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

[Avi Kivity <avi@qumranet.com>]

Alan Cox wrote:
> Ar Sul, 2006-10-22 am 10:37 +0200, ysgrifennodd Avi Kivity:
>   
>> I like this.  Since we plan to support multiple vcpus per vm, the fs 
>> structure might look like:
>>     
>
> Three times the syscall overhead is bad for an emulation very bad 

Why? You would usually just call kvm_run(). get/set regs are not needed 
normally.

> for an
> emulation of a CPU whose virtualisation is half baked.
>
>   

Blood rare.  The thing can't even virtualize the first instruction executed.

>> It's certainly a lot more code though, and requires new syscalls.  Since 
>> this is a little esoteric does it warrant new syscalls?
>>     
>
> I think not - ioctl exists to avoid adding a billion esoteric one user
> syscalls. The idea of a VFS sysfs type view of the running vm is great
> for tools however so I wouldn't throw it out entirely or see it as ioctl
> versus fs.
>   

I still want a separate object per vcpu:


    kvm_fd = open("/dev/kvm")
    for (i = 0; i < n; ++i)
        vcpu_fds[i] = ioctl(kvm_fd, KVM_CREATE_VCPU, i)

so the refcounting doesn't bounce cachelines too much.  In effect it's a 
mini filesystem.

-- 
error compiling committee.c: too many arguments to function