Re: [patch, try#2] kvm: fix GFP_KERNEL allocation in atomic section in kvm_dev_ioctl_create_vcpu()

[Ingo Molnar <mingo@elte.hu>]

* Avi Kivity <avi@qumranet.com> wrote:

> Ingo Molnar wrote:
> >Subject: [patch] kvm: fix GFP_KERNEL allocation in atomic section in 
> >kvm_dev_ioctl_create_vcpu()
> >From: Ingo Molnar <mingo@elte.hu>
> >
> >fix an GFP_KERNEL allocation in atomic section: 
> >kvm_dev_ioctl_create_vcpu() called kvm_mmu_init(), which calls 
> >alloc_pages(), while holding the vcpu.
> >
> >The fix is to set up the MMU state in two phases: kvm_mmu_create() and 
> >kvm_mmu_setup().
> >
> >(NOTE: free_vcpus does an kvm_mmu_destroy() call so there's no need
> > for any extra teardown branch on allocation/init failure here.)
> >
> >Signed-off-by: Ingo Molnar <mingo@elte.hu>
> >  
> 
> Applied, thanks.

great!

I've got a security related question as well: vcpu_load() sets up a 
physical CPU's VM registers/state, and vcpu_put() drops that. But 
vcpu_put() only does a put_cpu() call - it does not tear down any VM 
state that has been loaded into the CPU. Is it guaranteed that (hostile) 
user-space cannot use that VM state in any unauthorized way? The state 
is still loaded while arbitrary tasks execute on the CPU. The next 
vcpu_load() will then override it, but the state lingers around forever.

The new x86 VM instructions: vmclear, vmlaunch, vmresume, vmptrld, 
vmread, vmwrite, vmxoff, vmxon are all privileged so i guess it should 
be mostly safe - i'm just wondering whether you thought about this 
attack angle.

ultimately we want to integrate VM state management into the scheduler 
and the context-switch lowlevel arch code, but right now CPU state 
management is done by the KVM 'driver' and there's nothing that isolates 
other tasks from possible side-effects of a loaded VMX/SVN state.

	Ingo