On Fri, May 10, 2002, Alan Cox <[EMAIL PROTECTED]> wrote:
> > > We use clear_bit and set_bit because it's atomic and makes sure we don't
> > > race with the HC updating the same data.
> > 
> > If the HC is updating the same data, it'd seem like such
> > races have been a longstanding risk.
> 
> They are only safe for SMP. Worse yet thats an x86 'quirk' that they are.
> I'm not sure some platforms are capable of an I/O view atomic bit set.
> Having said that this sort of implies that you need to provide hooks for
> awkward boxes, not screw performance for everyone by turning controllers
> off and on to poke bits

You mean it's a quirk that it's atomic on SMP x86? That isn't normal?

Worst case, if there is no way we can do this reliably across
architectures, then I guess i'll have to remove that portion of the
optimization.

JE


_______________________________________________________________

Have big pipes? SourceForge.net is looking for download mirrors. We supply
the hardware. You get the recognition. Email Us: [EMAIL PROTECTED]
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to