_ORDER_MAX];
splat_taskq_arg_t tq_arg;
int order[SPLAT_TASKQ_ORDER_MAX] = { 1,2,3,6,7,8,4,5 };
taskq_ent_t *tqes;
int i, rc = 0;
uint_t tflags;
tqes = kmem_alloc(sizeof(*tqes) * SPLAT_TASKQ_ORDER_MAX, KM_SLEEP);
memset(tqes, 0, sizeof(*tqes) * SPLAT_TASKQ_ORDER_MAX);
splat_vprint(file, SPLAT_TASKQ_TEST6_NAME,
"Taskq '%s' creating (%s dispatch)\n",
SPLAT_TASKQ_TEST6_NAME,
prealloc ? "prealloc" : "dynamic");
if ((tq = taskq_create(SPLAT_TASKQ_TEST6_NAME, 3, maxclsyspri,
50, INT_MAX, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST6_NAME,
"Taskq '%s' create failed\n",
SPLAT_TASKQ_TEST6_NAME);
return -EINVAL;
}
tq_arg.flag = 0;
memset(&tq_arg.order, 0, sizeof(int) * SPLAT_TASKQ_ORDER_MAX);
spin_lock_init(&tq_arg.lock);
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST6_NAME;
for (i = 0; i < SPLAT_TASKQ_ORDER_MAX; i++) {
taskq_init_ent(&tqes[i]);
tq_id[i].id = i + 1;
tq_id[i].arg = &tq_arg;
tflags = TQ_SLEEP;
if (i > 4)
tflags |= TQ_FRONT;
if (prealloc) {
taskq_dispatch_ent(tq, splat_taskq_test6_func,
&tq_id[i], tflags, &tqes[i]);
id = tqes[i].tqent_id;
} else {
id = taskq_dispatch(tq, splat_taskq_test6_func,
&tq_id[i], tflags);
}
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST6_NAME,
"Taskq '%s' function '%s' dispatch failed\n",
tq_arg.name, sym2str(splat_taskq_test6_func));
rc = -EINVAL;
goto out;
}
if (tq_id[i].id != id) {
splat_vprint(file, SPLAT_TASKQ_TEST6_NAME,
"Taskq '%s' expected taskqid %d got %d\n",
tq_arg.name, (int)tq_id[i].id, (int)id);
rc = -EINVAL;
goto out;
}
/* Sleep to let tasks 1-3 start executing. */
if ( i == 2 )
msleep(100);
}
splat_vprint(file, SPLAT_TASKQ_TEST6_NAME, "Taskq '%s' "
"waiting for taskqid %d completion\n", tq_arg.name,
SPLAT_TASKQ_ORDER_MAX);
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
taskq_wait_all(tq, SPLAT_TASKQ_ORDER_MAX);
rc = splat_taskq_test_order(&tq_arg, order);
out:
splat_vprint(file, SPLAT_TASKQ_TEST6_NAME,
"Taskq '%s' destroying\n", tq_arg.name);
taskq_destroy(tq);
kmem_free(tqes, sizeof(*tqes) * SPLAT_TASKQ_ORDER_MAX);
return rc;
}
static int
splat_taskq_test6(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test6_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test6_impl(file, arg, B_TRUE);
return rc;
}
static void
splat_taskq_test7_func(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
taskqid_t id;
ASSERT(tq_arg);
if (tq_arg->depth >= SPLAT_TASKQ_DEPTH_MAX)
return;
tq_arg->depth++;
splat_vprint(tq_arg->file, SPLAT_TASKQ_TEST7_NAME,
"Taskq '%s' function '%s' dispatching (depth = %u)\n",
tq_arg->name, sym2str(splat_taskq_test7_func),
tq_arg->depth);
if (tq_arg->tqe) {
VERIFY(taskq_empty_ent(tq_arg->tqe));
taskq_dispatch_ent(tq_arg->tq, splat_taskq_test7_func,
tq_arg, TQ_SLEEP, tq_arg->tqe);
id = tq_arg->tqe->tqent_id;
} else {
id = taskq_dispatch(tq_arg->tq, splat_taskq_test7_func,
tq_arg, TQ_SLEEP);
}
if (id == 0) {
splat_vprint(tq_arg->file, SPLAT_TASKQ_TEST7_NAME,
"Taskq '%s' function '%s' dispatch failed "
"(depth = %u)\n", tq_arg->name,
sym2str(splat_taskq_test7_func), tq_arg->depth);
tq_arg->flag = -EINVAL;
return;
}
}
static int
splat_taskq_test7_impl(struct file *file, void *arg, boolean_t prealloc)
{
taskq_t *tq;
taskq_ent_t tqe;
splat_taskq_arg_t tq_arg;
splat_vprint(file, SPLAT_TASKQ_TEST7_NAME,
"Taskq '%s' creating (%s dispatch)\n",
SPLAT_TASKQ_TEST7_NAME,
prealloc ? "prealloc" : "dynamic");
if ((tq = taskq_create(SPLAT_TASKQ_TEST7_NAME, 1, maxclsyspri,
50, INT_MAX, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST7_NAME,
"Taskq '%s' create failed\n",
SPLAT_TASKQ_TEST7_NAME);
return -EINVAL;
}
tq_arg.depth = 0;
tq_arg.flag = 0;
tq_arg.id = 0;
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST7_NAME;
tq_arg.tq = tq;
if (prealloc) {
taskq_init_ent(&tqe);
tq_arg.tqe = &tqe;
} else {
tq_arg.tqe = NULL;
}
splat_taskq_test7_func(&tq_arg);
if (tq_arg.flag == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST7_NAME,
"Taskq '%s' waiting\n", tq_arg.name);
taskq delay/cancel functionality Add the ability to dispatch a delayed task to a taskq. The desired behavior is for the task to be queued but not executed by a worker thread until the expiration time is reached. To achieve this two new functions were added. * taskq_dispatch_delay() - This function behaves exactly like taskq_dispatch() however it takes a third 'expire_time' argument. The caller should pass the desired time the task should be executed as an absolute value in jiffies. The task is guarenteed not to run before this time, it may run slightly latter if all the worker threads are busy. * taskq_cancel_id() - Given a task id attempt to cancel the task before it gets executed. This is primarily useful for canceling delay tasks but can be used for canceling any previously dispatched task. There are three possible return values. 0 - The task was found and canceled before it was executed. ENOENT - The task was not found, either it was already run or an invalid task id was supplied by the caller. EBUSY - The task is currently executing any may not be canceled. This function will block until the task has been completed. * taskq_wait_all() - The taskq_wait_id() function was renamed taskq_wait_all() to more clearly reflect its actual behavior. It is only curreny used by the splat taskq regression tests. * taskq_wait_id() - Historically, the only difference between this function and taskq_wait() was that you passed the task id. In both functions you would block until ALL lower task ids which executed. This was semantically correct but could be very slow particularly if there were delay tasks submitted. To better accomidate the delay tasks this function was reimplemnted. It will now only block until the passed task id has been completed. This is actually a fairly low risk change for a few reasons. * Only new ZFS callers will make use of the new interfaces and very little common code was changed to support the new functions. * The existing taskq_wait() implementation was not changed just slightly refactored. * The newly optimized taskq_wait_id() implementation was never used by ZFS we can't accidentally introduce a new bug there. NOTE: This functionality does not exist in the Illumos taskqs. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-07 00:38:19 +04:00
taskq_wait_all(tq, SPLAT_TASKQ_DEPTH_MAX);
}
splat_vprint(file, SPLAT_TASKQ_TEST7_NAME,
"Taskq '%s' destroying\n", tq_arg.name);
taskq_destroy(tq);
return tq_arg.depth == SPLAT_TASKQ_DEPTH_MAX ? 0 : -EINVAL;
}
static int
splat_taskq_test7(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test7_impl(file, arg, B_FALSE);
if (rc)
return rc;
rc = splat_taskq_test7_impl(file, arg, B_TRUE);
return rc;
}
/*
* Create a taskq with 100 threads and dispatch a huge number of trivial
* tasks to generate contention on tq->tq_lock. This test should always
* pass. The purpose is to provide a benchmark for measuring the
* effectiveness of taskq optimizations.
*/
static void
splat_taskq_test8_func(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
ASSERT(tq_arg);
atomic_inc(tq_arg->count);
}
#define TEST8_NUM_TASKS 0x20000
#define TEST8_THREADS_PER_TASKQ 100
static int
splat_taskq_test8_common(struct file *file, void *arg, int minalloc,
int maxalloc)
{
taskq_t *tq;
taskqid_t id;
splat_taskq_arg_t tq_arg;
taskq_ent_t **tqes;
atomic_t count;
int i, j, rc = 0;
tqes = vmalloc(sizeof(*tqes) * TEST8_NUM_TASKS);
if (tqes == NULL)
return -ENOMEM;
memset(tqes, 0, sizeof(*tqes) * TEST8_NUM_TASKS);
splat_vprint(file, SPLAT_TASKQ_TEST8_NAME,
"Taskq '%s' creating (%d/%d/%d)\n",
SPLAT_TASKQ_TEST8_NAME,
minalloc, maxalloc, TEST8_NUM_TASKS);
if ((tq = taskq_create(SPLAT_TASKQ_TEST8_NAME, TEST8_THREADS_PER_TASKQ,
maxclsyspri, minalloc, maxalloc,
TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST8_NAME,
"Taskq '%s' create failed\n",
SPLAT_TASKQ_TEST8_NAME);
rc = -EINVAL;
goto out_free;
}
tq_arg.file = file;
tq_arg.name = SPLAT_TASKQ_TEST8_NAME;
tq_arg.count = &count;
atomic_set(tq_arg.count, 0);
for (i = 0; i < TEST8_NUM_TASKS; i++) {
tqes[i] = kmalloc(sizeof(taskq_ent_t), GFP_KERNEL);
if (tqes[i] == NULL) {
rc = -ENOMEM;
goto out;
}
taskq_init_ent(tqes[i]);
taskq_dispatch_ent(tq, splat_taskq_test8_func,
&tq_arg, TQ_SLEEP, tqes[i]);
id = tqes[i]->tqent_id;
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST8_NAME,
"Taskq '%s' function '%s' dispatch "
"%d failed\n", tq_arg.name,
sym2str(splat_taskq_test8_func), i);
rc = -EINVAL;
goto out;
}
}
splat_vprint(file, SPLAT_TASKQ_TEST8_NAME, "Taskq '%s' "
"waiting for %d dispatches\n", tq_arg.name,
TEST8_NUM_TASKS);
taskq_wait(tq);
splat_vprint(file, SPLAT_TASKQ_TEST8_NAME, "Taskq '%s' "
"%d/%d dispatches finished\n", tq_arg.name,
atomic_read(tq_arg.count), TEST8_NUM_TASKS);
if (atomic_read(tq_arg.count) != TEST8_NUM_TASKS)
rc = -ERANGE;
out:
splat_vprint(file, SPLAT_TASKQ_TEST8_NAME, "Taskq '%s' destroying\n",
tq_arg.name);
taskq_destroy(tq);
out_free:
for (j = 0; j < TEST8_NUM_TASKS && tqes[j] != NULL; j++)
kfree(tqes[j]);
vfree(tqes);
return rc;
}
static int
splat_taskq_test8(struct file *file, void *arg)
{
int rc;
rc = splat_taskq_test8_common(file, arg, 1, 100);
return rc;
}
/*
* Create a taskq and dispatch a number of delayed tasks to the queue.
* For each task verify that it was run no early than requested.
*/
static void
splat_taskq_test9_func(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
ASSERT(tq_arg);
if (ddi_get_lbolt() >= tq_arg->expire)
atomic_inc(tq_arg->count);
kmem_free(tq_arg, sizeof(splat_taskq_arg_t));
}
static int
splat_taskq_test9(struct file *file, void *arg)
{
taskq_t *tq;
atomic_t count;
int i, rc = 0;
int minalloc = 1;
int maxalloc = 10;
int nr_tasks = 100;
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME,
"Taskq '%s' creating (%s dispatch) (%d/%d/%d)\n",
SPLAT_TASKQ_TEST9_NAME, "delay", minalloc, maxalloc, nr_tasks);
if ((tq = taskq_create(SPLAT_TASKQ_TEST9_NAME, 3, maxclsyspri,
minalloc, maxalloc, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME,
"Taskq '%s' create failed\n", SPLAT_TASKQ_TEST9_NAME);
return -EINVAL;
}
atomic_set(&count, 0);
for (i = 1; i <= nr_tasks; i++) {
splat_taskq_arg_t *tq_arg;
taskqid_t id;
uint32_t rnd;
/* A random timeout in jiffies of at most 5 seconds */
get_random_bytes((void *)&rnd, 4);
rnd = rnd % (5 * HZ);
tq_arg = kmem_alloc(sizeof(splat_taskq_arg_t), KM_SLEEP);
tq_arg->file = file;
tq_arg->name = SPLAT_TASKQ_TEST9_NAME;
tq_arg->expire = ddi_get_lbolt() + rnd;
tq_arg->count = &count;
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME,
"Taskq '%s' delay dispatch %u jiffies\n",
SPLAT_TASKQ_TEST9_NAME, rnd);
id = taskq_dispatch_delay(tq, splat_taskq_test9_func,
tq_arg, TQ_SLEEP, ddi_get_lbolt() + rnd);
if (id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME,
"Taskq '%s' delay dispatch failed\n",
SPLAT_TASKQ_TEST9_NAME);
kmem_free(tq_arg, sizeof(splat_taskq_arg_t));
taskq_wait(tq);
rc = -EINVAL;
goto out;
}
}
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME, "Taskq '%s' waiting for "
"%d delay dispatches\n", SPLAT_TASKQ_TEST9_NAME, nr_tasks);
taskq_wait(tq);
if (atomic_read(&count) != nr_tasks)
rc = -ERANGE;
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME, "Taskq '%s' %d/%d delay "
"dispatches finished on time\n", SPLAT_TASKQ_TEST9_NAME,
atomic_read(&count), nr_tasks);
splat_vprint(file, SPLAT_TASKQ_TEST9_NAME, "Taskq '%s' destroying\n",
SPLAT_TASKQ_TEST9_NAME);
out:
taskq_destroy(tq);
return rc;
}
/*
* Create a taskq and dispatch then cancel tasks in the queue.
*/
static void
splat_taskq_test10_func(void *arg)
{
splat_taskq_arg_t *tq_arg = (splat_taskq_arg_t *)arg;
uint8_t rnd;
if (ddi_get_lbolt() >= tq_arg->expire)
atomic_inc(tq_arg->count);
/* Randomly sleep to further perturb the system */
get_random_bytes((void *)&rnd, 1);
msleep(1 + (rnd % 9));
}
static int
splat_taskq_test10(struct file *file, void *arg)
{
taskq_t *tq;
splat_taskq_arg_t **tqas;
atomic_t count;
int i, j, rc = 0;
int minalloc = 1;
int maxalloc = 10;
int nr_tasks = 100;
int canceled = 0;
int completed = 0;
int blocked = 0;
unsigned long start, cancel;
tqas = vmalloc(sizeof(*tqas) * nr_tasks);
if (tqas == NULL)
return -ENOMEM;
memset(tqas, 0, sizeof(*tqas) * nr_tasks);
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME,
"Taskq '%s' creating (%s dispatch) (%d/%d/%d)\n",
SPLAT_TASKQ_TEST10_NAME, "delay", minalloc, maxalloc, nr_tasks);
if ((tq = taskq_create(SPLAT_TASKQ_TEST10_NAME, 3, maxclsyspri,
minalloc, maxalloc, TASKQ_PREPOPULATE)) == NULL) {
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME,
"Taskq '%s' create failed\n", SPLAT_TASKQ_TEST10_NAME);
rc = -EINVAL;
goto out_free;
}
atomic_set(&count, 0);
for (i = 0; i < nr_tasks; i++) {
splat_taskq_arg_t *tq_arg;
uint32_t rnd;
/* A random timeout in jiffies of at most 5 seconds */
get_random_bytes((void *)&rnd, 4);
rnd = rnd % (5 * HZ);
tq_arg = kmem_alloc(sizeof(splat_taskq_arg_t), KM_SLEEP);
tq_arg->file = file;
tq_arg->name = SPLAT_TASKQ_TEST10_NAME;
tq_arg->count = &count;
tqas[i] = tq_arg;
/*
* Dispatch every 1/3 one immediately to mix it up, the cancel
* code is inherently racy and we want to try and provoke any
* subtle concurrently issues.
*/
if ((i % 3) == 0) {
tq_arg->expire = ddi_get_lbolt();
tq_arg->id = taskq_dispatch(tq, splat_taskq_test10_func,
tq_arg, TQ_SLEEP);
} else {
tq_arg->expire = ddi_get_lbolt() + rnd;
tq_arg->id = taskq_dispatch_delay(tq,
splat_taskq_test10_func,
tq_arg, TQ_SLEEP, ddi_get_lbolt() + rnd);
}
if (tq_arg->id == 0) {
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME,
"Taskq '%s' dispatch failed\n",
SPLAT_TASKQ_TEST10_NAME);
kmem_free(tq_arg, sizeof(splat_taskq_arg_t));
taskq_wait(tq);
rc = -EINVAL;
goto out;
} else {
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME,
"Taskq '%s' dispatch %lu in %lu jiffies\n",
SPLAT_TASKQ_TEST10_NAME, (unsigned long)tq_arg->id,
!(i % 3) ? 0 : tq_arg->expire - ddi_get_lbolt());
}
}
/*
* Start randomly canceling tasks for the duration of the test. We
* happen to know the valid task id's will be in the range 1..nr_tasks
* because the taskq is private and was just created. However, we
* have no idea of a particular task has already executed or not.
*/
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME, "Taskq '%s' randomly "
"canceling task ids\n", SPLAT_TASKQ_TEST10_NAME);
start = ddi_get_lbolt();
i = 0;
while (ddi_get_lbolt() < start + 5 * HZ) {
taskqid_t id;
uint32_t rnd;
i++;
cancel = ddi_get_lbolt();
get_random_bytes((void *)&rnd, 4);
id = 1 + (rnd % nr_tasks);
rc = taskq_cancel_id(tq, id);
/*
* Keep track of the results of the random cancels.
*/
if (rc == 0) {
canceled++;
} else if (rc == ENOENT) {
completed++;
} else if (rc == EBUSY) {
blocked++;
} else {
rc = -EINVAL;
break;
}
/*
* Verify we never get blocked to long in taskq_cancel_id().
* The worst case is 10ms if we happen to cancel the task
* which is currently executing. We allow a factor of 2x.
*/
if (ddi_get_lbolt() - cancel > HZ / 50) {
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME,
"Taskq '%s' cancel for %lu took %lu\n",
SPLAT_TASKQ_TEST10_NAME, (unsigned long)id,
ddi_get_lbolt() - cancel);
rc = -ETIMEDOUT;
break;
}
get_random_bytes((void *)&rnd, 4);
msleep(1 + (rnd % 100));
rc = 0;
}
taskq_wait(tq);
/*
* Cross check the results of taskq_cancel_id() with the number of
* times the dispatched function actually ran successfully.
*/
if ((rc == 0) && (nr_tasks - canceled != atomic_read(&count)))
rc = -EDOM;
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME, "Taskq '%s' %d attempts, "
"%d canceled, %d completed, %d blocked, %d/%d tasks run\n",
SPLAT_TASKQ_TEST10_NAME, i, canceled, completed, blocked,
atomic_read(&count), nr_tasks);
splat_vprint(file, SPLAT_TASKQ_TEST10_NAME, "Taskq '%s' destroying %d\n",
SPLAT_TASKQ_TEST10_NAME, rc);
out:
taskq_destroy(tq);
out_free:
for (j = 0; j < nr_tasks && tqas[j] != NULL; j++)
kmem_free(tqas[j], sizeof(splat_taskq_arg_t));
vfree(tqas);
return rc;
}
splat_subsystem_t *
splat_taskq_init(void)
{
splat_subsystem_t *sub;
sub = kmalloc(sizeof(*sub), GFP_KERNEL);
if (sub == NULL)
return NULL;
memset(sub, 0, sizeof(*sub));
strncpy(sub->desc.name, SPLAT_TASKQ_NAME, SPLAT_NAME_SIZE);
strncpy(sub->desc.desc, SPLAT_TASKQ_DESC, SPLAT_DESC_SIZE);
INIT_LIST_HEAD(&sub->subsystem_list);
INIT_LIST_HEAD(&sub->test_list);
spin_lock_init(&sub->test_lock);
sub->desc.id = SPLAT_SUBSYSTEM_TASKQ;
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST1_NAME, SPLAT_TASKQ_TEST1_DESC,
SPLAT_TASKQ_TEST1_ID, splat_taskq_test1);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST2_NAME, SPLAT_TASKQ_TEST2_DESC,
SPLAT_TASKQ_TEST2_ID, splat_taskq_test2);
2009-01-06 02:08:03 +03:00
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST3_NAME, SPLAT_TASKQ_TEST3_DESC,
SPLAT_TASKQ_TEST3_ID, splat_taskq_test3);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST4_NAME, SPLAT_TASKQ_TEST4_DESC,
SPLAT_TASKQ_TEST4_ID, splat_taskq_test4);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST5_NAME, SPLAT_TASKQ_TEST5_DESC,
SPLAT_TASKQ_TEST5_ID, splat_taskq_test5);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST6_NAME, SPLAT_TASKQ_TEST6_DESC,
SPLAT_TASKQ_TEST6_ID, splat_taskq_test6);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST7_NAME, SPLAT_TASKQ_TEST7_DESC,
SPLAT_TASKQ_TEST7_ID, splat_taskq_test7);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST8_NAME, SPLAT_TASKQ_TEST8_DESC,
SPLAT_TASKQ_TEST8_ID, splat_taskq_test8);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST9_NAME, SPLAT_TASKQ_TEST9_DESC,
SPLAT_TASKQ_TEST9_ID, splat_taskq_test9);
SPLAT_TEST_INIT(sub, SPLAT_TASKQ_TEST10_NAME, SPLAT_TASKQ_TEST10_DESC,
SPLAT_TASKQ_TEST10_ID, splat_taskq_test10);
return sub;
}
void
splat_taskq_fini(splat_subsystem_t *sub)
{
ASSERT(sub);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST10_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST9_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST8_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST7_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST6_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST5_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST4_ID);
2009-01-06 02:08:03 +03:00
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST3_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST2_ID);
SPLAT_TEST_FINI(sub, SPLAT_TASKQ_TEST1_ID);
kfree(sub);
}
int
splat_taskq_id(void) {
return SPLAT_SUBSYSTEM_TASKQ;
}