Indexing problem with address number of a USB device

[John-Mark Gurney]

Naina Patrascu wrote this message on Tue, May 11, 2021 at 18:47 +0300:
> We are working on the USB passthrough functionality for bhyve. We are
> currently focusing on the passthrough of a USB stick and a USB keyboard.
> 
> We are facing an indexing problem when trying to assign a USB device to a
> virtual machine.
> For example, we have the USB keyboard at busnum=1 and devaddr=2 on the
> FreeBSD host virtual machine:
> 
> # usbconfig -d ugen1.2
> ugen1.2: <Gaming KB Gaming KB> at usbus1, cfg=0 md=HOST spd=LOW (1.5Mbps)
> pwr=ON (100mA)
> 
> We specify to "bhyve" command to pass through the device to a guest virtual
> machine: -s 30,xhci,1-2
> 
> As part of the enumeration process, the host controller executes the
> internal scan to discover the physical devices and to compare them with the
> ones assigned with the bhyve command.
> But the USB device is discovered at bus 1 address 1, and we have to adjust
> accordingly when parsing the bus-port to match the port number:
> 
> 04f3:152e 1-1 connecting.
> 04f3:152e 1-1 belong to this vm.
> 04F3:152E 1-1 is attached to virtual port 5.
> 
> We hardcoded when parsing bus-port: port -= 1.
> When trying to pass through a USB stick, we have to increase the port
> number to match the one discovered in the internal scan:
> 
> # usbconfig -d ugen3.2
> ugen3.2: <Kingston DataTraveler 3.0> at usbus3, cfg=0 md=HOST spd=HIGH
> (480Mbps) pwr=ON (300mA)
> 
> 0951:1666 3-3 connecting.
> 0951:1666 3-3 belong to this vm.
> 0951:1666 3-3 is attached to virtual port 1.
> 
> In our last investigation, we discovered that when increasing the number of
> USB devices on the FreeBSD host vm, the port indexes of the devices
> discovered by the internal scan, also varies.
> 
> In function [1] we get the physical devices using libusb library and then
> we execute the internal scan using function [2]. We get the information
> about the physical device in function [3], also using libusb library.
> 
> Can you help us understand where this problem could come from and how to
> debug it?

Yes, the ugen named does not map to port naming.  I ran across this when
I did the work to make serial devices consistent based upon the port
number.  The review for that work is here:
https://reviews.freebsd.org/D21886#554613

The device address is a different enumeration from the port address
information.  The script linked in the review creates a mapping from
the device name to the proper port number mapping.

To see this, you can look at devinfo -v, and look at the port value
vs the devaddr (device address assigned at enumeration).

Hope this helps.


-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."