Re: [coreboot] Powersavings: 8W of difference between bios and coreboot

[Felix Held]

Hi!
> I don't know if Charlotte has added the ID of the dGPU to 
> src/drivers/lenovo/hybrid_graphics.c. Does the dGPU consume power 
> after hybrid_graphics.c disable the dGPU?

OPTIMUS_ENABLE is a PCH GPIO and controls the muxes that select if the 
internal display is connected to the iGPU or the dGPU; that's done in 
hybrid_graphics.c.
GFXCORE_ON_D (and another signal that controls the power supply of the 
VRAM) are driven by the Thinker-1 chip; those switch on/off the power 
supply of the GPU. So to really disable the GPU you probably have to ask 
the EC to make that chip turn off the voltage for the GPU and VRAM.

Regards
Felix

Powersavings: 8W of difference between bios and coreboot

[Charlotte Plusplus]

Hello

Super interesting, I didn't know all that!

Currently, I have only set in devicetree:

device pci 00.0 on end # host bridge
device pci 01.0 off end # NVidia
device pci 02.0 on end # Intel

and in my nvram options I have:
hybrid_graphics_mode = Integrated Only

I assumed that would be sufficient to turn off the power for the dGPU until
I figure out a way to make it work. I will enable the NVidia and do more
power tests.

After reading hybrid_graphics.c, I understand a bit more, but I also have
more questions:
 - can I add register "pcie_hotplug_map" = "{ 0, 1, 0}" to make the NVidia
removable?
(so that the operating system will not freak out when the NVidia disappears
from the PCI bus if I find a way to control the power by talking to the EC
and send
GFXCORE_ON_D + the other signal for the VRAM)

- can you change the connection of the displayport? Based on the
specsheets, it is connected to the dGPU, while the internal display is
connected to the iGPU. If is it possible to control the muxes for the
internal display, I suppose it is possible for the other displays as well.
(I have just tested and I do not have any video on the displayport, and
xrandr does not detect anything)

- there seem to be some missing IDs in pci_device_ids_nvidia : cf
http://envytools.readthedocs.io/en/latest/hw/pciid.html which agrees with
the W530 : 0x0ffb, so I will  propose a patch:
                0x0dfa, /* Nvidia NVS Quadro 1000m Lenovo W520 */
                0x0ffb, /* Nvidia NVS Quadro K1000m Lenovo W530
*/
                0x0ffc, /* Nvidia NVS Quadro K2000m Lenovo W530 */

It may also be the reason why the Nvidia is still getting power, as Iru
noted that hybrid_graphics should turn off the power. I will test that
separately.

- until I can find a better solution, I am thinking of letting the Nvidia
show on on the PCIe bus and then sending commands to get in into advanced
sleep - like on https://wireless.wiki.kernel.org/en/users/documentation/aspm

It should be possible as the w530 lspci -v shows:
        Capabilities: [60] Power Management version 3
                Flags: PMEClk- DSI- D1- D2- AuxCurrent=0mA
PME(D0-,D1-,D2-,D3hot-,D3cold-)
                Status: D0 NoSoftRst+ PME-Enable- DSel=0 DScale=0 PME-

Would you have a better idea?

Thanks
Charlotte

On Fri, Nov 18, 2016 at 4:52 AM, Felix Held 
wrote:
> Hi!
>
> I don't know if Charlotte has added the ID of the dGPU to
> > src/drivers/lenovo/hybrid_graphics.c. Does the dGPU consume power after
> > hybrid_graphics.c disable the dGPU?
>
> OPTIMUS_ENABLE is a PCH GPIO and controls the muxes that select if the
> internal display is connected to the iGPU or the dGPU; that's done in
> hybrid_graphics.c.
> GFXCORE_ON_D (and another signal that controls the power supply of the
> VRAM) are driven by the Thinker-1 chip; those switch on/off the power
> supply of the GPU. So to really disable the GPU you probably have to ask
> the EC to make that chip turn off the voltage for the GPU and VRAM.
>
> Regards
> Felix