Re: dwc3 Disable Compliance Mode

[Rob Weber <rob@gnarbox.com>]

Hello Felipe,

On Thu, Jul 11, 2019 at 10:37:11AM +0300, Felipe Balbi wrote:
> Rob Weber <rob@gnarbox.com> writes:
> > I hope you are doing well. My team and I are frequently experiencing an
> > issue with the dwc3 in our CherryTrail SoC where we encounter an LFPS
> > Polling timeout while our device is being enumerated.
> >
> > We configured the dwc3 as an ethernet gadget using configfs and the ecm
> > and RNDIS functions. The dwc3 transitions to U3 after configuration as
> > expected. Only once we connect our device to a USB host do we see the
> > link state transition to Polling. We are assuming LFPS Polling times
> > out because the link_state file in debugfs shows the link has
> > transitioned to compliance mode only after entering LFPS.Polling, and we
> > recently learned that compliance mode is triggered by a timeout during
> > LFPS.Polling.
> >
> > This issue is not 100% reproducible, but is occuring rather frequently
> > at the moment. We're unsure of the root cause of the issue as well. One
> > culprit might be the USB SuperSpeed Redriver we use in our design.
> >
> > We would like to disable compliance mode in the meantime to allow other
> > team members to continue developing and testing USB device mode features
> > while we dig into the root cause of the issue. Is there a proper way to
> > disable compliance mode entirely?
> 
> That's not something that can be done, unfortunately.

Okay, understood.
> 
> > If not, is there some mechanism we could implement to reset the
> > dwc3 when we enter compliance mode? I attempted some sort of mechanism
> > to reset the link state, but it does not seem to help the issue. I've
> > attached my patch and the trace events for my attempted workaround to
> > this email. My initial approach was to transition the link from
> > Compliance -> SS.Disabled -> Rx.Detect when we detect we've entered
> > compliance mode. The traces show that the dwc3 just enters LFPS.Polling
> > and subsequently enters compliance mode, despite the link being reset.
> 
> I think you would have to go through the entire Power On Reset
> sequence, but that's likely to be flakey.

Okay, good to know. I am not confident in this approach as well because of
the state management / recovery we might have to perform.

I wanted to explain our goal a little bit more in depth in case some
other apprach might come to mind. Our product supports 3 ways in which
the user can work with the USB port: Host mode, Mass Storage Mode, and
Ethernet Mode. Host mode is pretty straightforward. Users will generally
work with USB mass storage devices and USB-Ethernet adapters in host
mode. Mass storage mode exposes an internal user data partition using
f_mass_storage and configfs so the user can connect our product to their
host computer for backing up data. Ethernet uses the f_ecm and f_rndis
functions creating an ethernet connection with USB hosts, particularly
mobile devices, to interact with our mobile applications and servers
through a wired connection.

We control this functionality through a userspace application written in Go
that creates the gadget/function configuration in configfs. Given this
background information, are there any approaches that come to mind, such
as reconfiguring the gadget? Thanks in advance for your input.

> > Do you have any ideas on how we might work around compliance mode in the
> > meantime?
> 
> We have a few quirk flags that may help. snps,u2exit_lfps_quirk comes to
> mind, but I suggest trying a few of them and see if any helps.
> 
> All flags are described in
> Documentation/devicetree/bindings/usb/dwc3.txt

Great, thank you. I will give these a shot and let you know if I have
further questions.

> >  /* -------------------------------------------------------------------------- */
> > # tracer: nop
> > #
> > # entries-in-buffer/entries-written: 220/220   #P:4
> > #
> > #                              _-----=> irqs-off
> > #                             / _----=> need-resched
> > #                            | / _---=> hardirq/softirq
> > #                            || / _--=> preempt-depth
> > #                            ||| /     delay
> > #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
> > #              | |       |   ||||       |         |
> >           <idle>-0     [003] ..s1   178.444339: dwc3_core: Tick! Checking for compliance mode
> >
> >           <idle>-0     [003] d.s2   178.444352: dwc3_core: Compliance Mode detected. Attempting recovery routine
> 
> >      irq/23-dwc3-1115  [002] d..1   178.445380: dwc3_event: event (00140301): Link Change [SS.Disabled]
> >      irq/23-dwc3-1115  [002] d..1   178.445826: dwc3_event: event (00150301): Link Change [RX.Detect]
> >      irq/23-dwc3-1115  [002] d..1   178.445832: dwc3_event: event (00170301): Link Change [Polling]
> >           <idle>-0     [003] ..s1   180.492293: dwc3_core: Tick! Checking for compliance mode
> >
> >           <idle>-0     [003] d.s2   180.492306: dwc3_core: Compliance Mode detected. Attempting recovery routine
> >
> >      irq/23-dwc3-1115  [002] d..1   180.493333: dwc3_event: event (00140301): Link Change [SS.Disabled]
> >      irq/23-dwc3-1115  [002] d..1   180.493636: dwc3_event: event (00150301): Link Change [RX.Detect]
> >      irq/23-dwc3-1115  [002] d..1   180.493641: dwc3_event: event (00170301): Link Change [Polling]
> >           <idle>-0     [003] ..s1   182.540350: dwc3_core: Tick! Checking for compliance mode
> >
> >           <idle>-0     [003] d.s2   182.540362: dwc3_core: Compliance Mode detected. Attempting recovery routine
> >
> >      irq/23-dwc3-1115  [002] d..1   182.541392: dwc3_event: event (00140301): Link Change [SS.Disabled]
> >      irq/23-dwc3-1115  [002] d..1   182.541976: dwc3_event: event (00150301): Link Change [RX.Detect]
> >      irq/23-dwc3-1115  [002] d..1   182.541982: dwc3_event: event (00170301): Link Change [Polling]
> >           <idle>-0     [003] .Ns1   184.588211: dwc3_core: Tick! Checking for compliance mode
> >
> >           <idle>-0     [003] dNs2   184.588232: dwc3_core: Compliance Mode detected. Attempting recovery routine
> 
> Don't we get an interrupt for Compliance mode entry?

Not that I've seen, surprisingly. My compliance mode recovery mechanism
looks for both compliance mode and loopback mode, neither of which I
have ever seen in the link state change events.

Cheers,
Rob Weber