James Mastros wrote:
> I think if we get this in well, we can shave our shimms down pretty thin,
> and get some decent performace out of this beast.
I think so. Total throughput should be decent. For ring0 guest
code, there will be a good percentage of instructions translated 1:1.
Others which use handler functions or inline code will be 1:N, with
varying Ns. Considering we can run the ring3 guest code
without translation, that'll be 1:1. Centralization of the
handler routines should keep them nicely in code cache.
We'd also have to factor in things like the frequency of
transitions between guest user and kernel space (like for
system calls and scheduling), overhead from switching between
the host/monitor, translation overhead, and other stuff.
But I think the DT mods will put us on the "other side" of the
performance fence, the one where users would deem plex86
as "usable" performance-wise. Then, special drivers, and VM core
optimizations will enhance things from there.
-Kevin
--
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Kevin Lawton [EMAIL PROTECTED]
MandrakeSoft, Inc. Plex86 developer
http://www.linux-mandrake.com/ http://www.plex86.org/
