2.6.20 poweroff regression on se7525gp2

With 2.6.19.2, init 0 works
with 2.6.20, it causes this box to reset when it should poweroff, and thus it reboots.

any suggestions on where to look first?

thanks,
-Len

Success case:

Linux version 2.6.19.2 (lenb@d975xbx2) (gcc version 4.1.2 20061115 (prerelease) (SUSE Linux)) #5 SMP Sun Feb Command line: root=/dev/sda2 console=tty0 console=ttyS0,115200n8 debug
...

# init 0
...
The system will be halted immediately.
Synchronizing SCSI cache for disk sda:
ACPI: PCI interrupt for device 0000:06:03.0 disabled
ACPI: PCI interrupt for device 0000:05:06.1 disabled
ACPI: PCI interrupt for device 0000:05:06.0 disabled
ACPI: PCI interrupt for device 0000:05:04.1 disabled
ACPI: PCI interrupt for device 0000:05:04.0 disabled
Power down.
acpi_power_off called
 hwsleep-0285 [02] enter_sleep_state     : Entering sleep state [S5]

Failure case:

Linux version 2.6.20 (lenb@d975xbx2) (gcc version 4.1.2 20061115 (prerelease) (SUSE Linux)) #1 SMP Sun Feb Command line: root=/dev/sda2 console=tty0 console=ttyS0,115200n8 debug
...

# init 0
...
The system will be halted immediately.
Synchronizing SCSI cache for disk sda:
ACPI: PCI interrupt for device 0000:06:03.0 disabled
ACPI: PCI interrupt for device 0000:05:06.1 disabled
ACPI: PCI interrupt for device 0000:05:06.0 disabled
ACPI: PCI interrupt for device 0000:05:04.1 disabled
ACPI: PCI interrupt for device 0000:05:04.0 disabled
Power down.
acpi_power_off called
 hwsleep-0285 [02] enter_sleep_state     : Entering sleep state [S5]

Sounds like a regression of patch from linus

ACPI: Allow setting SCI_EN bit in PM1_CONTROL register
Linus Torvalds [Wed, 11 Oct 2006 00:14:44 +0000 (17:14 -0700)] This is needed by at least the Mac Mini's, which (incorrectly) come back from suspend with SCI_EN clear.

Turns out that USB broke poweroff:
http://bugzilla.kernel.org/show_bug.cgi?id=7945

-Len