Nouveau - Oct 2016 - Acer Aspire V7-582PG (Haswell, GTX 750M) fails to power off GPU via Power Resources

Hi, Mika
> From: Mika Westerberg [mailto:mika.westerberg at linux.intel.com]
> Subject: Re: Acer Aspire V7-582PG (Haswell, GTX 750M) fails to power off
GPU via Power Resources
> 
> On Thu, Oct 27, 2016 at 06:06:48PM +0200, Peter Wu wrote:
> > On Thu, Oct 27, 2016 at 12:55:08PM +0300, Mika Westerberg wrote:
> > > On Thu, Oct 27, 2016 at 09:42:28AM +0000, Rick Kerkhof wrote:
> > > >    No, there are not. Here is the recursive directory
listing:
> > >
> > > Are you able to try the following patch and send me dmesg (or
attach it
> > > to that bug)? It should show if the ACPI core even tries to add
those
> > > power resources.
> >
> > So Rick has tested this patch now on top of 4.8.4 (mainline fails to
> > boot due to a kbuild issue which I reported elsewhere), but the output
> > is empty. That seems to indicate that flags.power_resources is unset.
> 
> Is it completely empty or is it empty just for RP05? It should print out
> all devices with power resources.
> 
> > Given that _PS3 exists and is indeed a package with some elements, it
> > seems that acpi_extract_power_resources is failing. Note that in the
> > SSDT, the power resource NVP3 was referenced before it was defined,
> > could that result in this enumeration failure? Relevant SSDT excerpt:
> >
> >     Scope (\_SB.PCI0.RP05)
> >     {
> >         Name (_PR3, Package (0x01)  // _PR3: Power Resources for D3hot
> >         {
> >             NVP3
> >         })
> >         // ...
> >     }
> >
> >     PowerResource (NVP3, 0x00, 0x0000)
Looks wrong order to me.

However, _PR3 is a package, for AML opcode that contains PkgLength grammar
primitive, forward reference may be OK (for example Method).
DefPackage := PackageOp PkgLength NumElements PackageElementList
DefMethod := MethodOp PkgLength NameString MethodFlags TermList
As when it is encountered, AML interpreter may only saves entire packaged
object.

I need to test behaviors around Package with qemu but I don't have
environment now.
Can you help to give it a try?
By adding customize an ssdt with the above code putting under root scope,
DefinitionBlock ("ssdt-package.aml", "SSDT", 2, "INTEL
", "PACKAGE ", 0x00003000)
{
    Scope (\)
    {
        Name (_PR3, Package (0x01) { NVP3 })
    }
    PowerResource (NVP3, 0x00, 0x0000) {}
}
Running Windows images on qemu with this ssdt appended "-acpitable
file=ssdt-package.aml".
To see if blue screen can be resulted.

Thanks
Lv
> 
> That and the fact that they come from an SSDT instead of DSDT may cause
> this. However, I'm not expert in ACPICA so adding Bob and Lv if they
> have ideas.
> 
> Bob, Lv, the bug in question is:
https://bugs.freedesktop.org/show_bug.cgi?id=98398

On Sat, Oct 29, 2016 at 12:42:34AM +0000, Zheng, Lv
wrote:
> Hi, Mika
> 
> > From: Mika Westerberg [mailto:mika.westerberg at linux.intel.com]
> > Subject: Re: Acer Aspire V7-582PG (Haswell, GTX 750M) fails to power
off GPU via Power Resources
> > 
> > On Thu, Oct 27, 2016 at 06:06:48PM +0200, Peter Wu wrote:
> > > On Thu, Oct 27, 2016 at 12:55:08PM +0300, Mika Westerberg wrote:
> > > > On Thu, Oct 27, 2016 at 09:42:28AM +0000, Rick Kerkhof
wrote:
> > > > >    No, there are not. Here is the recursive directory
listing:
> > > >
> > > > Are you able to try the following patch and send me dmesg
(or attach it
> > > > to that bug)? It should show if the ACPI core even tries to
add those
> > > > power resources.
> > >
> > > So Rick has tested this patch now on top of 4.8.4 (mainline fails
to
> > > boot due to a kbuild issue which I reported elsewhere), but the
output
> > > is empty. That seems to indicate that flags.power_resources is
unset.
> > 
> > Is it completely empty or is it empty just for RP05? It should print
out
> > all devices with power resources.
> > 
> > > Given that _PS3 exists and is indeed a package with some
elements, it
> > > seems that acpi_extract_power_resources is failing. Note that in
the
> > > SSDT, the power resource NVP3 was referenced before it was
defined,
> > > could that result in this enumeration failure? Relevant SSDT
excerpt:
> > >
> > >     Scope (\_SB.PCI0.RP05)
> > >     {
> > >         Name (_PR3, Package (0x01)  // _PR3: Power Resources for
D3hot
> > >         {
> > >             NVP3
> > >         })
> > >         // ...
> > >     }
> > >
> > >     PowerResource (NVP3, 0x00, 0x0000)
> 
> Looks wrong order to me.
> 
> However, _PR3 is a package, for AML opcode that contains PkgLength grammar
primitive, forward reference may be OK (for example Method).
> DefPackage := PackageOp PkgLength NumElements PackageElementList
> DefMethod := MethodOp PkgLength NameString MethodFlags TermList
> As when it is encountered, AML interpreter may only saves entire packaged
object.
> 
> I need to test behaviors around Package with qemu but I don't have
environment now.
> Can you help to give it a try?
> By adding customize an ssdt with the above code putting under root scope,
> DefinitionBlock ("ssdt-package.aml", "SSDT", 2,
"INTEL ", "PACKAGE ", 0x00003000)
> {
>     Scope (\)
>     {
>         Name (_PR3, Package (0x01) { NVP3 })
>     }
>     PowerResource (NVP3, 0x00, 0x0000) {}
> }
> Running Windows images on qemu with this ssdt appended "-acpitable
file=ssdt-package.aml".
> To see if blue screen can be resulted.
> 
> Thanks
> Lv
Testing this code with Windows 10 gives a BSOD, changing the order of
the PowerResource and Scope does not make a difference. If I take my
previous SSDT and change the ordering of NVP3 definition vs _PR3 use,
there is no change.

Kind regards,
Peter
> > 
> > That and the fact that they come from an SSDT instead of DSDT may
cause
> > this. However, I'm not expert in ACPICA so adding Bob and Lv if
they
> > have ideas.
> > 
> > Bob, Lv, the bug in question is:
https://bugs.freedesktop.org/show_bug.cgi?id=98398