Re: kern/47648: NetBSD 6.1_RC1 ACPI interupt routing problem + other ACPI lossage

[Chuck Silvers <chuq%chuq.com@localhost>]

On Sat, Mar 23, 2013 at 09:44:16PM +0100, Reinoud Zandijk wrote:
> On Sat, Mar 23, 2013 at 08:04:00AM -0700, Chuck Silvers wrote:
> > On Mon, Mar 18, 2013 at 10:09:33PM +0100, Reinoud Zandijk wrote:
> > > The patch workes and fixes the problem! I get a neat 100 interrupts/sec 
> > > and my
> > > impression is that it is smoother than before, but thats most likely my
> > > imagination :)
> > 
> > the previous patch had several problems, please try the new one that's 
> > attached.
> > I dug through all my x86 boxes and finally found one with an "interesting"
> > ACPI setup, and this latest patch works on that one too.
> > could all the folks who have had trouble with my ACPI changes please
> > retest with this latest patch to make sure I haven't broken any of those
> > systems that I fixed previously?  this patch applies to either -current or 
> > netbsd-6.
> 
> I've tried the patch on NetBSD-6 and it indeed fixed the issue here. Just like
> your previous patch it also introduced a worrying behavior: when i press the
> power button the machine powers down *immediately* and is not calling the
> shutdown hooks, neither in userland nor in the kernel. This used to work fine
> but i am not sure it is related to this PR.

when did this work previously for you?  it works as expected on one of my 
systems
with netbsd-6 and -current.  could you try it again with some older kernels
to see if you can find the point when it stopped working?


> Since i dont know ACPI that intimately, y working hypothesis is if i press the
> power button, the ACPI subsystem is passing that interrupt to NetBSD and it is
> not processed correctly it is shutting down the machine hard immediately. So
> are the ACPI interrupts reaching NetBSD?

I don't know exactly how the power button stuff works either.  my understanding
is that the OS registers with ACPI somehow to receive power button
notifications, and pressing the power button results in the OS being notified
if it has registered, or in the power shutting off immediately if the OS has
not registered.  it sounds like this registration isn't working, so that
the ACPI and lower-level stuff doesn't know that the OS is interested in
doing something special for power button events.

if the system were trying to notify the OS of a power button event and we just
weren't getting the message, I would expect the result to be that nothing 
happens,
not that the power shuts off immediately.

-Chuck