[Toybox] Proposal: Add ucsicontrol (USB Type-C UCSI utility) to toybox
Rob Landley
rob at landley.net
Mon Jul 13 20:59:17 UTC 2026
On 7/10/26 21:18, M, Madhu via Toybox wrote:
> Hi Rob,
>
> I think there are a couple of point getting mixed together here.
>
> USB-C / USB Type-C is indeed the connector and cable ecosystem, and the USB data
> protocol over it may still be USB 2.0, USB 3.x, DisplayPort alt mode,
> Thunderbolt, etc. So I agree that calling something a “USB-C wire
> protocol” can be confusing if what we mean is not traffic between two
> devices over the cable.
It's arm7 vs armv7 all over again...
> It provides a more readable, user-friendly way to inspect and control Type-C connector
> and USB Power Delivery state from the OS.
Ha! I was looking for something over the pandemic that would let me
power cycle a USB-connected device from a Linux server, and couldn't
find anything. I'm presuming this is that thing?
>> Please consider adding the ucsicontrol utility to toybox and share review comment on attached patch
I've applied it to toys/pending (pushed to codeberg and my website), and
I'm taking a look at the code (in small chunks, it'll take a while to
get through).
It starts with a LOT of bitfield structure definitions, which seem to be
the inputs and outputs to the hardware? Hmmm... get_error_status is the
argument to print_error_status() which is filled out by ucsi_xfer()
which writes to the filehandle then calls read_resp() to read data from
the filehandle, which does a hex->binary conversion but doesn't try to
parse contents.
A) There's no existing #include <linux/blah.h> header or something for
this? (The structure definitions are 1/4 of the whole file. It already
won't work on bsd because of the debugfs requirement...)
B) Four of your structs say attribute packed and the rest don't. Why?
(Did you annotate those during debugging when you hit errors on a
specific target with a specific compiler, or is there a systematic reason?)
C) I've fought bitfields before, the problem is there's no specification
of which order the bits go in (is it highest bit first or lowest bit
first), and this turns out to be a GCC internal architecture-specific
flag that ISN'T THE SAME as byte endianness. And at least at the time I
was researching it, there was no way to specify at at runtime which you
wanted (it's hardwired in at compile time based on the --target tuple,
but DOES VARY). Meanwhile, you've got:
// Response is printed MSB-first; reverse to little-endian struct order.
for (i = 0; i < di; i++) data[i] = tmp[di-1-i];
So the hardware is (consistently?) producing little bit endian within
big byte endian? (Hmmm, I'd prefer to use the existing endianness
conversion macros but capability_data is over 64 bits. Seems to be the
only one...)
I'm wondering if I can get a test environment to play with this. Does it
require specific hardware and/or specific kernel config? (Who mounts the
debugfs?)
My laptop's an ancient dell e6230, which says it's got two 8087:0024 in
it (the port on the left is a hybrid USB2/eSATA thing, the two on the
right are USB3). Would this talk to them? Or do I need to dig up less
obsolete hardware? (I've got some Orange Pi 3b boards lying around I've
been meaning to resurrect, they have USB3 but I don't remember what;
rk3566 is a hairball SOC of chinese IP, I have no idea what rockchip
licenses or "licenses", or what's an external PHY chip...)
I could of course boot all sorts of stuff under QEMU
(https://landley.net/bin/mkroot/0.8.14), but the likelihood of this
working there seems tiny. :)
Rob
P.S. It would be amazing if I could get "power down the port" and "power
back up the port" commands. I have a USE for those: turtle board
automated testing with the gdbserver kernel load protocol over USB serial.
P.P.S. I have NO IDEA how to do a regression tests/ucsicontrol.test for
this. Is there a fake kernel module that hallucinates a virtual USB
device for testing purposes, maybe? That should work fine under qemu...
More information about the Toybox
mailing list