Re: [linux-usb-devel] Re: Fw: Re: oops in choose_configuration()

[David Brownell <david-b@pacbell.net>]

I wonder if that SCSI fix (restoring a wrongly deleted mem clear) helps
get rid of this oops too?


On Wednesday 08 March 2006 7:30 am, Alan Stern wrote:
> > > a) How come we're only considering the zeroth slot in that array in here?
> > 
> > We start out with the first interface setting, as we always know we have
> > one of them as per the USB spec (I think, anyone from linux-usb-devel
> > want to verify this?)
> 
> In this case it wouldn't make any difference, since all the altsettings
> for a particular interface are supposed to have the same bInterfaceClass,
> bInterfaceSubClass, and bInterfaceProtocol.  Although I don't think the
> USB spec actually says this anywhere..

I'd have stopped at "wouldn't make any difference"; the kernel must make
some initial choice, but userspace is free to revise it.  Agreed it would
be odd if altsettings had different class/subclass/protocol, but I don't
see any good reason to make that illegal.


> The bMaxPower value could be different for different altsettings. 

Erm, no; that's a per-configuration thing, not a per-altsetting thing.
It's checking the config descriptor, not the interface descriptor,
for that particular concern.


> > > b) How do we know that there's actually anything _there_?  The length of
> > >    that variable-sized array doesn't seem to have been stored anywhere
> > >    obvious by usb_parse_configuration() and choose_configuration() doesn't
> > >    check.  What happens if the length was zero?
> > 
> > I don't think it is allowed to be, as all USB devices have to have at
> > least 1 interface.

I think that's not true, and it would be worth verifying that it's not
a no-interfaces device even if the USB spec required it.  It's trivial
to create device firmware that advertises no-interfaces, and those should
never be able to make Linux hiccup (much less oops).


> The code in usb_parse_configuration() guarantees that the number of
> entries in the altsettings array is at least 1, because it sets nalts[n]
> to 1 initially and never decreases it.  The whole idea of an interface
> without altsettings makes no sense...

Right; there's always at least one setting.  Calling them "alt" settings
can be confusing; any one of them could be the "main" setting.

- Dave