* [PATCH 01/17] workqueue: unexport work_cpu()
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 02/17] workqueue: use std_ prefix for the standard per-cpu pools Tejun Heo
` (20 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
This function no longer has any external users. Unexport it. It will
be removed later on.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
include/linux/workqueue.h | 1 -
kernel/workqueue.c | 4 ++--
2 files changed, 2 insertions(+), 3 deletions(-)
diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index 2b58905..ff68b1d 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -426,7 +426,6 @@ extern bool cancel_delayed_work_sync(struct delayed_work *dwork);
extern void workqueue_set_max_active(struct workqueue_struct *wq,
int max_active);
extern bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq);
-extern unsigned int work_cpu(struct work_struct *work);
extern unsigned int work_busy(struct work_struct *work);
/*
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 7967f34..4275ae7 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -476,6 +476,7 @@ static atomic_t unbound_pool_nr_running[NR_WORKER_POOLS] = {
};
static int worker_thread(void *__worker);
+static unsigned int work_cpu(struct work_struct *work);
static int worker_pool_pri(struct worker_pool *pool)
{
@@ -3428,13 +3429,12 @@ EXPORT_SYMBOL_GPL(workqueue_congested);
* RETURNS:
* CPU number if @work was ever queued. WORK_CPU_NONE otherwise.
*/
-unsigned int work_cpu(struct work_struct *work)
+static unsigned int work_cpu(struct work_struct *work)
{
struct global_cwq *gcwq = get_work_gcwq(work);
return gcwq ? gcwq->cpu : WORK_CPU_NONE;
}
-EXPORT_SYMBOL_GPL(work_cpu);
/**
* work_busy - test whether a work is currently pending or running
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 02/17] workqueue: use std_ prefix for the standard per-cpu pools
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
2013-01-17 1:42 ` [PATCH 01/17] workqueue: unexport work_cpu() Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 03/17] workqueue: make GCWQ_DISASSOCIATED a pool flag Tejun Heo
` (19 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
There are currently two worker pools per cpu (including the unbound
cpu) and they are the only pools in use. New class of pools are
scheduled to be added and some pool related APIs will be added
inbetween. Call the existing pools the standard pools and prefix them
with std_. Do this early so that new APIs can use std_ prefix from
the beginning.
This patch doesn't introduce any functional difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 22 +++++++++++-----------
1 file changed, 11 insertions(+), 11 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 4275ae7..5e74cd2 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -80,7 +80,7 @@ enum {
WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND |
WORKER_CPU_INTENSIVE,
- NR_WORKER_POOLS = 2, /* # worker pools per gcwq */
+ NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */
BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */
@@ -183,7 +183,7 @@ struct global_cwq {
DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER);
/* L: hash of busy workers */
- struct worker_pool pools[NR_WORKER_POOLS];
+ struct worker_pool pools[NR_STD_WORKER_POOLS];
/* normal and highpri pools */
} ____cacheline_aligned_in_smp;
@@ -282,7 +282,7 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
#define for_each_worker_pool(pool, gcwq) \
for ((pool) = &(gcwq)->pools[0]; \
- (pool) < &(gcwq)->pools[NR_WORKER_POOLS]; (pool)++)
+ (pool) < &(gcwq)->pools[NR_STD_WORKER_POOLS]; (pool)++)
#define for_each_busy_worker(worker, i, pos, gcwq) \
hash_for_each(gcwq->busy_hash, i, pos, worker, hentry)
@@ -463,7 +463,7 @@ static bool workqueue_freezing; /* W: have wqs started freezing? */
* try_to_wake_up(). Put it in a separate cacheline.
*/
static DEFINE_PER_CPU(struct global_cwq, global_cwq);
-static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_WORKER_POOLS]);
+static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
/*
* Global cpu workqueue and nr_running counter for unbound gcwq. The
@@ -471,14 +471,14 @@ static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_WORKER_POOLS])
* workers have WORKER_UNBOUND set.
*/
static struct global_cwq unbound_global_cwq;
-static atomic_t unbound_pool_nr_running[NR_WORKER_POOLS] = {
- [0 ... NR_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
+static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
+ [0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
};
static int worker_thread(void *__worker);
static unsigned int work_cpu(struct work_struct *work);
-static int worker_pool_pri(struct worker_pool *pool)
+static int std_worker_pool_pri(struct worker_pool *pool)
{
return pool - pool->gcwq->pools;
}
@@ -494,7 +494,7 @@ static struct global_cwq *get_gcwq(unsigned int cpu)
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
{
int cpu = pool->gcwq->cpu;
- int idx = worker_pool_pri(pool);
+ int idx = std_worker_pool_pri(pool);
if (cpu != WORK_CPU_UNBOUND)
return &per_cpu(pool_nr_running, cpu)[idx];
@@ -1707,7 +1707,7 @@ static void rebind_workers(struct global_cwq *gcwq)
* wq doesn't really matter but let's keep @worker->pool
* and @cwq->pool consistent for sanity.
*/
- if (worker_pool_pri(worker->pool))
+ if (std_worker_pool_pri(worker->pool))
wq = system_highpri_wq;
else
wq = system_wq;
@@ -1750,7 +1750,7 @@ static struct worker *alloc_worker(void)
static struct worker *create_worker(struct worker_pool *pool)
{
struct global_cwq *gcwq = pool->gcwq;
- const char *pri = worker_pool_pri(pool) ? "H" : "";
+ const char *pri = std_worker_pool_pri(pool) ? "H" : "";
struct worker *worker = NULL;
int id = -1;
@@ -1780,7 +1780,7 @@ static struct worker *create_worker(struct worker_pool *pool)
if (IS_ERR(worker->task))
goto fail;
- if (worker_pool_pri(pool))
+ if (std_worker_pool_pri(pool))
set_user_nice(worker->task, HIGHPRI_NICE_LEVEL);
/*
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 03/17] workqueue: make GCWQ_DISASSOCIATED a pool flag
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
2013-01-17 1:42 ` [PATCH 01/17] workqueue: unexport work_cpu() Tejun Heo
2013-01-17 1:42 ` [PATCH 02/17] workqueue: use std_ prefix for the standard per-cpu pools Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 04/17] workqueue: make GCWQ_FREEZING " Tejun Heo
` (18 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Make GCWQ_DISASSOCIATED a pool flag POOL_DISASSOCIATED. This patch
doesn't change locking - DISASSOCIATED on both pools of a CPU are set
or clear together while holding gcwq->lock. It shouldn't cause any
functional difference.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 66 +++++++++++++++++++++++++++++-------------------------
1 file changed, 35 insertions(+), 31 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 5e74cd2..8feef7e 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -48,26 +48,28 @@
enum {
/*
* global_cwq flags
+ */
+ GCWQ_FREEZING = 1 << 1, /* freeze in progress */
+
+ /*
+ * worker_pool flags
*
- * A bound gcwq is either associated or disassociated with its CPU.
+ * A bound pool is either associated or disassociated with its CPU.
* While associated (!DISASSOCIATED), all workers are bound to the
* CPU and none has %WORKER_UNBOUND set and concurrency management
* is in effect.
*
* While DISASSOCIATED, the cpu may be offline and all workers have
* %WORKER_UNBOUND set and concurrency management disabled, and may
- * be executing on any CPU. The gcwq behaves as an unbound one.
+ * be executing on any CPU. The pool behaves as an unbound one.
*
* Note that DISASSOCIATED can be flipped only while holding
- * assoc_mutex of all pools on the gcwq to avoid changing binding
- * state while create_worker() is in progress.
+ * assoc_mutex to avoid changing binding state while
+ * create_worker() is in progress.
*/
- GCWQ_DISASSOCIATED = 1 << 0, /* cpu can't serve workers */
- GCWQ_FREEZING = 1 << 1, /* freeze in progress */
-
- /* pool flags */
POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */
POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */
+ POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */
/* worker flags */
WORKER_STARTED = 1 << 0, /* started */
@@ -115,7 +117,7 @@ enum {
* X: During normal operation, modification requires gcwq->lock and
* should be done only from local cpu. Either disabling preemption
* on local cpu or grabbing gcwq->lock is enough for read access.
- * If GCWQ_DISASSOCIATED is set, it's identical to L.
+ * If POOL_DISASSOCIATED is set, it's identical to L.
*
* F: wq->flush_mutex protected.
*
@@ -165,7 +167,7 @@ struct worker_pool {
struct timer_list idle_timer; /* L: worker idle timeout */
struct timer_list mayday_timer; /* L: SOS timer for workers */
- struct mutex assoc_mutex; /* protect GCWQ_DISASSOCIATED */
+ struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */
struct ida worker_ida; /* L: for worker IDs */
};
@@ -466,9 +468,9 @@ static DEFINE_PER_CPU(struct global_cwq, global_cwq);
static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
/*
- * Global cpu workqueue and nr_running counter for unbound gcwq. The
- * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its
- * workers have WORKER_UNBOUND set.
+ * Global cpu workqueue and nr_running counter for unbound gcwq. The pools
+ * for online CPUs have POOL_DISASSOCIATED set, and all their workers have
+ * WORKER_UNBOUND set.
*/
static struct global_cwq unbound_global_cwq;
static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
@@ -1493,7 +1495,6 @@ EXPORT_SYMBOL_GPL(mod_delayed_work);
static void worker_enter_idle(struct worker *worker)
{
struct worker_pool *pool = worker->pool;
- struct global_cwq *gcwq = pool->gcwq;
BUG_ON(worker->flags & WORKER_IDLE);
BUG_ON(!list_empty(&worker->entry) &&
@@ -1516,7 +1517,7 @@ static void worker_enter_idle(struct worker *worker)
* nr_running, the warning may trigger spuriously. Check iff
* unbind is not in progress.
*/
- WARN_ON_ONCE(!(gcwq->flags & GCWQ_DISASSOCIATED) &&
+ WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) &&
pool->nr_workers == pool->nr_idle &&
atomic_read(get_pool_nr_running(pool)));
}
@@ -1557,7 +1558,7 @@ static void worker_leave_idle(struct worker *worker)
* [dis]associated in the meantime.
*
* This function tries set_cpus_allowed() and locks gcwq and verifies the
- * binding against %GCWQ_DISASSOCIATED which is set during
+ * binding against %POOL_DISASSOCIATED which is set during
* %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker
* enters idle state or fetches works without dropping lock, it can
* guarantee the scheduling requirement described in the first paragraph.
@@ -1573,7 +1574,8 @@ static void worker_leave_idle(struct worker *worker)
static bool worker_maybe_bind_and_lock(struct worker *worker)
__acquires(&gcwq->lock)
{
- struct global_cwq *gcwq = worker->pool->gcwq;
+ struct worker_pool *pool = worker->pool;
+ struct global_cwq *gcwq = pool->gcwq;
struct task_struct *task = worker->task;
while (true) {
@@ -1581,13 +1583,13 @@ __acquires(&gcwq->lock)
* The following call may fail, succeed or succeed
* without actually migrating the task to the cpu if
* it races with cpu hotunplug operation. Verify
- * against GCWQ_DISASSOCIATED.
+ * against POOL_DISASSOCIATED.
*/
- if (!(gcwq->flags & GCWQ_DISASSOCIATED))
+ if (!(pool->flags & POOL_DISASSOCIATED))
set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu));
spin_lock_irq(&gcwq->lock);
- if (gcwq->flags & GCWQ_DISASSOCIATED)
+ if (pool->flags & POOL_DISASSOCIATED)
return false;
if (task_cpu(task) == gcwq->cpu &&
cpumask_equal(¤t->cpus_allowed,
@@ -1785,14 +1787,14 @@ static struct worker *create_worker(struct worker_pool *pool)
/*
* Determine CPU binding of the new worker depending on
- * %GCWQ_DISASSOCIATED. The caller is responsible for ensuring the
+ * %POOL_DISASSOCIATED. The caller is responsible for ensuring the
* flag remains stable across this function. See the comments
* above the flag definition for details.
*
* As an unbound worker may later become a regular one if CPU comes
* online, make sure every worker has %PF_THREAD_BOUND set.
*/
- if (!(gcwq->flags & GCWQ_DISASSOCIATED)) {
+ if (!(pool->flags & POOL_DISASSOCIATED)) {
kthread_bind(worker->task, gcwq->cpu);
} else {
worker->task->flags |= PF_THREAD_BOUND;
@@ -2153,10 +2155,10 @@ __acquires(&gcwq->lock)
/*
* Ensure we're on the correct CPU. DISASSOCIATED test is
* necessary to avoid spurious warnings from rescuers servicing the
- * unbound or a disassociated gcwq.
+ * unbound or a disassociated pool.
*/
WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) &&
- !(gcwq->flags & GCWQ_DISASSOCIATED) &&
+ !(pool->flags & POOL_DISASSOCIATED) &&
raw_smp_processor_id() != gcwq->cpu);
/*
@@ -3481,7 +3483,7 @@ EXPORT_SYMBOL_GPL(work_busy);
* gcwqs serve mix of short, long and very long running works making
* blocked draining impractical.
*
- * This is solved by allowing a gcwq to be disassociated from the CPU
+ * This is solved by allowing the pools to be disassociated from the CPU
* running as an unbound one and allowing it to be reattached later if the
* cpu comes back online.
*/
@@ -3531,7 +3533,8 @@ static void gcwq_unbind_fn(struct work_struct *work)
for_each_busy_worker(worker, i, pos, gcwq)
worker->flags |= WORKER_UNBOUND;
- gcwq->flags |= GCWQ_DISASSOCIATED;
+ for_each_worker_pool(pool, gcwq)
+ pool->flags |= POOL_DISASSOCIATED;
gcwq_release_assoc_and_unlock(gcwq);
@@ -3590,7 +3593,8 @@ static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb,
case CPU_DOWN_FAILED:
case CPU_ONLINE:
gcwq_claim_assoc_and_lock(gcwq);
- gcwq->flags &= ~GCWQ_DISASSOCIATED;
+ for_each_worker_pool(pool, gcwq)
+ pool->flags &= ~POOL_DISASSOCIATED;
rebind_workers(gcwq);
gcwq_release_assoc_and_unlock(gcwq);
break;
@@ -3815,12 +3819,12 @@ static int __init init_workqueues(void)
spin_lock_init(&gcwq->lock);
gcwq->cpu = cpu;
- gcwq->flags |= GCWQ_DISASSOCIATED;
hash_init(gcwq->busy_hash);
for_each_worker_pool(pool, gcwq) {
pool->gcwq = gcwq;
+ pool->flags |= POOL_DISASSOCIATED;
INIT_LIST_HEAD(&pool->worklist);
INIT_LIST_HEAD(&pool->idle_list);
@@ -3841,12 +3845,12 @@ static int __init init_workqueues(void)
struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
- if (cpu != WORK_CPU_UNBOUND)
- gcwq->flags &= ~GCWQ_DISASSOCIATED;
-
for_each_worker_pool(pool, gcwq) {
struct worker *worker;
+ if (cpu != WORK_CPU_UNBOUND)
+ pool->flags &= ~POOL_DISASSOCIATED;
+
worker = create_worker(pool);
BUG_ON(!worker);
spin_lock_irq(&gcwq->lock);
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 04/17] workqueue: make GCWQ_FREEZING a pool flag
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (2 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 03/17] workqueue: make GCWQ_DISASSOCIATED a pool flag Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 05/17] workqueue: introduce WORK_OFFQ_CPU_NONE Tejun Heo
` (17 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Make GCWQ_FREEZING a pool flag POOL_FREEZING. This patch doesn't
change locking - FREEZING on both pools of a CPU are set or clear
together while holding gcwq->lock. It shouldn't cause any functional
difference.
This leaves gcwq->flags w/o any flags. Removed.
While at it, convert BUG_ON()s in freeze_workqueue_begin() and
thaw_workqueues() to WARN_ON_ONCE().
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 28 +++++++++++++++-------------
1 file changed, 15 insertions(+), 13 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 8feef7e..6e8bad6 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -47,11 +47,6 @@
enum {
/*
- * global_cwq flags
- */
- GCWQ_FREEZING = 1 << 1, /* freeze in progress */
-
- /*
* worker_pool flags
*
* A bound pool is either associated or disassociated with its CPU.
@@ -70,6 +65,7 @@ enum {
POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */
POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */
POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */
+ POOL_FREEZING = 1 << 3, /* freeze in progress */
/* worker flags */
WORKER_STARTED = 1 << 0, /* started */
@@ -179,7 +175,6 @@ struct worker_pool {
struct global_cwq {
spinlock_t lock; /* the gcwq lock */
unsigned int cpu; /* I: the associated cpu */
- unsigned int flags; /* L: GCWQ_* flags */
/* workers are chained either in busy_hash or pool idle_list */
DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER);
@@ -3389,13 +3384,15 @@ void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
wq->saved_max_active = max_active;
for_each_cwq_cpu(cpu, wq) {
- struct global_cwq *gcwq = get_gcwq(cpu);
+ struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+ struct worker_pool *pool = cwq->pool;
+ struct global_cwq *gcwq = pool->gcwq;
spin_lock_irq(&gcwq->lock);
if (!(wq->flags & WQ_FREEZABLE) ||
- !(gcwq->flags & GCWQ_FREEZING))
- cwq_set_max_active(get_cwq(gcwq->cpu, wq), max_active);
+ !(pool->flags & POOL_FREEZING))
+ cwq_set_max_active(cwq, max_active);
spin_unlock_irq(&gcwq->lock);
}
@@ -3685,12 +3682,15 @@ void freeze_workqueues_begin(void)
for_each_gcwq_cpu(cpu) {
struct global_cwq *gcwq = get_gcwq(cpu);
+ struct worker_pool *pool;
struct workqueue_struct *wq;
spin_lock_irq(&gcwq->lock);
- BUG_ON(gcwq->flags & GCWQ_FREEZING);
- gcwq->flags |= GCWQ_FREEZING;
+ for_each_worker_pool(pool, gcwq) {
+ WARN_ON_ONCE(pool->flags & POOL_FREEZING);
+ pool->flags |= POOL_FREEZING;
+ }
list_for_each_entry(wq, &workqueues, list) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
@@ -3776,8 +3776,10 @@ void thaw_workqueues(void)
spin_lock_irq(&gcwq->lock);
- BUG_ON(!(gcwq->flags & GCWQ_FREEZING));
- gcwq->flags &= ~GCWQ_FREEZING;
+ for_each_worker_pool(pool, gcwq) {
+ WARN_ON_ONCE(!(pool->flags & POOL_FREEZING));
+ pool->flags &= ~POOL_FREEZING;
+ }
list_for_each_entry(wq, &workqueues, list) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 05/17] workqueue: introduce WORK_OFFQ_CPU_NONE
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (3 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 04/17] workqueue: make GCWQ_FREEZING " Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 06/17] workqueue: add worker_pool->id Tejun Heo
` (16 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Currently, when a work item is off queue, high bits of its data
encodes the last CPU it was on. This is scheduled to be changed to
pool ID, which will make it impossible to use WORK_CPU_NONE to
indicate no association.
This patch limits the number of bits which are used for off-queue cpu
number to 31 (so that the max fits in an int) and uses the highest
possible value - WORK_OFFQ_CPU_NONE - to indicate no association.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
include/linux/workqueue.h | 10 +++++++++-
kernel/workqueue.c | 4 ++--
2 files changed, 11 insertions(+), 3 deletions(-)
diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index ff68b1d..f8b3576 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -73,13 +73,21 @@ enum {
WORK_OFFQ_CANCELING = (1 << WORK_OFFQ_FLAG_BASE),
+ /*
+ * When a work item is off queue, its high bits point to the last
+ * cpu it was on. Cap at 31 bits and use the highest number to
+ * indicate that no cpu is associated.
+ */
WORK_OFFQ_FLAG_BITS = 1,
WORK_OFFQ_CPU_SHIFT = WORK_OFFQ_FLAG_BASE + WORK_OFFQ_FLAG_BITS,
+ WORK_OFFQ_LEFT = BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT,
+ WORK_OFFQ_CPU_BITS = WORK_OFFQ_LEFT <= 31 ? WORK_OFFQ_LEFT : 31,
+ WORK_OFFQ_CPU_NONE = (1LU << WORK_OFFQ_CPU_BITS) - 1,
/* convenience constants */
WORK_STRUCT_FLAG_MASK = (1UL << WORK_STRUCT_FLAG_BITS) - 1,
WORK_STRUCT_WQ_DATA_MASK = ~WORK_STRUCT_FLAG_MASK,
- WORK_STRUCT_NO_CPU = (unsigned long)WORK_CPU_NONE << WORK_OFFQ_CPU_SHIFT,
+ WORK_STRUCT_NO_CPU = (unsigned long)WORK_OFFQ_CPU_NONE << WORK_OFFQ_CPU_SHIFT,
/* bit mask for work_busy() return values */
WORK_BUSY_PENDING = 1 << 0,
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 6e8bad6..9da3db2 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -600,7 +600,7 @@ static struct global_cwq *get_work_gcwq(struct work_struct *work)
(data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq;
cpu = data >> WORK_OFFQ_CPU_SHIFT;
- if (cpu == WORK_CPU_NONE)
+ if (cpu == WORK_OFFQ_CPU_NONE)
return NULL;
BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND);
@@ -610,7 +610,7 @@ static struct global_cwq *get_work_gcwq(struct work_struct *work)
static void mark_work_canceling(struct work_struct *work)
{
struct global_cwq *gcwq = get_work_gcwq(work);
- unsigned long cpu = gcwq ? gcwq->cpu : WORK_CPU_NONE;
+ unsigned long cpu = gcwq ? gcwq->cpu : WORK_OFFQ_CPU_NONE;
set_work_data(work, (cpu << WORK_OFFQ_CPU_SHIFT) | WORK_OFFQ_CANCELING,
WORK_STRUCT_PENDING);
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 06/17] workqueue: add worker_pool->id
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (4 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 05/17] workqueue: introduce WORK_OFFQ_CPU_NONE Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 07/17] workqueue: record pool ID instead of CPU in work->data when off-queue Tejun Heo
` (15 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Add worker_pool->id which is allocated from worker_pool_idr. This
will be used to record the last associated worker_pool in work->data.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 9da3db2..fbe2e24 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -151,6 +151,7 @@ struct worker {
struct worker_pool {
struct global_cwq *gcwq; /* I: the owning gcwq */
+ int id; /* I: pool ID */
unsigned int flags; /* X: flags */
struct list_head worklist; /* L: list of pending works */
@@ -472,6 +473,10 @@ static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
[0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
};
+/* idr of all pools */
+static DEFINE_MUTEX(worker_pool_idr_mutex);
+static DEFINE_IDR(worker_pool_idr);
+
static int worker_thread(void *__worker);
static unsigned int work_cpu(struct work_struct *work);
@@ -488,6 +493,19 @@ static struct global_cwq *get_gcwq(unsigned int cpu)
return &unbound_global_cwq;
}
+/* allocate ID and assign it to @pool */
+static int worker_pool_assign_id(struct worker_pool *pool)
+{
+ int ret;
+
+ mutex_lock(&worker_pool_idr_mutex);
+ idr_pre_get(&worker_pool_idr, GFP_KERNEL);
+ ret = idr_get_new(&worker_pool_idr, pool, &pool->id);
+ mutex_unlock(&worker_pool_idr_mutex);
+
+ return ret;
+}
+
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
{
int cpu = pool->gcwq->cpu;
@@ -3839,6 +3857,9 @@ static int __init init_workqueues(void)
mutex_init(&pool->assoc_mutex);
ida_init(&pool->worker_ida);
+
+ /* alloc pool ID */
+ BUG_ON(worker_pool_assign_id(pool));
}
}
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 07/17] workqueue: record pool ID instead of CPU in work->data when off-queue
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (5 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 06/17] workqueue: add worker_pool->id Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 08/17] workqueue: move busy_hash from global_cwq to worker_pool Tejun Heo
` (14 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Currently, when a work item is off-queue, work->data records the CPU
it was last on, which is used to locate the last executing instance
for non-reentrance, flushing, etc.
We're in the process of removing global_cwq and making worker_pool the
top level abstraction. This patch makes work->data point to the pool
it was last associated with instead of CPU.
After the previous WORK_OFFQ_POOL_CPU and worker_poo->id additions,
the conversion is fairly straight-forward. WORK_OFFQ constants and
functions are modified to record and read back pool ID instead.
worker_pool_by_id() is added to allow looking up pool from ID.
get_work_pool() replaces get_work_gcwq(), which is reimplemented using
get_work_pool(). get_work_pool_id() replaces work_cpu().
This patch shouldn't introduce any observable behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
include/linux/workqueue.h | 18 ++++----
kernel/workqueue.c | 111 ++++++++++++++++++++++++++++------------------
2 files changed, 76 insertions(+), 53 deletions(-)
diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index f8b3576..a94e4e8 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -75,19 +75,19 @@ enum {
/*
* When a work item is off queue, its high bits point to the last
- * cpu it was on. Cap at 31 bits and use the highest number to
- * indicate that no cpu is associated.
+ * pool it was on. Cap at 31 bits and use the highest number to
+ * indicate that no pool is associated.
*/
WORK_OFFQ_FLAG_BITS = 1,
- WORK_OFFQ_CPU_SHIFT = WORK_OFFQ_FLAG_BASE + WORK_OFFQ_FLAG_BITS,
- WORK_OFFQ_LEFT = BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT,
- WORK_OFFQ_CPU_BITS = WORK_OFFQ_LEFT <= 31 ? WORK_OFFQ_LEFT : 31,
- WORK_OFFQ_CPU_NONE = (1LU << WORK_OFFQ_CPU_BITS) - 1,
+ WORK_OFFQ_POOL_SHIFT = WORK_OFFQ_FLAG_BASE + WORK_OFFQ_FLAG_BITS,
+ WORK_OFFQ_LEFT = BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT,
+ WORK_OFFQ_POOL_BITS = WORK_OFFQ_LEFT <= 31 ? WORK_OFFQ_LEFT : 31,
+ WORK_OFFQ_POOL_NONE = (1LU << WORK_OFFQ_POOL_BITS) - 1,
/* convenience constants */
WORK_STRUCT_FLAG_MASK = (1UL << WORK_STRUCT_FLAG_BITS) - 1,
WORK_STRUCT_WQ_DATA_MASK = ~WORK_STRUCT_FLAG_MASK,
- WORK_STRUCT_NO_CPU = (unsigned long)WORK_OFFQ_CPU_NONE << WORK_OFFQ_CPU_SHIFT,
+ WORK_STRUCT_NO_POOL = (unsigned long)WORK_OFFQ_POOL_NONE << WORK_OFFQ_POOL_SHIFT,
/* bit mask for work_busy() return values */
WORK_BUSY_PENDING = 1 << 0,
@@ -103,9 +103,9 @@ struct work_struct {
#endif
};
-#define WORK_DATA_INIT() ATOMIC_LONG_INIT(WORK_STRUCT_NO_CPU)
+#define WORK_DATA_INIT() ATOMIC_LONG_INIT(WORK_STRUCT_NO_POOL)
#define WORK_DATA_STATIC_INIT() \
- ATOMIC_LONG_INIT(WORK_STRUCT_NO_CPU | WORK_STRUCT_STATIC)
+ ATOMIC_LONG_INIT(WORK_STRUCT_NO_POOL | WORK_STRUCT_STATIC)
struct delayed_work {
struct work_struct work;
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index fbe2e24..3b4d8a7 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -478,7 +478,6 @@ static DEFINE_MUTEX(worker_pool_idr_mutex);
static DEFINE_IDR(worker_pool_idr);
static int worker_thread(void *__worker);
-static unsigned int work_cpu(struct work_struct *work);
static int std_worker_pool_pri(struct worker_pool *pool)
{
@@ -506,6 +505,15 @@ static int worker_pool_assign_id(struct worker_pool *pool)
return ret;
}
+/*
+ * Lookup worker_pool by id. The idr currently is built during boot and
+ * never modified. Don't worry about locking for now.
+ */
+static struct worker_pool *worker_pool_by_id(int pool_id)
+{
+ return idr_find(&worker_pool_idr, pool_id);
+}
+
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
{
int cpu = pool->gcwq->cpu;
@@ -547,17 +555,17 @@ static int work_next_color(int color)
/*
* While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data
* contain the pointer to the queued cwq. Once execution starts, the flag
- * is cleared and the high bits contain OFFQ flags and CPU number.
+ * is cleared and the high bits contain OFFQ flags and pool ID.
*
- * set_work_cwq(), set_work_cpu_and_clear_pending(), mark_work_canceling()
- * and clear_work_data() can be used to set the cwq, cpu or clear
+ * set_work_cwq(), set_work_pool_and_clear_pending(), mark_work_canceling()
+ * and clear_work_data() can be used to set the cwq, pool or clear
* work->data. These functions should only be called while the work is
* owned - ie. while the PENDING bit is set.
*
- * get_work_[g]cwq() can be used to obtain the gcwq or cwq corresponding to
- * a work. gcwq is available once the work has been queued anywhere after
- * initialization until it is sync canceled. cwq is available only while
- * the work item is queued.
+ * get_work_pool() and get_work_cwq() can be used to obtain the pool or cwq
+ * corresponding to a work. Pool is available once the work has been
+ * queued anywhere after initialization until it is sync canceled. cwq is
+ * available only while the work item is queued.
*
* %WORK_OFFQ_CANCELING is used to mark a work item which is being
* canceled. While being canceled, a work item may have its PENDING set
@@ -579,8 +587,8 @@ static void set_work_cwq(struct work_struct *work,
WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags);
}
-static void set_work_cpu_and_clear_pending(struct work_struct *work,
- unsigned int cpu)
+static void set_work_pool_and_clear_pending(struct work_struct *work,
+ int pool_id)
{
/*
* The following wmb is paired with the implied mb in
@@ -589,13 +597,13 @@ static void set_work_cpu_and_clear_pending(struct work_struct *work,
* owner.
*/
smp_wmb();
- set_work_data(work, (unsigned long)cpu << WORK_OFFQ_CPU_SHIFT, 0);
+ set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0);
}
static void clear_work_data(struct work_struct *work)
{
- smp_wmb(); /* see set_work_cpu_and_clear_pending() */
- set_work_data(work, WORK_STRUCT_NO_CPU, 0);
+ smp_wmb(); /* see set_work_pool_and_clear_pending() */
+ set_work_data(work, WORK_STRUCT_NO_POOL, 0);
}
static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work)
@@ -608,30 +616,58 @@ static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work)
return NULL;
}
-static struct global_cwq *get_work_gcwq(struct work_struct *work)
+/**
+ * get_work_pool - return the worker_pool a given work was associated with
+ * @work: the work item of interest
+ *
+ * Return the worker_pool @work was last associated with. %NULL if none.
+ */
+static struct worker_pool *get_work_pool(struct work_struct *work)
{
unsigned long data = atomic_long_read(&work->data);
- unsigned int cpu;
+ struct worker_pool *pool;
+ int pool_id;
if (data & WORK_STRUCT_CWQ)
return ((struct cpu_workqueue_struct *)
- (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq;
+ (data & WORK_STRUCT_WQ_DATA_MASK))->pool;
- cpu = data >> WORK_OFFQ_CPU_SHIFT;
- if (cpu == WORK_OFFQ_CPU_NONE)
+ pool_id = data >> WORK_OFFQ_POOL_SHIFT;
+ if (pool_id == WORK_OFFQ_POOL_NONE)
return NULL;
- BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND);
- return get_gcwq(cpu);
+ pool = worker_pool_by_id(pool_id);
+ WARN_ON_ONCE(!pool);
+ return pool;
+}
+
+/**
+ * get_work_pool_id - return the worker pool ID a given work is associated with
+ * @work: the work item of interest
+ *
+ * Return the worker_pool ID @work was last associated with.
+ * %WORK_OFFQ_POOL_NONE if none.
+ */
+static int get_work_pool_id(struct work_struct *work)
+{
+ struct worker_pool *pool = get_work_pool(work);
+
+ return pool ? pool->id : WORK_OFFQ_POOL_NONE;
+}
+
+static struct global_cwq *get_work_gcwq(struct work_struct *work)
+{
+ struct worker_pool *pool = get_work_pool(work);
+
+ return pool ? pool->gcwq : NULL;
}
static void mark_work_canceling(struct work_struct *work)
{
- struct global_cwq *gcwq = get_work_gcwq(work);
- unsigned long cpu = gcwq ? gcwq->cpu : WORK_OFFQ_CPU_NONE;
+ unsigned long pool_id = get_work_pool_id(work);
- set_work_data(work, (cpu << WORK_OFFQ_CPU_SHIFT) | WORK_OFFQ_CANCELING,
- WORK_STRUCT_PENDING);
+ pool_id <<= WORK_OFFQ_POOL_SHIFT;
+ set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING);
}
static bool work_is_canceling(struct work_struct *work)
@@ -2211,12 +2247,12 @@ __acquires(&gcwq->lock)
wake_up_worker(pool);
/*
- * Record the last CPU and clear PENDING which should be the last
+ * Record the last pool and clear PENDING which should be the last
* update to @work. Also, do this inside @gcwq->lock so that
* PENDING and queued state changes happen together while IRQ is
* disabled.
*/
- set_work_cpu_and_clear_pending(work, gcwq->cpu);
+ set_work_pool_and_clear_pending(work, pool->id);
spin_unlock_irq(&gcwq->lock);
@@ -2977,7 +3013,8 @@ bool cancel_delayed_work(struct delayed_work *dwork)
if (unlikely(ret < 0))
return false;
- set_work_cpu_and_clear_pending(&dwork->work, work_cpu(&dwork->work));
+ set_work_pool_and_clear_pending(&dwork->work,
+ get_work_pool_id(&dwork->work));
local_irq_restore(flags);
return ret;
}
@@ -3440,20 +3477,6 @@ bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq)
EXPORT_SYMBOL_GPL(workqueue_congested);
/**
- * work_cpu - return the last known associated cpu for @work
- * @work: the work of interest
- *
- * RETURNS:
- * CPU number if @work was ever queued. WORK_CPU_NONE otherwise.
- */
-static unsigned int work_cpu(struct work_struct *work)
-{
- struct global_cwq *gcwq = get_work_gcwq(work);
-
- return gcwq ? gcwq->cpu : WORK_CPU_NONE;
-}
-
-/**
* work_busy - test whether a work is currently pending or running
* @work: the work to be tested
*
@@ -3825,9 +3848,9 @@ static int __init init_workqueues(void)
{
unsigned int cpu;
- /* make sure we have enough bits for OFFQ CPU number */
- BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT)) <
- WORK_CPU_LAST);
+ /* make sure we have enough bits for OFFQ pool ID */
+ BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) <
+ WORK_CPU_LAST * NR_STD_WORKER_POOLS);
cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP);
hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN);
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 08/17] workqueue: move busy_hash from global_cwq to worker_pool
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (6 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 07/17] workqueue: record pool ID instead of CPU in work->data when off-queue Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 09/17] workqueue: move global_cwq->cpu " Tejun Heo
` (13 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
There's no functional necessity for the two pools on the same CPU to
share the busy hash table. It's also likely to be a bottleneck when
implementing pools with user-specified attributes.
This patch makes busy_hash per-pool. The conversion is mostly
straight-forward. Changes worth noting are,
* Large block of changes in rebind_workers() is moving the block
inside for_each_worker_pool() as now there are separate hash tables
for each pool. This changes the order of operations but doesn't
break anything.
* Thre for_each_worker_pool() loops in gcwq_unbind_fn() are combined
into one. This again changes the order of operaitons but doesn't
break anything.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 111 ++++++++++++++++++++++++++++-------------------------
1 file changed, 59 insertions(+), 52 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 3b4d8a7..092d14aa 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -164,6 +164,10 @@ struct worker_pool {
struct timer_list idle_timer; /* L: worker idle timeout */
struct timer_list mayday_timer; /* L: SOS timer for workers */
+ /* workers are chained either in busy_hash or idle_list */
+ DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER);
+ /* L: hash of busy workers */
+
struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */
struct ida worker_ida; /* L: for worker IDs */
};
@@ -177,10 +181,6 @@ struct global_cwq {
spinlock_t lock; /* the gcwq lock */
unsigned int cpu; /* I: the associated cpu */
- /* workers are chained either in busy_hash or pool idle_list */
- DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER);
- /* L: hash of busy workers */
-
struct worker_pool pools[NR_STD_WORKER_POOLS];
/* normal and highpri pools */
} ____cacheline_aligned_in_smp;
@@ -282,8 +282,8 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
for ((pool) = &(gcwq)->pools[0]; \
(pool) < &(gcwq)->pools[NR_STD_WORKER_POOLS]; (pool)++)
-#define for_each_busy_worker(worker, i, pos, gcwq) \
- hash_for_each(gcwq->busy_hash, i, pos, worker, hentry)
+#define for_each_busy_worker(worker, i, pos, pool) \
+ hash_for_each(pool->busy_hash, i, pos, worker, hentry)
static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask,
unsigned int sw)
@@ -911,11 +911,11 @@ static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
/**
* find_worker_executing_work - find worker which is executing a work
- * @gcwq: gcwq of interest
+ * @pool: pool of interest
* @work: work to find worker for
*
- * Find a worker which is executing @work on @gcwq by searching
- * @gcwq->busy_hash which is keyed by the address of @work. For a worker
+ * Find a worker which is executing @work on @pool by searching
+ * @pool->busy_hash which is keyed by the address of @work. For a worker
* to match, its current execution should match the address of @work and
* its work function. This is to avoid unwanted dependency between
* unrelated work executions through a work item being recycled while still
@@ -943,13 +943,13 @@ static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
* Pointer to worker which is executing @work if found, NULL
* otherwise.
*/
-static struct worker *find_worker_executing_work(struct global_cwq *gcwq,
+static struct worker *find_worker_executing_work(struct worker_pool *pool,
struct work_struct *work)
{
struct worker *worker;
struct hlist_node *tmp;
- hash_for_each_possible(gcwq->busy_hash, worker, tmp, hentry,
+ hash_for_each_possible(pool->busy_hash, worker, tmp, hentry,
(unsigned long)work)
if (worker->current_work == work &&
worker->current_func == work->func)
@@ -1210,13 +1210,15 @@ static bool is_chained_work(struct workqueue_struct *wq)
unsigned int cpu;
for_each_gcwq_cpu(cpu) {
- struct global_cwq *gcwq = get_gcwq(cpu);
+ struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+ struct worker_pool *pool = cwq->pool;
+ struct global_cwq *gcwq = pool->gcwq;
struct worker *worker;
struct hlist_node *pos;
int i;
spin_lock_irqsave(&gcwq->lock, flags);
- for_each_busy_worker(worker, i, pos, gcwq) {
+ for_each_busy_worker(worker, i, pos, pool) {
if (worker->task != current)
continue;
spin_unlock_irqrestore(&gcwq->lock, flags);
@@ -1257,7 +1259,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
/* determine gcwq to use */
if (!(wq->flags & WQ_UNBOUND)) {
- struct global_cwq *last_gcwq;
+ struct worker_pool *last_pool;
if (cpu == WORK_CPU_UNBOUND)
cpu = raw_smp_processor_id();
@@ -1269,14 +1271,15 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
* non-reentrancy.
*/
gcwq = get_gcwq(cpu);
- last_gcwq = get_work_gcwq(work);
+ last_pool = get_work_pool(work);
- if (last_gcwq && last_gcwq != gcwq) {
+ if (last_pool && last_pool->gcwq != gcwq) {
+ struct global_cwq *last_gcwq = last_pool->gcwq;
struct worker *worker;
spin_lock(&last_gcwq->lock);
- worker = find_worker_executing_work(last_gcwq, work);
+ worker = find_worker_executing_work(last_pool, work);
if (worker && worker->current_cwq->wq == wq)
gcwq = last_gcwq;
@@ -1741,31 +1744,32 @@ static void rebind_workers(struct global_cwq *gcwq)
*/
wake_up_process(worker->task);
}
- }
- /* rebind busy workers */
- for_each_busy_worker(worker, i, pos, gcwq) {
- struct work_struct *rebind_work = &worker->rebind_work;
- struct workqueue_struct *wq;
+ /* rebind busy workers */
+ for_each_busy_worker(worker, i, pos, pool) {
+ struct work_struct *rebind_work = &worker->rebind_work;
+ struct workqueue_struct *wq;
- if (test_and_set_bit(WORK_STRUCT_PENDING_BIT,
- work_data_bits(rebind_work)))
- continue;
+ if (test_and_set_bit(WORK_STRUCT_PENDING_BIT,
+ work_data_bits(rebind_work)))
+ continue;
- debug_work_activate(rebind_work);
+ debug_work_activate(rebind_work);
- /*
- * wq doesn't really matter but let's keep @worker->pool
- * and @cwq->pool consistent for sanity.
- */
- if (std_worker_pool_pri(worker->pool))
- wq = system_highpri_wq;
- else
- wq = system_wq;
-
- insert_work(get_cwq(gcwq->cpu, wq), rebind_work,
- worker->scheduled.next,
- work_color_to_flags(WORK_NO_COLOR));
+ /*
+ * wq doesn't really matter but let's keep
+ * @worker->pool and @cwq->pool consistent for
+ * sanity.
+ */
+ if (std_worker_pool_pri(worker->pool))
+ wq = system_highpri_wq;
+ else
+ wq = system_wq;
+
+ insert_work(get_cwq(gcwq->cpu, wq), rebind_work,
+ worker->scheduled.next,
+ work_color_to_flags(WORK_NO_COLOR));
+ }
}
}
@@ -2216,7 +2220,7 @@ __acquires(&gcwq->lock)
* already processing the work. If so, defer the work to the
* currently executing one.
*/
- collision = find_worker_executing_work(gcwq, work);
+ collision = find_worker_executing_work(pool, work);
if (unlikely(collision)) {
move_linked_works(work, &collision->scheduled, NULL);
return;
@@ -2224,7 +2228,7 @@ __acquires(&gcwq->lock)
/* claim and dequeue */
debug_work_deactivate(work);
- hash_add(gcwq->busy_hash, &worker->hentry, (unsigned long)work);
+ hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work);
worker->current_work = work;
worker->current_func = work->func;
worker->current_cwq = cwq;
@@ -2843,13 +2847,15 @@ EXPORT_SYMBOL_GPL(drain_workqueue);
static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
{
struct worker *worker = NULL;
+ struct worker_pool *pool;
struct global_cwq *gcwq;
struct cpu_workqueue_struct *cwq;
might_sleep();
- gcwq = get_work_gcwq(work);
- if (!gcwq)
+ pool = get_work_pool(work);
+ if (!pool)
return false;
+ gcwq = pool->gcwq;
spin_lock_irq(&gcwq->lock);
if (!list_empty(&work->entry)) {
@@ -2863,7 +2869,7 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
if (unlikely(!cwq || gcwq != cwq->pool->gcwq))
goto already_gone;
} else {
- worker = find_worker_executing_work(gcwq, work);
+ worker = find_worker_executing_work(pool, work);
if (!worker)
goto already_gone;
cwq = worker->current_cwq;
@@ -3491,18 +3497,20 @@ EXPORT_SYMBOL_GPL(workqueue_congested);
*/
unsigned int work_busy(struct work_struct *work)
{
- struct global_cwq *gcwq = get_work_gcwq(work);
+ struct worker_pool *pool = get_work_pool(work);
+ struct global_cwq *gcwq;
unsigned long flags;
unsigned int ret = 0;
- if (!gcwq)
+ if (!pool)
return 0;
+ gcwq = pool->gcwq;
spin_lock_irqsave(&gcwq->lock, flags);
if (work_pending(work))
ret |= WORK_BUSY_PENDING;
- if (find_worker_executing_work(gcwq, work))
+ if (find_worker_executing_work(pool, work))
ret |= WORK_BUSY_RUNNING;
spin_unlock_irqrestore(&gcwq->lock, flags);
@@ -3564,15 +3572,15 @@ static void gcwq_unbind_fn(struct work_struct *work)
* ones which are still executing works from before the last CPU
* down must be on the cpu. After this, they may become diasporas.
*/
- for_each_worker_pool(pool, gcwq)
+ for_each_worker_pool(pool, gcwq) {
list_for_each_entry(worker, &pool->idle_list, entry)
worker->flags |= WORKER_UNBOUND;
- for_each_busy_worker(worker, i, pos, gcwq)
- worker->flags |= WORKER_UNBOUND;
+ for_each_busy_worker(worker, i, pos, pool)
+ worker->flags |= WORKER_UNBOUND;
- for_each_worker_pool(pool, gcwq)
pool->flags |= POOL_DISASSOCIATED;
+ }
gcwq_release_assoc_and_unlock(gcwq);
@@ -3863,13 +3871,12 @@ static int __init init_workqueues(void)
spin_lock_init(&gcwq->lock);
gcwq->cpu = cpu;
- hash_init(gcwq->busy_hash);
-
for_each_worker_pool(pool, gcwq) {
pool->gcwq = gcwq;
pool->flags |= POOL_DISASSOCIATED;
INIT_LIST_HEAD(&pool->worklist);
INIT_LIST_HEAD(&pool->idle_list);
+ hash_init(pool->busy_hash);
init_timer_deferrable(&pool->idle_timer);
pool->idle_timer.function = idle_worker_timeout;
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 09/17] workqueue: move global_cwq->cpu to worker_pool
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (7 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 08/17] workqueue: move busy_hash from global_cwq to worker_pool Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 10/17] workqueue: move global_cwq->lock " Tejun Heo
` (12 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Move gcwq->cpu to pool->cpu. This introduces a couple places where
gcwq->pools[0].cpu is used. These will soon go away as gcwq is
further reduced.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
include/trace/events/workqueue.h | 2 +-
kernel/workqueue.c | 42 ++++++++++++++++++++--------------------
2 files changed, 22 insertions(+), 22 deletions(-)
diff --git a/include/trace/events/workqueue.h b/include/trace/events/workqueue.h
index f28d1b6..4e798e3 100644
--- a/include/trace/events/workqueue.h
+++ b/include/trace/events/workqueue.h
@@ -54,7 +54,7 @@ TRACE_EVENT(workqueue_queue_work,
__entry->function = work->func;
__entry->workqueue = cwq->wq;
__entry->req_cpu = req_cpu;
- __entry->cpu = cwq->pool->gcwq->cpu;
+ __entry->cpu = cwq->pool->cpu;
),
TP_printk("work struct=%p function=%pf workqueue=%p req_cpu=%u cpu=%u",
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 092d14aa..1720da7 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -151,6 +151,7 @@ struct worker {
struct worker_pool {
struct global_cwq *gcwq; /* I: the owning gcwq */
+ unsigned int cpu; /* I: the associated cpu */
int id; /* I: pool ID */
unsigned int flags; /* X: flags */
@@ -179,7 +180,6 @@ struct worker_pool {
*/
struct global_cwq {
spinlock_t lock; /* the gcwq lock */
- unsigned int cpu; /* I: the associated cpu */
struct worker_pool pools[NR_STD_WORKER_POOLS];
/* normal and highpri pools */
@@ -516,7 +516,7 @@ static struct worker_pool *worker_pool_by_id(int pool_id)
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
{
- int cpu = pool->gcwq->cpu;
+ int cpu = pool->cpu;
int idx = std_worker_pool_pri(pool);
if (cpu != WORK_CPU_UNBOUND)
@@ -791,7 +791,7 @@ void wq_worker_waking_up(struct task_struct *task, unsigned int cpu)
struct worker *worker = kthread_data(task);
if (!(worker->flags & WORKER_NOT_RUNNING)) {
- WARN_ON_ONCE(worker->pool->gcwq->cpu != cpu);
+ WARN_ON_ONCE(worker->pool->cpu != cpu);
atomic_inc(get_pool_nr_running(worker->pool));
}
}
@@ -1297,7 +1297,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
}
/* gcwq determined, get cwq and queue */
- cwq = get_cwq(gcwq->cpu, wq);
+ cwq = get_cwq(gcwq->pools[0].cpu, wq);
trace_workqueue_queue_work(req_cpu, cwq, work);
if (WARN_ON(!list_empty(&work->entry))) {
@@ -1404,20 +1404,20 @@ static void __queue_delayed_work(int cpu, struct workqueue_struct *wq,
/*
* This stores cwq for the moment, for the timer_fn. Note that the
- * work's gcwq is preserved to allow reentrance detection for
+ * work's pool is preserved to allow reentrance detection for
* delayed works.
*/
if (!(wq->flags & WQ_UNBOUND)) {
- struct global_cwq *gcwq = get_work_gcwq(work);
+ struct worker_pool *pool = get_work_pool(work);
/*
- * If we cannot get the last gcwq from @work directly,
+ * If we cannot get the last pool from @work directly,
* select the last CPU such that it avoids unnecessarily
* triggering non-reentrancy check in __queue_work().
*/
lcpu = cpu;
- if (gcwq)
- lcpu = gcwq->cpu;
+ if (pool)
+ lcpu = pool->cpu;
if (lcpu == WORK_CPU_UNBOUND)
lcpu = raw_smp_processor_id();
} else {
@@ -1638,14 +1638,14 @@ __acquires(&gcwq->lock)
* against POOL_DISASSOCIATED.
*/
if (!(pool->flags & POOL_DISASSOCIATED))
- set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu));
+ set_cpus_allowed_ptr(task, get_cpu_mask(pool->cpu));
spin_lock_irq(&gcwq->lock);
if (pool->flags & POOL_DISASSOCIATED)
return false;
- if (task_cpu(task) == gcwq->cpu &&
+ if (task_cpu(task) == pool->cpu &&
cpumask_equal(¤t->cpus_allowed,
- get_cpu_mask(gcwq->cpu)))
+ get_cpu_mask(pool->cpu)))
return true;
spin_unlock_irq(&gcwq->lock);
@@ -1766,7 +1766,7 @@ static void rebind_workers(struct global_cwq *gcwq)
else
wq = system_wq;
- insert_work(get_cwq(gcwq->cpu, wq), rebind_work,
+ insert_work(get_cwq(pool->cpu, wq), rebind_work,
worker->scheduled.next,
work_color_to_flags(WORK_NO_COLOR));
}
@@ -1825,10 +1825,10 @@ static struct worker *create_worker(struct worker_pool *pool)
worker->pool = pool;
worker->id = id;
- if (gcwq->cpu != WORK_CPU_UNBOUND)
+ if (pool->cpu != WORK_CPU_UNBOUND)
worker->task = kthread_create_on_node(worker_thread,
- worker, cpu_to_node(gcwq->cpu),
- "kworker/%u:%d%s", gcwq->cpu, id, pri);
+ worker, cpu_to_node(pool->cpu),
+ "kworker/%u:%d%s", pool->cpu, id, pri);
else
worker->task = kthread_create(worker_thread, worker,
"kworker/u:%d%s", id, pri);
@@ -1848,7 +1848,7 @@ static struct worker *create_worker(struct worker_pool *pool)
* online, make sure every worker has %PF_THREAD_BOUND set.
*/
if (!(pool->flags & POOL_DISASSOCIATED)) {
- kthread_bind(worker->task, gcwq->cpu);
+ kthread_bind(worker->task, pool->cpu);
} else {
worker->task->flags |= PF_THREAD_BOUND;
worker->flags |= WORKER_UNBOUND;
@@ -1955,7 +1955,7 @@ static bool send_mayday(struct work_struct *work)
return false;
/* mayday mayday mayday */
- cpu = cwq->pool->gcwq->cpu;
+ cpu = cwq->pool->cpu;
/* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */
if (cpu == WORK_CPU_UNBOUND)
cpu = 0;
@@ -2212,7 +2212,7 @@ __acquires(&gcwq->lock)
*/
WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) &&
!(pool->flags & POOL_DISASSOCIATED) &&
- raw_smp_processor_id() != gcwq->cpu);
+ raw_smp_processor_id() != pool->cpu);
/*
* A single work shouldn't be executed concurrently by
@@ -3562,7 +3562,7 @@ static void gcwq_unbind_fn(struct work_struct *work)
struct hlist_node *pos;
int i;
- BUG_ON(gcwq->cpu != smp_processor_id());
+ BUG_ON(gcwq->pools[0].cpu != smp_processor_id());
gcwq_claim_assoc_and_lock(gcwq);
@@ -3869,10 +3869,10 @@ static int __init init_workqueues(void)
struct worker_pool *pool;
spin_lock_init(&gcwq->lock);
- gcwq->cpu = cpu;
for_each_worker_pool(pool, gcwq) {
pool->gcwq = gcwq;
+ pool->cpu = cpu;
pool->flags |= POOL_DISASSOCIATED;
INIT_LIST_HEAD(&pool->worklist);
INIT_LIST_HEAD(&pool->idle_list);
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 10/17] workqueue: move global_cwq->lock to worker_pool
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (8 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 09/17] workqueue: move global_cwq->cpu " Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 11/17] workqueue: make hotplug processing per-pool Tejun Heo
` (11 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Move gcwq->lock to pool->lock. The conversion is mostly
straight-forward. Things worth noting are
* In many places, this removes the need to use gcwq completely. pool
is used directly instead. get_std_worker_pool() is added to help
some of these conversions. This also leaves get_work_gcwq() without
any user. Removed.
* In hotplug and freezer paths, the pools belonging to a CPU are often
processed together. This patch makes those paths hold locks of all
pools, with highpri lock nested inside, to keep the conversion
straight-forward. These nested lockings will be removed by
following patches.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 316 ++++++++++++++++++++++++++---------------------------
1 file changed, 154 insertions(+), 162 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 1720da7..755d2c4 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -108,12 +108,12 @@ enum {
* P: Preemption protected. Disabling preemption is enough and should
* only be modified and accessed from the local cpu.
*
- * L: gcwq->lock protected. Access with gcwq->lock held.
+ * L: pool->lock protected. Access with pool->lock held.
*
- * X: During normal operation, modification requires gcwq->lock and
- * should be done only from local cpu. Either disabling preemption
- * on local cpu or grabbing gcwq->lock is enough for read access.
- * If POOL_DISASSOCIATED is set, it's identical to L.
+ * X: During normal operation, modification requires pool->lock and should
+ * be done only from local cpu. Either disabling preemption on local
+ * cpu or grabbing pool->lock is enough for read access. If
+ * POOL_DISASSOCIATED is set, it's identical to L.
*
* F: wq->flush_mutex protected.
*
@@ -151,6 +151,7 @@ struct worker {
struct worker_pool {
struct global_cwq *gcwq; /* I: the owning gcwq */
+ spinlock_t lock; /* the pool lock */
unsigned int cpu; /* I: the associated cpu */
int id; /* I: pool ID */
unsigned int flags; /* X: flags */
@@ -179,8 +180,6 @@ struct worker_pool {
* target workqueues.
*/
struct global_cwq {
- spinlock_t lock; /* the gcwq lock */
-
struct worker_pool pools[NR_STD_WORKER_POOLS];
/* normal and highpri pools */
} ____cacheline_aligned_in_smp;
@@ -514,6 +513,13 @@ static struct worker_pool *worker_pool_by_id(int pool_id)
return idr_find(&worker_pool_idr, pool_id);
}
+static struct worker_pool *get_std_worker_pool(int cpu, bool highpri)
+{
+ struct global_cwq *gcwq = get_gcwq(cpu);
+
+ return &gcwq->pools[highpri];
+}
+
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
{
int cpu = pool->cpu;
@@ -655,13 +661,6 @@ static int get_work_pool_id(struct work_struct *work)
return pool ? pool->id : WORK_OFFQ_POOL_NONE;
}
-static struct global_cwq *get_work_gcwq(struct work_struct *work)
-{
- struct worker_pool *pool = get_work_pool(work);
-
- return pool ? pool->gcwq : NULL;
-}
-
static void mark_work_canceling(struct work_struct *work)
{
unsigned long pool_id = get_work_pool_id(work);
@@ -680,7 +679,7 @@ static bool work_is_canceling(struct work_struct *work)
/*
* Policy functions. These define the policies on how the global worker
* pools are managed. Unless noted otherwise, these functions assume that
- * they're being called with gcwq->lock held.
+ * they're being called with pool->lock held.
*/
static bool __need_more_worker(struct worker_pool *pool)
@@ -765,7 +764,7 @@ static struct worker *first_worker(struct worker_pool *pool)
* Wake up the first idle worker of @pool.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void wake_up_worker(struct worker_pool *pool)
{
@@ -832,7 +831,7 @@ struct task_struct *wq_worker_sleeping(struct task_struct *task,
* NOT_RUNNING is clear. This means that we're bound to and
* running on the local cpu w/ rq lock held and preemption
* disabled, which in turn means that none else could be
- * manipulating idle_list, so dereferencing idle_list without gcwq
+ * manipulating idle_list, so dereferencing idle_list without pool
* lock is safe.
*/
if (atomic_dec_and_test(nr_running) && !list_empty(&pool->worklist))
@@ -851,7 +850,7 @@ struct task_struct *wq_worker_sleeping(struct task_struct *task,
* woken up.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock)
+ * spin_lock_irq(pool->lock)
*/
static inline void worker_set_flags(struct worker *worker, unsigned int flags,
bool wakeup)
@@ -888,7 +887,7 @@ static inline void worker_set_flags(struct worker *worker, unsigned int flags,
* Clear @flags in @worker->flags and adjust nr_running accordingly.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock)
+ * spin_lock_irq(pool->lock)
*/
static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
{
@@ -937,7 +936,7 @@ static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
* function.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*
* RETURNS:
* Pointer to worker which is executing @work if found, NULL
@@ -973,7 +972,7 @@ static struct worker *find_worker_executing_work(struct worker_pool *pool,
* nested inside outer list_for_each_entry_safe().
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void move_linked_works(struct work_struct *work, struct list_head *head,
struct work_struct **nextp)
@@ -1026,7 +1025,7 @@ static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq)
* decrement nr_in_flight of its cwq and handle workqueue flushing.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color)
{
@@ -1090,7 +1089,7 @@ static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color)
static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
unsigned long *flags)
{
- struct global_cwq *gcwq;
+ struct worker_pool *pool;
local_irq_save(*flags);
@@ -1115,19 +1114,19 @@ static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
* The queueing is in progress, or it is already queued. Try to
* steal it from ->worklist without clearing WORK_STRUCT_PENDING.
*/
- gcwq = get_work_gcwq(work);
- if (!gcwq)
+ pool = get_work_pool(work);
+ if (!pool)
goto fail;
- spin_lock(&gcwq->lock);
+ spin_lock(&pool->lock);
if (!list_empty(&work->entry)) {
/*
- * This work is queued, but perhaps we locked the wrong gcwq.
- * In that case we must see the new value after rmb(), see
- * insert_work()->wmb().
+ * This work is queued, but perhaps we locked the wrong
+ * pool. In that case we must see the new value after
+ * rmb(), see insert_work()->wmb().
*/
smp_rmb();
- if (gcwq == get_work_gcwq(work)) {
+ if (pool == get_work_pool(work)) {
debug_work_deactivate(work);
/*
@@ -1145,11 +1144,11 @@ static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
cwq_dec_nr_in_flight(get_work_cwq(work),
get_work_color(work));
- spin_unlock(&gcwq->lock);
+ spin_unlock(&pool->lock);
return 1;
}
}
- spin_unlock(&gcwq->lock);
+ spin_unlock(&pool->lock);
fail:
local_irq_restore(*flags);
if (work_is_canceling(work))
@@ -1169,7 +1168,7 @@ fail:
* @extra_flags is or'd to work_struct flags.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void insert_work(struct cpu_workqueue_struct *cwq,
struct work_struct *work, struct list_head *head,
@@ -1212,23 +1211,22 @@ static bool is_chained_work(struct workqueue_struct *wq)
for_each_gcwq_cpu(cpu) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
struct worker_pool *pool = cwq->pool;
- struct global_cwq *gcwq = pool->gcwq;
struct worker *worker;
struct hlist_node *pos;
int i;
- spin_lock_irqsave(&gcwq->lock, flags);
+ spin_lock_irqsave(&pool->lock, flags);
for_each_busy_worker(worker, i, pos, pool) {
if (worker->task != current)
continue;
- spin_unlock_irqrestore(&gcwq->lock, flags);
+ spin_unlock_irqrestore(&pool->lock, flags);
/*
* I'm @worker, no locking necessary. See if @work
* is headed to the same workqueue.
*/
return worker->current_cwq->wq == wq;
}
- spin_unlock_irqrestore(&gcwq->lock, flags);
+ spin_unlock_irqrestore(&pool->lock, flags);
}
return false;
}
@@ -1236,7 +1234,8 @@ static bool is_chained_work(struct workqueue_struct *wq)
static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
struct work_struct *work)
{
- struct global_cwq *gcwq;
+ bool highpri = wq->flags & WQ_HIGHPRI;
+ struct worker_pool *pool;
struct cpu_workqueue_struct *cwq;
struct list_head *worklist;
unsigned int work_flags;
@@ -1257,7 +1256,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
WARN_ON_ONCE(!is_chained_work(wq)))
return;
- /* determine gcwq to use */
+ /* determine pool to use */
if (!(wq->flags & WQ_UNBOUND)) {
struct worker_pool *last_pool;
@@ -1270,38 +1269,37 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
* work needs to be queued on that cpu to guarantee
* non-reentrancy.
*/
- gcwq = get_gcwq(cpu);
+ pool = get_std_worker_pool(cpu, highpri);
last_pool = get_work_pool(work);
- if (last_pool && last_pool->gcwq != gcwq) {
- struct global_cwq *last_gcwq = last_pool->gcwq;
+ if (last_pool && last_pool != pool) {
struct worker *worker;
- spin_lock(&last_gcwq->lock);
+ spin_lock(&last_pool->lock);
worker = find_worker_executing_work(last_pool, work);
if (worker && worker->current_cwq->wq == wq)
- gcwq = last_gcwq;
+ pool = last_pool;
else {
/* meh... not running there, queue here */
- spin_unlock(&last_gcwq->lock);
- spin_lock(&gcwq->lock);
+ spin_unlock(&last_pool->lock);
+ spin_lock(&pool->lock);
}
} else {
- spin_lock(&gcwq->lock);
+ spin_lock(&pool->lock);
}
} else {
- gcwq = get_gcwq(WORK_CPU_UNBOUND);
- spin_lock(&gcwq->lock);
+ pool = get_std_worker_pool(WORK_CPU_UNBOUND, highpri);
+ spin_lock(&pool->lock);
}
- /* gcwq determined, get cwq and queue */
- cwq = get_cwq(gcwq->pools[0].cpu, wq);
+ /* pool determined, get cwq and queue */
+ cwq = get_cwq(pool->cpu, wq);
trace_workqueue_queue_work(req_cpu, cwq, work);
if (WARN_ON(!list_empty(&work->entry))) {
- spin_unlock(&gcwq->lock);
+ spin_unlock(&pool->lock);
return;
}
@@ -1319,7 +1317,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
insert_work(cwq, work, worklist, work_flags);
- spin_unlock(&gcwq->lock);
+ spin_unlock(&pool->lock);
}
/**
@@ -1542,7 +1540,7 @@ EXPORT_SYMBOL_GPL(mod_delayed_work);
* necessary.
*
* LOCKING:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void worker_enter_idle(struct worker *worker)
{
@@ -1565,7 +1563,7 @@ static void worker_enter_idle(struct worker *worker)
/*
* Sanity check nr_running. Because gcwq_unbind_fn() releases
- * gcwq->lock between setting %WORKER_UNBOUND and zapping
+ * pool->lock between setting %WORKER_UNBOUND and zapping
* nr_running, the warning may trigger spuriously. Check iff
* unbind is not in progress.
*/
@@ -1581,7 +1579,7 @@ static void worker_enter_idle(struct worker *worker)
* @worker is leaving idle state. Update stats.
*
* LOCKING:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void worker_leave_idle(struct worker *worker)
{
@@ -1616,7 +1614,7 @@ static void worker_leave_idle(struct worker *worker)
* guarantee the scheduling requirement described in the first paragraph.
*
* CONTEXT:
- * Might sleep. Called without any lock but returns with gcwq->lock
+ * Might sleep. Called without any lock but returns with pool->lock
* held.
*
* RETURNS:
@@ -1624,10 +1622,9 @@ static void worker_leave_idle(struct worker *worker)
* bound), %false if offline.
*/
static bool worker_maybe_bind_and_lock(struct worker *worker)
-__acquires(&gcwq->lock)
+__acquires(&pool->lock)
{
struct worker_pool *pool = worker->pool;
- struct global_cwq *gcwq = pool->gcwq;
struct task_struct *task = worker->task;
while (true) {
@@ -1640,14 +1637,14 @@ __acquires(&gcwq->lock)
if (!(pool->flags & POOL_DISASSOCIATED))
set_cpus_allowed_ptr(task, get_cpu_mask(pool->cpu));
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (pool->flags & POOL_DISASSOCIATED)
return false;
if (task_cpu(task) == pool->cpu &&
cpumask_equal(¤t->cpus_allowed,
get_cpu_mask(pool->cpu)))
return true;
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
/*
* We've raced with CPU hot[un]plug. Give it a breather
@@ -1666,15 +1663,13 @@ __acquires(&gcwq->lock)
*/
static void idle_worker_rebind(struct worker *worker)
{
- struct global_cwq *gcwq = worker->pool->gcwq;
-
/* CPU may go down again inbetween, clear UNBOUND only on success */
if (worker_maybe_bind_and_lock(worker))
worker_clr_flags(worker, WORKER_UNBOUND);
/* rebind complete, become available again */
list_add(&worker->entry, &worker->pool->idle_list);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&worker->pool->lock);
}
/*
@@ -1686,12 +1681,11 @@ static void idle_worker_rebind(struct worker *worker)
static void busy_worker_rebind_fn(struct work_struct *work)
{
struct worker *worker = container_of(work, struct worker, rebind_work);
- struct global_cwq *gcwq = worker->pool->gcwq;
if (worker_maybe_bind_and_lock(worker))
worker_clr_flags(worker, WORKER_UNBOUND);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&worker->pool->lock);
}
/**
@@ -1723,10 +1717,10 @@ static void rebind_workers(struct global_cwq *gcwq)
struct hlist_node *pos;
int i;
- lockdep_assert_held(&gcwq->lock);
-
- for_each_worker_pool(pool, gcwq)
+ for_each_worker_pool(pool, gcwq) {
lockdep_assert_held(&pool->assoc_mutex);
+ lockdep_assert_held(&pool->lock);
+ }
/* dequeue and kick idle ones */
for_each_worker_pool(pool, gcwq) {
@@ -1804,19 +1798,18 @@ static struct worker *alloc_worker(void)
*/
static struct worker *create_worker(struct worker_pool *pool)
{
- struct global_cwq *gcwq = pool->gcwq;
const char *pri = std_worker_pool_pri(pool) ? "H" : "";
struct worker *worker = NULL;
int id = -1;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
while (ida_get_new(&pool->worker_ida, &id)) {
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL))
goto fail;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
}
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
worker = alloc_worker();
if (!worker)
@@ -1857,9 +1850,9 @@ static struct worker *create_worker(struct worker_pool *pool)
return worker;
fail:
if (id >= 0) {
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
ida_remove(&pool->worker_ida, id);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
kfree(worker);
return NULL;
@@ -1872,7 +1865,7 @@ fail:
* Make the gcwq aware of @worker and start it.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void start_worker(struct worker *worker)
{
@@ -1889,12 +1882,11 @@ static void start_worker(struct worker *worker)
* Destroy @worker and adjust @gcwq stats accordingly.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock) which is released and regrabbed.
+ * spin_lock_irq(pool->lock) which is released and regrabbed.
*/
static void destroy_worker(struct worker *worker)
{
struct worker_pool *pool = worker->pool;
- struct global_cwq *gcwq = pool->gcwq;
int id = worker->id;
/* sanity check frenzy */
@@ -1909,21 +1901,20 @@ static void destroy_worker(struct worker *worker)
list_del_init(&worker->entry);
worker->flags |= WORKER_DIE;
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
kthread_stop(worker->task);
kfree(worker);
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
ida_remove(&pool->worker_ida, id);
}
static void idle_worker_timeout(unsigned long __pool)
{
struct worker_pool *pool = (void *)__pool;
- struct global_cwq *gcwq = pool->gcwq;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (too_many_workers(pool)) {
struct worker *worker;
@@ -1942,7 +1933,7 @@ static void idle_worker_timeout(unsigned long __pool)
}
}
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
static bool send_mayday(struct work_struct *work)
@@ -1967,10 +1958,9 @@ static bool send_mayday(struct work_struct *work)
static void gcwq_mayday_timeout(unsigned long __pool)
{
struct worker_pool *pool = (void *)__pool;
- struct global_cwq *gcwq = pool->gcwq;
struct work_struct *work;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (need_to_create_worker(pool)) {
/*
@@ -1983,7 +1973,7 @@ static void gcwq_mayday_timeout(unsigned long __pool)
send_mayday(work);
}
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL);
}
@@ -2002,24 +1992,22 @@ static void gcwq_mayday_timeout(unsigned long __pool)
* may_start_working() true.
*
* LOCKING:
- * spin_lock_irq(gcwq->lock) which may be released and regrabbed
+ * spin_lock_irq(pool->lock) which may be released and regrabbed
* multiple times. Does GFP_KERNEL allocations. Called only from
* manager.
*
* RETURNS:
- * false if no action was taken and gcwq->lock stayed locked, true
+ * false if no action was taken and pool->lock stayed locked, true
* otherwise.
*/
static bool maybe_create_worker(struct worker_pool *pool)
-__releases(&gcwq->lock)
-__acquires(&gcwq->lock)
+__releases(&pool->lock)
+__acquires(&pool->lock)
{
- struct global_cwq *gcwq = pool->gcwq;
-
if (!need_to_create_worker(pool))
return false;
restart:
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
/* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */
mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT);
@@ -2030,7 +2018,7 @@ restart:
worker = create_worker(pool);
if (worker) {
del_timer_sync(&pool->mayday_timer);
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
start_worker(worker);
BUG_ON(need_to_create_worker(pool));
return true;
@@ -2047,7 +2035,7 @@ restart:
}
del_timer_sync(&pool->mayday_timer);
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (need_to_create_worker(pool))
goto restart;
return true;
@@ -2061,11 +2049,11 @@ restart:
* IDLE_WORKER_TIMEOUT.
*
* LOCKING:
- * spin_lock_irq(gcwq->lock) which may be released and regrabbed
+ * spin_lock_irq(pool->lock) which may be released and regrabbed
* multiple times. Called only from manager.
*
* RETURNS:
- * false if no action was taken and gcwq->lock stayed locked, true
+ * false if no action was taken and pool->lock stayed locked, true
* otherwise.
*/
static bool maybe_destroy_workers(struct worker_pool *pool)
@@ -2104,12 +2092,12 @@ static bool maybe_destroy_workers(struct worker_pool *pool)
* and may_start_working() is true.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock) which may be released and regrabbed
+ * spin_lock_irq(pool->lock) which may be released and regrabbed
* multiple times. Does GFP_KERNEL allocations.
*
* RETURNS:
- * false if no action was taken and gcwq->lock stayed locked, true if
- * some action was taken.
+ * spin_lock_irq(pool->lock) which may be released and regrabbed
+ * multiple times. Does GFP_KERNEL allocations.
*/
static bool manage_workers(struct worker *worker)
{
@@ -2131,10 +2119,10 @@ static bool manage_workers(struct worker *worker)
* manager against CPU hotplug.
*
* assoc_mutex would always be free unless CPU hotplug is in
- * progress. trylock first without dropping @gcwq->lock.
+ * progress. trylock first without dropping @pool->lock.
*/
if (unlikely(!mutex_trylock(&pool->assoc_mutex))) {
- spin_unlock_irq(&pool->gcwq->lock);
+ spin_unlock_irq(&pool->lock);
mutex_lock(&pool->assoc_mutex);
/*
* CPU hotplug could have happened while we were waiting
@@ -2181,15 +2169,14 @@ static bool manage_workers(struct worker *worker)
* call this function to process a work.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock) which is released and regrabbed.
+ * spin_lock_irq(pool->lock) which is released and regrabbed.
*/
static void process_one_work(struct worker *worker, struct work_struct *work)
-__releases(&gcwq->lock)
-__acquires(&gcwq->lock)
+__releases(&pool->lock)
+__acquires(&pool->lock)
{
struct cpu_workqueue_struct *cwq = get_work_cwq(work);
struct worker_pool *pool = worker->pool;
- struct global_cwq *gcwq = pool->gcwq;
bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE;
int work_color;
struct worker *collision;
@@ -2244,7 +2231,7 @@ __acquires(&gcwq->lock)
worker_set_flags(worker, WORKER_CPU_INTENSIVE, true);
/*
- * Unbound gcwq isn't concurrency managed and work items should be
+ * Unbound pool isn't concurrency managed and work items should be
* executed ASAP. Wake up another worker if necessary.
*/
if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool))
@@ -2252,13 +2239,13 @@ __acquires(&gcwq->lock)
/*
* Record the last pool and clear PENDING which should be the last
- * update to @work. Also, do this inside @gcwq->lock so that
+ * update to @work. Also, do this inside @pool->lock so that
* PENDING and queued state changes happen together while IRQ is
* disabled.
*/
set_work_pool_and_clear_pending(work, pool->id);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
lock_map_acquire_read(&cwq->wq->lockdep_map);
lock_map_acquire(&lockdep_map);
@@ -2281,7 +2268,7 @@ __acquires(&gcwq->lock)
dump_stack();
}
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
/* clear cpu intensive status */
if (unlikely(cpu_intensive))
@@ -2304,7 +2291,7 @@ __acquires(&gcwq->lock)
* fetches a work from the top and executes it.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock) which may be released and regrabbed
+ * spin_lock_irq(pool->lock) which may be released and regrabbed
* multiple times.
*/
static void process_scheduled_works(struct worker *worker)
@@ -2330,16 +2317,15 @@ static int worker_thread(void *__worker)
{
struct worker *worker = __worker;
struct worker_pool *pool = worker->pool;
- struct global_cwq *gcwq = pool->gcwq;
/* tell the scheduler that this is a workqueue worker */
worker->task->flags |= PF_WQ_WORKER;
woke_up:
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
/* we are off idle list if destruction or rebind is requested */
if (unlikely(list_empty(&worker->entry))) {
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
/* if DIE is set, destruction is requested */
if (worker->flags & WORKER_DIE) {
@@ -2398,15 +2384,15 @@ sleep:
goto recheck;
/*
- * gcwq->lock is held and there's no work to process and no
- * need to manage, sleep. Workers are woken up only while
- * holding gcwq->lock or from local cpu, so setting the
- * current state before releasing gcwq->lock is enough to
- * prevent losing any event.
+ * pool->lock is held and there's no work to process and no need to
+ * manage, sleep. Workers are woken up only while holding
+ * pool->lock or from local cpu, so setting the current state
+ * before releasing pool->lock is enough to prevent losing any
+ * event.
*/
worker_enter_idle(worker);
__set_current_state(TASK_INTERRUPTIBLE);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
schedule();
goto woke_up;
}
@@ -2455,7 +2441,6 @@ repeat:
unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu;
struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq);
struct worker_pool *pool = cwq->pool;
- struct global_cwq *gcwq = pool->gcwq;
struct work_struct *work, *n;
__set_current_state(TASK_RUNNING);
@@ -2477,14 +2462,14 @@ repeat:
process_scheduled_works(rescuer);
/*
- * Leave this gcwq. If keep_working() is %true, notify a
+ * Leave this pool. If keep_working() is %true, notify a
* regular worker; otherwise, we end up with 0 concurrency
* and stalling the execution.
*/
if (keep_working(pool))
wake_up_worker(pool);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
schedule();
@@ -2524,7 +2509,7 @@ static void wq_barrier_func(struct work_struct *work)
* underneath us, so we can't reliably determine cwq from @target.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
struct wq_barrier *barr,
@@ -2534,7 +2519,7 @@ static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
unsigned int linked = 0;
/*
- * debugobject calls are safe here even with gcwq->lock locked
+ * debugobject calls are safe here even with pool->lock locked
* as we know for sure that this will not trigger any of the
* checks and call back into the fixup functions where we
* might deadlock.
@@ -2607,9 +2592,9 @@ static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq,
for_each_cwq_cpu(cpu, wq) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
- struct global_cwq *gcwq = cwq->pool->gcwq;
+ struct worker_pool *pool = cwq->pool;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (flush_color >= 0) {
BUG_ON(cwq->flush_color != -1);
@@ -2626,7 +2611,7 @@ static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq,
cwq->work_color = work_color;
}
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush))
@@ -2823,9 +2808,9 @@ reflush:
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
bool drained;
- spin_lock_irq(&cwq->pool->gcwq->lock);
+ spin_lock_irq(&cwq->pool->lock);
drained = !cwq->nr_active && list_empty(&cwq->delayed_works);
- spin_unlock_irq(&cwq->pool->gcwq->lock);
+ spin_unlock_irq(&cwq->pool->lock);
if (drained)
continue;
@@ -2848,25 +2833,23 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
{
struct worker *worker = NULL;
struct worker_pool *pool;
- struct global_cwq *gcwq;
struct cpu_workqueue_struct *cwq;
might_sleep();
pool = get_work_pool(work);
if (!pool)
return false;
- gcwq = pool->gcwq;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (!list_empty(&work->entry)) {
/*
* See the comment near try_to_grab_pending()->smp_rmb().
- * If it was re-queued to a different gcwq under us, we
+ * If it was re-queued to a different pool under us, we
* are not going to wait.
*/
smp_rmb();
cwq = get_work_cwq(work);
- if (unlikely(!cwq || gcwq != cwq->pool->gcwq))
+ if (unlikely(!cwq || pool != cwq->pool))
goto already_gone;
} else {
worker = find_worker_executing_work(pool, work);
@@ -2876,7 +2859,7 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
}
insert_wq_barrier(cwq, barr, work, worker);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
/*
* If @max_active is 1 or rescuer is in use, flushing another work
@@ -2892,7 +2875,7 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
return true;
already_gone:
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
return false;
}
@@ -3413,7 +3396,7 @@ EXPORT_SYMBOL_GPL(destroy_workqueue);
* increased.
*
* CONTEXT:
- * spin_lock_irq(gcwq->lock).
+ * spin_lock_irq(pool->lock).
*/
static void cwq_set_max_active(struct cpu_workqueue_struct *cwq, int max_active)
{
@@ -3447,15 +3430,14 @@ void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
for_each_cwq_cpu(cpu, wq) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
struct worker_pool *pool = cwq->pool;
- struct global_cwq *gcwq = pool->gcwq;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
if (!(wq->flags & WQ_FREEZABLE) ||
!(pool->flags & POOL_FREEZING))
cwq_set_max_active(cwq, max_active);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
spin_unlock(&workqueue_lock);
@@ -3498,22 +3480,20 @@ EXPORT_SYMBOL_GPL(workqueue_congested);
unsigned int work_busy(struct work_struct *work)
{
struct worker_pool *pool = get_work_pool(work);
- struct global_cwq *gcwq;
unsigned long flags;
unsigned int ret = 0;
if (!pool)
return 0;
- gcwq = pool->gcwq;
- spin_lock_irqsave(&gcwq->lock, flags);
+ spin_lock_irqsave(&pool->lock, flags);
if (work_pending(work))
ret |= WORK_BUSY_PENDING;
if (find_worker_executing_work(pool, work))
ret |= WORK_BUSY_RUNNING;
- spin_unlock_irqrestore(&gcwq->lock, flags);
+ spin_unlock_irqrestore(&pool->lock, flags);
return ret;
}
@@ -3541,7 +3521,10 @@ static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq)
for_each_worker_pool(pool, gcwq)
mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools);
- spin_lock_irq(&gcwq->lock);
+
+ local_irq_disable();
+ for_each_worker_pool(pool, gcwq)
+ spin_lock_nested(&pool->lock, pool - gcwq->pools);
}
/* release manager positions */
@@ -3549,7 +3532,10 @@ static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq)
{
struct worker_pool *pool;
- spin_unlock_irq(&gcwq->lock);
+ for_each_worker_pool(pool, gcwq)
+ spin_unlock(&pool->lock);
+ local_irq_enable();
+
for_each_worker_pool(pool, gcwq)
mutex_unlock(&pool->assoc_mutex);
}
@@ -3630,9 +3616,9 @@ static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb,
if (!worker)
return NOTIFY_BAD;
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
start_worker(worker);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
break;
@@ -3718,7 +3704,7 @@ EXPORT_SYMBOL_GPL(work_on_cpu);
* gcwq->worklist.
*
* CONTEXT:
- * Grabs and releases workqueue_lock and gcwq->lock's.
+ * Grabs and releases workqueue_lock and pool->lock's.
*/
void freeze_workqueues_begin(void)
{
@@ -3734,9 +3720,11 @@ void freeze_workqueues_begin(void)
struct worker_pool *pool;
struct workqueue_struct *wq;
- spin_lock_irq(&gcwq->lock);
+ local_irq_disable();
for_each_worker_pool(pool, gcwq) {
+ spin_lock_nested(&pool->lock, pool - gcwq->pools);
+
WARN_ON_ONCE(pool->flags & POOL_FREEZING);
pool->flags |= POOL_FREEZING;
}
@@ -3748,7 +3736,9 @@ void freeze_workqueues_begin(void)
cwq->max_active = 0;
}
- spin_unlock_irq(&gcwq->lock);
+ for_each_worker_pool(pool, gcwq)
+ spin_unlock(&pool->lock);
+ local_irq_enable();
}
spin_unlock(&workqueue_lock);
@@ -3807,7 +3797,7 @@ out_unlock:
* frozen works are transferred to their respective gcwq worklists.
*
* CONTEXT:
- * Grabs and releases workqueue_lock and gcwq->lock's.
+ * Grabs and releases workqueue_lock and pool->lock's.
*/
void thaw_workqueues(void)
{
@@ -3823,9 +3813,11 @@ void thaw_workqueues(void)
struct worker_pool *pool;
struct workqueue_struct *wq;
- spin_lock_irq(&gcwq->lock);
+ local_irq_disable();
for_each_worker_pool(pool, gcwq) {
+ spin_lock_nested(&pool->lock, pool - gcwq->pools);
+
WARN_ON_ONCE(!(pool->flags & POOL_FREEZING));
pool->flags &= ~POOL_FREEZING;
}
@@ -3840,10 +3832,11 @@ void thaw_workqueues(void)
cwq_set_max_active(cwq, wq->saved_max_active);
}
- for_each_worker_pool(pool, gcwq)
+ for_each_worker_pool(pool, gcwq) {
wake_up_worker(pool);
-
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock(&pool->lock);
+ }
+ local_irq_enable();
}
workqueue_freezing = false;
@@ -3868,10 +3861,9 @@ static int __init init_workqueues(void)
struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
- spin_lock_init(&gcwq->lock);
-
for_each_worker_pool(pool, gcwq) {
pool->gcwq = gcwq;
+ spin_lock_init(&pool->lock);
pool->cpu = cpu;
pool->flags |= POOL_DISASSOCIATED;
INIT_LIST_HEAD(&pool->worklist);
@@ -3906,9 +3898,9 @@ static int __init init_workqueues(void)
worker = create_worker(pool);
BUG_ON(!worker);
- spin_lock_irq(&gcwq->lock);
+ spin_lock_irq(&pool->lock);
start_worker(worker);
- spin_unlock_irq(&gcwq->lock);
+ spin_unlock_irq(&pool->lock);
}
}
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 11/17] workqueue: make hotplug processing per-pool
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (9 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 10/17] workqueue: move global_cwq->lock " Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 12/17] workqueue: make freezing/thawing per-pool Tejun Heo
` (10 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Instead of holding locks from both pools and then processing the pools
together, make hotplug processing per-pool - grab locks of one pool,
process it, release it and then proceed to the next pool.
rebind_workers() is updated to take and process @pool instead of @gcwq
which results in a lot of de-indentation. gcwq_claim_assoc_and_lock()
and its counterpart are replaced with in-line per-pool locking.
While this patch changes processing order across pools, order within
each pool remains the same. As each pool is independent, this
shouldn't break anything.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 149 ++++++++++++++++++++++-------------------------------
1 file changed, 62 insertions(+), 87 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 755d2c4..92e4d99 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -1689,10 +1689,10 @@ static void busy_worker_rebind_fn(struct work_struct *work)
}
/**
- * rebind_workers - rebind all workers of a gcwq to the associated CPU
- * @gcwq: gcwq of interest
+ * rebind_workers - rebind all workers of a pool to the associated CPU
+ * @pool: pool of interest
*
- * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding
+ * @pool->cpu is coming online. Rebind all workers to the CPU. Rebinding
* is different for idle and busy ones.
*
* Idle ones will be removed from the idle_list and woken up. They will
@@ -1710,60 +1710,53 @@ static void busy_worker_rebind_fn(struct work_struct *work)
* including the manager will not appear on @idle_list until rebind is
* complete, making local wake-ups safe.
*/
-static void rebind_workers(struct global_cwq *gcwq)
+static void rebind_workers(struct worker_pool *pool)
{
- struct worker_pool *pool;
struct worker *worker, *n;
struct hlist_node *pos;
int i;
- for_each_worker_pool(pool, gcwq) {
- lockdep_assert_held(&pool->assoc_mutex);
- lockdep_assert_held(&pool->lock);
- }
+ lockdep_assert_held(&pool->assoc_mutex);
+ lockdep_assert_held(&pool->lock);
/* dequeue and kick idle ones */
- for_each_worker_pool(pool, gcwq) {
- list_for_each_entry_safe(worker, n, &pool->idle_list, entry) {
- /*
- * idle workers should be off @pool->idle_list
- * until rebind is complete to avoid receiving
- * premature local wake-ups.
- */
- list_del_init(&worker->entry);
+ list_for_each_entry_safe(worker, n, &pool->idle_list, entry) {
+ /*
+ * idle workers should be off @pool->idle_list until rebind
+ * is complete to avoid receiving premature local wake-ups.
+ */
+ list_del_init(&worker->entry);
- /*
- * worker_thread() will see the above dequeuing
- * and call idle_worker_rebind().
- */
- wake_up_process(worker->task);
- }
+ /*
+ * worker_thread() will see the above dequeuing and call
+ * idle_worker_rebind().
+ */
+ wake_up_process(worker->task);
+ }
- /* rebind busy workers */
- for_each_busy_worker(worker, i, pos, pool) {
- struct work_struct *rebind_work = &worker->rebind_work;
- struct workqueue_struct *wq;
+ /* rebind busy workers */
+ for_each_busy_worker(worker, i, pos, pool) {
+ struct work_struct *rebind_work = &worker->rebind_work;
+ struct workqueue_struct *wq;
- if (test_and_set_bit(WORK_STRUCT_PENDING_BIT,
- work_data_bits(rebind_work)))
- continue;
+ if (test_and_set_bit(WORK_STRUCT_PENDING_BIT,
+ work_data_bits(rebind_work)))
+ continue;
- debug_work_activate(rebind_work);
+ debug_work_activate(rebind_work);
- /*
- * wq doesn't really matter but let's keep
- * @worker->pool and @cwq->pool consistent for
- * sanity.
- */
- if (std_worker_pool_pri(worker->pool))
- wq = system_highpri_wq;
- else
- wq = system_wq;
-
- insert_work(get_cwq(pool->cpu, wq), rebind_work,
- worker->scheduled.next,
- work_color_to_flags(WORK_NO_COLOR));
- }
+ /*
+ * wq doesn't really matter but let's keep @worker->pool
+ * and @cwq->pool consistent for sanity.
+ */
+ if (std_worker_pool_pri(worker->pool))
+ wq = system_highpri_wq;
+ else
+ wq = system_wq;
+
+ insert_work(get_cwq(pool->cpu, wq), rebind_work,
+ worker->scheduled.next,
+ work_color_to_flags(WORK_NO_COLOR));
}
}
@@ -3506,7 +3499,7 @@ EXPORT_SYMBOL_GPL(work_busy);
* are a lot of assumptions on strong associations among work, cwq and
* gcwq which make migrating pending and scheduled works very
* difficult to implement without impacting hot paths. Secondly,
- * gcwqs serve mix of short, long and very long running works making
+ * worker pools serve mix of short, long and very long running works making
* blocked draining impractical.
*
* This is solved by allowing the pools to be disassociated from the CPU
@@ -3514,32 +3507,6 @@ EXPORT_SYMBOL_GPL(work_busy);
* cpu comes back online.
*/
-/* claim manager positions of all pools */
-static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq)
-{
- struct worker_pool *pool;
-
- for_each_worker_pool(pool, gcwq)
- mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools);
-
- local_irq_disable();
- for_each_worker_pool(pool, gcwq)
- spin_lock_nested(&pool->lock, pool - gcwq->pools);
-}
-
-/* release manager positions */
-static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq)
-{
- struct worker_pool *pool;
-
- for_each_worker_pool(pool, gcwq)
- spin_unlock(&pool->lock);
- local_irq_enable();
-
- for_each_worker_pool(pool, gcwq)
- mutex_unlock(&pool->assoc_mutex);
-}
-
static void gcwq_unbind_fn(struct work_struct *work)
{
struct global_cwq *gcwq = get_gcwq(smp_processor_id());
@@ -3548,17 +3515,19 @@ static void gcwq_unbind_fn(struct work_struct *work)
struct hlist_node *pos;
int i;
- BUG_ON(gcwq->pools[0].cpu != smp_processor_id());
+ for_each_worker_pool(pool, gcwq) {
+ BUG_ON(pool->cpu != smp_processor_id());
- gcwq_claim_assoc_and_lock(gcwq);
+ mutex_lock(&pool->assoc_mutex);
+ spin_lock_irq(&pool->lock);
- /*
- * We've claimed all manager positions. Make all workers unbound
- * and set DISASSOCIATED. Before this, all workers except for the
- * ones which are still executing works from before the last CPU
- * down must be on the cpu. After this, they may become diasporas.
- */
- for_each_worker_pool(pool, gcwq) {
+ /*
+ * We've claimed all manager positions. Make all workers
+ * unbound and set DISASSOCIATED. Before this, all workers
+ * except for the ones which are still executing works from
+ * before the last CPU down must be on the cpu. After
+ * this, they may become diasporas.
+ */
list_for_each_entry(worker, &pool->idle_list, entry)
worker->flags |= WORKER_UNBOUND;
@@ -3566,9 +3535,10 @@ static void gcwq_unbind_fn(struct work_struct *work)
worker->flags |= WORKER_UNBOUND;
pool->flags |= POOL_DISASSOCIATED;
- }
- gcwq_release_assoc_and_unlock(gcwq);
+ spin_unlock_irq(&pool->lock);
+ mutex_unlock(&pool->assoc_mutex);
+ }
/*
* Call schedule() so that we cross rq->lock and thus can guarantee
@@ -3624,11 +3594,16 @@ static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb,
case CPU_DOWN_FAILED:
case CPU_ONLINE:
- gcwq_claim_assoc_and_lock(gcwq);
- for_each_worker_pool(pool, gcwq)
+ for_each_worker_pool(pool, gcwq) {
+ mutex_lock(&pool->assoc_mutex);
+ spin_lock_irq(&pool->lock);
+
pool->flags &= ~POOL_DISASSOCIATED;
- rebind_workers(gcwq);
- gcwq_release_assoc_and_unlock(gcwq);
+ rebind_workers(pool);
+
+ spin_unlock_irq(&pool->lock);
+ mutex_unlock(&pool->assoc_mutex);
+ }
break;
}
return NOTIFY_OK;
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 12/17] workqueue: make freezing/thawing per-pool
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (10 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 11/17] workqueue: make hotplug processing per-pool Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 13/17] workqueue: replace for_each_worker_pool() with for_each_std_worker_pool() Tejun Heo
` (9 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Instead of holding locks from both pools and then processing the pools
together, make freezing/thwaing per-pool - grab locks of one pool,
process it, release it and then proceed to the next pool.
While this patch changes processing order across pools, order within
each pool remains the same. As each pool is independent, this
shouldn't break anything.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 46 ++++++++++++++++++++--------------------------
1 file changed, 20 insertions(+), 26 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 92e4d99..766ca67 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -3695,25 +3695,22 @@ void freeze_workqueues_begin(void)
struct worker_pool *pool;
struct workqueue_struct *wq;
- local_irq_disable();
-
for_each_worker_pool(pool, gcwq) {
- spin_lock_nested(&pool->lock, pool - gcwq->pools);
+ spin_lock_irq(&pool->lock);
WARN_ON_ONCE(pool->flags & POOL_FREEZING);
pool->flags |= POOL_FREEZING;
- }
- list_for_each_entry(wq, &workqueues, list) {
- struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+ list_for_each_entry(wq, &workqueues, list) {
+ struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
- if (cwq && wq->flags & WQ_FREEZABLE)
- cwq->max_active = 0;
- }
+ if (cwq && cwq->pool == pool &&
+ (wq->flags & WQ_FREEZABLE))
+ cwq->max_active = 0;
+ }
- for_each_worker_pool(pool, gcwq)
- spin_unlock(&pool->lock);
- local_irq_enable();
+ spin_unlock_irq(&pool->lock);
+ }
}
spin_unlock(&workqueue_lock);
@@ -3788,30 +3785,27 @@ void thaw_workqueues(void)
struct worker_pool *pool;
struct workqueue_struct *wq;
- local_irq_disable();
-
for_each_worker_pool(pool, gcwq) {
- spin_lock_nested(&pool->lock, pool - gcwq->pools);
+ spin_lock_irq(&pool->lock);
WARN_ON_ONCE(!(pool->flags & POOL_FREEZING));
pool->flags &= ~POOL_FREEZING;
- }
- list_for_each_entry(wq, &workqueues, list) {
- struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+ list_for_each_entry(wq, &workqueues, list) {
+ struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
- if (!cwq || !(wq->flags & WQ_FREEZABLE))
- continue;
+ if (!cwq || cwq->pool != pool ||
+ !(wq->flags & WQ_FREEZABLE))
+ continue;
- /* restore max_active and repopulate worklist */
- cwq_set_max_active(cwq, wq->saved_max_active);
- }
+ /* restore max_active and repopulate worklist */
+ cwq_set_max_active(cwq, wq->saved_max_active);
+ }
- for_each_worker_pool(pool, gcwq) {
wake_up_worker(pool);
- spin_unlock(&pool->lock);
+
+ spin_unlock_irq(&pool->lock);
}
- local_irq_enable();
}
workqueue_freezing = false;
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 13/17] workqueue: replace for_each_worker_pool() with for_each_std_worker_pool()
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (11 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 12/17] workqueue: make freezing/thawing per-pool Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 14/17] workqueue: remove worker_pool->gcwq Tejun Heo
` (8 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
for_each_std_worker_pool() takes @cpu instead of @gcwq.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 39 +++++++++++++++++----------------------
1 file changed, 17 insertions(+), 22 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 766ca67..81f041f 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -277,9 +277,9 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
#define CREATE_TRACE_POINTS
#include <trace/events/workqueue.h>
-#define for_each_worker_pool(pool, gcwq) \
- for ((pool) = &(gcwq)->pools[0]; \
- (pool) < &(gcwq)->pools[NR_STD_WORKER_POOLS]; (pool)++)
+#define for_each_std_worker_pool(pool, cpu) \
+ for ((pool) = &get_gcwq((cpu))->pools[0]; \
+ (pool) < &get_gcwq((cpu))->pools[NR_STD_WORKER_POOLS]; (pool)++)
#define for_each_busy_worker(worker, i, pos, pool) \
hash_for_each(pool->busy_hash, i, pos, worker, hentry)
@@ -3509,14 +3509,14 @@ EXPORT_SYMBOL_GPL(work_busy);
static void gcwq_unbind_fn(struct work_struct *work)
{
- struct global_cwq *gcwq = get_gcwq(smp_processor_id());
+ int cpu = smp_processor_id();
struct worker_pool *pool;
struct worker *worker;
struct hlist_node *pos;
int i;
- for_each_worker_pool(pool, gcwq) {
- BUG_ON(pool->cpu != smp_processor_id());
+ for_each_std_worker_pool(pool, cpu) {
+ BUG_ON(cpu != smp_processor_id());
mutex_lock(&pool->assoc_mutex);
spin_lock_irq(&pool->lock);
@@ -3550,15 +3550,15 @@ static void gcwq_unbind_fn(struct work_struct *work)
/*
* Sched callbacks are disabled now. Zap nr_running. After this,
* nr_running stays zero and need_more_worker() and keep_working()
- * are always true as long as the worklist is not empty. @gcwq now
- * behaves as unbound (in terms of concurrency management) gcwq
- * which is served by workers tied to the CPU.
+ * are always true as long as the worklist is not empty. Pools on
+ * @cpu now behave as unbound (in terms of concurrency management)
+ * pools which are served by workers tied to the CPU.
*
* On return from this function, the current worker would trigger
* unbound chain execution of pending work items if other workers
* didn't already.
*/
- for_each_worker_pool(pool, gcwq)
+ for_each_std_worker_pool(pool, cpu)
atomic_set(get_pool_nr_running(pool), 0);
}
@@ -3571,12 +3571,11 @@ static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb,
void *hcpu)
{
unsigned int cpu = (unsigned long)hcpu;
- struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
switch (action & ~CPU_TASKS_FROZEN) {
case CPU_UP_PREPARE:
- for_each_worker_pool(pool, gcwq) {
+ for_each_std_worker_pool(pool, cpu) {
struct worker *worker;
if (pool->nr_workers)
@@ -3594,7 +3593,7 @@ static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb,
case CPU_DOWN_FAILED:
case CPU_ONLINE:
- for_each_worker_pool(pool, gcwq) {
+ for_each_std_worker_pool(pool, cpu) {
mutex_lock(&pool->assoc_mutex);
spin_lock_irq(&pool->lock);
@@ -3691,11 +3690,10 @@ void freeze_workqueues_begin(void)
workqueue_freezing = true;
for_each_gcwq_cpu(cpu) {
- struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
struct workqueue_struct *wq;
- for_each_worker_pool(pool, gcwq) {
+ for_each_std_worker_pool(pool, cpu) {
spin_lock_irq(&pool->lock);
WARN_ON_ONCE(pool->flags & POOL_FREEZING);
@@ -3781,11 +3779,10 @@ void thaw_workqueues(void)
goto out_unlock;
for_each_gcwq_cpu(cpu) {
- struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
struct workqueue_struct *wq;
- for_each_worker_pool(pool, gcwq) {
+ for_each_std_worker_pool(pool, cpu) {
spin_lock_irq(&pool->lock);
WARN_ON_ONCE(!(pool->flags & POOL_FREEZING));
@@ -3827,11 +3824,10 @@ static int __init init_workqueues(void)
/* initialize gcwqs */
for_each_gcwq_cpu(cpu) {
- struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
- for_each_worker_pool(pool, gcwq) {
- pool->gcwq = gcwq;
+ for_each_std_worker_pool(pool, cpu) {
+ pool->gcwq = get_gcwq(cpu);
spin_lock_init(&pool->lock);
pool->cpu = cpu;
pool->flags |= POOL_DISASSOCIATED;
@@ -3856,10 +3852,9 @@ static int __init init_workqueues(void)
/* create the initial worker */
for_each_online_gcwq_cpu(cpu) {
- struct global_cwq *gcwq = get_gcwq(cpu);
struct worker_pool *pool;
- for_each_worker_pool(pool, gcwq) {
+ for_each_std_worker_pool(pool, cpu) {
struct worker *worker;
if (cpu != WORK_CPU_UNBOUND)
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 14/17] workqueue: remove worker_pool->gcwq
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (12 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 13/17] workqueue: replace for_each_worker_pool() with for_each_std_worker_pool() Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 15/17] workqueue: remove global_cwq Tejun Heo
` (7 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
The only remaining user of pool->gcwq is std_worker_pool_pri().
Reimplement it using get_gcwq() and remove worker_pool->gcwq.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 12 +++++-------
1 file changed, 5 insertions(+), 7 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 81f041f..d37db53 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -150,7 +150,6 @@ struct worker {
};
struct worker_pool {
- struct global_cwq *gcwq; /* I: the owning gcwq */
spinlock_t lock; /* the pool lock */
unsigned int cpu; /* I: the associated cpu */
int id; /* I: pool ID */
@@ -478,11 +477,6 @@ static DEFINE_IDR(worker_pool_idr);
static int worker_thread(void *__worker);
-static int std_worker_pool_pri(struct worker_pool *pool)
-{
- return pool - pool->gcwq->pools;
-}
-
static struct global_cwq *get_gcwq(unsigned int cpu)
{
if (cpu != WORK_CPU_UNBOUND)
@@ -491,6 +485,11 @@ static struct global_cwq *get_gcwq(unsigned int cpu)
return &unbound_global_cwq;
}
+static int std_worker_pool_pri(struct worker_pool *pool)
+{
+ return pool - get_gcwq(pool->cpu)->pools;
+}
+
/* allocate ID and assign it to @pool */
static int worker_pool_assign_id(struct worker_pool *pool)
{
@@ -3827,7 +3826,6 @@ static int __init init_workqueues(void)
struct worker_pool *pool;
for_each_std_worker_pool(pool, cpu) {
- pool->gcwq = get_gcwq(cpu);
spin_lock_init(&pool->lock);
pool->cpu = cpu;
pool->flags |= POOL_DISASSOCIATED;
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 15/17] workqueue: remove global_cwq
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (13 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 14/17] workqueue: remove worker_pool->gcwq Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-22 6:50 ` Joonsoo Kim
2013-01-23 18:09 ` [PATCH v2 " Tejun Heo
2013-01-17 1:42 ` [PATCH 16/17] workqueue: rename nr_running variables Tejun Heo
` (6 subsequent siblings)
21 siblings, 2 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
global_cwq is now nothing but a container for per-pcu standard
worker_pools. Declare the worker pools directly as
cpu/unbound_std_worker_pools[] and remove global_cwq.
* get_gcwq() is replaced with std_worker_pools() which returns the
pointer to the standard pool array for a given CPU.
* __alloc_workqueue_key() updated to use get_std_worker_pool() instead
of open-coding pool determination.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 47 +++++++++++++++++------------------------------
1 file changed, 17 insertions(+), 30 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index d37db53..4bddf52 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -120,7 +120,6 @@ enum {
* W: workqueue_lock protected.
*/
-struct global_cwq;
struct worker_pool;
/*
@@ -174,16 +173,6 @@ struct worker_pool {
};
/*
- * Global per-cpu workqueue. There's one and only one for each cpu
- * and all works are queued and processed here regardless of their
- * target workqueues.
- */
-struct global_cwq {
- struct worker_pool pools[NR_STD_WORKER_POOLS];
- /* normal and highpri pools */
-} ____cacheline_aligned_in_smp;
-
-/*
* The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of
* work_struct->data are used for flags and thus cwqs need to be
* aligned at two's power of the number of flag bits.
@@ -277,8 +266,8 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
#include <trace/events/workqueue.h>
#define for_each_std_worker_pool(pool, cpu) \
- for ((pool) = &get_gcwq((cpu))->pools[0]; \
- (pool) < &get_gcwq((cpu))->pools[NR_STD_WORKER_POOLS]; (pool)++)
+ for ((pool) = &std_worker_pools(cpu)[0]; \
+ (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++)
#define for_each_busy_worker(worker, i, pos, pool) \
hash_for_each(pool->busy_hash, i, pos, worker, hentry)
@@ -454,19 +443,19 @@ static LIST_HEAD(workqueues);
static bool workqueue_freezing; /* W: have wqs started freezing? */
/*
- * The almighty global cpu workqueues. nr_running is the only field
- * which is expected to be used frequently by other cpus via
- * try_to_wake_up(). Put it in a separate cacheline.
+ * The CPU standard worker pools. nr_running is the only field which is
+ * expected to be used frequently by other cpus via try_to_wake_up(). Put
+ * it in a separate cacheline.
*/
-static DEFINE_PER_CPU(struct global_cwq, global_cwq);
+static DEFINE_PER_CPU_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
+ cpu_std_worker_pools);
static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
/*
- * Global cpu workqueue and nr_running counter for unbound gcwq. The pools
- * for online CPUs have POOL_DISASSOCIATED set, and all their workers have
- * WORKER_UNBOUND set.
+ * Standard worker pools and nr_running counter for unbound CPU. The pools
+ * have POOL_DISASSOCIATED set, and all workers have WORKER_UNBOUND set.
*/
-static struct global_cwq unbound_global_cwq;
+static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS];
static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
[0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
};
@@ -477,17 +466,17 @@ static DEFINE_IDR(worker_pool_idr);
static int worker_thread(void *__worker);
-static struct global_cwq *get_gcwq(unsigned int cpu)
+static struct worker_pool *std_worker_pools(int cpu)
{
if (cpu != WORK_CPU_UNBOUND)
- return &per_cpu(global_cwq, cpu);
+ return per_cpu(cpu_std_worker_pools, cpu);
else
- return &unbound_global_cwq;
+ return unbound_std_worker_pools;
}
static int std_worker_pool_pri(struct worker_pool *pool)
{
- return pool - get_gcwq(pool->cpu)->pools;
+ return pool - std_worker_pools(pool->cpu);
}
/* allocate ID and assign it to @pool */
@@ -514,9 +503,9 @@ static struct worker_pool *worker_pool_by_id(int pool_id)
static struct worker_pool *get_std_worker_pool(int cpu, bool highpri)
{
- struct global_cwq *gcwq = get_gcwq(cpu);
+ struct worker_pool *pools = std_worker_pools(cpu);
- return &gcwq->pools[highpri];
+ return &pools[highpri];
}
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
@@ -3279,11 +3268,9 @@ struct workqueue_struct *__alloc_workqueue_key(const char *fmt,
for_each_cwq_cpu(cpu, wq) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
- struct global_cwq *gcwq = get_gcwq(cpu);
- int pool_idx = (bool)(flags & WQ_HIGHPRI);
BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK);
- cwq->pool = &gcwq->pools[pool_idx];
+ cwq->pool = get_std_worker_pool(cpu, flags & WQ_HIGHPRI);
cwq->wq = wq;
cwq->flush_color = -1;
cwq->max_active = max_active;
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* Re: [PATCH 15/17] workqueue: remove global_cwq
2013-01-17 1:42 ` [PATCH 15/17] workqueue: remove global_cwq Tejun Heo
@ 2013-01-22 6:50 ` Joonsoo Kim
2013-01-23 1:09 ` Tejun Heo
2013-01-23 18:09 ` [PATCH v2 " Tejun Heo
1 sibling, 1 reply; 31+ messages in thread
From: Joonsoo Kim @ 2013-01-22 6:50 UTC (permalink / raw)
To: Tejun Heo; +Cc: linux-kernel, Lai Jiangshan
On Wed, Jan 16, 2013 at 05:42:47PM -0800, Tejun Heo wrote:
> global_cwq is now nothing but a container for per-pcu standard
s/per-pcu/per-cpu/
> worker_pools. Declare the worker pools directly as
> cpu/unbound_std_worker_pools[] and remove global_cwq.
>
> * get_gcwq() is replaced with std_worker_pools() which returns the
> pointer to the standard pool array for a given CPU.
>
> * __alloc_workqueue_key() updated to use get_std_worker_pool() instead
> of open-coding pool determination.
>
> This is part of an effort to remove global_cwq and make worker_pool
> the top level abstraction, which in turn will help implementing worker
> pools with user-specified attributes.
>
> Signed-off-by: Tejun Heo <tj@kernel.org>
> ---
> kernel/workqueue.c | 47 +++++++++++++++++------------------------------
> 1 file changed, 17 insertions(+), 30 deletions(-)
>
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index d37db53..4bddf52 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -120,7 +120,6 @@ enum {
> * W: workqueue_lock protected.
> */
>
> -struct global_cwq;
> struct worker_pool;
>
> /*
> @@ -174,16 +173,6 @@ struct worker_pool {
> };
>
> /*
> - * Global per-cpu workqueue. There's one and only one for each cpu
> - * and all works are queued and processed here regardless of their
> - * target workqueues.
> - */
> -struct global_cwq {
> - struct worker_pool pools[NR_STD_WORKER_POOLS];
> - /* normal and highpri pools */
> -} ____cacheline_aligned_in_smp;
> -
> -/*
> * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of
> * work_struct->data are used for flags and thus cwqs need to be
> * aligned at two's power of the number of flag bits.
> @@ -277,8 +266,8 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
> #include <trace/events/workqueue.h>
>
> #define for_each_std_worker_pool(pool, cpu) \
> - for ((pool) = &get_gcwq((cpu))->pools[0]; \
> - (pool) < &get_gcwq((cpu))->pools[NR_STD_WORKER_POOLS]; (pool)++)
> + for ((pool) = &std_worker_pools(cpu)[0]; \
> + (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++)
>
> #define for_each_busy_worker(worker, i, pos, pool) \
> hash_for_each(pool->busy_hash, i, pos, worker, hentry)
> @@ -454,19 +443,19 @@ static LIST_HEAD(workqueues);
> static bool workqueue_freezing; /* W: have wqs started freezing? */
>
> /*
> - * The almighty global cpu workqueues. nr_running is the only field
> - * which is expected to be used frequently by other cpus via
> - * try_to_wake_up(). Put it in a separate cacheline.
> + * The CPU standard worker pools. nr_running is the only field which is
> + * expected to be used frequently by other cpus via try_to_wake_up(). Put
> + * it in a separate cacheline.
> */
> -static DEFINE_PER_CPU(struct global_cwq, global_cwq);
> +static DEFINE_PER_CPU_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
> + cpu_std_worker_pools);
> static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
Why worker_pool is defined as DEFINE_PER_CPU_ALIGNED?
And this makes only worker_pool[0] aligned with cacheline.
worker_pool[1] is not aligned with cacheline.
Now, we have a spin_lock for each instance of worker_pool and
each one is independent instance.
So, IMHO, it is better to align worker_pool[1] with cacheline.
Thanks.
> /*
> - * Global cpu workqueue and nr_running counter for unbound gcwq. The pools
> - * for online CPUs have POOL_DISASSOCIATED set, and all their workers have
> - * WORKER_UNBOUND set.
> + * Standard worker pools and nr_running counter for unbound CPU. The pools
> + * have POOL_DISASSOCIATED set, and all workers have WORKER_UNBOUND set.
> */
> -static struct global_cwq unbound_global_cwq;
> +static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS];
> static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
> [0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
> };
> @@ -477,17 +466,17 @@ static DEFINE_IDR(worker_pool_idr);
>
> static int worker_thread(void *__worker);
>
> -static struct global_cwq *get_gcwq(unsigned int cpu)
> +static struct worker_pool *std_worker_pools(int cpu)
> {
> if (cpu != WORK_CPU_UNBOUND)
> - return &per_cpu(global_cwq, cpu);
> + return per_cpu(cpu_std_worker_pools, cpu);
> else
> - return &unbound_global_cwq;
> + return unbound_std_worker_pools;
> }
>
> static int std_worker_pool_pri(struct worker_pool *pool)
> {
> - return pool - get_gcwq(pool->cpu)->pools;
> + return pool - std_worker_pools(pool->cpu);
> }
>
> /* allocate ID and assign it to @pool */
> @@ -514,9 +503,9 @@ static struct worker_pool *worker_pool_by_id(int pool_id)
>
> static struct worker_pool *get_std_worker_pool(int cpu, bool highpri)
> {
> - struct global_cwq *gcwq = get_gcwq(cpu);
> + struct worker_pool *pools = std_worker_pools(cpu);
>
> - return &gcwq->pools[highpri];
> + return &pools[highpri];
> }
>
> static atomic_t *get_pool_nr_running(struct worker_pool *pool)
> @@ -3279,11 +3268,9 @@ struct workqueue_struct *__alloc_workqueue_key(const char *fmt,
>
> for_each_cwq_cpu(cpu, wq) {
> struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
> - struct global_cwq *gcwq = get_gcwq(cpu);
> - int pool_idx = (bool)(flags & WQ_HIGHPRI);
>
> BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK);
> - cwq->pool = &gcwq->pools[pool_idx];
> + cwq->pool = get_std_worker_pool(cpu, flags & WQ_HIGHPRI);
> cwq->wq = wq;
> cwq->flush_color = -1;
> cwq->max_active = max_active;
> --
> 1.8.0.2
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCH 15/17] workqueue: remove global_cwq
2013-01-22 6:50 ` Joonsoo Kim
@ 2013-01-23 1:09 ` Tejun Heo
0 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-23 1:09 UTC (permalink / raw)
To: Joonsoo Kim; +Cc: linux-kernel, Lai Jiangshan
On Tue, Jan 22, 2013 at 03:50:39PM +0900, Joonsoo Kim wrote:
> > -static DEFINE_PER_CPU(struct global_cwq, global_cwq);
> > +static DEFINE_PER_CPU_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
> > + cpu_std_worker_pools);
> > static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
>
> Why worker_pool is defined as DEFINE_PER_CPU_ALIGNED?
>
> And this makes only worker_pool[0] aligned with cacheline.
> worker_pool[1] is not aligned with cacheline.
> Now, we have a spin_lock for each instance of worker_pool and
> each one is independent instance.
> So, IMHO, it is better to align worker_pool[1] with cacheline.
Right, it's just carried over from the time when there was single pool
implemented as global_gcwq. We probably should define struct
worker_pool as cache aligned.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread
* [PATCH v2 15/17] workqueue: remove global_cwq
2013-01-17 1:42 ` [PATCH 15/17] workqueue: remove global_cwq Tejun Heo
2013-01-22 6:50 ` Joonsoo Kim
@ 2013-01-23 18:09 ` Tejun Heo
2013-01-24 9:29 ` Joonsoo Kim
1 sibling, 1 reply; 31+ messages in thread
From: Tejun Heo @ 2013-01-23 18:09 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan
global_cwq is now nothing but a container for per-cpu standard
worker_pools. Declare the worker pools directly as
cpu/unbound_std_worker_pools[] and remove global_cwq.
* ____cacheline_aligned_in_smp moved from global_cwq to worker_pool.
This probably would have made sense even before this change as we
want each pool to be aligned.
* get_gcwq() is replaced with std_worker_pools() which returns the
pointer to the standard pool array for a given CPU.
* __alloc_workqueue_key() updated to use get_std_worker_pool() instead
of open-coding pool determination.
This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.
v2: Joonsoo pointed out that it'd better to align struct worker_pool
rather than the array so that every pool is aligned.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Joonsoo Kim <js1304@gmail.com>
---
Rebased on top of the current wq/for-3.9 and Joonsoo's comments
applied.
Thanks.
kernel/workqueue.c | 46 ++++++++++++++++----------------------------
kernel/workqueue_internal.h | 1
2 files changed, 17 insertions(+), 30 deletions(-)
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -144,16 +144,6 @@ struct worker_pool {
struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */
struct ida worker_ida; /* L: for worker IDs */
-};
-
-/*
- * Global per-cpu workqueue. There's one and only one for each cpu
- * and all works are queued and processed here regardless of their
- * target workqueues.
- */
-struct global_cwq {
- struct worker_pool pools[NR_STD_WORKER_POOLS];
- /* normal and highpri pools */
} ____cacheline_aligned_in_smp;
/*
@@ -250,8 +240,8 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
#include <trace/events/workqueue.h>
#define for_each_std_worker_pool(pool, cpu) \
- for ((pool) = &get_gcwq((cpu))->pools[0]; \
- (pool) < &get_gcwq((cpu))->pools[NR_STD_WORKER_POOLS]; (pool)++)
+ for ((pool) = &std_worker_pools(cpu)[0]; \
+ (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++)
#define for_each_busy_worker(worker, i, pos, pool) \
hash_for_each(pool->busy_hash, i, pos, worker, hentry)
@@ -427,19 +417,19 @@ static LIST_HEAD(workqueues);
static bool workqueue_freezing; /* W: have wqs started freezing? */
/*
- * The almighty global cpu workqueues. nr_running is the only field
- * which is expected to be used frequently by other cpus via
- * try_to_wake_up(). Put it in a separate cacheline.
+ * The CPU standard worker pools. nr_running is the only field which is
+ * expected to be used frequently by other cpus via try_to_wake_up(). Put
+ * it in a separate cacheline.
*/
-static DEFINE_PER_CPU(struct global_cwq, global_cwq);
+static DEFINE_PER_CPU(struct worker_pool [NR_STD_WORKER_POOLS],
+ cpu_std_worker_pools);
static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
/*
- * Global cpu workqueue and nr_running counter for unbound gcwq. The pools
- * for online CPUs have POOL_DISASSOCIATED set, and all their workers have
- * WORKER_UNBOUND set.
+ * Standard worker pools and nr_running counter for unbound CPU. The pools
+ * have POOL_DISASSOCIATED set, and all workers have WORKER_UNBOUND set.
*/
-static struct global_cwq unbound_global_cwq;
+static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS];
static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
[0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
};
@@ -450,17 +440,17 @@ static DEFINE_IDR(worker_pool_idr);
static int worker_thread(void *__worker);
-static struct global_cwq *get_gcwq(unsigned int cpu)
+static struct worker_pool *std_worker_pools(int cpu)
{
if (cpu != WORK_CPU_UNBOUND)
- return &per_cpu(global_cwq, cpu);
+ return per_cpu(cpu_std_worker_pools, cpu);
else
- return &unbound_global_cwq;
+ return unbound_std_worker_pools;
}
static int std_worker_pool_pri(struct worker_pool *pool)
{
- return pool - get_gcwq(pool->cpu)->pools;
+ return pool - std_worker_pools(pool->cpu);
}
/* allocate ID and assign it to @pool */
@@ -487,9 +477,9 @@ static struct worker_pool *worker_pool_b
static struct worker_pool *get_std_worker_pool(int cpu, bool highpri)
{
- struct global_cwq *gcwq = get_gcwq(cpu);
+ struct worker_pool *pools = std_worker_pools(cpu);
- return &gcwq->pools[highpri];
+ return &pools[highpri];
}
static atomic_t *get_pool_nr_running(struct worker_pool *pool)
@@ -3269,11 +3259,9 @@ struct workqueue_struct *__alloc_workque
for_each_cwq_cpu(cpu, wq) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
- struct global_cwq *gcwq = get_gcwq(cpu);
- int pool_idx = (bool)(flags & WQ_HIGHPRI);
BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK);
- cwq->pool = &gcwq->pools[pool_idx];
+ cwq->pool = get_std_worker_pool(cpu, flags & WQ_HIGHPRI);
cwq->wq = wq;
cwq->flush_color = -1;
cwq->max_active = max_active;
--- a/kernel/workqueue_internal.h
+++ b/kernel/workqueue_internal.h
@@ -10,7 +10,6 @@
#include <linux/workqueue.h>
#include <linux/kthread.h>
-struct global_cwq;
struct worker_pool;
/*
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCH v2 15/17] workqueue: remove global_cwq
2013-01-23 18:09 ` [PATCH v2 " Tejun Heo
@ 2013-01-24 9:29 ` Joonsoo Kim
2013-01-24 18:44 ` Tejun Heo
0 siblings, 1 reply; 31+ messages in thread
From: Joonsoo Kim @ 2013-01-24 9:29 UTC (permalink / raw)
To: Tejun Heo; +Cc: linux-kernel, Lai Jiangshan
Hello, Tejun.
On Wed, Jan 23, 2013 at 10:09:57AM -0800, Tejun Heo wrote:
> global_cwq is now nothing but a container for per-cpu standard
> worker_pools. Declare the worker pools directly as
> cpu/unbound_std_worker_pools[] and remove global_cwq.
>
> * ____cacheline_aligned_in_smp moved from global_cwq to worker_pool.
> This probably would have made sense even before this change as we
> want each pool to be aligned.
>
> * get_gcwq() is replaced with std_worker_pools() which returns the
> pointer to the standard pool array for a given CPU.
>
> * __alloc_workqueue_key() updated to use get_std_worker_pool() instead
> of open-coding pool determination.
>
> This is part of an effort to remove global_cwq and make worker_pool
> the top level abstraction, which in turn will help implementing worker
> pools with user-specified attributes.
>
> v2: Joonsoo pointed out that it'd better to align struct worker_pool
> rather than the array so that every pool is aligned.
>
> Signed-off-by: Tejun Heo <tj@kernel.org>
> Cc: Joonsoo Kim <js1304@gmail.com>
> ---
> Rebased on top of the current wq/for-3.9 and Joonsoo's comments
> applied.
>
> Thanks.
>
> kernel/workqueue.c | 46 ++++++++++++++++----------------------------
> kernel/workqueue_internal.h | 1
> 2 files changed, 17 insertions(+), 30 deletions(-)
>
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -144,16 +144,6 @@ struct worker_pool {
>
> struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */
> struct ida worker_ida; /* L: for worker IDs */
> -};
> -
> -/*
> - * Global per-cpu workqueue. There's one and only one for each cpu
> - * and all works are queued and processed here regardless of their
> - * target workqueues.
> - */
> -struct global_cwq {
> - struct worker_pool pools[NR_STD_WORKER_POOLS];
> - /* normal and highpri pools */
> } ____cacheline_aligned_in_smp;
>
> /*
> @@ -250,8 +240,8 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
> #include <trace/events/workqueue.h>
>
> #define for_each_std_worker_pool(pool, cpu) \
> - for ((pool) = &get_gcwq((cpu))->pools[0]; \
> - (pool) < &get_gcwq((cpu))->pools[NR_STD_WORKER_POOLS]; (pool)++)
> + for ((pool) = &std_worker_pools(cpu)[0]; \
> + (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++)
>
> #define for_each_busy_worker(worker, i, pos, pool) \
> hash_for_each(pool->busy_hash, i, pos, worker, hentry)
> @@ -427,19 +417,19 @@ static LIST_HEAD(workqueues);
> static bool workqueue_freezing; /* W: have wqs started freezing? */
>
> /*
> - * The almighty global cpu workqueues. nr_running is the only field
> - * which is expected to be used frequently by other cpus via
> - * try_to_wake_up(). Put it in a separate cacheline.
> + * The CPU standard worker pools. nr_running is the only field which is
> + * expected to be used frequently by other cpus via try_to_wake_up(). Put
> + * it in a separate cacheline.
> */
> -static DEFINE_PER_CPU(struct global_cwq, global_cwq);
> +static DEFINE_PER_CPU(struct worker_pool [NR_STD_WORKER_POOLS],
> + cpu_std_worker_pools);
> static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
AFAIK, worker_pool can be accessed by other cpus.
So, I think that we also need "DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [], xxx)".
> /*
> - * Global cpu workqueue and nr_running counter for unbound gcwq. The pools
> - * for online CPUs have POOL_DISASSOCIATED set, and all their workers have
> - * WORKER_UNBOUND set.
> + * Standard worker pools and nr_running counter for unbound CPU. The pools
> + * have POOL_DISASSOCIATED set, and all workers have WORKER_UNBOUND set.
> */
> -static struct global_cwq unbound_global_cwq;
> +static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS];
> static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
> [0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
> };
> @@ -450,17 +440,17 @@ static DEFINE_IDR(worker_pool_idr);
>
> static int worker_thread(void *__worker);
>
> -static struct global_cwq *get_gcwq(unsigned int cpu)
> +static struct worker_pool *std_worker_pools(int cpu)
> {
> if (cpu != WORK_CPU_UNBOUND)
> - return &per_cpu(global_cwq, cpu);
> + return per_cpu(cpu_std_worker_pools, cpu);
> else
> - return &unbound_global_cwq;
> + return unbound_std_worker_pools;
> }
>
> static int std_worker_pool_pri(struct worker_pool *pool)
> {
> - return pool - get_gcwq(pool->cpu)->pools;
> + return pool - std_worker_pools(pool->cpu);
> }
>
> /* allocate ID and assign it to @pool */
> @@ -487,9 +477,9 @@ static struct worker_pool *worker_pool_b
>
> static struct worker_pool *get_std_worker_pool(int cpu, bool highpri)
> {
> - struct global_cwq *gcwq = get_gcwq(cpu);
> + struct worker_pool *pools = std_worker_pools(cpu);
>
> - return &gcwq->pools[highpri];
> + return &pools[highpri];
> }
>
> static atomic_t *get_pool_nr_running(struct worker_pool *pool)
> @@ -3269,11 +3259,9 @@ struct workqueue_struct *__alloc_workque
>
> for_each_cwq_cpu(cpu, wq) {
> struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
> - struct global_cwq *gcwq = get_gcwq(cpu);
> - int pool_idx = (bool)(flags & WQ_HIGHPRI);
>
> BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK);
> - cwq->pool = &gcwq->pools[pool_idx];
> + cwq->pool = get_std_worker_pool(cpu, flags & WQ_HIGHPRI);
> cwq->wq = wq;
> cwq->flush_color = -1;
> cwq->max_active = max_active;
> --- a/kernel/workqueue_internal.h
> +++ b/kernel/workqueue_internal.h
> @@ -10,7 +10,6 @@
> #include <linux/workqueue.h>
> #include <linux/kthread.h>
>
> -struct global_cwq;
> struct worker_pool;
>
> /*
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCH v2 15/17] workqueue: remove global_cwq
2013-01-24 9:29 ` Joonsoo Kim
@ 2013-01-24 18:44 ` Tejun Heo
0 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-24 18:44 UTC (permalink / raw)
To: Joonsoo Kim; +Cc: linux-kernel, Lai Jiangshan
Hey, Joonsoo.
On Thu, Jan 24, 2013 at 06:29:36PM +0900, Joonsoo Kim wrote:
> > -static DEFINE_PER_CPU(struct global_cwq, global_cwq);
> > +static DEFINE_PER_CPU(struct worker_pool [NR_STD_WORKER_POOLS],
> > + cpu_std_worker_pools);
> > static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
>
> AFAIK, worker_pool can be accessed by other cpus.
> So, I think that we also need "DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [], xxx)".
Hmmm... yean. I think I'll just move pool_nr_running into
worker_pool. Will prep a separate patch.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread
* [PATCH 16/17] workqueue: rename nr_running variables
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (14 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 15/17] workqueue: remove global_cwq Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:42 ` [PATCH 17/17] workqueue: post global_cwq removal cleanups Tejun Heo
` (5 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Rename per-cpu and unbound nr_running variables such that they match
the pool variables.
This patch doesn't introduce any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 4bddf52..52ca95b 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -449,14 +449,15 @@ static bool workqueue_freezing; /* W: have wqs started freezing? */
*/
static DEFINE_PER_CPU_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
cpu_std_worker_pools);
-static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]);
+static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t [NR_STD_WORKER_POOLS],
+ cpu_std_pool_nr_running);
/*
* Standard worker pools and nr_running counter for unbound CPU. The pools
* have POOL_DISASSOCIATED set, and all workers have WORKER_UNBOUND set.
*/
static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS];
-static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = {
+static atomic_t unbound_std_pool_nr_running[NR_STD_WORKER_POOLS] = {
[0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */
};
@@ -514,9 +515,9 @@ static atomic_t *get_pool_nr_running(struct worker_pool *pool)
int idx = std_worker_pool_pri(pool);
if (cpu != WORK_CPU_UNBOUND)
- return &per_cpu(pool_nr_running, cpu)[idx];
+ return &per_cpu(cpu_std_pool_nr_running, cpu)[idx];
else
- return &unbound_pool_nr_running[idx];
+ return &unbound_std_pool_nr_running[idx];
}
static struct cpu_workqueue_struct *get_cwq(unsigned int cpu,
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* [PATCH 17/17] workqueue: post global_cwq removal cleanups
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (15 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 16/17] workqueue: rename nr_running variables Tejun Heo
@ 2013-01-17 1:42 ` Tejun Heo
2013-01-17 1:48 ` [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (4 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:42 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Tejun Heo
Remove remaining references to gcwq.
* __next_gcwq_cpu() steals __next_wq_cpu() name. The original
__next_wq_cpu() became __next_cwq_cpu().
* s/for_each_gcwq_cpu/for_each_wq_cpu/
s/for_each_online_gcwq_cpu/for_each_online_wq_cpu/
* s/gcwq_mayday_timeout/pool_mayday_timeout/
* s/gcwq_unbind_fn/wq_unbind_fn/
* Drop references to gcwq in comments.
This patch doesn't introduce any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
---
kernel/workqueue.c | 104 ++++++++++++++++++++++++++---------------------------
1 file changed, 52 insertions(+), 52 deletions(-)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 52ca95b..7507609 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -272,8 +272,8 @@ EXPORT_SYMBOL_GPL(system_freezable_wq);
#define for_each_busy_worker(worker, i, pos, pool) \
hash_for_each(pool->busy_hash, i, pos, worker, hentry)
-static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask,
- unsigned int sw)
+static inline int __next_wq_cpu(int cpu, const struct cpumask *mask,
+ unsigned int sw)
{
if (cpu < nr_cpu_ids) {
if (sw & 1) {
@@ -287,39 +287,39 @@ static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask,
return WORK_CPU_NONE;
}
-static inline int __next_wq_cpu(int cpu, const struct cpumask *mask,
- struct workqueue_struct *wq)
+static inline int __next_cwq_cpu(int cpu, const struct cpumask *mask,
+ struct workqueue_struct *wq)
{
- return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2);
+ return __next_wq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2);
}
/*
* CPU iterators
*
- * An extra gcwq is defined for an invalid cpu number
+ * An extra cpu number is defined using an invalid cpu number
* (WORK_CPU_UNBOUND) to host workqueues which are not bound to any
- * specific CPU. The following iterators are similar to
- * for_each_*_cpu() iterators but also considers the unbound gcwq.
+ * specific CPU. The following iterators are similar to for_each_*_cpu()
+ * iterators but also considers the unbound CPU.
*
- * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND
- * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND
+ * for_each_wq_cpu() : possible CPUs + WORK_CPU_UNBOUND
+ * for_each_online_wq_cpu() : online CPUs + WORK_CPU_UNBOUND
* for_each_cwq_cpu() : possible CPUs for bound workqueues,
* WORK_CPU_UNBOUND for unbound workqueues
*/
-#define for_each_gcwq_cpu(cpu) \
- for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \
+#define for_each_wq_cpu(cpu) \
+ for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, 3); \
(cpu) < WORK_CPU_NONE; \
- (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3))
+ (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, 3))
-#define for_each_online_gcwq_cpu(cpu) \
- for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \
+#define for_each_online_wq_cpu(cpu) \
+ for ((cpu) = __next_wq_cpu(-1, cpu_online_mask, 3); \
(cpu) < WORK_CPU_NONE; \
- (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3))
+ (cpu) = __next_wq_cpu((cpu), cpu_online_mask, 3))
#define for_each_cwq_cpu(cpu, wq) \
- for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \
+ for ((cpu) = __next_cwq_cpu(-1, cpu_possible_mask, (wq)); \
(cpu) < WORK_CPU_NONE; \
- (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq)))
+ (cpu) = __next_cwq_cpu((cpu), cpu_possible_mask, (wq)))
#ifdef CONFIG_DEBUG_OBJECTS_WORK
@@ -681,7 +681,7 @@ static bool __need_more_worker(struct worker_pool *pool)
* running workers.
*
* Note that, because unbound workers never contribute to nr_running, this
- * function will always return %true for unbound gcwq as long as the
+ * function will always return %true for unbound pools as long as the
* worklist isn't empty.
*/
static bool need_more_worker(struct worker_pool *pool)
@@ -1147,14 +1147,14 @@ fail:
}
/**
- * insert_work - insert a work into gcwq
+ * insert_work - insert a work into a pool
* @cwq: cwq @work belongs to
* @work: work to insert
* @head: insertion point
* @extra_flags: extra WORK_STRUCT_* flags to set
*
- * Insert @work which belongs to @cwq into @gcwq after @head.
- * @extra_flags is or'd to work_struct flags.
+ * Insert @work which belongs to @cwq after @head. @extra_flags is or'd to
+ * work_struct flags.
*
* CONTEXT:
* spin_lock_irq(pool->lock).
@@ -1197,7 +1197,7 @@ static bool is_chained_work(struct workqueue_struct *wq)
unsigned long flags;
unsigned int cpu;
- for_each_gcwq_cpu(cpu) {
+ for_each_wq_cpu(cpu) {
struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
struct worker_pool *pool = cwq->pool;
struct worker *worker;
@@ -1551,7 +1551,7 @@ static void worker_enter_idle(struct worker *worker)
mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT);
/*
- * Sanity check nr_running. Because gcwq_unbind_fn() releases
+ * Sanity check nr_running. Because wq_unbind_fn() releases
* pool->lock between setting %WORKER_UNBOUND and zapping
* nr_running, the warning may trigger spuriously. Check iff
* unbind is not in progress.
@@ -1581,7 +1581,7 @@ static void worker_leave_idle(struct worker *worker)
}
/**
- * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq
+ * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock pool
* @worker: self
*
* Works which are scheduled while the cpu is online must at least be
@@ -1593,10 +1593,10 @@ static void worker_leave_idle(struct worker *worker)
* themselves to the target cpu and may race with cpu going down or
* coming online. kthread_bind() can't be used because it may put the
* worker to already dead cpu and set_cpus_allowed_ptr() can't be used
- * verbatim as it's best effort and blocking and gcwq may be
+ * verbatim as it's best effort and blocking and pool may be
* [dis]associated in the meantime.
*
- * This function tries set_cpus_allowed() and locks gcwq and verifies the
+ * This function tries set_cpus_allowed() and locks pool and verifies the
* binding against %POOL_DISASSOCIATED which is set during
* %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker
* enters idle state or fetches works without dropping lock, it can
@@ -1607,7 +1607,7 @@ static void worker_leave_idle(struct worker *worker)
* held.
*
* RETURNS:
- * %true if the associated gcwq is online (@worker is successfully
+ * %true if the associated pool is online (@worker is successfully
* bound), %false if offline.
*/
static bool worker_maybe_bind_and_lock(struct worker *worker)
@@ -1844,7 +1844,7 @@ fail:
* start_worker - start a newly created worker
* @worker: worker to start
*
- * Make the gcwq aware of @worker and start it.
+ * Make the pool aware of @worker and start it.
*
* CONTEXT:
* spin_lock_irq(pool->lock).
@@ -1861,7 +1861,7 @@ static void start_worker(struct worker *worker)
* destroy_worker - destroy a workqueue worker
* @worker: worker to be destroyed
*
- * Destroy @worker and adjust @gcwq stats accordingly.
+ * Destroy @worker and adjust @pool stats accordingly.
*
* CONTEXT:
* spin_lock_irq(pool->lock) which is released and regrabbed.
@@ -1937,7 +1937,7 @@ static bool send_mayday(struct work_struct *work)
return true;
}
-static void gcwq_mayday_timeout(unsigned long __pool)
+static void pool_mayday_timeout(unsigned long __pool)
{
struct worker_pool *pool = (void *)__pool;
struct work_struct *work;
@@ -2065,9 +2065,9 @@ static bool maybe_destroy_workers(struct worker_pool *pool)
* manage_workers - manage worker pool
* @worker: self
*
- * Assume the manager role and manage gcwq worker pool @worker belongs
+ * Assume the manager role and manage the worker pool @worker belongs
* to. At any given time, there can be only zero or one manager per
- * gcwq. The exclusion is handled automatically by this function.
+ * pool. The exclusion is handled automatically by this function.
*
* The caller can safely start processing works on false return. On
* true return, it's guaranteed that need_to_create_worker() is false
@@ -2110,11 +2110,11 @@ static bool manage_workers(struct worker *worker)
* CPU hotplug could have happened while we were waiting
* for assoc_mutex. Hotplug itself can't handle us
* because manager isn't either on idle or busy list, and
- * @gcwq's state and ours could have deviated.
+ * @pool's state and ours could have deviated.
*
* As hotplug is now excluded via assoc_mutex, we can
* simply try to bind. It will succeed or fail depending
- * on @gcwq's current state. Try it and adjust
+ * on @pool's current state. Try it and adjust
* %WORKER_UNBOUND accordingly.
*/
if (worker_maybe_bind_and_lock(worker))
@@ -2289,8 +2289,8 @@ static void process_scheduled_works(struct worker *worker)
* worker_thread - the worker thread function
* @__worker: self
*
- * The gcwq worker thread function. There's a single dynamic pool of
- * these per each cpu. These workers process all works regardless of
+ * The worker thread function. There are NR_CPU_WORKER_POOLS dynamic pools
+ * of these per each cpu. These workers process all works regardless of
* their specific target workqueue. The only exception is works which
* belong to workqueues with a rescuer which will be explained in
* rescuer_thread().
@@ -2386,14 +2386,14 @@ sleep:
* Workqueue rescuer thread function. There's one rescuer for each
* workqueue which has WQ_RESCUER set.
*
- * Regular work processing on a gcwq may block trying to create a new
+ * Regular work processing on a pool may block trying to create a new
* worker which uses GFP_KERNEL allocation which has slight chance of
* developing into deadlock if some works currently on the same queue
* need to be processed to satisfy the GFP_KERNEL allocation. This is
* the problem rescuer solves.
*
- * When such condition is possible, the gcwq summons rescuers of all
- * workqueues which have works queued on the gcwq and let them process
+ * When such condition is possible, the pool summons rescuers of all
+ * workqueues which have works queued on the pool and let them process
* those works so that forward progress can be guaranteed.
*
* This should happen rarely.
@@ -3484,7 +3484,7 @@ EXPORT_SYMBOL_GPL(work_busy);
*
* There are two challenges in supporting CPU hotplug. Firstly, there
* are a lot of assumptions on strong associations among work, cwq and
- * gcwq which make migrating pending and scheduled works very
+ * pool which make migrating pending and scheduled works very
* difficult to implement without impacting hot paths. Secondly,
* worker pools serve mix of short, long and very long running works making
* blocked draining impractical.
@@ -3494,7 +3494,7 @@ EXPORT_SYMBOL_GPL(work_busy);
* cpu comes back online.
*/
-static void gcwq_unbind_fn(struct work_struct *work)
+static void wq_unbind_fn(struct work_struct *work)
{
int cpu = smp_processor_id();
struct worker_pool *pool;
@@ -3609,7 +3609,7 @@ static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb,
switch (action & ~CPU_TASKS_FROZEN) {
case CPU_DOWN_PREPARE:
/* unbinding should happen on the local CPU */
- INIT_WORK_ONSTACK(&unbind_work, gcwq_unbind_fn);
+ INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn);
queue_work_on(cpu, system_highpri_wq, &unbind_work);
flush_work(&unbind_work);
break;
@@ -3662,7 +3662,7 @@ EXPORT_SYMBOL_GPL(work_on_cpu);
*
* Start freezing workqueues. After this function returns, all freezable
* workqueues will queue new works to their frozen_works list instead of
- * gcwq->worklist.
+ * pool->worklist.
*
* CONTEXT:
* Grabs and releases workqueue_lock and pool->lock's.
@@ -3676,7 +3676,7 @@ void freeze_workqueues_begin(void)
BUG_ON(workqueue_freezing);
workqueue_freezing = true;
- for_each_gcwq_cpu(cpu) {
+ for_each_wq_cpu(cpu) {
struct worker_pool *pool;
struct workqueue_struct *wq;
@@ -3723,7 +3723,7 @@ bool freeze_workqueues_busy(void)
BUG_ON(!workqueue_freezing);
- for_each_gcwq_cpu(cpu) {
+ for_each_wq_cpu(cpu) {
struct workqueue_struct *wq;
/*
* nr_active is monotonically decreasing. It's safe
@@ -3751,7 +3751,7 @@ out_unlock:
* thaw_workqueues - thaw workqueues
*
* Thaw workqueues. Normal queueing is restored and all collected
- * frozen works are transferred to their respective gcwq worklists.
+ * frozen works are transferred to their respective pool worklists.
*
* CONTEXT:
* Grabs and releases workqueue_lock and pool->lock's.
@@ -3765,7 +3765,7 @@ void thaw_workqueues(void)
if (!workqueue_freezing)
goto out_unlock;
- for_each_gcwq_cpu(cpu) {
+ for_each_wq_cpu(cpu) {
struct worker_pool *pool;
struct workqueue_struct *wq;
@@ -3809,8 +3809,8 @@ static int __init init_workqueues(void)
cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP);
hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN);
- /* initialize gcwqs */
- for_each_gcwq_cpu(cpu) {
+ /* initialize CPU pools */
+ for_each_wq_cpu(cpu) {
struct worker_pool *pool;
for_each_std_worker_pool(pool, cpu) {
@@ -3825,7 +3825,7 @@ static int __init init_workqueues(void)
pool->idle_timer.function = idle_worker_timeout;
pool->idle_timer.data = (unsigned long)pool;
- setup_timer(&pool->mayday_timer, gcwq_mayday_timeout,
+ setup_timer(&pool->mayday_timer, pool_mayday_timeout,
(unsigned long)pool);
mutex_init(&pool->assoc_mutex);
@@ -3837,7 +3837,7 @@ static int __init init_workqueues(void)
}
/* create the initial worker */
- for_each_online_gcwq_cpu(cpu) {
+ for_each_online_wq_cpu(cpu) {
struct worker_pool *pool;
for_each_std_worker_pool(pool, cpu) {
--
1.8.0.2
^ permalink raw reply related [flat|nested] 31+ messages in thread* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (16 preceding siblings ...)
2013-01-17 1:42 ` [PATCH 17/17] workqueue: post global_cwq removal cleanups Tejun Heo
@ 2013-01-17 1:48 ` Tejun Heo
2013-01-17 3:25 ` Wanlong Gao
` (3 subsequent siblings)
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 1:48 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Jens Axboe, jmoyer
Forgot to cc Jens and Jeff.
Jens, Jeff, the patchset itself probably won't interest you guys too
much but it's part of effort towards worker pool w/ custom attributes.
I'm working toward a design where the custom pools are integral part
of workqueue which share all the interface and semantics, which
probably is a bit more involved than implementing something on the
side but I think the end result will be much better this way.
I'm aiming 3.9 merge window but that could be too ambitious. Will
keep you guys in the loop.
Thanks.
On Wed, Jan 16, 2013 at 05:42:32PM -0800, Tejun Heo wrote:
> Hello,
>
> Currently, on the backend side, there are two layers of abstraction.
> For each CPU and the special unbound wq-specific CPU, there's one
> global_cwq. gcwq in turn hosts two worker_pools - one for normal
> priority, the other for highpri - each of which actually serves the
> work items.
>
> worker_pool is the later addition to support separate pool of workers
> for highpri workqueues. Stuff was moved to worker_pool on as-needed
> basis and, as a result, the two pools belonging to the same CPU share
> some stuff in the gcwq - most notably the lock and the hash table for
> work items currently being executed.
>
> It seems like we'll need to support worker pools with custom
> attributes, which is planned to be implemented as extra worker_pools
> for the unbound CPU. Removing gcwq and having worker_pool as the top
> level abstraction makes things much simpler for such designs. Also,
> there's scalability benefit to not sharing locking and busy hash among
> different worker pools as worker pools w/ custom attributes are likely
> to have widely different memory / cpu locality characteristics.
>
> In retrospect, it might have been less churn if we just converted to
> have multiple gcwqs per CPU when we were adding highpri pool support.
> Oh well, such is life and the name worker_pool fits the role much
> better anyway at this point.
>
> This patchset moves the remaining stuff in gcwq to worker_pool and
> then removes gcwq entirely making worker_pool the top level and the
> only backend abstraction. In the process, this patchset also prepares
> for later addition of worker_pools with custom attributes.
>
> This patchset shouldn't introduce any visible differences outside of
> workqueue proper and contains the following 17 patches.
>
> 0001-workqueue-unexport-work_cpu.patch
> 0002-workqueue-use-std_-prefix-for-the-standard-per-cpu-p.patch
> 0003-workqueue-make-GCWQ_DISASSOCIATED-a-pool-flag.patch
> 0004-workqueue-make-GCWQ_FREEZING-a-pool-flag.patch
> 0005-workqueue-introduce-WORK_OFFQ_CPU_NONE.patch
> 0006-workqueue-add-worker_pool-id.patch
> 0007-workqueue-record-pool-ID-instead-of-CPU-in-work-data.patch
> 0008-workqueue-move-busy_hash-from-global_cwq-to-worker_p.patch
> 0009-workqueue-move-global_cwq-cpu-to-worker_pool.patch
> 0010-workqueue-move-global_cwq-lock-to-worker_pool.patch
> 0011-workqueue-make-hotplug-processing-per-pool.patch
> 0012-workqueue-make-freezing-thawing-per-pool.patch
> 0013-workqueue-replace-for_each_worker_pool-with-for_each.patch
> 0014-workqueue-remove-worker_pool-gcwq.patch
> 0015-workqueue-remove-global_cwq.patch
> 0016-workqueue-rename-nr_running-variables.patch
> 0017-workqueue-post-global_cwq-removal-cleanups.patch
>
> 0001-0002 are misc preps.
>
> 0003-0004 move flags from gcwq to pool.
>
> 0005-0007 make work->data off-queue backlink point to worker_pools
> instead of CPUs, which is necessary to move busy_hash to pool.
>
> 0008-0010 move busy_hash, cpu and locking to pool.
>
> 0011-0014 make operations per-pool and remove gcwq usages.
>
> 0015-0017 remove gcwq and cleanup afterwards.
>
> This patchset is on top of wq/for-3.9 023f27d3d6f ("workqueue: fix
> find_worker_executing_work() brekage from hashtable conversion") and
> available in the following git branch.
>
> git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git for-3.9-remove-gcwq
>
> Thanks.
>
> include/linux/workqueue.h | 17
> include/trace/events/workqueue.h | 2
> kernel/workqueue.c | 897 +++++++++++++++++++--------------------
> 3 files changed, 461 insertions(+), 455 deletions(-)
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (17 preceding siblings ...)
2013-01-17 1:48 ` [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
@ 2013-01-17 3:25 ` Wanlong Gao
2013-01-17 19:11 ` Tejun Heo
2013-01-22 5:37 ` Joonsoo Kim
` (2 subsequent siblings)
21 siblings, 1 reply; 31+ messages in thread
From: Wanlong Gao @ 2013-01-17 3:25 UTC (permalink / raw)
To: Tejun Heo; +Cc: linux-kernel, Lai Jiangshan
On 01/17/2013 09:42 AM, Tejun Heo wrote:
> Hello,
>
> Currently, on the backend side, there are two layers of abstraction.
> For each CPU and the special unbound wq-specific CPU, there's one
> global_cwq. gcwq in turn hosts two worker_pools - one for normal
> priority, the other for highpri - each of which actually serves the
> work items.
>
> worker_pool is the later addition to support separate pool of workers
> for highpri workqueues. Stuff was moved to worker_pool on as-needed
> basis and, as a result, the two pools belonging to the same CPU share
> some stuff in the gcwq - most notably the lock and the hash table for
> work items currently being executed.
>
> It seems like we'll need to support worker pools with custom
> attributes, which is planned to be implemented as extra worker_pools
> for the unbound CPU. Removing gcwq and having worker_pool as the top
> level abstraction makes things much simpler for such designs. Also,
> there's scalability benefit to not sharing locking and busy hash among
> different worker pools as worker pools w/ custom attributes are likely
> to have widely different memory / cpu locality characteristics.
>
> In retrospect, it might have been less churn if we just converted to
> have multiple gcwqs per CPU when we were adding highpri pool support.
> Oh well, such is life and the name worker_pool fits the role much
> better anyway at this point.
>
> This patchset moves the remaining stuff in gcwq to worker_pool and
> then removes gcwq entirely making worker_pool the top level and the
> only backend abstraction. In the process, this patchset also prepares
> for later addition of worker_pools with custom attributes.
>
> This patchset shouldn't introduce any visible differences outside of
> workqueue proper and contains the following 17 patches.
>
> 0001-workqueue-unexport-work_cpu.patch
> 0002-workqueue-use-std_-prefix-for-the-standard-per-cpu-p.patch
> 0003-workqueue-make-GCWQ_DISASSOCIATED-a-pool-flag.patch
> 0004-workqueue-make-GCWQ_FREEZING-a-pool-flag.patch
> 0005-workqueue-introduce-WORK_OFFQ_CPU_NONE.patch
> 0006-workqueue-add-worker_pool-id.patch
> 0007-workqueue-record-pool-ID-instead-of-CPU-in-work-data.patch
> 0008-workqueue-move-busy_hash-from-global_cwq-to-worker_p.patch
> 0009-workqueue-move-global_cwq-cpu-to-worker_pool.patch
> 0010-workqueue-move-global_cwq-lock-to-worker_pool.patch
> 0011-workqueue-make-hotplug-processing-per-pool.patch
> 0012-workqueue-make-freezing-thawing-per-pool.patch
> 0013-workqueue-replace-for_each_worker_pool-with-for_each.patch
> 0014-workqueue-remove-worker_pool-gcwq.patch
> 0015-workqueue-remove-global_cwq.patch
> 0016-workqueue-rename-nr_running-variables.patch
> 0017-workqueue-post-global_cwq-removal-cleanups.patch
>
> 0001-0002 are misc preps.
>
> 0003-0004 move flags from gcwq to pool.
>
> 0005-0007 make work->data off-queue backlink point to worker_pools
> instead of CPUs, which is necessary to move busy_hash to pool.
>
> 0008-0010 move busy_hash, cpu and locking to pool.
>
> 0011-0014 make operations per-pool and remove gcwq usages.
>
> 0015-0017 remove gcwq and cleanup afterwards.
>
> This patchset is on top of wq/for-3.9 023f27d3d6f ("workqueue: fix
> find_worker_executing_work() brekage from hashtable conversion") and
> available in the following git branch.
>
> git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git for-3.9-remove-gcwq
Forgot to push out this branch?
Thanks,
Wanlong Gao
>
> Thanks.
>
> include/linux/workqueue.h | 17
> include/trace/events/workqueue.h | 2
> kernel/workqueue.c | 897 +++++++++++++++++++--------------------
> 3 files changed, 461 insertions(+), 455 deletions(-)
>
> --
> tejun
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-17 3:25 ` Wanlong Gao
@ 2013-01-17 19:11 ` Tejun Heo
0 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-17 19:11 UTC (permalink / raw)
To: Wanlong Gao; +Cc: linux-kernel, Lai Jiangshan
On Thu, Jan 17, 2013 at 11:25:28AM +0800, Wanlong Gao wrote:
> > git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git for-3.9-remove-gcwq
>
> Forgot to push out this branch?
Yeap, I did. Pushed out last night from phone so should be there now.
Thanks!
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread
* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (18 preceding siblings ...)
2013-01-17 3:25 ` Wanlong Gao
@ 2013-01-22 5:37 ` Joonsoo Kim
2013-01-23 1:07 ` Tejun Heo
2013-01-24 13:36 ` Lai Jiangshan
2013-01-24 19:03 ` Tejun Heo
21 siblings, 1 reply; 31+ messages in thread
From: Joonsoo Kim @ 2013-01-22 5:37 UTC (permalink / raw)
To: Tejun Heo; +Cc: linux-kernel, Lai Jiangshan
Hello, Tejun.
On Wed, Jan 16, 2013 at 05:42:32PM -0800, Tejun Heo wrote:
> Hello,
>
> Currently, on the backend side, there are two layers of abstraction.
> For each CPU and the special unbound wq-specific CPU, there's one
> global_cwq. gcwq in turn hosts two worker_pools - one for normal
> priority, the other for highpri - each of which actually serves the
> work items.
>
> worker_pool is the later addition to support separate pool of workers
> for highpri workqueues. Stuff was moved to worker_pool on as-needed
> basis and, as a result, the two pools belonging to the same CPU share
> some stuff in the gcwq - most notably the lock and the hash table for
> work items currently being executed.
>
> It seems like we'll need to support worker pools with custom
> attributes, which is planned to be implemented as extra worker_pools
> for the unbound CPU. Removing gcwq and having worker_pool as the top
> level abstraction makes things much simpler for such designs. Also,
> there's scalability benefit to not sharing locking and busy hash among
> different worker pools as worker pools w/ custom attributes are likely
> to have widely different memory / cpu locality characteristics.
Could you tell me why extra worker_pools with custom attributes are needed?
Or could you give a reference link for this?
Thanks.
> In retrospect, it might have been less churn if we just converted to
> have multiple gcwqs per CPU when we were adding highpri pool support.
> Oh well, such is life and the name worker_pool fits the role much
> better anyway at this point.
>
> This patchset moves the remaining stuff in gcwq to worker_pool and
> then removes gcwq entirely making worker_pool the top level and the
> only backend abstraction. In the process, this patchset also prepares
> for later addition of worker_pools with custom attributes.
>
> This patchset shouldn't introduce any visible differences outside of
> workqueue proper and contains the following 17 patches.
>
> 0001-workqueue-unexport-work_cpu.patch
> 0002-workqueue-use-std_-prefix-for-the-standard-per-cpu-p.patch
> 0003-workqueue-make-GCWQ_DISASSOCIATED-a-pool-flag.patch
> 0004-workqueue-make-GCWQ_FREEZING-a-pool-flag.patch
> 0005-workqueue-introduce-WORK_OFFQ_CPU_NONE.patch
> 0006-workqueue-add-worker_pool-id.patch
> 0007-workqueue-record-pool-ID-instead-of-CPU-in-work-data.patch
> 0008-workqueue-move-busy_hash-from-global_cwq-to-worker_p.patch
> 0009-workqueue-move-global_cwq-cpu-to-worker_pool.patch
> 0010-workqueue-move-global_cwq-lock-to-worker_pool.patch
> 0011-workqueue-make-hotplug-processing-per-pool.patch
> 0012-workqueue-make-freezing-thawing-per-pool.patch
> 0013-workqueue-replace-for_each_worker_pool-with-for_each.patch
> 0014-workqueue-remove-worker_pool-gcwq.patch
> 0015-workqueue-remove-global_cwq.patch
> 0016-workqueue-rename-nr_running-variables.patch
> 0017-workqueue-post-global_cwq-removal-cleanups.patch
>
> 0001-0002 are misc preps.
>
> 0003-0004 move flags from gcwq to pool.
>
> 0005-0007 make work->data off-queue backlink point to worker_pools
> instead of CPUs, which is necessary to move busy_hash to pool.
>
> 0008-0010 move busy_hash, cpu and locking to pool.
>
> 0011-0014 make operations per-pool and remove gcwq usages.
>
> 0015-0017 remove gcwq and cleanup afterwards.
>
> This patchset is on top of wq/for-3.9 023f27d3d6f ("workqueue: fix
> find_worker_executing_work() brekage from hashtable conversion") and
> available in the following git branch.
>
> git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git for-3.9-remove-gcwq
>
> Thanks.
>
> include/linux/workqueue.h | 17
> include/trace/events/workqueue.h | 2
> kernel/workqueue.c | 897 +++++++++++++++++++--------------------
> 3 files changed, 461 insertions(+), 455 deletions(-)
>
> --
> tejun
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-22 5:37 ` Joonsoo Kim
@ 2013-01-23 1:07 ` Tejun Heo
0 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-23 1:07 UTC (permalink / raw)
To: Joonsoo Kim; +Cc: linux-kernel, Lai Jiangshan
Hey.
On Tue, Jan 22, 2013 at 02:37:02PM +0900, Joonsoo Kim wrote:
> > It seems like we'll need to support worker pools with custom
> > attributes, which is planned to be implemented as extra worker_pools
> > for the unbound CPU. Removing gcwq and having worker_pool as the top
> > level abstraction makes things much simpler for such designs. Also,
> > there's scalability benefit to not sharing locking and busy hash among
> > different worker pools as worker pools w/ custom attributes are likely
> > to have widely different memory / cpu locality characteristics.
>
> Could you tell me why extra worker_pools with custom attributes are needed?
> Or could you give a reference link for this?
Currently, there are two expected users - writeback and crypto. The
former currently implements its own worker pool and the latter is
using per-cpu workqueue but not particularly happy with it. Being
bound to the issuing CPU seems a bit too limiting.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread
* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (19 preceding siblings ...)
2013-01-22 5:37 ` Joonsoo Kim
@ 2013-01-24 13:36 ` Lai Jiangshan
2013-01-24 18:51 ` Tejun Heo
2013-01-24 19:03 ` Tejun Heo
21 siblings, 1 reply; 31+ messages in thread
From: Lai Jiangshan @ 2013-01-24 13:36 UTC (permalink / raw)
To: Tejun Heo; +Cc: linux-kernel, Lai Jiangshan
On 17/01/13 09:42, Tejun Heo wrote:
> Hello,
>
> Currently, on the backend side, there are two layers of abstraction.
> For each CPU and the special unbound wq-specific CPU, there's one
> global_cwq. gcwq in turn hosts two worker_pools - one for normal
> priority, the other for highpri - each of which actually serves the
> work items.
>
> worker_pool is the later addition to support separate pool of workers
> for highpri workqueues. Stuff was moved to worker_pool on as-needed
> basis and, as a result, the two pools belonging to the same CPU share
> some stuff in the gcwq - most notably the lock and the hash table for
> work items currently being executed.
>
> It seems like we'll need to support worker pools with custom
> attributes, which is planned to be implemented as extra worker_pools
> for the unbound CPU. Removing gcwq and having worker_pool as the top
> level abstraction makes things much simpler for such designs. Also,
> there's scalability benefit to not sharing locking and busy hash among
> different worker pools as worker pools w/ custom attributes are likely
> to have widely different memory / cpu locality characteristics.
>
> In retrospect, it might have been less churn if we just converted to
> have multiple gcwqs per CPU when we were adding highpri pool support.
> Oh well, such is life and the name worker_pool fits the role much
> better anyway at this point.
>
> This patchset moves the remaining stuff in gcwq to worker_pool and
> then removes gcwq entirely making worker_pool the top level and the
> only backend abstraction. In the process, this patchset also prepares
> for later addition of worker_pools with custom attributes.
>
> This patchset shouldn't introduce any visible differences outside of
> workqueue proper and contains the following 17 patches.
>
> 0001-workqueue-unexport-work_cpu.patch
> 0002-workqueue-use-std_-prefix-for-the-standard-per-cpu-p.patch
> 0003-workqueue-make-GCWQ_DISASSOCIATED-a-pool-flag.patch
> 0004-workqueue-make-GCWQ_FREEZING-a-pool-flag.patch
> 0005-workqueue-introduce-WORK_OFFQ_CPU_NONE.patch
> 0006-workqueue-add-worker_pool-id.patch
> 0007-workqueue-record-pool-ID-instead-of-CPU-in-work-data.patch
> 0008-workqueue-move-busy_hash-from-global_cwq-to-worker_p.patch
> 0009-workqueue-move-global_cwq-cpu-to-worker_pool.patch
> 0010-workqueue-move-global_cwq-lock-to-worker_pool.patch
> 0011-workqueue-make-hotplug-processing-per-pool.patch
> 0012-workqueue-make-freezing-thawing-per-pool.patch
> 0013-workqueue-replace-for_each_worker_pool-with-for_each.patch
> 0014-workqueue-remove-worker_pool-gcwq.patch
> 0015-workqueue-remove-global_cwq.patch
> 0016-workqueue-rename-nr_running-variables.patch
> 0017-workqueue-post-global_cwq-removal-cleanups.patch
>
> 0001-0002 are misc preps.
>
> 0003-0004 move flags from gcwq to pool.
>
> 0005-0007 make work->data off-queue backlink point to worker_pools
> instead of CPUs, which is necessary to move busy_hash to pool.
>
> 0008-0010 move busy_hash, cpu and locking to pool.
>
> 0011-0014 make operations per-pool and remove gcwq usages.
>
> 0015-0017 remove gcwq and cleanup afterwards.
>
> This patchset is on top of wq/for-3.9 023f27d3d6f ("workqueue: fix
> find_worker_executing_work() brekage from hashtable conversion") and
> available in the following git branch.
>
> git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git for-3.9-remove-gcwq
>
For the whole patchset
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
The only concern: get_work_pool() may slow down __queue_work().
I think we can save the pool->id at work_struct->entry.next, It will
simply the code a little. More aggressive, we can save the work_pool
pointer at work_struct->entry.next, it will simply more code and
__queue_work() will not be slowed down. (It is the user's responsibility
not to modify work_struct if the user want to pass it to workqueue API
later)
Thanks,
Lai
^ permalink raw reply [flat|nested] 31+ messages in thread* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-24 13:36 ` Lai Jiangshan
@ 2013-01-24 18:51 ` Tejun Heo
0 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-24 18:51 UTC (permalink / raw)
To: Lai Jiangshan; +Cc: linux-kernel, Lai Jiangshan
Hey, Lai.
On Thu, Jan 24, 2013 at 09:36:39PM +0800, Lai Jiangshan wrote:
> For the whole patchset
> Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Thanks for reviewing it.
> The only concern: get_work_pool() may slow down __queue_work().
Yeap, I'm currently working on to optimize idr so that the fast path
basically comes down to
if (id < FIRST_LAYER_LIMIT)
return idr->first_layer[id];
> I think we can save the pool->id at work_struct->entry.next, It will
> simply the code a little.
Hmm... I'm not sure whether saving id somewhere else would make any
noticeable difference. Decoding ->id probalby comes down to a couple
bit shuffling instructions.
> More aggressive, we can save the work_pool
> pointer at work_struct->entry.next, it will simply more code and
> __queue_work() will not be slowed down. (It is the user's
> responsibility not to modify work_struct if the user want to pass it
> to workqueue API later)
We probably can't store pointers to pools directly because custom
pools will need to be dynamically created and destroyed.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread
* Re: [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction
2013-01-17 1:42 [PATCHSET] workqueue: remove gcwq and make worker_pool the only backend abstraction Tejun Heo
` (20 preceding siblings ...)
2013-01-24 13:36 ` Lai Jiangshan
@ 2013-01-24 19:03 ` Tejun Heo
21 siblings, 0 replies; 31+ messages in thread
From: Tejun Heo @ 2013-01-24 19:03 UTC (permalink / raw)
To: linux-kernel; +Cc: Lai Jiangshan, Joonsoo Kim
On Wed, Jan 16, 2013 at 05:42:32PM -0800, Tejun Heo wrote:
> Currently, on the backend side, there are two layers of abstraction.
> For each CPU and the special unbound wq-specific CPU, there's one
> global_cwq. gcwq in turn hosts two worker_pools - one for normal
> priority, the other for highpri - each of which actually serves the
> work items.
>
> worker_pool is the later addition to support separate pool of workers
> for highpri workqueues. Stuff was moved to worker_pool on as-needed
> basis and, as a result, the two pools belonging to the same CPU share
> some stuff in the gcwq - most notably the lock and the hash table for
> work items currently being executed.
Applied to wq/for-3.9.
Thanks!
--
tejun
^ permalink raw reply [flat|nested] 31+ messages in thread