Re: [PATCH v0.9.1 3/6] sched/umcg: implement UMCG syscalls

[Peter Oskolkov]

On Mon, Dec 6, 2021 at 3:47 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, Nov 29, 2021 at 09:34:49AM -0800, Peter Oskolkov wrote:
> > On Mon, Nov 29, 2021 at 8:41 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> > > Also, timeout on sys_umcg_wait() gets you the exact same situation (or
> > > worse, multiple running workers).
> >
> > It should not. Timed out workers should be added to the runnable list
> > and not become running unless a server chooses so. So sys_umcg_wait()
> > with a timeout should behave similarly to a normal sleep, in that the
> > server is woken upon the worker blocking, and upon the worker wakeup
> > the worker is added to the woken workers list and waits for a server
> > to run it. The only difference is that in a sleep the worker becomes
> > BLOCKED, while in sys_umcg_wait() the worker is RUNNABLE the whole
> > time.
> >
> > Why then have sys_umcg_wait() with a timeout at all, instead of
> > calling nanosleep()? Because the worker in sys_umcg_wait() can be
> > context-switched into by another worker, or made running by a server;
> > if the worker is in nanosleep(), it just sleeps.
>
> I've been trying to figure out the semantics of that timeout thing, and
> I can't seem to make sense of it.
>
> Consider two workers:
>
>         S0 running A                            S1 running B
>
> therefore:
>
>         S0::state == RUNNABLE                   S1::state == RUNNABLE
>         A::server_tid == S0.tid                 B::server_tid = S1.tid
>         A::state == RUNNING                     B::state == RUNNING
>
> Doing:
>
>         self->state = RUNNABLE;                 self->state = RUNNABLE;
>         sys_umcg_wait(0);                       sys_umcg_wait(10);
>           umcg_enqueue_runnable()                 umcg_enqueue_runnable()

sys_umcg_wait() should not enqueue the worker as runnable; workers are
enqueued to indicate wakeup events.

>           umcg_wake()                             umcg_wake()
>           umcg_wait()                             umcg_wait()
>                                                     hrtimer_start()
>
> In both cases we get the exact same outcome:
>
>         A::state == RUNNABLE                    B::state == RUNNABLE
>         S0::state == RUNNING                    S1::state == RUNNING
>         S0::runnable_ptr == &A                  S1::runnable_ptr = &B

So without sys_umcg_wait enqueueing into the queue, the state now is

         A::state == RUNNABLE                    B::state == RUNNABLE
         S0::state == RUNNING                    S1::state == RUNNING
         S0::runnable_ptr == NULL                  S1::runnable_ptr = NULL

>
>
> Which is, AFAICT, the exact state you wanted to achieve, except B now
> has an active timer, but what do you want it to do when that goes?

When the timer goes off, _then_ B is enqueued into the queue, so the
state becomes

         A::state == RUNNABLE                    B::state == RUNNABLE
         S0::state == RUNNING                    S1::state == RUNNING
         S0::runnable_ptr == NULL                  S1::runnable_ptr = &B

So worker timeouts in sys_umcg_wait are treated as wakeup events, with
the difference that when the worker is eventually scheduled by a
server, sys_umcg_wait returns with ETIMEDOUT.

>
> I'm tempted to say workers cannot have timeout, and servers can use it
> to wake themselves.