Date: Tue, 17 May 2011 09:20:40 -0700 From: Max Laier <max@love2party.net> To: John Baldwin <jhb@FreeBSD.org> Cc: neel@freebsd.org, Andriy Gapon <avg@freebsd.org>, Attilio Rao <attilio@freebsd.org>, FreeBSD current <freebsd-current@freebsd.org>, Stephan Uphoff <ups@FreeBSD.org>, Peter Grehan <grehan@freebsd.org> Subject: Re: proposed smp_rendezvous change Message-ID: <4DD2A058.6050400@love2party.net> In-Reply-To: <4DD26720.3000001@FreeBSD.org> References: <4DCD357D.6000109@FreeBSD.org> <201105161738.53366.max@love2party.net> <BANLkTik8CqtiP9OgvBpL08dqK6Aj%2BLQ3OA@mail.gmail.com> <201105161805.51188.max@love2party.net> <4DD26720.3000001@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 05/17/2011 05:16 AM, John Baldwin wrote: ... > Index: kern/kern_switch.c > =================================================================== > --- kern/kern_switch.c (revision 221536) > +++ kern/kern_switch.c (working copy) > @@ -192,15 +192,22 @@ > critical_exit(void) > { > struct thread *td; > - int flags; > + int flags, owepreempt; > > td = curthread; > KASSERT(td->td_critnest != 0, > ("critical_exit: td_critnest == 0")); > > if (td->td_critnest == 1) { > + owepreempt = td->td_owepreempt; > + td->td_owepreempt = 0; > + /* > + * XXX: Should move compiler_memory_barrier() from > + * rmlock to a header. > + */ XXX: If we get an interrupt at this point and td_owepreempt was zero, the new interrupt will re-set it, because td_critnest is still non-zero. So we still end up with a thread that is leaking an owepreempt *and* lose a preemption. We really need an atomic_readandclear() which gives us a local copy of td_owepreempt *and* clears critnest in the same operation. Sadly, that is rather expensive. It is possible to implement with a flag for owepreempt, but that means that all writes to critnest must then be atomic. Either because we know we have interrupts disabled (i.e. setting owepreempt can be a RMW), or with a proper atomic_add/set/... I'm not sure what the performance impact of this will be. One would hope that atomic_add without a memory barrier isn't much more expensive than a compiler generated read-modify-write, tho. Especially, since this cacheline should be local and exclusive to us, anyway. > + __asm __volatile("":::"memory"); > td->td_critnest = 0; > - if (td->td_owepreempt) { > + if (owepreempt) { > td->td_critnest = 1; > thread_lock(td); > td->td_critnest--;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4DD2A058.6050400>