[Qemu-devel] Possible derivative project : a simple I386 virtualizer ?

[Qemu-devel] Possible derivative project : a simple I386 virtualizer ?

[Emmanuel Charpentier]

Dear list,

My first attemps with QEMY weren't very impressive : while Windows 98 was 
(somewhat) instalable, the display was so limited and so slow that it was 
in practice very hard to use.

I've been extremely impressed with my attempts to test a W2K installation : 
the system is in fact usable for practical purposes. This emulator, being 
real fast, allows for a lot of testing situations.

Another possibility is to actually use the guest system for production. The 
classical cases, in the i386 world, are using Windows applications on a 
Linux workstation or, conversively, use Linux applications on Windows 
desktops. This is the target of virtualizers such as VMWare and Win4Lin.

Kevin Lawton, author of the Plex86 virtualizer, noted that coupling its 
virtualizer with Bochs could lead to fast execution of user-level programs 
by leaving least-privileged code (i. e. anything that doesn't touch 
hardware or system) running on the actual hardware, and trapping any 
instruction escaping this level (privileged instructions, system calls, 
access to no-owned memory location or IO ports, etc ...). Unfortunately, 
Kevin didn't go further in this direction, and concentrates now on the idea 
  of a virtualizer dedicated to running a specialized version of the Linux 
kernels on virtual machines. (BTW, Kevin states that the development of 
this idea probably requires outside funding, and won't touch it. Being laid 
off by Mandrakesoft was probably a hard experience to him, and I fully 
respect his point of view ...).

Could Kevin's trick be used with QEMU ? The guest OS as a whole would run 
in an emulator, who would leave the user-level code run on the host CPU, 
but trapping any and all non-user-level code and running it on the emulator ?

This idea sounds simple. So simple in fact that I'm probably reinventing 
the wheel (and a square wheel wit an excentered axis, at that ...). 
Furthermore, while the idea souns extremely simple, the realization 
probably isn't ...

Furthermore, my abilities at this level of programming are almost zilch : 
while I have been able to write (loooong ago) some assembly code for Z80 
and 6502, while I'd probably still able to write for the 8086, my last 
assembly gigs were some toy programs to undetstand the 80286, and my 
knowledge of the 386 and later is strictly theoretical. I migh be 
pipedreaming in a big way.

Could you sound off your reactions to this (ammitedly simplistic) idea ?

					Emmanuel Charpentier