The patch titled
workqueue: rework threads/hotplug management
has been added to the -mm tree. Its filename is
workqueue-rework-threads-hotplug-management.patch
*** Remember to use Documentation/SubmitChecklist when testing your code ***
See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
out what to do about this
------------------------------------------------------
Subject: workqueue: rework threads/hotplug management
From: Oleg Nesterov <oleg@xxxxxxxxxx>
With this patch workqueue_mutex is used only to protect workqueues list,
all workqueue operations can run in parallel with cpuhotplug callback path.
take_over_work(), migrate_sequence, CPU_LOCK_ACQUIRE/RELEASE go away.
CPU_DEAD just sets cwq->should_stop and returns. After that cwq->thread
runs unbound until it flushes cwq->worklist, then exits. So this patch
slightly increases the probability that work_struct will be processed by
the "wrong" CPU, but we can't avoid this with CONFIG_HOTPLUG_CPU anyway.
CPU_UP_PREPARE creates the new cwq->thread if it's not NULL, CPU_ONLINE binds
it to the new cpu.
This all greatly simplifies the workqueues/cpu-hotplug interaction and imho
makes the code better. Shrinks both the source and compiled code (430
bytes). In particular, we can take workqueue_mutex in work->func() or
create/destroy workqueues from the cpuhotplug callback path.
The ugly part of this patch is that it adds "static int embryonic_cpu", it's
used by __create_workqueue() when it runs between CPU_UP_PREPARE/CPU_ONLINE.
cpu_populated_map was introduced to optimize for_each_possible_cpu(), it is
not strictly needed, and it is more a documentation in fact.
Further possible changes:
- don't use kthread_create(), we don't use kthread_stop()
- don't do kthread_bind() when create_workqueue_thread()
is called by CPU_UP_PREPARE, this is noop because of the
wake_up_process() below.
- make cpu_populated_map per workqueue_struct, this allows
us to remove some "is_single_threaded(wq)" checks, and we
can clear the bit when cwq->thread exits.
Signed-off-by: Oleg Nesterov <oleg@xxxxxxxxxx>
Cc: Srivatsa Vaddagiri <vatsa@xxxxxxxxxx>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@xxxxxxxxx>
Cc: Gautham shenoy <ego@xxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
kernel/workqueue.c | 441 +++++++++++++++++++++----------------------
1 files changed, 220 insertions(+), 221 deletions(-)
diff -puN kernel/workqueue.c~workqueue-rework-threads-hotplug-management kernel/workqueue.c
--- a/kernel/workqueue.c~workqueue-rework-threads-hotplug-management
+++ a/kernel/workqueue.c
@@ -43,10 +43,11 @@ struct cpu_workqueue_struct {
struct list_head worklist;
wait_queue_head_t more_work;
+ struct work_struct *current_work;
struct workqueue_struct *wq;
struct task_struct *thread;
- struct work_struct *current_work;
+ int should_stop;
int run_depth; /* Detect run_workqueue() recursion depth */
} ____cacheline_aligned;
@@ -64,11 +65,10 @@ struct workqueue_struct {
/* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
threads to each one as cpus come/go. */
-static long migrate_sequence __read_mostly;
static DEFINE_MUTEX(workqueue_mutex);
static LIST_HEAD(workqueues);
-static int singlethread_cpu;
+static int singlethread_cpu __read_mostly;
/* If it's single threaded, it isn't in the list of workqueues. */
static inline int is_single_threaded(struct workqueue_struct *wq)
@@ -344,10 +344,28 @@ static void run_workqueue(struct cpu_wor
spin_unlock_irqrestore(&cwq->lock, flags);
}
+/*
+ * NOTE: the caller must not touch *cwq if this func returns true
+ */
+static int cwq_should_stop(struct cpu_workqueue_struct *cwq)
+{
+ int should_stop = cwq->should_stop;
+
+ if (unlikely(should_stop)) {
+ spin_lock_irq(&cwq->lock);
+ should_stop = cwq->should_stop && list_empty(&cwq->worklist);
+ if (should_stop)
+ cwq->thread = NULL;
+ spin_unlock_irq(&cwq->lock);
+ }
+
+ return should_stop;
+}
+
static int worker_thread(void *__cwq)
{
struct cpu_workqueue_struct *cwq = __cwq;
- DECLARE_WAITQUEUE(wait, current);
+ DEFINE_WAIT(wait);
struct k_sigaction sa;
sigset_t blocked;
@@ -373,23 +391,21 @@ static int worker_thread(void *__cwq)
siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
- set_current_state(TASK_INTERRUPTIBLE);
- while (!kthread_should_stop()) {
+ for (;;) {
if (cwq->wq->freezeable)
try_to_freeze();
- add_wait_queue(&cwq->more_work, &wait);
- if (list_empty(&cwq->worklist))
+ prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
+ if (!cwq->should_stop && list_empty(&cwq->worklist))
schedule();
- else
- __set_current_state(TASK_RUNNING);
- remove_wait_queue(&cwq->more_work, &wait);
-
- if (!list_empty(&cwq->worklist))
- run_workqueue(cwq);
- set_current_state(TASK_INTERRUPTIBLE);
+ finish_wait(&cwq->more_work, &wait);
+
+ if (cwq_should_stop(cwq))
+ break;
+
+ run_workqueue(cwq);
}
- __set_current_state(TASK_RUNNING);
+
return 0;
}
@@ -415,6 +431,9 @@ static void insert_wq_barrier(struct cpu
insert_work(cwq, &barr->work, tail);
}
+/* optimization, we could use cpu_possible_map */
+static cpumask_t cpu_populated_map __read_mostly;
+
static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
{
if (cwq->thread == current) {
@@ -454,20 +473,13 @@ static void flush_cpu_workqueue(struct c
*/
void fastcall flush_workqueue(struct workqueue_struct *wq)
{
- if (is_single_threaded(wq)) {
- /* Always use first cpu's area. */
+ if (is_single_threaded(wq))
flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, singlethread_cpu));
- } else {
- long sequence;
+ else {
int cpu;
-again:
- sequence = migrate_sequence;
- for_each_possible_cpu(cpu)
+ for_each_cpu_mask(cpu, cpu_populated_map)
flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
-
- if (unlikely(sequence != migrate_sequence))
- goto again;
}
}
EXPORT_SYMBOL_GPL(flush_workqueue);
@@ -485,11 +497,8 @@ static void wait_on_work(struct cpu_work
}
spin_unlock_irq(&cwq->lock);
- if (unlikely(running)) {
- mutex_unlock(&workqueue_mutex);
+ if (unlikely(running))
wait_for_completion(&barr.done);
- mutex_lock(&workqueue_mutex);
- }
}
/**
@@ -510,155 +519,31 @@ void flush_work(struct workqueue_struct
{
struct cpu_workqueue_struct *cwq;
- mutex_lock(&workqueue_mutex);
cwq = get_wq_data(work);
/* Was it ever queued ? */
if (!cwq)
- goto out;
+ return;
/*
- * This work can't be re-queued, and the lock above protects us
- * from take_over_work(), no need to re-check that get_wq_data()
- * is still the same when we take cwq->lock.
+ * This work can't be re-queued, no need to re-check that
+ * get_wq_data() is still the same when we take cwq->lock.
*/
spin_lock_irq(&cwq->lock);
list_del_init(&work->entry);
work_release(work);
spin_unlock_irq(&cwq->lock);
- if (is_single_threaded(wq)) {
- /* Always use first cpu's area. */
+ if (is_single_threaded(wq))
wait_on_work(per_cpu_ptr(wq->cpu_wq, singlethread_cpu), work);
- } else {
+ else {
int cpu;
- for_each_online_cpu(cpu)
+ for_each_cpu_mask(cpu, cpu_populated_map)
wait_on_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
}
-out:
- mutex_unlock(&workqueue_mutex);
}
EXPORT_SYMBOL_GPL(flush_work);
-static void init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
-{
- struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
-
- cwq->wq = wq;
- spin_lock_init(&cwq->lock);
- INIT_LIST_HEAD(&cwq->worklist);
- init_waitqueue_head(&cwq->more_work);
-}
-
-static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
- int cpu)
-{
- struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
- struct task_struct *p;
-
- if (is_single_threaded(wq))
- p = kthread_create(worker_thread, cwq, "%s", wq->name);
- else
- p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
- if (IS_ERR(p))
- return NULL;
- cwq->thread = p;
- return p;
-}
-
-struct workqueue_struct *__create_workqueue(const char *name,
- int singlethread, int freezeable)
-{
- int cpu, destroy = 0;
- struct workqueue_struct *wq;
- struct task_struct *p;
-
- wq = kzalloc(sizeof(*wq), GFP_KERNEL);
- if (!wq)
- return NULL;
-
- wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
- if (!wq->cpu_wq) {
- kfree(wq);
- return NULL;
- }
-
- wq->name = name;
- wq->freezeable = freezeable;
-
- mutex_lock(&workqueue_mutex);
- if (singlethread) {
- INIT_LIST_HEAD(&wq->list);
- init_cpu_workqueue(wq, singlethread_cpu);
- p = create_workqueue_thread(wq, singlethread_cpu);
- if (!p)
- destroy = 1;
- else
- wake_up_process(p);
- } else {
- list_add(&wq->list, &workqueues);
- for_each_possible_cpu(cpu) {
- init_cpu_workqueue(wq, cpu);
- if (!cpu_online(cpu))
- continue;
-
- p = create_workqueue_thread(wq, cpu);
- if (p) {
- kthread_bind(p, cpu);
- wake_up_process(p);
- } else
- destroy = 1;
- }
- }
- mutex_unlock(&workqueue_mutex);
-
- /*
- * Was there any error during startup? If yes then clean up:
- */
- if (destroy) {
- destroy_workqueue(wq);
- wq = NULL;
- }
- return wq;
-}
-EXPORT_SYMBOL_GPL(__create_workqueue);
-
-static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
-{
- struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
-
- if (cwq->thread) {
- kthread_stop(cwq->thread);
- cwq->thread = NULL;
- }
-}
-
-/**
- * destroy_workqueue - safely terminate a workqueue
- * @wq: target workqueue
- *
- * Safely destroy a workqueue. All work currently pending will be done first.
- */
-void destroy_workqueue(struct workqueue_struct *wq)
-{
- int cpu;
-
- flush_workqueue(wq);
-
- /* We don't need the distraction of CPUs appearing and vanishing. */
- mutex_lock(&workqueue_mutex);
- if (is_single_threaded(wq))
- cleanup_workqueue_thread(wq, singlethread_cpu);
- else {
- for_each_online_cpu(cpu)
- cleanup_workqueue_thread(wq, cpu);
- list_del(&wq->list);
- }
- mutex_unlock(&workqueue_mutex);
- free_percpu(wq->cpu_wq);
- kfree(wq);
-}
-EXPORT_SYMBOL_GPL(destroy_workqueue);
static struct workqueue_struct *keventd_wq;
@@ -822,95 +707,209 @@ int current_is_keventd(void)
}
-/* Take the work from this (downed) CPU. */
-static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
+static struct cpu_workqueue_struct *
+init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
{
struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
- struct list_head list;
- struct work_struct *work;
+
+ cwq->wq = wq;
+ spin_lock_init(&cwq->lock);
+ INIT_LIST_HEAD(&cwq->worklist);
+ init_waitqueue_head(&cwq->more_work);
+
+ return cwq;
+}
+
+static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
+{
+ struct task_struct *p;
spin_lock_irq(&cwq->lock);
- list_replace_init(&cwq->worklist, &list);
- migrate_sequence++;
+ cwq->should_stop = 0;
+ p = cwq->thread;
+ spin_unlock_irq(&cwq->lock);
- while (!list_empty(&list)) {
- printk("Taking work for %s\n", wq->name);
- work = list_entry(list.next,struct work_struct,entry);
- list_del(&work->entry);
- __queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
+ if (!p) {
+ struct workqueue_struct *wq = cwq->wq;
+ const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d";
+
+ p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
+ /*
+ * Nobody can add the work_struct to this cwq,
+ * if (caller is __create_workqueue)
+ * nobody should see this wq
+ * else // caller is CPU_UP_PREPARE
+ * cpu is not on cpu_online_map
+ * so we can abort safely.
+ */
+ if (IS_ERR(p))
+ return PTR_ERR(p);
+
+ cwq->thread = p;
+ if (!is_single_threaded(wq))
+ kthread_bind(p, cpu);
+ /*
+ * Cancels affinity if the caller is CPU_UP_PREPARE.
+ * Needs a cleanup, but OK.
+ */
+ wake_up_process(p);
}
- spin_unlock_irq(&cwq->lock);
+
+ return 0;
}
-/* We're holding the cpucontrol mutex here */
-static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
- unsigned long action,
- void *hcpu)
+static int embryonic_cpu __read_mostly = -1;
+
+struct workqueue_struct *__create_workqueue(const char *name,
+ int singlethread, int freezeable)
{
- unsigned int hotcpu = (unsigned long)hcpu;
struct workqueue_struct *wq;
+ struct cpu_workqueue_struct *cwq;
+ int err = 0, cpu;
- switch (action) {
- case CPU_UP_PREPARE:
- mutex_lock(&workqueue_mutex);
- /* Create a new workqueue thread for it. */
- list_for_each_entry(wq, &workqueues, list) {
- if (!create_workqueue_thread(wq, hotcpu)) {
- printk("workqueue for %i failed\n", hotcpu);
- return NOTIFY_BAD;
- }
- }
- break;
+ wq = kzalloc(sizeof(*wq), GFP_KERNEL);
+ if (!wq)
+ return NULL;
- case CPU_ONLINE:
- /* Kick off worker threads. */
- list_for_each_entry(wq, &workqueues, list) {
- struct cpu_workqueue_struct *cwq;
-
- cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
- kthread_bind(cwq->thread, hotcpu);
- wake_up_process(cwq->thread);
- }
- mutex_unlock(&workqueue_mutex);
- break;
+ wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
+ if (!wq->cpu_wq) {
+ kfree(wq);
+ return NULL;
+ }
- case CPU_UP_CANCELED:
- list_for_each_entry(wq, &workqueues, list) {
- if (!per_cpu_ptr(wq->cpu_wq, hotcpu)->thread)
+ wq->name = name;
+ wq->freezeable = freezeable;
+
+ if (singlethread) {
+ INIT_LIST_HEAD(&wq->list);
+ cwq = init_cpu_workqueue(wq, singlethread_cpu);
+ err = create_workqueue_thread(cwq, singlethread_cpu);
+ } else {
+ mutex_lock(&workqueue_mutex);
+ list_add(&wq->list, &workqueues);
+
+ for_each_possible_cpu(cpu) {
+ cwq = init_cpu_workqueue(wq, cpu);
+ if (err || !(cpu_online(cpu) || cpu == embryonic_cpu))
continue;
- /* Unbind so it can run. */
- kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
- any_online_cpu(cpu_online_map));
- cleanup_workqueue_thread(wq, hotcpu);
+ err = create_workqueue_thread(cwq, cpu);
}
mutex_unlock(&workqueue_mutex);
- break;
+ }
- case CPU_DOWN_PREPARE:
- mutex_lock(&workqueue_mutex);
- break;
+ if (err) {
+ destroy_workqueue(wq);
+ wq = NULL;
+ }
+ return wq;
+}
+EXPORT_SYMBOL_GPL(__create_workqueue);
- case CPU_DOWN_FAILED:
- mutex_unlock(&workqueue_mutex);
- break;
+static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
+{
+ struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
+ struct wq_barrier barr;
+ int alive = 0;
+
+ spin_lock_irq(&cwq->lock);
+ if (cwq->thread != NULL) {
+ insert_wq_barrier(cwq, &barr, 1);
+ cwq->should_stop = 1;
+ alive = 1;
+ }
+ spin_unlock_irq(&cwq->lock);
+
+ if (alive) {
+ wait_for_completion(&barr.done);
+
+ while (unlikely(cwq->thread != NULL))
+ cpu_relax();
+ /*
+ * Wait until cwq->thread unlocks cwq->lock,
+ * it won't touch *cwq after that.
+ */
+ smp_rmb();
+ spin_unlock_wait(&cwq->lock);
+ }
+}
+
+/**
+ * destroy_workqueue - safely terminate a workqueue
+ * @wq: target workqueue
+ *
+ * Safely destroy a workqueue. All work currently pending will be done first.
+ */
+void destroy_workqueue(struct workqueue_struct *wq)
+{
+ if (is_single_threaded(wq))
+ cleanup_workqueue_thread(wq, singlethread_cpu);
+ else {
+ int cpu;
- case CPU_DEAD:
- list_for_each_entry(wq, &workqueues, list)
- cleanup_workqueue_thread(wq, hotcpu);
- list_for_each_entry(wq, &workqueues, list)
- take_over_work(wq, hotcpu);
+ mutex_lock(&workqueue_mutex);
+ list_del(&wq->list);
mutex_unlock(&workqueue_mutex);
- break;
+
+ for_each_cpu_mask(cpu, cpu_populated_map)
+ cleanup_workqueue_thread(wq, cpu);
+ }
+
+ free_percpu(wq->cpu_wq);
+ kfree(wq);
+}
+EXPORT_SYMBOL_GPL(destroy_workqueue);
+
+static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
+ unsigned long action,
+ void *hcpu)
+{
+ struct workqueue_struct *wq;
+ struct cpu_workqueue_struct *cwq;
+ unsigned int cpu = (unsigned long)hcpu;
+ int ret = NOTIFY_OK;
+
+ mutex_lock(&workqueue_mutex);
+ embryonic_cpu = -1;
+ if (action == CPU_UP_PREPARE) {
+ cpu_set(cpu, cpu_populated_map);
+ embryonic_cpu = cpu;
+ }
+
+ list_for_each_entry(wq, &workqueues, list) {
+ cwq = per_cpu_ptr(wq->cpu_wq, cpu);
+
+ switch (action) {
+ case CPU_UP_PREPARE:
+ if (create_workqueue_thread(cwq, cpu))
+ ret = NOTIFY_BAD;
+ break;
+
+ case CPU_ONLINE:
+ set_cpus_allowed(cwq->thread, cpumask_of_cpu(cpu));
+ break;
+
+ case CPU_UP_CANCELED:
+ case CPU_DEAD:
+ cwq->should_stop = 1;
+ wake_up(&cwq->more_work);
+ break;
+ }
+
+ if (ret != NOTIFY_OK) {
+ printk(KERN_ERR "workqueue for %i failed\n", cpu);
+ break;
+ }
}
+ mutex_unlock(&workqueue_mutex);
- return NOTIFY_OK;
+ return ret;
}
void init_workqueues(void)
{
+ cpu_populated_map = cpu_online_map;
singlethread_cpu = first_cpu(cpu_possible_map);
hotcpu_notifier(workqueue_cpu_callback, 0);
keventd_wq = create_workqueue("events");
BUG_ON(!keventd_wq);
}
-
_
Patches currently in -mm which might be from oleg@xxxxxxxxxx are
videobuf_qbuf-fix-possible-videobuf_queue-stream-corruption-and-lockup.patch
git-block.patch
doc-atomic_add_unless-doesnt-imply-mb-on-failure.patch
vt-refactor-console-sak-processing.patch
procfs-fix-race-between-proc_readdir-and-remove_proc_entry.patch
procfs-fix-race-between-proc_readdir-and-remove_proc_entry-fix.patch
kill_pid_info-kill-acquired_tasklist_lock.patch
clone-flag-clone_parent_tidptr-leaves-invalid-results-in-memory.patch
tty-make-__proc_set_tty-static.patch
tty-clarify-disassociate_ctty.patch
tty-fix-the-locking-for-signal-session-in-disassociate_ctty.patch
signal-use-kill_pgrp-not-kill_pg-in-the-sunos-compatibility-code.patch
signal-rewrite-kill_something_info-so-it-uses-newer-helpers.patch
pid-make-session_of_pgrp-use-struct-pid-instead-of-pid_t.patch
pid-use-struct-pid-for-talking-about-process-groups-in-exitc.patch
pid-replace-is_orphaned_pgrp-with-is_current_pgrp_orphaned.patch
tty-update-the-tty-layer-to-work-with-struct-pid.patch
pid-replace-do-while_each_task_pid-with-do-while_each_pid_task.patch
pid-remove-now-unused-do_each_task_pid-and-while_each_task_pid.patch
pid-remove-the-now-unused-kill_pg-kill_pg_info-and-__kill_pg_info.patch
reimplement-flush_workqueue.patch
implement-flush_work.patch
implement-flush_work-sanity.patch
implement-flush_work_keventd.patch
flush_workqueue-use-preempt_disable-to-hold-off-cpu-hotplug.patch
flush_cpu_workqueue-dont-flush-an-empty-worklist.patch
aio-use-flush_work.patch
kblockd-use-flush_work.patch
relayfs-use-flush_keventd_work.patch
tg3-use-flush_keventd_work.patch
e1000-use-flush_keventd_work.patch
libata-use-flush_work.patch
phy-use-flush_work.patch
call-cpu_chain-with-cpu_down_failed-if-cpu_down_prepare-failed.patch
slab-use-cpu_lock_.patch
workqueue-fix-freezeable-workqueues-implementation.patch
workqueue-fix-flush_workqueue-vs-cpu_dead-race.patch
workqueue-dont-clear-cwq-thread-until-it-exits.patch
workqueue-rework-threads-hotplug-management.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
