On Thu, Dec 20, 2007 at 03:57:16PM -0500, Jared D. McNeill wrote:
> On Thu, 20 Dec 2007, Bernd Ernesti wrote:
> >The kernel things it knows all power state and what it has to do and
> >you can't disable this behaviour?
> 
> Running the userspace power service, powerd(8), overrides the kernel's 
> behaviour.

Thats maybe nice, but is very unexpected. I wouldn't expect that the
system powers itself down if powerd is not running.

Was there a discussion to add such a 'kernel' feature and where?

> >This seeems to be very wrong. That code should be disabled by default.
> >At least for now, given all the acpi problems. Later we can add a switch
> >to disable that kernel interface and don't get into any acpi trouble.
> 
> All the ACPI problems? Give me a break, we just finished a major rototill 
> on a _development branch_ of NetBSD and had a ~1 week window where a small 
> number of machines had issues.
> 
> Can you point to a specific issue that causes invalid shutdown 
> notifications on -current that is caused by an ACPI problem?

We are in the middle of this thread.

Yes, I know that the branch was just merged and that there are problems
to be expected to show up. And btw, didn't this kind of problems happend
before on the branch, so this is not a problem of the merge itself.

> >I wouldn't trust such a system where the kernel decides to issue a shutdown
> >because of a wrong event.
> 
> I wouldn't trust a system that would rather trash my filesystem than 
> gracefully shutdown when it detects a low power condition.

You are talking about a notebook, but I see it from a server side where
such an action can't and shouldn't happen. At least we don't have the
infrastructure to deal with dual power supplies and the failure for of one
it, where this should't primary be decided in the kernel, because it is
not flexible enough and therefor not a good default action.

Bernd