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

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


On Sunday 22 October 2006 10:37, Avi Kivity wrote:
> I like this.  Since we plan to support multiple vcpus per vm, the fs
> structure might look like:
> /kvm/my_vm
>     |
>     +----memory          # mkdir to create memory slot.

Note that the way spufs does it, every directory is a reference-counted
object. Currently that includes single contexts and groups of
contexts that are supposed to be scheduled simultaneously.

The trick is that we use the special 'spu_create' syscall to
add a new object, while naming it, and return an open file
descriptor to it. When that file descriptor gets closed, the
object gets garbage-collected automatically.

This way you can simply kill a task, which also cleans up
all the special objects it allocated.

We ended up adding a lot more file than we initially planned,
but the interface is really handy, especially if you want to
create some procps-like tools for it.

>     |     |              #    how to set size and offset?
>     |     |
>     |     +---0          # guest physical memory slot
>     |         |
>     |         +-- dirty_bitmap  # read to get and atomically reset
>     |                           # the changed pages log

Have you thought about simply defining your guest to be a section
of the processes virtual address space? That way you could use
an anonymous mapping in the host as your guest address space, or
even use a file backed mapping in order to make the state persistant
over multiple runs. Or you could map the guest kernel into the
guest real address space with a private mapping and share the
text segment over multiple guests to save L2 and RAM.

>     |
>     |
>     +----cpu             # mkdir/rmdir to create/remove vcpu
>           |

I'd recommend not allowing mkdir or similar operations, although
it's not that far off. One option would be to let the user specify
the number of CPUs at kvm_create() time, another option might
be to allow kvm_create with a special flag or yet another syscall
to create the vcpu objects.

>           +----0
>           |     |
>           |     +--- irq     # write to inject an irq
>           |     |
>           |     +--- regs    # read/write to get/set registers
>           |     |
>           |     +--- debugger   # write to set breakpoints/singlestep mode
>           |
>           +----1
>                 [...]
> It's certainly a lot more code though, and requires new syscalls.  Since
> this is a little esoteric does it warrant new syscalls?

We've gone through a number of iterations on the spufs design regarding this,
and in the end decided that the garbage-collecting property of spu_create
was superior to any other option, and adding the spu_run syscall was then
the logical step. BTW, one inspiration for spu_run came from sys_vm86, which
as you are probably aware of is already doing a lot of what you do, just
not for protected mode guests.

	Arnd <><
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at
Please read the FAQ at

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux