Re: [PATCH] xhci: Prevent runtime suspend all the time with XHCI_RESET_ON_RESUME quirk

[Kai-Heng Feng]

> On May 4, 2020, at 18:13, Mathias Nyman <mathias.nyman@xxxxxxxxx> wrote:
> 
> On 4.5.2020 13.02, Kai-Heng Feng wrote:
>> 
>> 
>>> On May 4, 2020, at 17:47, Oliver Neukum <oneukum@xxxxxxxx> wrote:
>>> 
>>> Am Montag, den 04.05.2020, 17:19 +0800 schrieb Kai-Heng Feng:
>>>> Etron EJ168 USB 3.0 Host Controller stops working after S3, if it was
>>>> runtime suspended previously:
>>>> [  370.080359] pci 0000:02:00.0: can't change power state from D3cold to D0 (config space inaccessible)
>>> 
>>> Apparently this controller has issues with D3cold
>>> 
>>>> [  370.080477] xhci_hcd 0000:04:00.0: can't change power state from D3cold to D0 (config space inaccessible)
>>>> [  370.080532] pcieport 0000:00:1c.0: DPC: containment event, status:0x1f05 source:0x0200
>>>> [  370.080533] pcieport 0000:00:1c.0: DPC: ERR_FATAL detected
>>>> [  370.080536] xhci_hcd 0000:04:00.0: can't change power state from D3hot to D0 (config space inaccessible)
>>>> [  370.080552] xhci_hcd 0000:04:00.0: AER: can't recover (no error_detected callback)
>>>> [  370.080566] usb usb3: root hub lost power or was reset
>>>> [  370.080566] usb usb4: root hub lost power or was reset
>>>> [  370.080572] xhci_hcd 0000:04:00.0: Host halt failed, -19
>>>> [  370.080574] xhci_hcd 0000:04:00.0: Host not accessible, reset failed.
>>>> [  370.080575] xhci_hcd 0000:04:00.0: PCI post-resume error -19!
>>>> [  370.080586] xhci_hcd 0000:04:00.0: HC died; cleaning up
>>>> 
>>>> This can be fixed by not runtime suspend the controller at all.
>>>> 
>>>> So instead of conditionally runtime suspend the controller, always
>>>> prevent runtime suspend with XHCI_RESET_ON_RESUME quirk.
>>> 
>>> What does that do to other controllers that can do runtime suspend
>>> under the current scheme?
>> 
>> Ok, I'll add a new quirk specific to this controller.
>> 
>> Kai-Heng
> 
> Host shouldn't runtime suspend by default unless set by userspace, or
> it has XHCI_DEFAULT_PM_RUNTIME_ALLOW quirk set.  

In this case, runtime suspend is indeed enabled by userspace.
However we should still disable it at kernel side to prevent any bug.

Kai-Heng

> 
> -Mathias