Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 29 Jan 2007 10:07:24 -0800
From:      Nate Lawson <nate@root.org>
To:        Pawel Jakub Dawidek <pjd@FreeBSD.org>
Cc:        cvs-src@freebsd.org, src-committers@freebsd.org, cvs-all@freebsd.org
Subject:   Re: cvs commit: src/sys/geom/eli g_eli.c
Message-ID:  <45BE37DC.6080509@root.org>
In-Reply-To: <20070129175222.GA87767@garage.freebsd.pl>
References:  <20070128202917.5B67916A5A6@hub.freebsd.org> <45BD82D2.20301@root.org> <20070129175222.GA87767@garage.freebsd.pl>

next in thread | previous in thread | raw e-mail | index | archive | help
Pawel Jakub Dawidek wrote:
> On Sun, Jan 28, 2007 at 09:14:58PM -0800, Nate Lawson wrote:
>> Pawel Jakub Dawidek wrote:
>>> pjd         2007-01-28 20:29:12 UTC
>>>  FreeBSD src repository
>>>  Modified files:
>>>    sys/geom/eli         g_eli.c   Log:
>>>  It is possible that GEOM taste provider before SMP is started.
>>>  We can't bind to a CPU which is not yet on-line, so add code that wait for
>>>  CPUs to go on-line before binding to them.
>>>    Reported by:    Alin-Adrian Anton <aanton@spintech.ro>
>>>  MFC after:      2 weeks
>>>    Revision  Changes    Path
>>>  1.34      +7 -0      src/sys/geom/eli/g_eli.c
>>> Index: src/sys/geom/eli/g_eli.c
>>> diff -u src/sys/geom/eli/g_eli.c:1.33 src/sys/geom/eli/g_eli.c:1.34
>>> --- src/sys/geom/eli/g_eli.c:1.33	Thu Nov  2 09:01:34 2006
>>> +++ src/sys/geom/eli/g_eli.c	Sun Jan 28 20:29:12 2007
>>> @@ -324,6 +324,13 @@
>>>  	wr = arg;
>>> 	sc = wr->w_softc;
>>> +#ifdef SMP
>>> +	/* Before sched_bind() to a CPU, wait for all CPUs to go on-line. */
>>> +	if (sc->sc_crypto == G_ELI_CRYPTO_SW && g_eli_threads == 0) {
>>> +		while (!smp_started)
>>> +			tsleep(wr, 0, "geli:smp", hz / 4);
>>> +	}
>>> +#endif
>>> 	mtx_lock_spin(&sched_lock);
>>> 	sched_prio(curthread, PRIBIO);
>>> 	if (sc->sc_crypto == G_ELI_CRYPTO_SW && g_eli_threads == 0)
>> I thought tsleep() didn't work right before !cold.  Is that old knowledge?
> 
> Hmm, I thought that cold is zeroed before smp_started is set?
> 

I don't think that's guaranteed.  Besides, there's an easier way to fix 
your problem.  Instead of calling kthread_create() from your geom create 
routine directly, queue a stub request on the thread taskqueue to call 
kthread_create().

For example, see sys/dev/acpica/acpi_acad.c.  We call this from attach() 
    and then the system thread only runs and then calls acpi_acad_init() 
once all CPUs are initialized.  See sys/sys/taskqueue.h for how to add a 
task to the system thread taskqueue:

taskqueue_enqueue(taskqueue_thread, &task);

-- 
Nate



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?45BE37DC.6080509>