Re: [PATCH] usb: host: ehci: workaround PME bug on AMD EHCI controller

[Kai-Heng Feng <kai.heng.feng@canonical.com>]

On Thu, Jun 15, 2017 at 2:55 AM, Alan Stern <stern@rowland.harvard.edu> wrote:
> On Tue, 13 Jun 2017, Bjorn Helgaas wrote:
>
>> [+cc Rafael, linux-pm]
>>
>> On Tue, Jun 13, 2017 at 12:21:15PM +0800, Kai-Heng Feng wrote:
>> > On Mon, Jun 12, 2017 at 10:18 PM, Alan Stern <stern@rowland.harvard.edu> wrote:
>> > > Let's get some help from people who understand PCI well.
>> > >
>> > > Here's the general problem: Kai-Heng has a PCI-based USB host
>> > > controller that advertises wakeup capability from D3, but it doesn't
>> > > assert PME# from D3 when it should.  For "lspci -vv" output, see
>> > >
>> > >         http://marc.info/?l=linux-usb&m=149570231732519&w=2
>> > >
>> > > On Mon, 12 Jun 2017, Kai-Heng Feng wrote:
>> > >
>> > >> On Mon, Jun 12, 2017 at 3:04 PM, Kai-Heng Feng
>> > >> <kai.heng.feng@canonical.com> wrote:
>> > >> > On Fri, Jun 9, 2017 at 10:43 PM, Alan Stern <stern@rowland.harvard.edu> wrote:
>> > >> >> On Fri, 9 Jun 2017, Kai-Heng Feng wrote:
>> > >> >>
>> > >> >> Is this really the right solution?  Maybe it would be better to allow
>> > >> >> the controller to go into D3 provided no wakeup signal is needed.  You
>> > >> >> could do:
>> > >> >>
>> > >> >>                 device_set_wakeup_capable(&pdev->dev, 0);
>> > >> >
>> > >> > This doesn't work.
>> > >> > After applying this function, still nothing happens when devices get plugged in.
>> > >> > IIUC this function disable the wakeup function, but what I want to do
>> > >> > here is to have PME signal works even when runtime PM is enabled.
>> > >
>> > > This may indicate a bug in either the PCI or USB stacks (or both!).  If
>> > > a driver requires wakeup capability from runtime suspend but the device
>> > > does not provide it, the PCI core should not allow the device to go
>> > > into runtime suspend.  Or is that the driver's responsibility?
>> > >
>> > >> > I also saw some legacy PCI PM stuff, so I also tried:
>> > >> > device_set_wakeup_capable(&pdev->dev, 1);
>> > >> > ...doesn't work either.
>> > >> >
>> > >> >>
>> > >> >> Another alternative is to put the controller into D2 instead of D3, but
>> > >> >> (1) I don't know how to do that, and (2) we don't know if wakeup
>> > >> >> signalling works any better in D2 than it does in D3.
>> > >> >
>> > >> > I'll try if D2 works.
>> > >>
>> > >> Put the device into D2 instead of D3 can make the wakeup signaling
>> > >> work, i.e. USB devices can be correctly detected after plugged into
>> > >> EHCI port.
>> > >>
>> > >> Do you think this alternative an acceptable workaround?
>> > >
>> > > Yes, it is.  The difficulty is that I don't know how to tell the PCI
>> > > core that the device should go in D2 during runtime suspend instead of
>> > > D3.  Some sort of quirk may be needed -- perhaps Bjorn can help.
>
>> The lspci output [1] shows:
>>
>>   00:12.0 USB controller: Advanced Micro Devices, Inc. [AMD] FCH USB EHCI Controller (rev 39) (prog-if 20 [EHCI])
>>     Capabilities: [c0] Power Management version 2
>>       Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
>>       Status: D3 NoSoftRst- PME-Enable+ DSel=0 DScale=0 PME-
>>       Bridge: PM- B3+
>>
>> The device claims it can assert PME# from D3hot.  If it can't, that
>> sounds like a hardware defect that should be addressed with a quirk.
>> Ideally we would also have a pointer to the AMD hardware erratum.
>>
>> Is the following path involved here?
>>
>>   pci_finish_runtime_suspend
>>     target_state = pci_target_state()
>>       if (device_may_wakup())
>>       if (dev->pme_support)
>>         ...
>>     pci_set_power_state(..., target_state)
>>
>> If so, I would naively expect that a quirk could clear the
>> PCI_PM_CAP_PME_D3 and PCI_PM_CAP_PME_D3cold bits in dev->pme_support,
>> and pci_target_state() would then avoid selecting D3 or D3cold.  But
>> I'm not an expert in power management.
>
> That's a good idea.  However, we should apply the quirk only when it is
> needed.  Which means we need to know the numeric values for the PCI
> IDs.  Also, this will help searching for published errata.
>
> Kai-Heng, what does "lspci -nvs 00:12.0" show?

00:12.0 0c03: 1022:7808 (rev 39) (prog-if 20 [EHCI])
        Subsystem: 1028:0732
        Flags: bus master, 66MHz, medium devsel, latency 32, IRQ 18
        Memory at fe769000 (32-bit, non-prefetchable) [size=256]
        Capabilities: [c0] Power Management version 2
        Capabilities: [e4] Debug port: BAR=1 offset=00e0
        Kernel driver in use: ehci-pci

Here's the diff that can make it work:

diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 1a14ca8965e6..7bd278535ab3 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -2208,6 +2208,12 @@ void pci_pm_init(struct pci_dev *dev)
        }

        pmc &= PCI_PM_CAP_PME_MASK;
+
+       if (unlikely(dev->vendor == 0x1022 && dev->device == 0x7808))
{
+               dev_info(&dev->dev, "PME# does not work under D3,
disabling it\n");
+               pmc &= ~(PCI_PM_CAP_PME_D3 | PCI_PM_CAP_PME_D3cold);
+       }
+
        if (pmc) {
                dev_printk(KERN_DEBUG, &dev->dev,
                         "PME# supported from%s%s%s%s%s\n",

If you think this is OK, I'll resend the patch.