kern/162578: 9.0-RC2's regression in power management on VIA Samuel 2

[John Baldwin]

On Wednesday, November 23, 2011 4:09:25 pm kron wrote:
> On 2011/11/23 07:57, kron wrote:
> > Hello,
> >
> > I'm bringing this to -acpi@ as suggested by jhb at .
> >
> > Some time ago while testing 9.0-RC2 I noticed that power management
> > got broken (powerd doesn't start, Cx states disappeared) on a specific
> > class of our minirouters. I created kern/162578, bisected the issue
> > down to r216674 and I contacted the author - jhb at . John was kind to do
> > a further analysis. Verbose boots before and after r216674 differ this
> > way:
> >
> > -Calibrating TSC clock ... TSC clock: 532647138 Hz
> > +Calibrating TSC clock ... TSC clock: 532648183 Hz
> >
> > -ACPI timer: 0/4 0/5 0/4 0/5 0/4 0/5 0/4 0/5 0/4 0/4 -> 0
> > -Timecounter "ACPI-safe" frequency 3579545 Hz quality 850
> > -acpi_timer0: <24-bit timer at 3.579545MHz> port 0x4008-0x400b on acpi0
> > +acpi_timer0: couldn't allocate resource (port 0x4008)
> >
> > -acpi_throttle0: P_CNT from P_BLK 0x4010
> > +acpi_throttle0: failed to attach P_CNT
> > +device_attach: acpi_throttle0 attach returned 6
> >
> > John's comment:
> >  > So this is the issue, and it seems what happens is that your
> >  > BIOS assigns the resources for this range to the pcib0 device
> >  > when we expect them to be assigned as a system resource (if
> >  > at all):
> >
> >  > pcib0: <ACPI Host-PCI bridge> port
> >  > 0xcf8-0xcff,0x4000-0x407f,0x4080-0x40ff,0x5000-0x500f,0x6000-0x607f
> >  > on acpi0
> >
> >  > You could try hacking your ASL to not list the 0x4000-0x407f range
> >  > for now, but the real fix is probably to make resources attached
> >  > to Host-PCI bridge devices be treated as if they were system
> >  > resources and put into the ACPI system resource rman instead.
> >  > That requires a fair bit of work however.
> >
> > John also suggested to involve jkim@ and -acpi at .
> >
> > I'm going to experiment with ASL because it would be an acceptable
> > solution to me and the real fix is way above my skills ATM.
> 
> As promised, I fiddled with ASL. I did found something which
> resembled 0x4000-0x407f:
> ...
> DefinitionBlock ("/tmp/acpidump.aml", "DSDT", 1, "VIA601", "AWRDACPI", 
> 0x00001000)
> {
>      ....
>      Scope (\_SB)
>      {
>          ...
>          Device (PCI0)
>          {
>              ...
>              Method (_CRS, 0, NotSerialized)
>              {
>                  ...
>                  Name (BUF0, ResourceTemplate ()
>                  {
>                      ...
>                      IO (Decode16,
>                          0x4000,             // Range Minimum
>                          0x4000,             // Range Maximum
>                          0x01,               // Alignment
>                          0x80,               // Length
>                          )
>                      ...
> I removed the IO block, compiled, installed the AML, rebooted
> and voila - I have my power management back even with r216674.
> 
> Big thanks to jhb@ for guiding me.
> 
> As always, big thanks to all the good souls who write the
> Handbook, too. It helped me with ASL and AML.
> 
> As
> 1. this hack is good enough for me, and
> 2. the real fix John suggested is above my skills
> I think the PR can be closed.

Can you try an updated HEAD and see if a stock ASL works?

-- 
John Baldwin