Re: [PATCH 1/2] vga_switcheroo: add power support for windows 10 machines.

From: Dave Airlie
Date: Wed Mar 09 2016 - 17:02:32 EST

On 10 March 2016 at 06:17, Lukas Wunner <lukas@xxxxxxxxx> wrote:
> Hi,
> On Wed, Mar 09, 2016 at 11:52:33AM -0500, Alex Deucher wrote:
>> On Wed, Mar 9, 2016 at 9:33 AM, Lukas Wunner <lukas@xxxxxxxxx> wrote:
>> > On Wed, Mar 09, 2016 at 04:14:04PM +1000, Dave Airlie wrote:
>> >> From: Dave Airlie <airlied@xxxxxxxxxx>
>> >>
>> >> Windows 10 seems to have standardised power control for the
>> >> optimus/powerxpress laptops using PR3 power resource hooks.
>> >
>> > What happened to the Optimus DSM, does this still work? If not,
>> > echoing OFF to the vgaswitcheroo sysfs file won't power down the
>> > GPU now, right? (If runtime pm is disabled in nouveau.)
>> >
>> If the OS identifies itself as windows 10 or newer when initializing
>> ACPI, the standardized interfaces should be used if they exist. I'm
>> not sure if there is any explicit policy on the vendor specific ones,
>> but I suspect they are probably broken in that case since I doubt the
>> OEM validates them when the standardized interfaces are available.
> The vendor interface (Optimus DSM) must be present, otherwise the call
> to nouveau_is_optimus() in patch [2/2] would return false.
> But indeed it seems to not be working, otherwise why would the patches
> be necessary?
> My point is that the chosen approach does not square with vga_switcheroo's
> architecture: Normally it's the handler's job to power the GPU up/down.
> If runtime pm is enabled then vga_switcheroo_init_domain_pm_ops()
> activates runtime_suspend/_resume callbacks which ask the handler to
> flip the power switch.
> However these two patches add *additional* runtime_suspend/_resume
> callbacks which do not rely on the handler at all. The handler is thus
> useless. This becomes apparent when loading the nouveau module with
> runpm=0: Normally the user is then able to manually power the GPU
> up/down by echoing ON or OFF to the vgaswitcheroo sysfs file. With the
> chosen approach this will likely not work because the handler only knows
> how to invoke the DSM, it doesn't know anything about PR3.
> Hence my suggestion to solve this with an additional handler which
> gets used in lieu of the nouveau DSM handler.

I'll think about this a bit more, but really I don't care for vgaswitcheroo
manual power control anymore. It's in debugfs for a reason, it was a stopgap
until we got dynamic power control.

If dynamic power control isn't working for some people we should fix that,
but supporting nouveau.runpm=0 and manual power control is so far down
the list of things I care about.