Re: [PATCH] usb: gadget: f_rndis: fix usb_interface_descriptor for rndis

From: Heiko Schocher
Date: Mon Sep 29 2014 - 08:11:49 EST

Hello Lars,

sorry for my late answer ...

Am 24.09.2014 16:22, schrieb Lars Melin:
On 2014-09-24 20:12, Heiko Schocher wrote:
Hello Lars,

Am 24.09.2014 14:25, schrieb Lars Melin:
On 2014-09-24 13:48, Heiko Schocher wrote:
use the values for RNDIS over Ethernet as defined in
(search for RDNIS):

- baseclass: 0xef (miscellaneous)
- subclass: 0x04
- protocol: 0x01

That is usb class, it is not the same thing as communication device class.
--- a/include/uapi/linux/usb/cdc.h
+++ b/include/uapi/linux/usb/cdc.h
@@ -12,6 +12,7 @@
#include <linux/types.h>
No, no, no.
There is no CDC_SUBCLASS_RNDIS and you can not define one over an already used cdc subclass number, 0x04 is Multi-Channel Control Model

Ah, ok, so I have to define this values in a new header file, as there
is no current file for the USB_CLASS_MISC defines? Or is there a proper
place for them?

BTW: where do I find the "cdc subclass number, 0x04 is Multi-Channel
Control Model" define?


You can still find the original specification usbcdc11.pdf on the net if you google for it, it has been pulled from where you could download it until a few years ago.
It is old but covers a lot of what you need to know.

Hmm.. maybe I am to dummy for finding this docment...

does not find this document ... could you send me a direct link?

I found with the above search:

and this site, exactly describes the values for RNDIS over ethernet,
as my patch changes [1]

Linux has afaik only the cdc.h definition file, everything else is coded by class/subclass in respectively drivers when needed.

why not in header files? I thought, magical values are not welcome
in source code ...

As for the is_rndis() function case, this function is defined in
2 places:

- drivers/net/usb/cdc_ether.c
- drivers/usb/core/generic.c

Has this a special reason? This seems suboptimal to me ...

02/02/ff or e0/01/03 are the most common interface attribute for rndis, both of them together with a data interface with attributes 0a/00/00.

I must admit, I am not a USB nor a RNDIS expert ...

Please check the whitelisting in drivers/net/usb/rndis_host.c and also blacklistings in other net drivers under the same path, it should give you an idea how to bind an interface to a specific driver by interface attributes and/or usb vid:pid.
You should be able to do the same for your particular device.

Hmm.. I did not understand you here ... so, one step back:

I got from a customer this patch (in a similiar version) and
he did tests with [3] and saw, that a board which runs linux,
is seen in [3] with the values [2] ... so he changed the
values in drivers/usb/gadget/function/f_rndis.c to the
values [1], which are documented in [4] and with them
the test [3] is happy ... and the file
"Documentation/usb/linux.inf" is not longer needed on the
windows pc!

So he (and at the end I too) thought, that this is the proper
way to make [3] happy ... (maybe [3] is incorrect ? )

Is current ML code correct? And if yes, why?

If the values [2] in current ML linux are correct,
could you say me, where they are documented?

(and sorry for my stupid questions ...)



[1] values which my patch sets for RNDIS over ethernet
- baseclass: 0xef (miscellaneous)
- subclass: 0x04
- protocol: 0x01

[2] currently used values for RNDIS over ethernet
- baseclass: 0x02 (USB_CLASS_COMM)
- subclass: 0x02
- protocol: 0xff

[3] "USB Compliance test suite which runs Windows", see:

DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at