On Tue, Jan 20, 2026 at 02:16:17PM -0800, Elson Serrao wrote:

On 1/19/2026 11:20 PM, Krzysztof Kozlowski wrote:

quoted

On 19/01/2026 20:58, Bjorn Andersson wrote:

quoted

On Sat, Jan 17, 2026 at 12:57:58PM +0100, Krzysztof Kozlowski wrote:

quoted

On Fri, Jan 16, 2026 at 03:20:58PM -0800, Elson Serrao wrote:

quoted

The Qualcomm Embedded USB Debugger (EUD) hardware can intercept up to
two independent High-Speed UTMI data paths, depending on the SoC
configuration. Each path operates independently with:

- Dedicated PHY interface
- Distinct USB connector and controller associations
- Role dependent routing

Model these hardware paths as separate eud-path nodes to accurately
represent the physical topology and add below per-path properties:

phys: EUD exposes a High-Speed debug hub that relies on HS-PHY for its
operation. This property references the HS-PHY associated with the UTMI
path.

usb-role-switch: Indicates that the USB port on this UTMI path supports
role switching. In device role, debug mode inserts the EUD hub into the
UTMI path. In host role, the EUD hub is bypassed and UTMI traffic flows
directly between the PHY and the USB controller.

This change breaks backwards compatibility, but the previous binding
omitted critical resources like PHY and did not describe per-path
topology. Without these modifications EUD cannot be guaranteed to
function.

It was working for 3 years, so your guarantees are just imprecise. FUD
is not an argument.

Qualcomm task at 2022 was to post complete bindings. These were posted
and accepted. Three years later you say that previous posting was
bollocks and this cannot even work?

That is correct. The description of the hardware that was provided when
this was upstreamed and the binding that was accepted based on this
description is wrong.

There's absolutely a value in maintainting backwards compatibility in
general, but is this one of those cases?

quoted

Nah, take responsibility of what you did in the past.

In my view the responsible thing is to accept that we got it wrong and
make sure EUD is enabled end-to-end so people can actually use it.

I would expect to see what is not working. This is in mainline for three
years, so the assumption is that it was working for these three years.
If it wasn't, this should be described and "cannot be guaranteed to
function" is just imprecise.

I see, that I agree with.

Thanks, Bjorn and Krzysztof, for the feedback.

Let me clarify what I meant by “cannot be guaranteed to work”, as I agree
the phrasing can be improved to more precisely convey the intent.

The concern is not that EUD did not or could not work historically.
Rather, the issue is that the hardware description provided by the
binding does not explicitly describe ownership and control of hardware
resources that EUD depends on. As a result, correctness of EUD operation
relies on behavior outside of what is expressed in the description.

Concretely, the binding does not reference the HS-PHY. In practice,
EUD may function because the USB controller (e.g. DWC3) keeps the
PHY powered and configured.

However, this relationship is not described as a contract in the
binding. The USB controller may legitimately relinquish PHY control as
part of its own power-management or low-power transitions. The EUD
hardware is capable of operating independently, but doing so requires
EUD to have explicit control of the PHY.

The hardware specification lists the PHY as a required resource of the
EUD debug hub. Not modeling it in the binding therefore leaves the
description incomplete, because EUD resource requirements are being met
implicitly through another hardware block (USB controller) rather than
being described directly.

In addition, the hardware can expose multiple UTMI paths (dual-port
EUD), which the current binding cannot represent.

The intent of this patch is therefore to correct and tighten the
binding so that the hardware resources and topology EUD depends on are
explicitly modeled, rather than relying on side effects of USB controller.

I will update the backwards compatibility justification accordingly.

Please try to consolidate this whole explanation a bit and include it in
the commit message.

Regards,
Bjorn

Thanks
Elson