Hi On 5.10.2022 23.23, Mario Limonciello wrote:
For optimal power consumption of USB4 routers the XHCI PCIe endpoint used for tunneling must be in D3. Historically this is accomplished by a long list of PCIe IDs that correspond to these endpoints. The linux thunderbolt CM currently uses the `usb4-host-interface` ACPI property to create a device link between the USB4 host router PCIe endpoint and the XHCI PCIe endpoint. The device link will move the devices in out of D3 together. To avoid having to maintain this never ending list of PCIe IDs, use the existence of `usb4-host-interface` property on a USB port as a proxy to allow runtime PM for these controllers. The device links will continue to be created when the CM initializes the USB4 host router and also discovers this property.
Agree that maintaining this list isn't a long term solution.
Link: https://learn.microsoft.com/en-us/windows-hardware/design/component-guidelines/usb4-acpi-requirements#port-mapping-_dsd-for-usb-3x-and-pcie Signed-off-by: Mario Limonciello <mario.limonciello@xxxxxxx> --- RFC v1->PATCH v1 * Move this detection from Thunderbolt CM into USB core --- drivers/usb/core/usb-acpi.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/drivers/usb/core/usb-acpi.c b/drivers/usb/core/usb-acpi.c index 6d93428432f13..f91ab4fd84cf8 100644 --- a/drivers/usb/core/usb-acpi.c +++ b/drivers/usb/core/usb-acpi.c @@ -177,6 +177,15 @@ usb_acpi_find_companion_for_port(struct usb_port *port_dev) port_dev->connect_type = usb_acpi_get_connect_type(handle, pld); ACPI_FREE(pld); } + if (!acpi_dev_get_property(adev, "usb4-host-interface", + ACPI_TYPE_ANY, NULL)) { + struct device *dev = &port_dev->dev; + + while (dev && !dev_is_pci(dev)) + dev = dev->parent; + if (dev) + pm_runtime_allow(dev);
This would enable runtime pm for usb hosts during usb_acpi_bus acpi companion finding. I think host drivers should be the ones making this decision. Maybe it's time to enable runtime pm as default for most new xHC controllers. How about we enable it for all AMD and Intel PCI xHC hosts with version 1.2 or later? If it causes un-fixable regression to some controller we can add it to a denylist. Thanks -Mathias
