Re: [PATCH v2 7/9] x86/kvm/nVMX: introduce source data cache for kvm_init_shadow_ept_mmu()

[Vitaly Kuznetsov]

Sean Christopherson <sean.j.christopherson@xxxxxxxxx> writes:

> On Tue, Sep 25, 2018 at 07:58:42PM +0200, Vitaly Kuznetsov wrote:
>> MMU re-initialization is expensive, in particular,
>> update_permission_bitmask() and update_pkru_bitmask() are.
>> 
>> Cache the data used to setup shadow EPT MMU and avoid full re-init when
>> it is unchanged.
>> 
>> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
>> ---
>>  arch/x86/include/asm/kvm_host.h | 14 +++++++++
>>  arch/x86/kvm/mmu.c              | 51 ++++++++++++++++++++++++---------
>>  2 files changed, 52 insertions(+), 13 deletions(-)
>> 
>> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
>> index 1821b0215230..87ddaa1579e7 100644
>> --- a/arch/x86/include/asm/kvm_host.h
>> +++ b/arch/x86/include/asm/kvm_host.h
>> @@ -274,7 +274,21 @@ union kvm_mmu_page_role {
>>  };
>>  
>>  union kvm_mmu_extended_role {
>> +/*
>> + * This structure complements kvm_mmu_page_role caching everything needed for
>> + * MMU configuration. If nothing in both these structures changed, MMU
>> + * re-configuration can be skipped. @valid bit is set on first usage so we don't
>> + * treat all-zero structure as valid data.
>> + */
>>  	u32 word;
>> +	struct {
>> +		unsigned int valid:1;
>> +		unsigned int execonly:1;
>> +		unsigned int cr4_pse:1;
>> +		unsigned int cr4_pke:1;
>> +		unsigned int cr4_smap:1;
>> +		unsigned int cr4_smep:1;
>> +	};
>>  };
>>  
>>  union kvm_mmu_role {
>> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
>> index bb1ef0f68f8e..d8611914544a 100644
>> --- a/arch/x86/kvm/mmu.c
>> +++ b/arch/x86/kvm/mmu.c
>> @@ -4708,6 +4708,24 @@ static void paging32E_init_context(struct kvm_vcpu *vcpu,
>>  	paging64_init_context_common(vcpu, context, PT32E_ROOT_LEVEL);
>>  }
>>  
>> +static union kvm_mmu_role
>> +kvm_calc_mmu_role_common(struct kvm_vcpu *vcpu)
>> +{
>> +	union kvm_mmu_role role = {0};
>> +
>> +	role.base.access = ACC_ALL;
>> +	role.base.cr0_wp = is_write_protection(vcpu);
>> +
>> +	role.ext.cr4_smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP) != 0;
>> +	role.ext.cr4_smap = kvm_read_cr4_bits(vcpu, X86_CR4_SMAP) != 0;
>> +	role.ext.cr4_pse = !!is_pse(vcpu);
>> +	role.ext.cr4_pke = kvm_read_cr4_bits(vcpu, X86_CR4_PKE) != 0;
>> +
>> +	role.ext.valid = 1;
>> +
>> +	return role;
>> +}
>> +
>>  static union kvm_mmu_page_role
>>  kvm_calc_tdp_mmu_root_page_role(struct kvm_vcpu *vcpu)
>>  {
>> @@ -4814,16 +4832,18 @@ void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu)
>>  }
>>  EXPORT_SYMBOL_GPL(kvm_init_shadow_mmu);
>>  
>> -static union kvm_mmu_page_role
>> -kvm_calc_shadow_ept_root_page_role(struct kvm_vcpu *vcpu, bool accessed_dirty)
>> +static union kvm_mmu_role
>> +kvm_calc_shadow_ept_root_page_role(struct kvm_vcpu *vcpu, bool accessed_dirty,
>> +				   bool execonly)
>>  {
>> -	union kvm_mmu_page_role role = vcpu->arch.mmu->mmu_role.base;
>> +	union kvm_mmu_role role = kvm_calc_mmu_role_common(vcpu);
>
> kvm_calc_mmu_role_common() doesn't preserve the current mmu_role.base
> and kvm_calc_mmu_role_common() doesn't capture all base fields.  Won't
> @role will be incorrect for base fields that aren't set below, e.g.
> cr4_pae, smep_andnot_wp, smap_andnot_wp, etc...

Oh, I see what you mean. Actually, PATCH8 of this series adds some of
this stuff but smep_andnot_wp and smap_andnot_wp are still not set. I
think I'll enhance kvm_calc_mmu_role_common() and move some stuff from
PATCH8 to this one.
(The fact that @role is currently not fully re-initialized here is very
unobvious so I would definitely prefer to explicitly initialize
everything over inheriting something from previously initialized role).

Thanks!

>
>>  
>> -	role.level = PT64_ROOT_4LEVEL;
>> -	role.direct = false;
>> -	role.ad_disabled = !accessed_dirty;
>> -	role.guest_mode = true;
>> -	role.access = ACC_ALL;
>> +	role.base.level = PT64_ROOT_4LEVEL;
>> +	role.base.direct = false;
>> +	role.base.ad_disabled = !accessed_dirty;
>> +	role.base.guest_mode = true;
>> +
>> +	role.ext.execonly = execonly;
>>  
>>  	return role;
>>  }
>> @@ -4832,10 +4852,16 @@ void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
>>  			     bool accessed_dirty, gpa_t new_eptp)
>>  {
>>  	struct kvm_mmu *context = vcpu->arch.mmu;
>> -	union kvm_mmu_page_role root_page_role =
>> -		kvm_calc_shadow_ept_root_page_role(vcpu, accessed_dirty);
>> +	union kvm_mmu_role new_role =
>> +		kvm_calc_shadow_ept_root_page_role(vcpu, accessed_dirty,
>> +						   execonly);
>> +
>> +	__kvm_mmu_new_cr3(vcpu, new_eptp, new_role.base, false);
>> +
>> +	new_role.base.word &= mmu_base_role_mask.word;
>> +	if (new_role.as_u64 == context->mmu_role.as_u64)
>> +		return;
>>  
>> -	__kvm_mmu_new_cr3(vcpu, new_eptp, root_page_role, false);
>>  	context->shadow_root_level = PT64_ROOT_4LEVEL;
>>  
>>  	context->nx = true;
>> @@ -4847,8 +4873,7 @@ void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
>>  	context->update_pte = ept_update_pte;
>>  	context->root_level = PT64_ROOT_4LEVEL;
>>  	context->direct_map = false;
>> -	context->mmu_role.base.word =
>> -		root_page_role.word & mmu_base_role_mask.word;
>> +	context->mmu_role.as_u64 = new_role.as_u64;
>>  	context->get_pdptr = kvm_pdptr_read;
>>  
>>  	update_permission_bitmask(vcpu, context, true);
>> -- 
>> 2.17.1
>> 

-- 
Vitaly