Re: [PATCH 1/3] KVM: x86: Copy kvm_pmu_ops by value to eliminate layer of indirection

[Like Xu <like.xu.linux@gmail.com>]

On 5/11/2021 11:30 pm, Sean Christopherson wrote:
> On Wed, Nov 03, 2021, Like Xu wrote:
>> Replace the kvm_pmu_ops pointer in common x86 with an instance of the
>> struct to save one pointer dereference when invoking functions. Copy the
>> struct by value to set the ops during kvm_init().
>>
>> Using kvm_x86_ops.hardware_enable to track whether or not the
>> ops have been initialized, i.e. a vendor KVM module has been loaded.
>>
>> Signed-off-by: Like Xu <likexu@tencent.com>
>> ---
>>   arch/x86/kvm/pmu.c        | 41 +++++++++++++++++++++------------------
>>   arch/x86/kvm/pmu.h        |  4 +++-
>>   arch/x86/kvm/vmx/nested.c |  2 +-
>>   arch/x86/kvm/x86.c        |  3 +++
>>   4 files changed, 29 insertions(+), 21 deletions(-)
>>
>> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
>> index 0772bad9165c..0db1887137d9 100644
>> --- a/arch/x86/kvm/pmu.c
>> +++ b/arch/x86/kvm/pmu.c
>> @@ -47,6 +47,9 @@
>>    *        * AMD:   [0 .. AMD64_NUM_COUNTERS-1] <=> gp counters
>>    */
>>   
>> +struct kvm_pmu_ops kvm_pmu_ops __read_mostly;
>> +EXPORT_SYMBOL_GPL(kvm_pmu_ops);
>> +
> 
> ...
> 
>> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
>> index b4ee5e9f9e20..1e793e44b5ff 100644
>> --- a/arch/x86/kvm/vmx/nested.c
>> +++ b/arch/x86/kvm/vmx/nested.c
>> @@ -4796,7 +4796,7 @@ void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
>>   		return;
>>   
>>   	vmx = to_vmx(vcpu);
>> -	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
>> +	if (kvm_pmu_ops.is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
> 
> I would much prefer we export kvm_pmu_is_valid_msr() and go through that for nVMX
> than export all of kvm_pmu_ops for this one case.

Applied. Is it an abuse to export a function for only one case ?

> 
>>   		vmx->nested.msrs.entry_ctls_high |=
>>   				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
>>   		vmx->nested.msrs.exit_ctls_high |=
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index ac83d873d65b..72d286595012 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -11317,6 +11317,9 @@ int kvm_arch_hardware_setup(void *opaque)
>>   	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
>>   	kvm_ops_static_call_update();
>>   
>> +	if (kvm_x86_ops.hardware_enable)
> 
> Huh?  Did you intend this to be?
> 
> 	if (kvm_x86_ops.pmu_ops)
> 
> Either way, I don't see the point, VMX and SVM unconditionally provide the ops.

Let me drop it.

> 
> I would also say land this memcpy() above kvm_ops_static_call_update(), then the
> enabling patch can do the static call updates in kvm_ops_static_call_update()
> instead of adding another helper.

Both applied.

> 
>> +		memcpy(&kvm_pmu_ops, kvm_x86_ops.pmu_ops, sizeof(kvm_pmu_ops));
> 
> As part of this change, the pmu_ops should be moved to kvm_x86_init_ops and tagged
> as __initdata.  That'll save those precious few bytes, and more importantly make
> the original ops unreachable, i.e. make it harder to sneak in post-init modification
> bugs.

Applied. Opportunistically,

-struct kvm_pmu_ops [amd|intel]_pmu_ops = {
+struct kvm_pmu_ops [amd|intel]_pmu_ops __initdata = {

> 
>> +
>>   	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
>>   		supported_xss = 0;
>>   
>> -- 
>> 2.33.0
>>
>