xhciep: Transfer timed out with Circuitpython device

[diyelectromusic]

I'm using a Circuitpython v8 USB MIDI device plugged into a Pi 4 running Circle (MiniDexed) which seems to be recognised as the following USB devices:
int2-2-0
inta-0-0
int8-6-50
int1-1-0
int1-3-0

The only one I'm really interested in is int1-3-0 for audio/MIDI, but when accessing the SD card filesystem (this is on MiniDexed) for some reason I'm getting xhciep Transfer timeouts which causes the execution/UI to pause (presumably whilst the timeout occurs).

I've tried using usbignore= for some of the above - with int8-6-50 being the most suspect as a mass storage device, but it doesn't seem to make a difference.

I've also used the following boot.py code (more here):

import usb_hid, usb_midi
usb_hid.disable()
usb_midi.enable()

And also tried disabling storage completely too using:

import storage
storage.disable_usb_drive()

(which I know it warns not to do - I still had REPL access so could turn it back on again).

But so far, it doesn't seem to make any difference. As I'm largely trying things pretty much at random at the moment, I just wondered if anyone had any more concrete ideas for what might be going on?

I'll keep trying various configurations and see if anything changes.

Thanks,
Kevin

[rsta2]

I'm trying to reproduce this with sample/29-miniorgan and a RPi Pico. What would be helpful is a simple code.py for Circuitpython, which sends some MIDI events (notes) to the USB host. Unfortunately I'm not familiar with Circuitpython.

[diyelectromusic]

It was causing issues when changing performances in MiniDexed, so I took that as probably related to the SD card? MIDI was working fine, which is why I thought it perhaps a side effect of one of the other USB endpoints.

As I say, I'll see if I can get something down to a smaller test case, but not really knowing what it is relating to, it won't be easy... I really need a trace through the USB driver to know what transfer was triggering the timeout.

I was using a Circuitpython 8 device, so that is certainly something to try first - I'll grab the latest :)

Thanks,
Kevin

[rsta2]

With this patch it tells you the EP number, which causes the timeout message:

diff --git a/lib/usb/xhciendpoint.cpp b/lib/usb/xhciendpoint.cpp
index 449f1d2..ab6097a 100644
--- a/lib/usb/xhciendpoint.cpp
+++ b/lib/usb/xhciendpoint.cpp
@@ -223,7 +223,8 @@ boolean CXHCIEndpoint::Transfer (CUSBRequest *pURB, unsigned nTimeoutMs)
        {
                if (CTimer::Get ()->GetTicks () - nStartTicks >= HZ)
                {
-                       CLogger::Get ()->Write (From, LogDebug, "Transfer timed out");
+                       CLogger::Get ()->Write (From, LogDebug, "Transfer timed out (ep %u)",
+                                               (unsigned) pURB->GetEndpoint ()->GetNumber ());
 #ifdef XHCI_DEBUG
                        m_pDevice->DumpStatus ();
 #endif

You can use lsusb -v on Linux to search for this endpoint number for the Circuitpython device. The direction bit (0x80) is not included in the shown EP number.

[diyelectromusic]

Ah brilliant! Many thanks - will add that in and see what it does! :)

[diyelectromusic]

Ok. So, turns out it is the MIDI int1-3-0 endpoint - so the one I actually want. Then I realised this is happening on send - i.e. when MiniDexed tries to send some data back over the MIDI link.

Also, it turns out it is a lot worse when changing voices. Then I realised what is going on...

In MiniDexed there is an option someone added (that personally I don't agree with!) that automatically does a SysEx voice dump of all voice parameters whenever the voice changes. This happens when a performance is selected too, as it resets voices on each TG, but performances have to be selected to load, but voices change on menu changing - so every click of the Voice menu is dumping data out on all USB MIDI interfacces!

But the Circuitpython device isn't listening for USB data, it only sends.

So I think the timeout is because MiniDexed is spamming the USB MIDI interface and the CP device is ignoring it...

Testing again with the option to disable the auto-sending of the SysEx data and yep - all fine!?

Sorry for wasting your time, but also thanks for that clue - without that link I suspect it would have taken a lot longer to work out what was going on!

Time to update the MiniDexed wiki I think :)

Kevin

[rsta2]

I see. No problem, I will probably generally add this patch.

Rene