On Wed, Aug 28, 2019 at 10:34 AM Daniel Drake [off-list ref] wrote:

On Tue, Aug 27, 2019 at 3:48 PM Rafael J. Wysocki [off-list ref] wrote:

quoted

That depends on what exactly happens when you try to do the D0-D3-D0
with setpci.  If the device becomes unreachable (or worse) after that,
it indicates a platform issue.  It should not do any harm at the
least.

However, in principle D0-D3-D0 at the PCI level alone may not be
sufficient, because ACPI may need to be involved.

After using setpci to do D0-D3-D0 transitions, the xhci module fails to probe.

  xhci_hcd 0000:03:00.3: WARN: xHC restore state timeout
  xhci_hcd 0000:03:00.3: PCI post-resume error -110!

But maybe it's not a great test; as you say I'm not involving ACPI,
and also if Linux has a reason for not runtime suspending PCI devices
without drivers present then maybe I should also not be doing this.

quoted

I think that PM-runtime should suspend XHCI controllers without
anything on the bus under them, so I wonder what happens if
".../power/control" is set to "on" and then to "auto" for that device,
with the driver loaded.

Good hint.

# echo on > /sys/bus/pci/devices/0000\:03\:00.3/power/control
# echo auto > /sys/bus/pci/devices/0000\:03\:00.3/power/control
# echo 1 > /sys/bus/usb/devices/1-4/remove
# cat /sys/bus/pci/devices/0000\:03\:00.3/power/runtime_status
suspended
# echo on > /sys/bus/pci/devices/0000\:03\:00.3/power/control

The final command there triggers these messages (including a printk I
added in pci_raw_set_power_state):
 xhci_hcd 0000:03:00.3: pci_raw_set_power_state from 3 to 0
 xhci_hcd 0000:03:00.3: Refused to change power state, currently in D3
 xhci_hcd 0000:03:00.3: pci_raw_set_power_state from 3 to 0
 xhci_hcd 0000:03:00.3: enabling device (0000 -> 0002)
 xhci_hcd 0000:03:00.3: WARN: xHC restore state timeout
 xhci_hcd 0000:03:00.3: PCI post-resume error -110!
 xhci_hcd 0000:03:00.3: HC died; cleaning up

So we just reproduced the same issue using runtime PM, without having
to go through the whole suspend path.

I guess that points towards a platform issue, although the weird thing
is that Windows presumably does the D3-D0-D3 transition during
suspend/resume and that appears to work fine.

It looks like the platform expects the OS to do something that our
generic XHCI driver and the PCI/ACPI layer don't do.

A quirk or similar may be needed to address that.

I'll report it to the vendor,

Yes, please.  At least try to get the information on what the exact
platform expectations with respect to the OS are.  Quite evidently,
they aren't just "do the usual thing".

but if you have any other debug ideas they would be much appreciated.

With the git branch mentioned previously merged in, you can enable
dynamic debug in device_pm.c, repeat the PM-runtime test and collect
the log.  There should be some additional messages from the ACPI layer
in it.

Daniel