Re: [PATCH 1/1] x86/platform/intel-mid: Retrofit pci_platform_pm_ops ->get_power hook

[Lukas Wunner <lukas@wunner.de>]

On Sun, Oct 09, 2016 at 03:26:37PM +0300, Andy Shevchenko wrote:
> On Sat, 2016-10-08 at 15:49 +0200, Lukas Wunner wrote:
> >   I'm not familiar at all with Intel MID devices, do
> > you have a way to clearly identify if a PCI device is power managed
> > by the PWRMU versus PMCSR?  My guess is that at the very least,
> > intel_mid_pwr_get_lss_id() needs to be called and false needs to
> > be returned by mid_pci_power_manageable() if the return value is
> > negative.
> 
> PCI bus is kinda fake on those platforms (which implement SFI) and don't
> always follow [PCI] specification. The vendor register represents so
> called Logical Subsystem ID of the device in question. Some of them are
> related to PWRMU, some to P-Unit, some just has no such register.
> 
> In PWRMU/P-Unit cases PMCSR is present and writing to it is needed.
> For the rest I don't remember what is actual state, perhaps writing is
> just ignored and OS reads D0 all the time from it.

For devices with PM mechanisms not covered by the standard PCI code
(or ACPI platform code), a common pattern is to assign them a struct
dev_pm_domain using dev_pm_domain_set().  Then on suspend you'd
invoke the PCI bus suspend hook and afterwards ask the PWRMU to
power down.  On resume you'd first power up using the PWRMU, then
call the PCI bus resume hook.  See drivers/gpu/vga/vga_switcheroo.c:
vga_switcheroo_init_domain_pm_ops() for an example.  This would have
been an alternative approach to introducing a new PCI platform for
these devices.  I'm not sure which approach is more suitable, it's
just something that crossed my mind when looking at this.

Best regards,

Lukas