Re: [PATCH] x86/kvm/mmu: make mmu->prev_roots cache work for NPT case

[Vitaly Kuznetsov <vkuznets@redhat.com>]

Paolo Bonzini <pbonzini@redhat.com> writes:

> On 22/02/19 17:46, Vitaly Kuznetsov wrote:
>> I noticed that fast_cr3_switch() always fails when we switch back from L2
>> to L1 as it is not able to find a cached root. This is odd: host's CR3
>> usually stays the same, we expect to always follow the fast path. Turns
>> out the problem is that page role is always mismatched because
>> kvm_mmu_get_page() filters out cr4_pae when direct, the value is stored
>> in page header and later compared with new_role in cached_root_available().
>> As cr4_pae is always set in long mode prev_roots cache is dysfunctional.
>
> Really cr4_pae means "are the PTEs 8 bytes".  So I think your patch is
> correct but on top we should set it to 1 (not zero!!) for
> kvm_calc_shadow_ept_root_page_role, init_kvm_nested_mmu and
> kvm_calc_tdp_mmu_root_page_role.  Or maybe everything breaks with that
> change.
>

Yes, exactly. If we put '1' there kvm_mmu_get_page() will again filter
it out and we won't be able to find the root in prev_roots cache :-(

>> - Do not clear cr4_pae in kvm_mmu_get_page() and check direct on call sites
>>  (detect_write_misaligned(), get_written_sptes()).
>
> These only run with shadow page tables, by the way.
>

Yes, and that's why I think it may make sense to move the filtering
logic there. At least in other cases cr4_pae will always be equal to
is_pae().

It seems I know too little about shadow paging and all these corner
cases :-(

-- 
Vitaly