Re: [PATCH] x86/PCI: Remove D0 PME capability on AMD FCH xHCI

[Bjorn Helgaas <helgaas@kernel.org>]

On Wed, Oct 02, 2019 at 01:32:07PM +0800, Kai-Heng Feng wrote:
> On Oct 2, 2019, at 08:07, Bjorn Helgaas <helgaas@kernel.org> wrote:
> > On Mon, Sep 02, 2019 at 10:52:52PM +0800, Kai-Heng Feng wrote:
> >> There's an xHCI device that doesn't wake when a USB 2.0 device gets
> >> plugged to its USB 3.0 port. The driver's own runtime suspend callback
> >> was called, PME# signaling was enabled, but it stays at PCI D0:
> >> 
> >> 00:10.0 USB controller [0c03]: Advanced Micro Devices, Inc. [AMD] FCH USB XHCI Controller [1022:7914] (rev 20) (prog-if 30 [XHCI])
> >>        Subsystem: Dell FCH USB XHCI Controller [1028:087e]
> >>        Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx+
> >>        Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
> >>        Interrupt: pin A routed to IRQ 18
> >>        Region 0: Memory at f0b68000 (64-bit, non-prefetchable) [size=8K]
> >>        Capabilities: [50] 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-
> >> 
> >> A PCI device can be runtime suspended while still stays at D0 when it
> >> supports D0 PME# and its ACPI _S0W method reports D0. Though plugging
> >> USB 3.0 devices can wakeup the xHCI, it doesn't respond to USB 2.0
> >> devices.
> > 
> > I don't think _S0W and runtime suspend are relevant here.  What *is*
> > relevant is that the device advertises that it can generate PME from
> > D0, and it apparently does not do so.
> 
> Yes that's the case. It doesn't generate PME when USB2.0 or USB1.1
> device gets plugged.

OK, thanks.  I added a stable tag and applied this to pci/misc for v5.5.

> > Table 10 in the xHCI spec r1.0, sec 4.15.2.3, says the xHC should
> > assert PME# if enabled and the port's WCE bit is set.  Did you ever
> > confirm that WCE is set?
> 
> How do I check WCE when xHCI is suspended?  If I want to read WCE
> then I have the resume the device, but after resuming all USB
> devices get enumerated, and checking WCE doesn't matter anymore.

Yeah, that's a problem.  I guess you'd have to inspect or instrument
the driver to ensure WCE is set in that path.  But if you get PME# for
USB3 devices, it seems safe to assume WCE is set.

> > I assume WCE *is* set because plugging in a USB3 device *does*
> > generate a PME#, and I don't see anything in Table 10 that says it
> > would work for USB3 but not USB2.
> 
> It should work on all USB speeds, but it didn't.
> That's why the OEM/ODM use the _S0W workaround on Windows.
> 
> Kai-Heng
> 
> > 
> >> So let's disable D0 PME capability on this device to avoid the issue.
> >> 
> >> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=203673
> >> Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
> >> ---
> >> arch/x86/pci/fixup.c | 11 +++++++++++
> >> 1 file changed, 11 insertions(+)
> >> 
> >> diff --git a/arch/x86/pci/fixup.c b/arch/x86/pci/fixup.c
> >> index 527e69b12002..0851a05d092f 100644
> >> --- a/arch/x86/pci/fixup.c
> >> +++ b/arch/x86/pci/fixup.c
> >> @@ -588,6 +588,17 @@ static void pci_fixup_amd_ehci_pme(struct pci_dev *dev)
> >> }
> >> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, 0x7808, pci_fixup_amd_ehci_pme);
> >> 
> >> +/*
> >> + * Device [1022:7914]
> >> + * D0 PME# doesn't get asserted when plugging USB 2.0 device.
> >> + */
> >> +static void pci_fixup_amd_fch_xhci_pme(struct pci_dev *dev)
> >> +{
> >> +	dev_info(&dev->dev, "PME# does not work under D0, disabling it\n");
> > 
> > Use pci_info() as in the rest of the file.

Sorry, this was just wrong.  I was assuming this was for
drivers/pci/quirks.c, where dev_info() has been replaced by
pci_info().  But that's not the case for this file.

> >> +	dev->pme_support &= ~(PCI_PM_CAP_PME_D0 >> PCI_PM_CAP_PME_SHIFT);
> >> +}
> >> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, 0x7914, pci_fixup_amd_fch_xhci_pme);
> >> +
> >> /*
> >>  * Apple MacBook Pro: Avoid [mem 0x7fa00000-0x7fbfffff]
> >>  *
> >> -- 
> >> 2.17.1
>