Re: ACPI "Soft-off" power button only rebooting system, not powering off

[Simon Farnsworth <simon@farnz.org.uk>]

Stephen Clark wrote:
> Pavel Machek wrote:
>> On Wed 2008-09-17 08:33:50, Andrew Paprocki wrote:
>>> On Wed, Sep 17, 2008 at 6:18 AM, Pavel Machek <pavel@suse.cz> wrote:
>>>> On Wed 2008-09-17 00:50:03, Andrew Paprocki wrote:
>>>>> I have an AMD / Award BIOS based system which does not properly shut
>>>>> off when I hold down the power button for 4 seconds. The BIOS is
>>>>> configured to have "Soft-off" set to "Delay for 4 seconds" which is
>>>>> supposed to power the machine off if the button is held that long.
>>>>>
>>>>> I first thought this could be a BIOS bug, but this *only* appears to
>>>>> happen while Linux is running. If the computer is running WinXP, it
>>>> Well, so the BIOS is broken. 4second hold should power down the
>>>> machine regardless of the operating system.
>>> Yes, I agree.. I just find it strange that it works on the BIOS screen
>>> as well as in GRUB, but as soon as Linux boots, it no longer works.
>>> Since it works under WinXP, there must be some way to get it to work
>>> properly from the OS side even it if means working around the bug
>>> somehow.
>>
>> Try noapic/nolapic/nosmp etc... and ask your vendor to fix the bios ;-).
>>
> Why do you say it is the BIOS when it shuts down properly from the GRUB
> prompt or from in WINXP?
> 
Because it's only the BIOS that can cause this behaviour.

When you push the power button, the BIOS is invoked. If you release
within 4 seconds, the BIOS sends an ACPI event to the running OS,
telling it that the power button has been pushed.

If you hold it down for 4 or more seconds, the BIOS is supposed to shut
the machine down without the OS knowing what's happening (no ACPI
events, just a loss of power). From your problem description, the BIOS
is noticing *something* about the way Linux sets things up, and
*choosing* to reboot instead of cutting the power.

GRUB doesn't touch hardware without BIOS help, so is unlikely to change
the system state in a way that tickles this BIOS bug; similarly, BIOSes
are tested against current Windows versions by vendors, so they'll have
hacked it around until it worked there.
-- 
Hope this helps,

Simon Farnsworth