Re: [PATCH] USB: serial: simple: narrow OWON match to avoid breaking VDS1022
From: Johan Hovold
Date: Tue Nov 17 2026 - 09:05:20 EST
On Tue, Nov 03, 2026 at 10:23:11AM +0800, Huang Wei wrote:
> Hi Craig,
>
> Thanks for the additional data points - and for the comprehensive
> descriptor table. This changes the picture significantly.
>
> The VDS1022I reporting (5, 0, 0) - the same tuple as the HDS200 -
> means the full-tuple match in my v2 doesn't actually solve the
> problem it was designed for. It would:
>
> - correctly claim the HDS200 (which wants serial)
> - incorrectly claim the VDS1022I (which needs libusb, same
> florentbr tooling as the VDS1022)
>
> So the interface tuple is not a reliable discriminator for this
> VID:PID - OWON reuses the same tuples across device families with
> different userspace needs. That undermines any approach based on
> narrowing the match at the interface level.
>
> Given this, I think the options are:
>
> 1. Keep the original broad USB_DEVICE() match, accepting that
> VDS1022/VDS1022I users need to blacklist the driver (this was
> the status quo before my patches, and users have workarounds)
>
> 2. Remove the owon entry entirely and have HDS200 users move to
> libusb as well (uniform but breaks existing serial users)
>
> 3. Find a different discriminator (bcdDevice, string descriptors)
> - but these seem equally unlikely to be reliable given OWON's
> firmware practices
>
> Johan - given that the interface-level narrowing doesn't work,
> what's your preference? Should we revert to the original match and
> document the libusb incompatibility, or is there another approach?
As there apparently are now users of the in-kernel driver and libusb has
a mechanism for unbinding kernel drivers when userspace wants to handle
a device itself, my preference is option 1.
I've tried to skim these threads and IIUC the only issue here is that
some java-wrapper around libusb doesn't provide the unbind mechanism?
For the cases where the userspace driver does not unbind the driver
itself, users are currently already using a udev rule to unbind the
driver (or they use module blacklisting). And they probably already have
udev rules to set up permissions.
And there should be nothing preventing a userspace application from
using the tty interface even if a device does not speak SCPI, right?
Johan