Date:   Thu, 19 Jan 2017 11:12:57 +0800
From:   Baolin Wang <baolin.wang@...aro.org>
To:     John Youn <John.Youn@...opsys.com>
Cc:     Felipe Balbi <balbi@...nel.org>,
        Greg KH <gregkh@...uxfoundation.org>,
        Mark Brown <broonie@...nel.org>,
        USB <linux-usb@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] usb: dwc3: core: Disable USB2.0 phy suspend when dwc3
 acts as host role

Hi John,

On 19 January 2017 at 09:33, John Youn <John.Youn@...opsys.com> wrote:
> On 1/16/2017 2:38 AM, Felipe Balbi wrote:
>>
>> Hi,
>>
>> John Youn <John.Youn@...opsys.com> writes:
>>>> Baolin Wang <baolin.wang@...aro.org> writes:
>>>>>> Baolin Wang <baolin.wang@...aro.org> writes:
>>>>>>> When dwc3 controller acts as host role with attaching slow speed device
>>>>>>> (like mouse or keypad). Then if we plugged out the slow speed device,
>>>>>>> it will timeout to run the deconfiguration endpoint command to drop the
>>>>>>> endpoint's resources. Some xHCI command timeout log as below when
>>>>>>> disconnecting one slow device:
>>>>>>>
>>>>>>> [   99.807739] c0 xhci-hcd.0.auto: Port Status Change Event for port 1
>>>>>>> [   99.814699] c0 xhci-hcd.0.auto: resume root hub
>>>>>>> [   99.819992] c0 xhci-hcd.0.auto: handle_port_status: starting port
>>>>>>>                                  polling.
>>>>>>> [   99.827808] c0 xhci-hcd.0.auto: get port status, actual port 0 status
>>>>>>>                                  = 0x202a0
>>>>>>> [   99.835903] c0 xhci-hcd.0.auto: Get port status returned 0x10100
>>>>>>> [   99.850052] c0 xhci-hcd.0.auto: clear port connect change, actual
>>>>>>>                                  port 0 status  = 0x2a0
>>>>>>> [   99.859313] c0 xhci-hcd.0.auto: Cancel URB ffffffc01ed6cd00, dev 1,
>>>>>>>                                  ep 0x81, starting at offset 0xc406d210
>>>>>>> [   99.869645] c0 xhci-hcd.0.auto: // Ding dong!
>>>>>>> [   99.874776] c0 xhci-hcd.0.auto: Stopped on Transfer TRB
>>>>>>> [   99.880713] c0 xhci-hcd.0.auto: Removing canceled TD starting at
>>>>>>>                                  0xc406d210 (dma).
>>>>>>> [   99.889012] c0 xhci-hcd.0.auto: Finding endpoint context
>>>>>>> [   99.895069] c0 xhci-hcd.0.auto: Cycle state = 0x1
>>>>>>> [   99.900519] c0 xhci-hcd.0.auto: New dequeue segment =
>>>>>>>                                  ffffffc1112f0880 (virtual)
>>>>>>> [   99.908655] c0 xhci-hcd.0.auto: New dequeue pointer = 0xc406d220 (DMA)
>>>>>>> [   99.915927] c0 xhci-hcd.0.auto: Set TR Deq Ptr cmd, new deq seg =
>>>>>>>                                  ffffffc1112f0880 (0xc406d000 dma),
>>>>>>>                                  new deq ptr = ffffff8002175220
>>>>>>>                                  (0xc406d220 dma), new cycle = 1
>>>>>>> [   99.931242] c0 xhci-hcd.0.auto: // Ding dong!
>>>>>>> [   99.936360] c0 xhci-hcd.0.auto: Successful Set TR Deq Ptr cmd,
>>>>>>>                                  deq = @c406d220
>>>>>>> [   99.944458] c0 xhci-hcd.0.auto: xhci_hub_status_data: stopping port
>>>>>>>                                  polling.
>>>>>>> [  100.047619] c0 xhci-hcd.0.auto: xhci_drop_endpoint called for udev
>>>>>>>                                  ffffffc01ae08800
>>>>>>> [  100.057002] c0 xhci-hcd.0.auto: drop ep 0x81, slot id 1, new drop
>>>>>>>                                  flags = 0x8, new add flags = 0x0
>>>>>>> [  100.067878] c0 xhci-hcd.0.auto: xhci_check_bandwidth called for udev
>>>>>>>                                  ffffffc01ae08800
>>>>>>> [  100.076868] c0 xhci-hcd.0.auto: New Input Control Context:
>>>>>>>
>>>>>>> ......
>>>>>>>
>>>>>>> [  100.427252] c0 xhci-hcd.0.auto: // Ding dong!
>>>>>>> [  105.430728] c0 xhci-hcd.0.auto: Command timeout
>>>>>>> [  105.436029] c0 xhci-hcd.0.auto: Abort command ring
>>>>>>> [  113.558223] c0 xhci-hcd.0.auto: Command completion event does not match
>>>>>>>                                  command
>>>>>>> [  113.569778] c0 xhci-hcd.0.auto: Timeout while waiting for configure
>>>>>>>                                  endpoint command
>>>>>>>
>>>>>>> The reason is it will suspend USB phy to disable phy clock when
>>>>>>> disconnecting the slow USB decice, that will hang on the xHCI commands
>>>>>>> executing which depends on the phy clock.
>>>>>>>
>>>>>>> Thus we should disable USB2.0 phy suspend feature when dwc3 acts as host
>>>>>>> role.
>>>>>>>
>>>>>>> Signed-off-by: Baolin Wang <baolin.wang@...aro.org>
>>>>>>> ---
>>>>>>>  drivers/usb/dwc3/core.c |   14 ++++++++++++++
>>>>>>>  1 file changed, 14 insertions(+)
>>>>>>>
>>>>>>> diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
>>>>>>> index 9a4a5e4..0b646cf 100644
>>>>>>> --- a/drivers/usb/dwc3/core.c
>>>>>>> +++ b/drivers/usb/dwc3/core.c
>>>>>>> @@ -565,6 +565,20 @@ static int dwc3_phy_setup(struct dwc3 *dwc)
>>>>>>>       if (dwc->revision > DWC3_REVISION_194A)
>>>>>>>               reg |= DWC3_GUSB2PHYCFG_SUSPHY;
>>>>>>>
>>>>>>> +     /*
>>>>>>> +      * When dwc3 controller acts as host role with attaching one slow speed
>>>>>>> +      * device (like mouse or keypad). Then if we plugged out the slow speed
>>>>>>> +      * device, it will timeout to run the deconfiguration endpoint command.
>>>>>>> +      * The reason is it will suspend USB phy to disable phy clock when
>>>>>>> +      * disconnecting slow speed decice, which will affect the xHCI commands
>>>>>>> +      * executing.
>>>>>>> +      *
>>>>>>> +      * Thus we should disable USB 2.0 phy suspend feature when dwc3 acts as
>>>>>>> +      * host role.
>>>>>>> +      */
>>>>>>> +     if (dwc->dr_mode == USB_DR_MODE_HOST || dwc->dr_mode == USB_DR_MODE_OTG)
>>>>>>> +             reg &= ~DWC3_GUSB2PHYCFG_SUSPHY;
>>>>>>
>>>>>> which version of the core you're using? Recent version (since 1.94A,
>>>>>
>>>>> My version is 2.80a.
>>>>>
>>>>>> IIRC) can manage core suspend automatically. Also, this patch of yours
>>>>>> will cause a power consumption regression.
>>>>>
>>>>> Yes, it can manage core suspend automatically, that is the problem.
>>>>> When plugging out one mouse or keypad device, the phy will suspend
>>>>> automatically to disable the phy clock. But now the disconnecting
>>>>> process is not finished, and some xHCI commands (like deconfiguration
>>>>> endpoint command to drop endpoint resources) need depend on the phy
>>>>> clock, which will hang on the system to timeout the command or abort
>>>>> command ring to halt the xHCI.
>>>>>
>>>>> I agree with you it will cause a power consumption regression, but it
>>>>> will cause serious problem if not. Do you have some suggestion?
>>>>
>>>> sorry for the long delay. This was lost in my inbox.
>>>>
>>>> I'm not sure this patch is the best solution. There's no mention in
>>>> Databook that we should avoid PHY suspend when acting as host. Adding
>>>> John here to see if John has any idea of how to fix this.
>>>>
>>>
>>> I'm not familiar enough with XHCI side of things to say.
>>>
>>> I'll ask around to see if anyone has an idea.
>>
>
> Hi Felipe, Baolin,
>
> I talked with a couple engineers here and the behavior is not
> something that's expected in host mode.
>
> Can you check that the value of the GCTL.RAMCLKSEL is set
> appropriately? This affects where the core gets the clock signal

In host mode, the bit[6:7] for RAMCLKSEL is default value 0, which
means it selects bus clock.

> from. If it is getting it from the phy clock then you will likely have
> this problem and will need to adjust it. Otherwise you should probably
> use the existing quirk instead.

So the bus clock is from the phy clock, then it will have this problem
when suspending phy? Yes, I can use the existing quirk, but I am
afraid it is one common problem if we use the mainline kernel. Or we
can add some documentation for enabling the phy suspend feature to
remind other people.

-- 
Baolin.wang
Best Regards

Powered by blists - more mailing lists