Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 18 Oct 2010 17:05:24 -0400
From:      John Baldwin <jhb@freebsd.org>
To:        Marius Strobl <marius@alchemy.franken.de>
Cc:        svn-src-head@freebsd.org, Alexander Motin <mav@freebsd.org>, src-committers@freebsd.org, svn-src-all@freebsd.org
Subject:   Re: svn commit: r213985 - head/sys/sparc64/sparc64
Message-ID:  <201010181705.24879.jhb@freebsd.org>
In-Reply-To: <20101018205224.GO1416@alchemy.franken.de>
References:  <201010171646.o9HGks2U038501@svn.freebsd.org> <4CBCADDD.5070109@FreeBSD.org> <20101018205224.GO1416@alchemy.franken.de>

next in thread | previous in thread | raw e-mail | index | archive | help
On Monday, October 18, 2010 4:52:24 pm Marius Strobl wrote:
> On Mon, Oct 18, 2010 at 11:28:13PM +0300, Alexander Motin wrote:
> > Marius Strobl wrote:
> > > On Mon, Oct 18, 2010 at 10:03:12AM -0400, John Baldwin wrote:
> > >> On Sunday, October 17, 2010 12:46:54 pm Marius Strobl wrote:
> > >>> Author: marius
> > >>> Date: Sun Oct 17 16:46:54 2010
> > >>> New Revision: 213985
> > >>> URL: http://svn.freebsd.org/changeset/base/213985
> > >>>
> > >>> Log:
> > >>>   - In oneshot-mode it doesn't make sense to try to compensate the clock
> > >>>     drift in order to achieve a more stable clock as the tick intervals may
> > >>>     vary in the first place. In fact I haven't seen this code kick in when
> > >>>     in oneshot-mode so just skip it in that case.
> > >>>   - There's no need to explicitly stop the (S)TICK counter in oneshot-mode
> > >>>     with every tick as it just won't trigger again with the (S)TICK compare
> > >>>     register set to a value in the past (with a wrap-around once every ~195
> > >>>     years of uptime at 1.5 GHz this isn't something we have to worry about
> > >>>     in practice).
> > >>>   - Given that we'll disable interrupts completely anyway there's no
> > >>>     need to enter critical sections.
> > >> This last is not entirely true.  The purpose of the critical section is to 
> > >> prevent the kernel from preempting to the softclock swi thread until all of 
> > >> the hardclock handler has finished execution.  Thus, places that actually 
> > >> actually call hardclock() should probably still be wrapped in a critical
> > >> section.
> > > 
> > > It's currently unclear to me how on architectures converted to the
> > > event timer world order hardclock() is called eventually but in any case
> > > shouldn't it be the responsibility of the code actually calling it (or
> > > the equivalent code) to wrap it in a critical section instead then? After
> > > all the MD part just enrolls in calling _something_ in one-shot and/or
> > > periodic mode without knowing what it actually calls (and IMO it also
> > > should no longer need to). In handleevents() of kern_clocksource.c
> > > hardclock_anycpu() is called so i think that is what actually needs to
> > > be wrapped in a critical section.
> > 
> > At this time on most (all?) platforms critical section is grabbed by MD
> > interrupt code. It is important to be there, as soon as there touched
> > td_intr_nesting_level and td_intr_frame fields of curthread. We can't
> > allow thread migration until all counted interrupt handlers complete.
> > 
> 
> AFAICT this is not true; intr_event_handle() in sys/kern/kern_intr.c
> is what enters a critical section and f.e. on amd64 I don't see where
> anywhere in the path from ISR_VEC() to intr_execute_handlers()
> calling intr_event_handle() a critical section would be entered,
> which also means that in intr_execute_handlers() td_intr_nesting_level
> is incremented outside of a critical section.

Not all of the clock interrupts use intr_event_handle().  The local APIC
timer uses its own interrupt entry point on x86 for example and uses an
explicit critical section as a result.  I suspect the sparc64 tick interrupt
is closer to the local APIC timer case and doesn't use intr_event_handle().

The fact that some clock interrupts do use intr_event_handle() (e.g. the
atrtc driver on x86 now) does indicate that the low-level interrupt code
probably does not belong in the time events code but in the caller.

-- 
John Baldwin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201010181705.24879.jhb>