[PATCH] workqueue: Document (some) memory-ordering properties of {queue,schedule}_work()

[PATCH] workqueue: Document (some) memory-ordering properties of {queue,schedule}_work()

[Andrea Parri]

It's desirable to be able to rely on the following property:  All stores
preceding (in program order) a call to a successful queue_work() will be
visible from the CPU which will execute the queued work by the time such
work executes, e.g.,

  { x is initially 0 }

    CPU0                              CPU1

    WRITE_ONCE(x, 1);                 [ "work" is being executed ]
    r0 = queue_work(wq, work);          r1 = READ_ONCE(x);

  Forbids: r0 == true && r1 == 0

The current implementation of queue_work() provides such memory-ordering
property:

  - In __queue_work(), the ->lock spinlock is acquired.

  - On the other side, in worker_thread(), this same ->lock is held
    when dequeueing work.

So the locking ordering makes things work out.

Add this property to the DocBook headers of {queue,schedule}_work().

Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Andrea Parri <parri.andrea@gmail.com>
---
 include/linux/workqueue.h | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index 4261d1c6e87b1..4fef6c38b0536 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -487,6 +487,19 @@ extern void wq_worker_comm(char *buf, size_t size, struct task_struct *task);
  *
  * We queue the work to the CPU on which it was submitted, but if the CPU dies
  * it can be processed by another CPU.
+ *
+ * Memory-ordering properties:  If it returns %true, guarantees that all stores
+ * preceding the call to queue_work() in the program order will be visible from
+ * the CPU which will execute @work by the time such work executes, e.g.,
+ *
+ * { x is initially 0 }
+ *
+ *   CPU0				CPU1
+ *
+ *   WRITE_ONCE(x, 1);			[ @work is being executed ]
+ *   r0 = queue_work(wq, work);		  r1 = READ_ONCE(x);
+ *
+ * Forbids: r0 == true && r1 == 0
  */
 static inline bool queue_work(struct workqueue_struct *wq,
 			      struct work_struct *work)
@@ -546,6 +559,9 @@ static inline bool schedule_work_on(int cpu, struct work_struct *work)
  * This puts a job in the kernel-global workqueue if it was not already
  * queued and leaves it in the same position on the kernel-global
  * workqueue otherwise.
+ *
+ * Shares the same memory-ordering properties of queue_work(), c.f., the
+ * DocBook header of queue_work().
  */
 static inline bool schedule_work(struct work_struct *work)
 {
-- 
2.24.0

Re: [PATCH] workqueue: Document (some) memory-ordering properties of {queue,schedule}_work()

[Randy Dunlap]

Hi,

On 1/18/20 1:58 PM, Andrea Parri wrote:
> It's desirable to be able to rely on the following property:  All stores
> preceding (in program order) a call to a successful queue_work() will be
> visible from the CPU which will execute the queued work by the time such
> work executes, e.g.,
> 
>   { x is initially 0 }
> 
>     CPU0                              CPU1
> 
>     WRITE_ONCE(x, 1);                 [ "work" is being executed ]
>     r0 = queue_work(wq, work);          r1 = READ_ONCE(x);
> 
>   Forbids: r0 == true && r1 == 0
> 
> The current implementation of queue_work() provides such memory-ordering
> property:
> 
>   - In __queue_work(), the ->lock spinlock is acquired.
> 
>   - On the other side, in worker_thread(), this same ->lock is held
>     when dequeueing work.
> 
> So the locking ordering makes things work out.
> 
> Add this property to the DocBook headers of {queue,schedule}_work().
> 
> Suggested-by: Paul E. McKenney <paulmck@kernel.org>
> Signed-off-by: Andrea Parri <parri.andrea@gmail.com>
> ---
>  include/linux/workqueue.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
> index 4261d1c6e87b1..4fef6c38b0536 100644
> --- a/include/linux/workqueue.h
> +++ b/include/linux/workqueue.h
> @@ -487,6 +487,19 @@ extern void wq_worker_comm(char *buf, size_t size, struct task_struct *task);
>   *
>   * We queue the work to the CPU on which it was submitted, but if the CPU dies
>   * it can be processed by another CPU.
> + *
> + * Memory-ordering properties:  If it returns %true, guarantees that all stores
> + * preceding the call to queue_work() in the program order will be visible from
> + * the CPU which will execute @work by the time such work executes, e.g.,
> + *
> + * { x is initially 0 }
> + *
> + *   CPU0				CPU1
> + *
> + *   WRITE_ONCE(x, 1);			[ @work is being executed ]
> + *   r0 = queue_work(wq, work);		  r1 = READ_ONCE(x);
> + *
> + * Forbids: r0 == true && r1 == 0
>   */
>  static inline bool queue_work(struct workqueue_struct *wq,
>  			      struct work_struct *work)
> @@ -546,6 +559,9 @@ static inline bool schedule_work_on(int cpu, struct work_struct *work)
>   * This puts a job in the kernel-global workqueue if it was not already
>   * queued and leaves it in the same position on the kernel-global
>   * workqueue otherwise.
> + *
> + * Shares the same memory-ordering properties of queue_work(), c.f., the

nit:                                                              cf. the

> + * DocBook header of queue_work().
>   */
>  static inline bool schedule_work(struct work_struct *work)
>  {
> 


-- 
~Randy

Re: [PATCH] workqueue: Document (some) memory-ordering properties of {queue,schedule}_work()

[Paul E. McKenney]

On Sat, Jan 18, 2020 at 10:58:20PM +0100, Andrea Parri wrote:
> It's desirable to be able to rely on the following property:  All stores
> preceding (in program order) a call to a successful queue_work() will be
> visible from the CPU which will execute the queued work by the time such
> work executes, e.g.,
> 
>   { x is initially 0 }
> 
>     CPU0                              CPU1
> 
>     WRITE_ONCE(x, 1);                 [ "work" is being executed ]
>     r0 = queue_work(wq, work);          r1 = READ_ONCE(x);
> 
>   Forbids: r0 == true && r1 == 0
> 
> The current implementation of queue_work() provides such memory-ordering
> property:
> 
>   - In __queue_work(), the ->lock spinlock is acquired.
> 
>   - On the other side, in worker_thread(), this same ->lock is held
>     when dequeueing work.
> 
> So the locking ordering makes things work out.
> 
> Add this property to the DocBook headers of {queue,schedule}_work().
> 
> Suggested-by: Paul E. McKenney <paulmck@kernel.org>
> Signed-off-by: Andrea Parri <parri.andrea@gmail.com>

Acked-by: Paul E. McKenney <paulmck@kernel.org>

An alternative to Randy's suggestion of dropping the comma following
the "cf." is to just drop that whole phrase.  I will let you and Randy
work that one out, though.  ;-)

> ---
>  include/linux/workqueue.h | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
> index 4261d1c6e87b1..4fef6c38b0536 100644
> --- a/include/linux/workqueue.h
> +++ b/include/linux/workqueue.h
> @@ -487,6 +487,19 @@ extern void wq_worker_comm(char *buf, size_t size, struct task_struct *task);
>   *
>   * We queue the work to the CPU on which it was submitted, but if the CPU dies
>   * it can be processed by another CPU.
> + *
> + * Memory-ordering properties:  If it returns %true, guarantees that all stores
> + * preceding the call to queue_work() in the program order will be visible from
> + * the CPU which will execute @work by the time such work executes, e.g.,
> + *
> + * { x is initially 0 }
> + *
> + *   CPU0				CPU1
> + *
> + *   WRITE_ONCE(x, 1);			[ @work is being executed ]
> + *   r0 = queue_work(wq, work);		  r1 = READ_ONCE(x);
> + *
> + * Forbids: r0 == true && r1 == 0
>   */
>  static inline bool queue_work(struct workqueue_struct *wq,
>  			      struct work_struct *work)
> @@ -546,6 +559,9 @@ static inline bool schedule_work_on(int cpu, struct work_struct *work)
>   * This puts a job in the kernel-global workqueue if it was not already
>   * queued and leaves it in the same position on the kernel-global
>   * workqueue otherwise.
> + *
> + * Shares the same memory-ordering properties of queue_work(), c.f., the
> + * DocBook header of queue_work().
>   */
>  static inline bool schedule_work(struct work_struct *work)
>  {
> -- 
> 2.24.0
> 

Re: [PATCH] workqueue: Document (some) memory-ordering properties of {queue,schedule}_work()

[Andrea Parri]

On Sun, Jan 19, 2020 at 06:02:35PM -0800, Paul E. McKenney wrote:
> On Sat, Jan 18, 2020 at 10:58:20PM +0100, Andrea Parri wrote:
> > It's desirable to be able to rely on the following property:  All stores
> > preceding (in program order) a call to a successful queue_work() will be
> > visible from the CPU which will execute the queued work by the time such
> > work executes, e.g.,
> > 
> >   { x is initially 0 }
> > 
> >     CPU0                              CPU1
> > 
> >     WRITE_ONCE(x, 1);                 [ "work" is being executed ]
> >     r0 = queue_work(wq, work);          r1 = READ_ONCE(x);
> > 
> >   Forbids: r0 == true && r1 == 0
> > 
> > The current implementation of queue_work() provides such memory-ordering
> > property:
> > 
> >   - In __queue_work(), the ->lock spinlock is acquired.
> > 
> >   - On the other side, in worker_thread(), this same ->lock is held
> >     when dequeueing work.
> > 
> > So the locking ordering makes things work out.
> > 
> > Add this property to the DocBook headers of {queue,schedule}_work().
> > 
> > Suggested-by: Paul E. McKenney <paulmck@kernel.org>
> > Signed-off-by: Andrea Parri <parri.andrea@gmail.com>
> 
> Acked-by: Paul E. McKenney <paulmck@kernel.org>

Thanks!

>
> An alternative to Randy's suggestion of dropping the comma following
> the "cf." is to just drop that whole phrase.  I will let you and Randy
> work that one out, though.  ;-)

Either way works for me.

I'd give Tejun and Lai some more time to review this and send a non-RFC with
your Ack and this nit fixed later this week (unless I hear some objections).

Thanks,
  Andrea