On Tue, Apr 04, 2006 at 09:34:54PM +0200, Matthias Drochner wrote:
> 
> The recent discussion about kcopy() reminds me of something,
> also pcb related:
> The i/o permissions bitmap in the i386 TSS is utterly useless
> nowadays, except in special cases like vm86. Actual hardware
> isn't limited to the first 1024 I/O ports, and all userlevel
> software I've seen in the last years uses the iopl() mechanism
> to get I/O access.
> Besides that, the implementation is conceptually broken since
> LWPs were introduced: it should be per-process, but the per-lwp
> PCBs are never synchronized afaics.
> 
> So I'm for deprecating the -IOPERM part in the sysarch()
> system call. This would save us some bytes in the TSS needlessly
> copied on LWP creation.
> I'd suggest to make it an option for now which is off per default.
> 
> Opinions?

How will this affect binary emulation ? I have some linux binary that
talks directly to the hardware (parallel port) ...

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--