DragonFly kernel List (threaded) for 2003-08
Re: Vmware was Re: Announcing shell machine availability for dragonfly developers
I have used bochs too. It's been improved considerably from earlier
versions but, of course being an instruction-by-instruction emulator,
it is slower then VMWare. On the otherhand Bochs is entirely usermode
and has no chance of crashing your system, something I have managed
to do with vmware on occassion :-).
It occurs to me that we could implement a full address space abstraction
for DragonFly which would allow normal user programs to use the entire
address space. Once that were done it would only be a small step to
implement a usermode 'emulation' layer which allows one user program to
intercept privileged instructions and address accesses that another tries
to run. Then we could implement something similar to VMWare ourselves
and probably achieve 70-80% native operation. It would be an interesting
:> > Vmware is an excellent product, although expensive. I haven't tried
:> > Bochs but I imagine that would also work (although at a far lower speed)
:> > and it's OSS.
:> I can vouch for bochs, it works quite well on a 1Ghz Celeron. You can even
:> compile it with a remote gdb stub - essential for this kind of work.
:Also take a look at QEMU, which uses dynamic translation to
:virtualise the system. QEMU achieves around 25% of native
:speed, despite being fully emulated...
:It also makes it way easier to debug a kernel then using eg.
:vmware, because you can more easily look inside the qemu
:virtual machine. The disadvantage is that QEMU doesn't yet
:do full MMU emulation, which means you need to build a
:special kernel to use with QEMU (different address space
:layout, but that's basically it).