Re: [PATCH 3/3] KVM: x86/mmu: don't skip mmu initialization when mmu root level changes

[Maxim Levitsky <mlevitsk@redhat.com>]

On Wed, 2021-11-10 at 15:48 +0100, Vitaly Kuznetsov wrote:
> Maxim Levitsky <mlevitsk@redhat.com> writes:
> 
> > When running mix of 32 and 64 bit guests, it is possible to have mmu
> > reset with same mmu role but different root level (32 bit vs 64 bit paging)
> > 
> > Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
> > ---
> >  arch/x86/kvm/mmu/mmu.c | 14 ++++++++++----
> >  1 file changed, 10 insertions(+), 4 deletions(-)
> > 
> > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> > index 354d2ca92df4d..763867475860f 100644
> > --- a/arch/x86/kvm/mmu/mmu.c
> > +++ b/arch/x86/kvm/mmu/mmu.c
> > @@ -4745,7 +4745,10 @@ static void init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
> >  	union kvm_mmu_role new_role =
> >  		kvm_calc_tdp_mmu_root_page_role(vcpu, &regs, false);
> >  
> > -	if (new_role.as_u64 == context->mmu_role.as_u64)
> > +	u8 new_root_level = role_regs_to_root_level(&regs);
> > +
> > +	if (new_role.as_u64 == context->mmu_role.as_u64 &&
> > +	    context->root_level == new_root_level)
> >  		return;
> 
> role_regs_to_root_level() uses 3 things: CR0.PG, EFER.LMA and CR4.PAE
> and two of these three are already encoded into extended mmu role
> (kvm_calc_mmu_role_ext()). Could we achieve the same result by adding
> EFER.LMA there?

Absolutely. I just wanted your feedback on this to see if there is any reason to not
do this.

Also it seems that only basic role is compared here.

I don't 100% know the reason why we have basic and extended roles - there is a
comment about basic/extended mmu role to minimize the size of arch.gfn_track,
but I haven't yet studied in depth why.

Best regards,
	Maxim Levitsky

> 
> >  
> >  	context->mmu_role.as_u64 = new_role.as_u64;
> > @@ -4757,7 +4760,7 @@ static void init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
> >  	context->get_guest_pgd = get_cr3;
> >  	context->get_pdptr = kvm_pdptr_read;
> >  	context->inject_page_fault = kvm_inject_page_fault;
> > -	context->root_level = role_regs_to_root_level(&regs);
> > +	context->root_level = new_root_level;
> >  
> >  	if (!is_cr0_pg(context))
> >  		context->gva_to_gpa = nonpaging_gva_to_gpa;
> > @@ -4806,7 +4809,10 @@ static void shadow_mmu_init_context(struct kvm_vcpu *vcpu, struct kvm_mmu *conte
> >  				    struct kvm_mmu_role_regs *regs,
> >  				    union kvm_mmu_role new_role)
> >  {
> > -	if (new_role.as_u64 == context->mmu_role.as_u64)
> > +	u8 new_root_level = role_regs_to_root_level(regs);
> > +
> > +	if (new_role.as_u64 == context->mmu_role.as_u64 &&
> > +	    context->root_level == new_root_level)
> >  		return;
> >  
> >  	context->mmu_role.as_u64 = new_role.as_u64;
> > @@ -4817,8 +4823,8 @@ static void shadow_mmu_init_context(struct kvm_vcpu *vcpu, struct kvm_mmu *conte
> >  		paging64_init_context(context);
> >  	else
> >  		paging32_init_context(context);
> > -	context->root_level = role_regs_to_root_level(regs);
> >  
> > +	context->root_level = new_root_level;
> >  	reset_guest_paging_metadata(vcpu, context);
> >  	context->shadow_root_level = new_role.base.level;