svn commit: r199067 - in head/sys: amd64/amd64 i386/i386

[Kostik Belousov]

On Thu, Nov 12, 2009 at 12:33:40AM +0200, Kostik Belousov wrote:
> On Wed, Nov 11, 2009 at 10:56:51PM +0100, Ed Schouten wrote:
> > Hi,
> > 
> > * Jun Kuriyama <kuriyama at FreeBSD.org> wrote:
> > > Can you test with these patches?  Testing on only one of both system
> > > is enough.  "patch-1" forces disabling CLFLUSH feature even if SS bit
> > > exists.  "patch-2" forces no CLFLUSH tweak.  I'd like to know with
> > > which patch your system can live.
> > 
> > For some reason they both seem to boot. Yikes. I've done some more
> > testing and it seems the old version even hangs if I add some additional
> > printf's above and below, which makes me believe the problem is a bit
> > more complex than we realize...
> 
> The only thing I see now is that TUNABLE_INT declaration is not needed,
> since SYSINIT is started after hammer_time(), so TUNABLE_FETCH is
> processed after everything is done for BSP.
> 
> Wait, are your machines SMP ? Hmm, could you, please, remove TUNABLE_INT()
> and see how it ends up ?

[Selecting random email in the thread].

Luckily, my co-worker has laptop that runs HEAD and exhibited the
problem. Patch below fixed the hang for him. Will be happy to get
confirmation that patch fixes the hung for other people who reported it.

http://people.freebsd.org/~kib/misc/initcache.1.patch

Issue is that TUNABLE_INT_FETCH was called on AP while CPU was not
initialized properly, in particular, curthread was not set etc. But,
kernel environment is already switched to dynamic, so access needs to
lock a mutex.

I am currently looking at i386, that probably needs similar change.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/svn-src-head/attachments/20091112/208f6dab/attachment.pgp