On 10/15, Oleg Nesterov wrote:
>
> But, this can race with cpu_down(). I think this is solvable but needs
> more locking. I mean, the caller of queue_work_xxx() must not use the old
> get_wq_data(work) if this CPU is already dead, but a simple cpu_online()
> is not enough, we can race with workqueue_cpu_callback(CPU_POST_DEAD)
> flushing this cwq, in this case we should carefully insert this work
> into the almost-dead queue.
>
> Or, perhaps better, instead of new helper, we can probably use the free
> bit in work_struct->data to mark this work/dwork as "single-instance-work".
> In this case __queue_work and queue_delayed_work_on should check this bit.
Actually, this looks simple. Please see the patch below.
Of course! the horror in __queue_work() should be cleanuped somehow.
The change queue_delayed_work_on() needs a separate patch probably.
All, what do you think? Do we need this?
Oleg.
If the work_struct/delayed_work has WORK_STRUCT_XXX bit set, it can never
race with itself.
Note: queue_work_on() or queue_delayed_work_on() must not be used if it is
work_xxx().
Also, we can optimize flush/cancel operations to not scan all CPUs if this
work is "singlethreaded".
PROBLEM: work_xxx() work can block cpu_down() if it contsantly re-queues
itself, hopefully we shouldn't have such stupid users.
---
--- TTT_32/include/linux/workqueue.h~WORK_XXX 2009-09-23 21:12:03.000000000
+0200
+++ TTT_32/include/linux/workqueue.h 2009-10-15 16:49:25.000000000 +0200
@@ -24,7 +24,8 @@ typedef void (*work_func_t)(struct work_
struct work_struct {
atomic_long_t data;
-#define WORK_STRUCT_PENDING 0 /* T if work item pending execution */
+#define WORK_STRUCT_PENDING 0 /* T if work item pending execution */
+#define WORK_STRUCT_XXX 1 /* deny multiple running
instances */
#define WORK_STRUCT_FLAG_MASK (3UL)
#define WORK_STRUCT_WQ_DATA_MASK (~WORK_STRUCT_FLAG_MASK)
struct list_head entry;
@@ -148,6 +149,9 @@ struct execute_work {
#define work_pending(work) \
test_bit(WORK_STRUCT_PENDING, work_data_bits(work))
+#define work_xxx(work) \
+ test_bit(WORK_STRUCT_XXX, work_data_bits(work))
+
/**
* delayed_work_pending - Find out whether a delayable work item is currently
* pending
--- TTT_32/kernel/workqueue.c~WORK_XXX 2009-09-12 21:40:11.000000000 +0200
+++ TTT_32/kernel/workqueue.c 2009-10-15 17:09:51.000000000 +0200
@@ -145,6 +145,35 @@ static void __queue_work(struct cpu_work
{
unsigned long flags;
+ if (work_xxx(work)) {
+ struct cpu_workqueue_struct *old = get_wq_data(work);
+ bool done = false;
+
+ if (!old)
+ goto fallback;
+
+ // This lockless check is racy. We should either remove it
+ // or add mb__before_clear_bit() into run_workqueue().
+ if (old->current_work != work)
+ goto fallback;
+
+ // OK, we should keep this old cwq. But its CPU can be dead,
+ // we have to recheck under old->lock
+ spin_lock_irqsave(&old->lock, flags);
+ if (old->current_work == work) {
+ // It is stiill running, queue the work here.
+ // even if this CPU is dead, run_workqueue()
+ // can't return without noticing this work
+ insert_work(old, work, &old->worklist);
+ done = true;
+ }
+ spin_unlock_irqrestore(&cwq->lock, flags);
+
+ if (done)
+ return;
+ }
+
+fallback:
spin_lock_irqsave(&cwq->lock, flags);
insert_work(cwq, work, &cwq->worklist);
spin_unlock_irqrestore(&cwq->lock, flags);
@@ -246,7 +275,8 @@ int queue_delayed_work_on(int cpu, struc
timer_stats_timer_set_start_info(&dwork->timer);
/* This stores cwq for the moment, for the timer_fn */
- set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id()));
+ if (!get_wq_data(work))
+ set_wq_data(work, wq_per_cpu(wq,
raw_smp_processor_id()));
timer->expires = jiffies + delay;
timer->data = (unsigned long)dwork;
timer->function = delayed_work_timer_fn;
--
To unsubscribe from this list: send the line "unsubscribe kernel-testers" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
