Re: How to wait on cv_timedwait() correctly

Why not use cv_timedwaitbt_sig() in that case? It updates the time 
argument based on amount of time passed.

But I don't think cv_signal() unblocks all. Not sure why you think that. 
That would be cv_broadcast().

   Johnny

On 2025-01-30 15:37, Stephan wrote:
> Hello,
> 
> I think I have found a PR relating to my question:
> 
> http://gnats.netbsd.org/38554
> 
> I am implementing a semaphore system for applications. To wait on a
> semaphore to become ready, it does basically this:
> 
> do
>          {
>              khs->khs_waiters++;
>              error = cv_timedwait_sig(&khs->khs_cv,
> &khs->khs_interlock, mstohz((flags & SEM_RELATIVE_TIMEOUT) ? timeout :
> 0));
>              khs->khs_waiters--;
> 
>              [...]
> 
>          } while (khs->khs_count - count < 0);
> 
> 
> I have seen that cv_signal() on a CV unblocks all threads waiting on
> it, instead of just one. That this could happen is briefly mentioned
> in condvar(9).
> 
> I think the problem is that when cv_timedwait_sig() is called
> repeatedly, the original timeout is used again and again, adding up to
> an unpredictable amount of time.
> 
> If that is the case, how can it be solved best? Is
> cv_timedwaitbt_sig() suitable? Or should one track sleep and wakeup
> times and calculate a new timeout each time?
> 
> 
> Thanks,
> 
> Stephan

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: [email protected]             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol