Re: EHCI considered harmful?

On Saturday, 30. October 2004 21:32, Brian K. White wrote:
> > On Friday, 29. October 2004 10:08, Bruce M Simpson wrote:
> > > On Fri, Oct 29, 2004 at 09:59:30AM +0200, Thomas E. Zander wrote:
> > > > I'm just wondering why ehci doesn't make it into GENERIC for
> > > > RELENG_5. Are there unresolved show stoppers related to it?
> > >
> > > ehci(4) is not stable code and fails reproducibly with my ALi-based
> > > USB2 disk enclosure.
> >
> > Well, if we would take stability and general usefulness (even more so in
> > comparison to other USB implementations in mind) as the reference point,
> > we
> > would need to disable most of USB. I think enabling ehci in GENERIC would
> > be
> > a good idea, especially since there's no loadable module...
>
> I think that since it can't be unloaded and can crash or lock up a box
> before the kernel even finishes booting, that this idea is absolutely 
> backwards.

Every device driver has the potential to do that, some are reportedly doing it 
for some users (in particular those parts of usb which _are_ in GENERIC). 

It's always annoying when that happens, but that doesn't change reality: USB 
2.0 hardware is not exactly on the way out, very much on the contrary. Thus 
ehci support must find its way into GENERIC, and if it's buggy, it needs to 
be fixed.

Maybe I should clarify that I'm talking -CURRENT here. Of course we shouldn't 
put code with known serious bugs into -STABLE (although OTOH USB 1.1 support 
managed to sneak in there, too ... 8-) ).

-- 
   ,_,   | Michael Nottebrock               | lofi_at_freebsd.org
 (/^ ^\) | FreeBSD - The Power to Serve     | http://www.freebsd.org
   \u/   | K Desktop Environment on FreeBSD | http://freebsd.kde.org