Re: [Libmtp-discuss] udev and libmtp: detect a new mtp device based on its USB interface descriptor

[Marcus M. <ma...@je...>]

On Sun, Jan 23, 2011 at 10:08:42PM +0100, Linus Walleij wrote:
> 2011/1/23 Marcus Meissner <ma...@je...>:
> 
> > An important thing is btw to consider this is called for more devices than
> > just MTP ones.
> >
> > - querying the interface descriptor string should be safe even on
> >  cheap chinese made USB junk. (other things might crash the devices)
> 
> Sadly no, people actually have reported problems with that when I tested
> it in Fedora so I added a patch that:
> 
> - Only check class 0 (per interface), communications, PTP, and
>   vendor spec devices.
> - On per-interface devices, only check it if they have atleast one
>   vendor spec interface.
> - If it still fails to probe the OS decriptor, clear halt on EP0
>   (this was probably the real problem... the rest is perhaps
>   a bit overcautious.)
> 
> Microsoft likely has something similar, I remember the clear halt
> requirement from some document vaguely.
> 
> > - time... Querying the descriptors takes time during bootup if already plugged
> >  in and people love faster booting these days.
> >
> >  But I think udev is operating stuff in parallel, so this should not be an issue.
> 
> Dunno if it does that, anyway, since I implemented it in C now it
> will atleast not spawn a shell for each device :-P
> 
> So this was sort of the best I could come up with...
> 
> The only remaining alternative is to go into the kernel and add
> the probe code there (under much the same conditions I believe).

Probably...

Well lets see how it works out. :)

Ciao, Marcus