> > > Any fixes for this would be gladly accepted.
> >
> > Well, this isn't easy to do well without deep surgery.
> > Here's a patch that fixes the problem but isn't pretty.
> > It's just for review and please note, my kernel still
> > doesn't compile.
> > Perhaps we should allocate the buffers before we go down
> > the tree ? IMHO using BKL is more or less inevitable
> > in this case in any fix.
> > People from IBM will soon come down on me for
> > abuse of BKL ;-)
>
> Heh.
>
> Why can't we just use reference counting?
At least the initial increase has to be atomic for the
whole tree. If you don't meet that requirement
there's a window between determining a hub's children
and incrementing their count. As you have to have
mutual exclusion to khubd this means BKL.
So you'd have major surgery. You'd have to walk
the tree three times. The first time atomically
incrementing the counts, then building the output
and then decrementing the counts again, bottomup
or again atomically. And you'd have to make sure that
no devices are added during that time.
Doesn't sound easy to me.
Regards
Oliver
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
