Re: [PATCH 0/7] KVM: Kernel-based Virtual Machine

From: Avi Kivity
Date: Thu Oct 19 2006 - 14:00:39 EST


Muli Ben-Yehuda wrote:
Hi,

Looks pretty interesting! some comments:

- patch 4/7 hasn't made it to the list?

Probably too big. It's also the ugliest. I'll split it and resend (not through thunderbird though... ate all my tabs!).


- it would be useful for reviewing this if you could post example code
making use of the /dev/kvm interfaces - they seem fairly complex.

Working code is fairly hairy, since it's emulating a PC. That'll be on sourceforge once they approve my new project.

In general one does

open("/dev/kvm")
ioctl(KVM_SET_MEMORY_REGION) for main memory
ioctl(KVM_SET_MEMORY_REGION) for the framebuffer
ioctl(KVM_CREATE_VCPU) for the obvious reason
if (debugger)
ioctl(KVM_DEBUG_GUEST) to singlestep or breakpoint the guest
while (1) {
ioctl(KVM_RUN)
switch (exit reason) {
handle mmio, I/O etc. might call
ioctl(KVM_INTERRUPT) to queue an external interrupt
ioctl(KVM_{GET,SET}_{REGS,SREGS}) to query/modify registers
ioctl(KVM_GET_DIRTY_LOG) to see which guest memory pages have changed
}

I have some simple test code, I'll clean it up and post it.

- why do it this way rather than through a virtual machine monitor
such as Xen? what do you gain from having the virtual machines
encapsulated as Linux processes?

- architectural simplicity: instead of splitting memory management and scheduling between Xen and domain 0, use just the Linux memory management and scheduler
- use standard tools (top(1), kill(1)) and security model (permissions on /dev/kvm)
- much smaller codebase (although paravirtualization is not included (yet))
- no changes to core code
- easy to upgrade an existing system
- easier for drive-by virtualization (modprobe kvm; do-your-stuff; ctrl-C; rmmod kvm)
- longer term, better performance since there's no need to switch to domain 0 for I/O (instead just switch to user mode of the VM's process)

--
Do not meddle in the internals of kernels, for they are subtle and quick to panic.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/