From: Kevin Lo <kevlo@openbsd.org>
Subject: Re: umsm(4): don't deactivate the whole usb device if umsm claims an unworkable interface
To: David Gwynne <david@gwynne.id.au>
Cc: tech@openbsd.org
Date: Thu, 18 Jun 2026 14:38:23 +0800

On Thu, Jun 18, 2026 at 11:37:38AM +1000, David Gwynne wrote:
> 
> we got a bunch of Quectel EM060K-GL modems at work and, as i've come to
> expect from mobile modems, there was a new and exciting bug to fix.
> 
> when i plugged it in, it looked like this:
> 
> umsm0 at uhub1 port 1 configuration 1 interface 0 "Quectel EM060K-GL" rev 2.00/5.04 addr 2
> ucom0 at umsm0: usb1.0.00001.0
> umsm1 at uhub1 port 1 configuration 1 interface 1 "Quectel EM060K-GL" rev 2.00/5.04 addr 2
> ucom1 at umsm1: usb1.0.00001.1
> umsm2 at uhub1 port 1 configuration 1 interface 2 "Quectel EM060K-GL" rev 2.00/5.04 addr 2
> ucom2 at umsm2: usb1.0.00001.2
> umsm3 at uhub1 port 1 configuration 1 interface 3 "Quectel EM060K-GL" rev 2.00/5.04 addr 2
> ucom3 at umsm3: usb1.0.00001.3
> umb0 at uhub1 port 1 configuration 1 interface 8 "Quectel EM060K-GL" rev 2.00/5.04 addr 2
> umsm4 at uhub1 port 1 configuration 1 interface 12 "Quectel EM060K-GL" rev 2.00/5.04 addr 2
> umsm4: missing endpoint
> 
> but any attempt to use any of these devices resulted in errors.
> opening /dev/cuaU things would return EIO, and ifconfig against the
> umb device produced weird errors.
> 
> after a lot of trial and error, i figured out that if i prevented
> umsm4 from attaching, the remaining devices would work as expected.
> while using kevlo@ as a rubber duck, he pointed out that this device
> was working in the past (as per
> https://marc.info/?l=openbsd-tech&m=171644719119922&w=2).
> 
> the difference between then and now is that umsm used to only attach
> to low interfaces, but i found a quectel device that used high
> interfaces for umsm and got rid of that restriction, which is in
> line with the advice from quectel themselves about how to support
> these devices. however, this allows umsm(4) to attach to interface
> 12 on EM060K modems, which now breaks them.
> 
> the problem is that umsm_attach tries to find the necessary endpoints
> to drive the usb interface, and if it can't then it calls
> usbd_deactivate(). this kills the entire usb device, not just the
> interface it was working against. removing this line allows umsm_attach
> to fail gracefully on interface 12 above, but leaves the rest of
> them still working.

Yes, deactivating sc_udev on one failed interface takes everything down.
Failing this interface gracefully is the correct behavior.

> ok?

ok kevlo@

> ps, this is in line with the guidance from quectel on how to support
> these devices. it's worth pointing out that linux maintains a big
> table that lists specific interfaces it will support on specific
> versions of quectel modems. kevlo@ and i are considering moving
> toward following the linux example in the future.
> 
> Index: umsm.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/usb/umsm.c,v
> diff -u -p -r1.130 umsm.c
> --- umsm.c	18 Jun 2026 01:20:17 -0000	1.130
> +++ umsm.c	18 Jun 2026 01:21:39 -0000
> @@ -447,7 +447,6 @@ umsm_attach(struct device *parent, struc
>  	}
>  	if (uca.bulkin == -1 || uca.bulkout == -1) {
>  		printf("%s: missing endpoint\n", sc->sc_dev.dv_xname);
> -		usbd_deactivate(sc->sc_udev);
>  		return;
>  	}
>  
>