Tobias Diedrich wrote:

So my BIOS is not as borked as I thought and it should be possible
to wake up the machine even with platform.  Further debugging will
have to wait until at least next weekend though (maybe longer)...

Or maybe it doesn't have to wait, I was just too curious:

Summary first: I got platform mode to work!

After grepping and reading through kernel/power/disk.c and
(rather obfuscated) drivers/acpi/.* code, and reading up on
ACPI _GPE (General Purpose Event?), and having a look at my DSDT I
noticed two things:

1) The network controllers are assigned to their own _GPE bits(pins?):
|[...]
|    Scope (\_GPE)
|    {
|[...]
|        Method (_L0B, 0, NotSerialized)
|        {
|            Notify (\_SB.PCI0.MMAC, 0x02)
|        }
|
|        Method (_L0A, 0, NotSerialized)
|        {
|            Notify (\_SB.PCI0.MAC1, 0x02)
|        }
|[...]

2) drivers/acpi/sleep/proc.c registers a 'wakeup' file:
|        proc_create("wakeup", S_IFREG | S_IRUGO | S_IWUSR,
|                    acpi_root_dir, &acpi_system_wakeup_device_fops);

And I then remembered that someone said in
http://bugzilla.kernel.org/show_bug.cgi?id=8381
it works for him if he writes $MAGICVALUE into a proc file.

And yes, if I write 'MMAC' and 'MAC1' into /proc/acpi/wakeup, then
wake-on-lan works even in platform mode.

So...
AFAICS this bit of setup magic should not be required, because:

1) /proc/acpi/wakeup knows which pci device is associated to each GPE bit
|ranma@melchior:~$ cat /proc/acpi/wakeup 
|Device	S-state	  Status   Sysfs node
|HUB0	  S5	 disabled  pci:0000:00:06.0
|XVR0	  S5	 disabled  
|XVR1	  S5	 disabled  pci:0000:00:0e.0
|XVR2	  S5	 disabled  
|XVR3	  S5	 disabled  
|XVR4	  S5	 disabled  
|XVR5	  S5	 disabled  pci:0000:00:0a.0
|UAR1	  S5	 disabled  pnp:00:09
|PS2K	  S4	 disabled  pnp:00:0b
|USB0	  S4	 disabled  pci:0000:00:02.0
|USB2	  S4	 disabled  pci:0000:00:02.1
|AZAD	  S5	 disabled  pci:0000:00:06.1
|MMAC	  S5	 enabled   pci:0000:00:08.0
|MAC1	  S5	 enabled   pci:0000:00:09.0
(values after manually enabling MMAC and MAC1)

2) kernel/power/disk.c calls hibernation_ops->enter(), which is
acpi_suspend_enter, which calls acpi_enable_wakeup_device, which
sets up GPE wakup bits.  This _should_ take care of enabling MMAC
and MAC1 automatically, but apparently does not work correctly at
some point.

I guess someone more knowledgable in ACPI stuff should have a look
at this.

Hope that helps,

-- 
Tobias						PGP: http://9ac7e0bc.uguu.de
このメールは十割再利用されたビットで作られています。