Re: dwc3 gadget and USB3

[Thinh Nguyen <Thinh.Nguyen@synopsys.com>]

On Thu, Mar 09, 2023, Joakim Tjernlund wrote:
> On Thu, 2023-03-09 at 17:32 +0100, Joakim Tjernlund wrote:
> > On Wed, 2023-03-08 at 22:26 +0100, Joakim Tjernlund wrote:
> > > On Wed, 2023-03-08 at 19:58 +0100, gregkh@linuxfoundation.org wrote:
> > > > On Wed, Mar 08, 2023 at 06:12:51PM +0000, Joakim Tjernlund wrote:
> > > > > On Wed, 2023-03-08 at 18:25 +0100, Greg KH wrote:
> > > > > > On Wed, Mar 08, 2023 at 05:10:17PM +0000, Joakim Tjernlund wrote:
> > > > > > > we are using fsl-ls1043a-rdb based design but with a ls1023a SOC and
> > > > > > > use USB0 in gadget mode running either NCM or RNDIS ethernet on top.
> > > > > > > 
> > > > > > > When we connect the gadget to a PC(Linux of Windows) over an USB2 hub,
> > > > > > > networking(NCM or RNDIS) works well.
> > > > > > > 
> > > > > > > However, when we connect the gadget directly to the PC/laptop which uses USB3
> > > > > > > we see something odd:
> > > > > > >   Ping from PC to gadget works.
> > > > > > >   Ping from gadget to laptop does not. However if we also ping from PC at the same time we
> > > > > > >   see gadget to PC start working.
> > > > > > > Seems like ping from the PC tiggers the gadget to see incoming pkgs somehow.
> > > > > > > 
> > > > > > > Any idea what might be wrong or how to debug this?
> > > > > > > Kernel 5.15.87
> > > > > > 
> > > > > > 5.15.y is very old, does this also happen on 6.2?
> > > > > > 
> > > > > 
> > > > > I just tried 6.1.15 and the problem remains, I hope that is close enough ?
> > > > 
> > > > It's good enough :)
> > > > 
> > > > Have any logs at all that show any problems?
> > > > 
> > > No, don't know where to start. There are no errors logged.
> > > 
> > > >   Also, you might want to
> > > >  cc:  the dwc3 maintainer...
> > > 
> > > I thought I did but that look like old info, added Thinh Nguyen now, thanks
> > > 
> > >  Jocke
> > > 
> > > > 
> > > >  thanks,
> > > > 
> > > >  greg k-hj
> > > 
> > 
> > Found and USBC Dock and connected that between gadget an PC and this also works well.
> > Seems like a hub, regardless of USB2/USB3, make the usb network function in both directions.
> > 
> > Found out something interesting, on PC:
> > cd /sys/devices/pci0000:00/0000:00:14.0/usb1/1-1/power  # Where my gadget is connected
> > echo 0 > usb2_hardware_lpm
> > 
> > Now ping works normally.
> > 
> > So LPM does not seem to work properly on gadget. Can I disable LPM somehow
> > on gadget side? 
> > 

There's no option in gadget configfs to allow you to do that at the
moment. You can disable LPM in dwc3 controller in the devicetree with
"snps,dis_enblslpm_quirk" instead.

If the host puts the gadget in suspend, the gadget won't be able to
communicate with the host until the host wakes the gadget up and starts
talking to the gadget again. The gadget may be able to signal the host
to wakeup via remote wakeup. Did you check if the device is in suspend?
If it's in suspend, is the gadget enabled with remote wakeup? Did the
NCM driver sent a remote wakeup signal to the host? I didn't verify, but
I suspect the NCM gadget driver isn't configured/implemented with remote
wakeup.

You can work around this by disabling LPM, which removes any power
saving as you've tested.

BR,
Thinh

> >  Jocke
> 
> Found some DTS quirks to disable LPM, work fine :)
> One observation:
> 
> ping over NCM to Linux PC:
> PING 169.254.100.102 (169.254.100.102): 56 data bytes
> 64 bytes from 169.254.100.102: seq=0 ttl=64 time=2.166 ms
> 64 bytes from 169.254.100.102: seq=1 ttl=64 time=2.168 ms
> 64 bytes from 169.254.100.102: seq=2 ttl=64 time=2.333 ms
> 
> ping over NCM to Windows 10 PC:
> PING 169.254.100.102 (169.254.100.102): 56 data bytes
> 64 bytes from 169.254.100.102: seq=0 ttl=128 time=0.921 ms
> 64 bytes from 169.254.100.102: seq=1 ttl=128 time=0.963 ms
> 64 bytes from 169.254.100.102: seq=2 ttl=128 time=1.143 ms
> 64 bytes from 169.254.100.102: seq=3 ttl=128 time=1.161 ms
> 
> NCM to Windows appears to have much lower latency.
>