Experimental Physics and
Industrial Control System

Michael Davidsaver via Core-talk <[email protected]> · Wed, 12 Jun 2019 09:50:58 -0700

On 6/11/19 6:57 AM, Mark Rivers via Core-talk wrote:
>> If we want to sleep for a minimum amount of time we might want to call nanosleep(0) instead.
> 
> But that will not work on Windows or vxWorks.
> 
> Mark
> 
> 
> -----Original Message-----
> From: Konrad, Martin <[email protected]> 
> Sent: Tuesday, June 11, 2019 8:49 AM
> To: Mark Rivers <[email protected]>; [email protected]
> Subject: Re: epicsThreadSleep and epicsThreadSleepQuantum
> 
> Hi,
>> I started a tech-talk thread on this topic 11 years ago:
>>
>> https://epics.anl.gov/tech-talk/2008/msg00987.php
> Thanks Mark, I wasn't aware of this thread. I'm not aware of any way to detect the sleep quantum reliably on modern machines either. The fact that epicsThreadSleepQuantum() only works correctly on old machines without high resolution timers suggests that we should avoid using it. I think it would make sense to

I'll second this.  Minimum sleep time is not a meaningful concept for a modern general purpose OS.

https://en.wikipedia.org/wiki/Tickless_kernel

All the usages of sleep(0) I've seen have been shortcuts to avoid using real synchronization primitives.
This is something I would discourage.

> a) Add a warning to the App Dev Guide.
> b) Consider marking epicsThreadSleepQuantum(), timerQueueActive::quantum(), and fdManager::quantum() as deprecated.

I'd be fine with this, provided all internal usages can be removed.

> c) Remove the "(ppsl->period < 2 * quantum)" part of the test in [1].

There is also a usage in timer::privateStart() to bias expiration time.
I wonder if this tweak is still relevant?

>    this->exp = expire - ( this->queue.notify.quantum () / 2.0 );

https://github.com/epics-base/epics-base/blob/655c5bed925137619d4a538f7893400b38c7c907/modules/libcom/src/timer/timer.cpp#L69

originally introduced by:

https://github.com/epics-base/epics-base/commit/0792aebf952dbc4f7e7291e6175c27b72a80c338

The remaining usage I find in Base, outside of test code, is in various epicsThreadOnce()
implementations which spin to avoid coupling, or copying, the internals of osdEvent.c into osdThread.c.

> If we want to sleep for a minimum amount of time we might want to call
> nanosleep(0) instead. If I'm understanding the man page correctly this should result at least in a context switch.

Maybe add a dedicated call?

> /* Ensure your epicsThread is fresh and ready to start the day */
> void epicsSnooze();

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System