lkml.org 
[lkml]   [2014]   [Jul]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH 11/16] rcu: Check for spurious wakeup using return value
On Wed, Jul 23, 2014 at 11:43 PM, Paul E. McKenney
<paulmck@linux.vnet.ibm.com> wrote:
> On Wed, Jul 23, 2014 at 10:36:19PM -0400, Pranith Kumar wrote:
>> On Wed, Jul 23, 2014 at 8:26 AM, Paul E. McKenney
>> <paulmck@linux.vnet.ibm.com> wrote:
>> > On Wed, Jul 23, 2014 at 01:09:48AM -0400, Pranith Kumar wrote:
>> >> When the gp_kthread wakes up from the wait event, it returns 0 if the wake up is
>> >> due to the condition having been met. This commit checks this return value
>> >> for a spurious wake up before calling rcu_gp_init().
>> >>
>> >> Signed-off-by: Pranith Kumar <bobby.prani@gmail.com>
>> >
>> > How does this added check help? I don't see that it does. If the flag
>> > is set, we want to wake up. If we get a spurious wakeup, but then the
>> > flag gets set before we actually wake up, we still want to wake up.
>>
>> So I took a look at the docs again, and using the return value is the
>> recommended way to check for spurious wakeups.
>>
>> The condition in wait_event_interruptible() is checked when the task
>> is woken up (either due to stray signals or explicitly) and it returns
>> true if condition evaluates to true.

this should be returns '0' if the condition evaluates to true.

>>
>> In the current scenario, if we get a spurious wakeup, we take the
>> costly path of checking this condition again (with a barrier and lock)
>> before going back to wait.
>>
>> The scenario of getting an actual wakeup after getting a spurious
>> wakeup exists even today, this is the window after detecting a
>> spurious wakeup and before going back to wait. I am not sure if using
>> the return value enlarges that window as we are going back to sleep
>> immediately.
>>
>> Thoughts?
>
> If the flag is set, why should we care whether or not the wakeup was
> spurious? If the flag is not set, why should we care whether or not
> wait_event_interruptible() thought that the wakeup was not spurious?
>

A correction about the return value above: return will be 0 if the
condition is true, in this case if the flag is set.

If the flag is set, ret will be 0 and we will go ahead with
rcu_gp_init(). (no change wrt current behavior)

If the flag is not set, currently we go ahead and call rcu_gp_init()
from where we check if the flag is set (after a lock+barrier) and
return.

If we care about what wait_event_interruptible() returns, we can go
back and wait for an actual wakeup much earlier without the additional
overhead of calling rcu_gp_init().

--
Pranith


\
 
 \ /
  Last update: 2014-07-24 07:08    [W:0.081 / U:31.828 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site