On Tue, May 17, 2011 at 4:43 PM, Rafael J. Wysocki <rjw@xxxxxxx> wrote:
> On Tuesday, May 17, 2011, Dwight Schauer wrote:
>> On Tue, May 17, 2011 at 3:03 PM, Rafael J. Wysocki <rjw@xxxxxxx> wrote:
>> > On Tuesday, May 17, 2011, Rafael J. Wysocki wrote:
>> >> On Tuesday, May 17, 2011, Dwight Schauer wrote:
>> >> > >> > >> >> >> >> On Thursday, May 12, 2011, Alan Stern wrote:
>> >> > >> >
>> >> > >> > >> >> >> >> > For new readers: The problem is that an xHCI USB host controller does
>> >> > >> > >> >> >> >> > not wake up a suspended system properly.
>> >> > >> > ...
>> >> > >> > >> > So, clearly, you don't get any PCIe PME interrupts from root ports
>> >> > >> > >> > when the keyboard is plugged in. Without those interrupts the runtime
>> >> > >> > >> > resume of xhci won't work.
>> >> > >> > >> >
>> >> > >> > >> > Please attach the output of "lspci -vv" with "auto" in the (suspended) xhci's
>> >> > >> > >> > power/control file before and after you've plugged in the keyboard.
>> >> > >> > >> >
>> >> > >> > >> > Thanks,
>> >> > >> > >> > Rafael
>> >> > >> > >>
>> >> > >> > >> The lspci -vv before, after, and diff are attached.
>> >> > >> > >
>> >> > >> > > This means that the PME signaled by the xHCI doesn't cause the PMEStatus bit
>> >> > >> > > in its root port to be set, which is why the root port doesn't generate
>> >> > >> > > interrupts. This seriously looks like a hardware bug and the only thing
>> >> > >> > > we could do to work around it would be to poll the xHCI for the PME status
>> >> > >> > > periodically (while suspended).
>> >> > >> > >
>> >> > >> > > Can you see if the feature works after booting with pcie_ports=compat in
>> >> > >> > > the kernel command line?
>> >> > >> > >
>> >> > >> > > Rafael
>> >> > >> > >
>> >> > >> >
>> >> > >> > I'll try that on Monday (the pcie_ports=compat kernel option).
>> >> > >> >
>> >> > >> > Well, I've got 2 different systems (one Intel and one AMD based, both
>> >> > >> > exhibit the same behavior).
>> >> > >>
>> >> > >> Are both xHCI controllers from NEC?
>> >> > >>
>> >> > >> > I have a few other systems I can try it on as well on Monday.
>> >> > >>
>> >> > >> Please do if possible.
>> >> > >
>> >> > > If pcie_ports=compat doesn't help, does it help if you use pci=nomsi?
>> >> > > I'm wondering if the hardware bug shows up only when MSI or MSI-X is
>> >> > > enabled for the NEC hardware. Also, if you turn on
>> >> > > CONFIG_USB_XHCI_HCD_DEBUGGING, what NEC firmware version do you see in
>> >> > > dmesg?
>> >> > >
>> >> > > Sarah Sharp
>> >> > >
>> >> >
>> >> > pcie_ports=compat had no effect
>> >> > with pci_=nomsi I get upon boot.
>> >> >
>> >> > xhci_hcd 0000:02:00.0: Failed to enable MSI-X
>> >> > xhci_hcd 0000:02:00.0: failed to allocate MSI entry
>> >> > xhci_hcd 0000:06:00.0: Failed to enable MSI-X
>> >> > xhci_hcd 0000:06:00.0: failed to allocate MSI entry
>> >> >
>> >> > The 6:00.0 is for NEC device in the PCIe slot.
>> >> >
>> >> > With pci_=nomsi the rutime power management works, I can put auto in
>> >> > power/control and the device comes back on it's own.
>> >> >
>> >> > However, this does not address the wakeup from system suspend issue,
>> >> > enabled in power/wakeup still does not work. If the interrupt is now
>> >> > being polled, that would make sense as to why it still does not work.
>> >> >
>> >> > For TI's xHCI device I get "xhci_hcd 0000:02:00.0: failed to allocate
>> >> > MSI entry" with pci=nomsi.
>> >> >
>> >> > The system with TI's device in it is exhibiting the same symptoms
>> >> > (wakeup does not work and by default auto in power/control does not
>> >> > work). It is a different motherboard and chip-set than the system I
>> >> > have the NEC device in.
>> >> >
>> >> > pci=nomsi has no effect on the system with the TI xHCI device in it.
>> >> > pcie_ports=compat had no effect either as far as making auto in
>> >> > power/control work properly.
>> >> >
>> >> >
>> >> > As to the NEC firmware, both the on-board and off-board NEC devices
>> >> > have the same version:
>> >> >
>> >> > xhci_hcd 0000:02:00.0: NEC firmware version 30.21
>> >> > xhci_hcd 0000:06:00.0: NEC firmware version 30.21
>> >> >
>> >> > Attached are the dmesg boot logs for both systems.
>> >>
>> >> OK, so for the S3 issue, I think there's a problem with the PCIe link
>> >> on these devices such that the link is not present during resume
>> >> and that's why commands are not reaching the device. I'm not sure yet
>> >> what to do about it, but I wonder if xhci_hcd does something around
>> >> PCIe links on suspend?
>> >
>> > It doesn't seem so.
>> >
>> > Dwight, can you comment out the pci_disable_device(pci_dev); in
>> > drivers/usb/core/hcd-pci.c and see if that makes a difference?
>> >
>> > Rafael
>>
>> I just tried that, but it has no effect on wakeup from S3.
>
> OK, one more test, please.
>
> Try to do
>
> # echo core > /sys/power/pm_test
> # echo mem > /sys/power/state
>
> (that should simulate suspend, but without going into the BIOS, and it
> should return do the command prompt after 5-10 sec.) and check if the
> USB3 controllers work after that ("echo none > /sys/power/pm_test" resets
> to the normal suspend behavior).
>
> Thanks,
> Rafael
No problem.
The simulated suspend works fine. Also, waking up from S3 via a PS/2
keyboard works fine.
Dwight
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
