Re: [PATCH v2] USB: cdc-acm: add Whistler radio scanners TRX series support

From: Johan Hovold <johan@kernel.org>
To: Oliver Neukum <oneukum@suse.com>
Cc: Johan Hovold <johan@kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	linux-usb@vger.kernel.org,
	Daniel Caujolle-Bert <f1rmb.daniel@gmail.com>,
	stable@vger.kernel.org
Subject: Re: [PATCH v2] USB: cdc-acm: add Whistler radio scanners TRX series support
Date: Mon, 21 Sep 2020 15:43:47 +0200	[thread overview]
Message-ID: <20200921134347.GW24441@localhost> (raw)
In-Reply-To: <1600690627.2424.80.camel@suse.com>

On Mon, Sep 21, 2020 at 02:17:07PM +0200, Oliver Neukum wrote:
> Am Montag, den 21.09.2020, 14:03 +0200 schrieb Johan Hovold:
> > On Mon, Sep 21, 2020 at 01:49:14PM +0200, Oliver Neukum wrote:
> > > Am Montag, den 21.09.2020, 13:36 +0200 schrieb Johan Hovold:
> > > > On Mon, Sep 21, 2020 at 12:29:16PM +0200, Oliver Neukum wrote:
> > > 
> > > Hi,
> > > 
> > > > I meant that instead of falling back to "combined-interface" probing we
> > > > could assume that all interfaces with three endpoints are "combined" and
> > > > simply ignore the union and call managementy. descriptors and all the ways
> > > > that devices may have gotten those wrong.
> > > 
> > > I am afraid we would break the spec. I cannot recall a prohibition on
> > > having more endpoints than necessary. Heuristics and ignoring invalid
> > > descriptors is one things. Ignoring valid descriptors is something
> > > else.
> > 
> > That depends on how you read the spec (see "3.3.1 Communication Class
> > Interface"). But sure, it's probably be better to err on the safe-side.
> 
> You mean 3.4.1?

It's 3.3.1 in Version 1.1 at least.

> > > > I was thinking more of the individual entries in the device-id table
> > > > whose control interfaces may not even be of the Communication class. But
> > > > hopefully that was verified when adding them.
> > > 
> > > Now you are confusing me. In case of a quirky device, why change
> > > the current logic?
> > 
> > Just because they have a quirk defined, doesn't mean they don't rely on
> > the generic probe algorithm (e.g. a USB_DEVICE entry which matches all
> > interface classes and only specifies SEND_ZERO_PACKET).
> 
> Right, so let me be more specific. It would probably be unwise to
> change the decision tree in probe() as far as devices whose quirks
> affect decisions in that already are concerned.

But we need to draw line somewhere to keep the code maintainable and
ourselves sane, especially since a lot of these devices where added
without any record of their descriptors.

I guess we could add another test for the device-id fields, but I'm
reluctant to add more special casing before we know it's needed.

Johan