Date: Fri, 24 Aug 2018 15:37:27 +0200 From: Sebastian Huber <sebastian.huber@embedded-brains.de> To: FreeBSD <freebsd-hackers@freebsd.org> Subject: Re: epoch(9) background information? Message-ID: <b75359f8-f05e-b2c8-0b28-0c3e03bfa3fd@embedded-brains.de> In-Reply-To: <db397431-2c4c-64de-634a-20f38ce6a60e@embedded-brains.de> References: <db397431-2c4c-64de-634a-20f38ce6a60e@embedded-brains.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Hello,
I try currently to understand the epoch implementation in FreeBSD to=20
implement the API in RTEMS.
If I understood epoch_wait() correctly, then this function just busy=20
waits until all conditions are satisfied. This is all right, since all=20
other participants operate with thread dispatching disabled, so the time=20
can be only influenced by interrupts and actual work.
I have some questions to epoch_block_handler_preempt(). The caller of=20
epoch_wait_prempt() seems to depend on threads which are currently=20
blocked, so busy waiting is not a good idea. Some questions to the code:
/*
=C2=A0* epoch_block_handler_preempt is a callback from the ck code when=20
another thread is
=C2=A0* currently in an epoch section.
=C2=A0*/
static void
epoch_block_handler_preempt(struct ck_epoch *global __unused,=20
ck_epoch_record_t *cr,
=C2=A0=C2=A0=C2=A0 void *arg __unused)
{
=C2=A0=C2=A0=C2=A0 epoch_record_t record;
=C2=A0=C2=A0=C2=A0 struct thread *td, *owner, *curwaittd;
=C2=A0=C2=A0=C2=A0 struct epoch_thread *tdwait;
=C2=A0=C2=A0=C2=A0 struct turnstile *ts;
=C2=A0=C2=A0=C2=A0 struct lock_object *lock;
=C2=A0=C2=A0=C2=A0 int spincount, gen;
=C2=A0=C2=A0=C2=A0 int locksheld __unused;
=C2=A0=C2=A0=C2=A0 record =3D __containerof(cr, struct epoch_record, er_=
record);
=C2=A0=C2=A0=C2=A0 td =3D curthread;
=C2=A0=C2=A0=C2=A0 locksheld =3D td->td_locks;
=C2=A0=C2=A0=C2=A0 spincount =3D 0;
=C2=A0=C2=A0=C2=A0 counter_u64_add(block_count, 1);
=C2=A0=C2=A0=C2=A0 /*
=C2=A0=C2=A0=C2=A0 =C2=A0* We lost a race and there's no longer any thre=
ads
=C2=A0=C2=A0=C2=A0 =C2=A0* on the CPU in an epoch section.
=C2=A0=C2=A0=C2=A0 =C2=A0*/
=C2=A0=C2=A0=C2=A0 if (TAILQ_EMPTY(&record->er_tdlist))
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 return;
=C2=A0=C2=A0=C2=A0 if (record->er_cpuid !=3D curcpu) {
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 ...
### this seems clear
=C2=A0=C2=A0=C2=A0 }
=C2=A0=C2=A0=C2=A0 /*
=C2=A0=C2=A0=C2=A0 =C2=A0* Try to find a thread in an epoch section on t=
his CPU
=C2=A0=C2=A0=C2=A0 =C2=A0* waiting on a turnstile. Otherwise find the lo=
west
=C2=A0=C2=A0=C2=A0 =C2=A0* priority thread (highest prio value) and drop=
our priority
=C2=A0=C2=A0=C2=A0 =C2=A0* to match to allow it to run.
=C2=A0=C2=A0=C2=A0 =C2=A0*/
=C2=A0=C2=A0=C2=A0 TAILQ_FOREACH(tdwait, &record->er_tdlist, et_link) {
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 /*
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0* Propagate our priority to =
any other waiters to prevent us
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0* from starving them. They w=
ill have their original priority
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0* restore on exit from epoch=
_wait().
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0*/
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 curwaittd =3D tdwait->et_td;
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 if (!TD_IS_INHIBITED(curwaittd) &&=
curwaittd->td_priority >=20
td->td_priority) {
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 critical_enter(=
);
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 thread_unlock(t=
d);
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 thread_lock(cur=
waittd);
### could we accidentally lower the priority here?
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 sched_prio(curw=
aittd, td->td_priority);
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 thread_unlock(c=
urwaittd);
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 thread_lock(td)=
;
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 critical_exit()=
;
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 }
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 if (TD_IS_INHIBITED(curwaittd) && =
TD_ON_LOCK(curwaittd) &&
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 ((ts =3D curwai=
ttd->td_blocked) !=3D NULL)) {
=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 ...
### is this some sort of condition variable which wakes us up together=20
with the waiting thread?
=C2=A0=C2=A0=C2=A0 =C2=A0=C2=A0=C2=A0 }
=C2=A0=C2=A0=C2=A0 }
### is this a yield operation, e.g. in case we still the highest=20
priority thread, then we run immediately again?
=C2=A0=C2=A0=C2=A0 /*
=C2=A0=C2=A0=C2=A0 =C2=A0* We didn't find any threads actually blocked o=
n a lock
=C2=A0=C2=A0=C2=A0 =C2=A0* so we have nothing to do except context switc=
h away.
=C2=A0=C2=A0=C2=A0 =C2=A0*/
=C2=A0=C2=A0=C2=A0 counter_u64_add(switch_count, 1);
=C2=A0=C2=A0=C2=A0 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
=C2=A0=C2=A0=C2=A0 /*
=C2=A0=C2=A0=C2=A0 =C2=A0* Release the thread lock while yielding to
=C2=A0=C2=A0=C2=A0 =C2=A0* allow other threads to acquire the lock
=C2=A0=C2=A0=C2=A0 =C2=A0* pointed to by TDQ_LOCKPTR(td). Else a
=C2=A0=C2=A0=C2=A0 =C2=A0* deadlock like situation might happen. (HPS)
=C2=A0=C2=A0=C2=A0 =C2=A0*/
=C2=A0=C2=A0=C2=A0 thread_unlock(td);
=C2=A0=C2=A0=C2=A0 thread_lock(td);
}
--=20
Sebastian Huber, embedded brains GmbH
Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone : +49 89 189 47 41-16
Fax : +49 89 189 47 41-09
E-Mail : sebastian.huber@embedded-brains.de
PGP : Public key available on request.
Diese Nachricht ist keine gesch=C3=A4ftliche Mitteilung im Sinne des EHUG=
.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b75359f8-f05e-b2c8-0b28-0c3e03bfa3fd>