Hi, tl;dr : Should device allowed to do runtime pm that have no driver go to D3 on their own ? And should pci root bridge go to D3 when all the device on them are on D3 ? The whole story: Modern kernel ( > 3.15 ) report "Windows 2013" as an OSI string, even newer ones report "Windows 2015" as well. This triggers some changes in the ACPI code of the T440s laptop (and in others as well, but this is the one I'm using as reference here). This changes breaks the method used by Bumbblebee's bbswitch module and by the nouveau driver to shutdown the card (to save power). No error is reported, and some power is saved, but not as much as previously (hence users might not notice at all). The usual "solution" given to users is to use acpi_osi="!Windows 2013" (and now they'll have to add acpi_osi="!Windows 2015" as well), but I was interested to find out exactly what was causing that behavior and whathttps://bugs.freedesktop.org/attachment.cgi?id=117267 new method Win 8.1 and Win 10 were using to shutdown the card. The DSDT of the T440s running the latest 2.33 bios can be found here : https://bugs.freedesktop.org/attachment.cgi?id=117267 The current method used to shutdown the card relies on two actions : - Making a _DSM call to the video acpi node, this will prepare the shutdown and instructs ACPI to do the actual power down on the next device state transition to D3. This is just preparaton and just sets some variable but doesn't actually do the power off. - Actually make the device go to D3 with pci_set_power_state. This then executes the _PS3 method in ACPI and if the proper variable have been set by the DSM above, this will call the GPOF method which will turn off power. I looked at what changed when "Windows 2013" was present vs "Windows 2012". And here's what's triggered : - OSYS is set to 0x07DD. - This in turn cause VMSH to be set to 1. - This in turn causes _SB_.PCI0.PEG_.VID_.PS3 to _NOT call GPOF when the device is set to D3. - and so the card is never really turned off completely Now I looked at what other path could be taken to reach the GPOF method. And if you look at the table you can see that it can be called as part of the NVP3 power resource which is PR3 ... but on the node _SB.PCI0.PEG_ and not _SB.PCI0.PEG_.VID_ ! So the GPU gets turned off when the PCIe bridge/port it's connected to gets transitionned to D3hot. I did confirm this by manually patching bbswitch to put the pci_upstream_bridge(...) of the device to D3hot after putting the card itself in D3cold. Power saving measured matches what's expected, but I have no idea if it's the "right" thing to do. The DSM call doesn't seem to be required now, so everything is done with pure PM state which I guess is good. What I expected is that now that everything is purely PM, then linux should be able to do all of this automatically without the need for bbswitch at all. If no driver is loaded for the card (so obviously no users), then doing an "echo auto > .../power/control" on both the card node and the bridge port node it's attached to should allow runtime PM. And shince they're unused, everything should go to D3 on its own. But that doesn't seem to happen ? Are my expectations wrong ? Cheers, Sylvain -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
