-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On Nov 19, 2007, at 18:23, Andrew Doran wrote:

> On Mon, Nov 19, 2007 at 05:54:56PM -0500, Michael Lorenz wrote:
>
>> Not sure if that was the purpose but even with this I still get
>> deadlocks when opening a zs on SMP. Any idea what to look for? So far
>> I couldn't even figure out where exactly the deadlock occurs.
>
> Is there a head on this machine?

Yes.

> Have you tried with a LOCKDEBUG kernel?

Yes, it runs into a panic much earlier - apparently it manages to run  
into pool_get(), coming from uvm_init() with pp->pr_lock held. Sorry,  
I'm totally unfamiliar with that code.

> Not much use if there is only a serial port as console, though. The  
> only major
> change is that the driver now uses a mutex where before it used  
> simple_lock
> + splzs. Can you double check that IPL_TTY is actually enough to  
> block the
> serial interrupt? If it's not, it will deadlock pretty quickly.

Eh, probably not. IPL_SERIAL used to be very high on macppc, I  
lowered it somewhat but it's likely higher than IPL_TTY. So, in the  
New Order Of Things IPL_SERIAL should be lower (priority-wise) than  
IPL_TTY? At least sparc will have a very high priority IPL_SERIAL as  
well.

have fun
Michael
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)

iQEVAwUBR0IgMcpnzkX8Yg2nAQJqsAf8DjF5RnfIfwqp5eNGFDFFGxwIgBZbaI7a
EWEsi6fq4eAcpeevshy5WU0mfVw09TdpJ3GM8tyU3Do+ZnXUPeEzUJ7o8LWxPrxJ
qCzyOv47Iiwk3l9c8s6jfF25X4M1lW0MCKf55Rc0NIvGDRT4Zsn6yr/j3Jb8mATi
K3OAV4ek/Rnmoqli18QV26yPop7dV673Kw0d2NOdlZv/PR0/Ne8uC166il/QsnB9
RAlPDNXXPeMo2aFSM98icfYZ+04qKn61+urHIEi3UkS+Ts2/anUELkholFp2ROfo
cbUh4ewzCWeGXnQ3SGaDsosIy+7FihaazCZLQsiiiEBA3hcmyhsAhw==
=81u0
-----END PGP SIGNATURE-----