Re: [PATCH v2 06/15] kvm: x86: Add support for activate/de-activate APICv at runtime

[Alexander Graf <graf@amazon.com>]

On 27.08.19 09:29, Vitaly Kuznetsov wrote:
> "Suthikulpanit, Suravee" <Suravee.Suthikulpanit@amd.com> writes:
> 
>> Certain runtime conditions require APICv to be temporary deactivated.
>> However, current implementation only support permanently deactivate
>> APICv at runtime (mainly used when running Hyper-V guest).
>>
>> In addtion, for AMD, when activate / deactivate APICv during runtime,
>> all vcpus in the VM has to be operating in the same APICv mode, which
>> requires the requesting (main) vcpu to notify others.
>>
>> So, introduce interfaces to request all vcpus to activate/deactivate
>> APICv.
>>
>> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
>> ---
>>   arch/x86/include/asm/kvm_host.h |  9 +++++
>>   arch/x86/kvm/x86.c              | 76 +++++++++++++++++++++++++++++++++++++++++
>>   2 files changed, 85 insertions(+)
>>
>> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
>> index 04d7066..dfb7c3d 100644
>> --- a/arch/x86/include/asm/kvm_host.h
>> +++ b/arch/x86/include/asm/kvm_host.h
>> @@ -76,6 +76,10 @@
>>   #define KVM_REQ_HV_STIMER		KVM_ARCH_REQ(22)
>>   #define KVM_REQ_LOAD_EOI_EXITMAP	KVM_ARCH_REQ(23)
>>   #define KVM_REQ_GET_VMCS12_PAGES	KVM_ARCH_REQ(24)
>> +#define KVM_REQ_APICV_ACTIVATE		\
>> +	KVM_ARCH_REQ_FLAGS(25, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
>> +#define KVM_REQ_APICV_DEACTIVATE	\
>> +	KVM_ARCH_REQ_FLAGS(26, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
>>   
>>   #define CR0_RESERVED_BITS                                               \
>>   	(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
>> @@ -1089,6 +1093,7 @@ struct kvm_x86_ops {
>>   	void (*enable_irq_window)(struct kvm_vcpu *vcpu);
>>   	void (*update_cr8_intercept)(struct kvm_vcpu *vcpu, int tpr, int irr);
>>   	bool (*get_enable_apicv)(struct kvm *kvm);
>> +	void (*pre_update_apicv_exec_ctrl)(struct kvm_vcpu *vcpu, bool activate);
>>   	void (*refresh_apicv_exec_ctrl)(struct kvm_vcpu *vcpu);
>>   	void (*hwapic_irr_update)(struct kvm_vcpu *vcpu, int max_irr);
>>   	void (*hwapic_isr_update)(struct kvm_vcpu *vcpu, int isr);
>> @@ -1552,6 +1557,10 @@ int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
>>   
>>   void kvm_make_mclock_inprogress_request(struct kvm *kvm);
>>   void kvm_make_scan_ioapic_request(struct kvm *kvm);
>> +void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu);
>> +void kvm_vcpu_activate_apicv(struct kvm_vcpu *vcpu);
>> +void kvm_make_apicv_activate_request(struct kvm_vcpu *vcpu);
>> +void kvm_make_apicv_deactivate_request(struct kvm_vcpu *vcpu, bool disable);
>>   
>>   void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
>>   				     struct kvm_async_pf *work);
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index f9c3f63..40a20bf 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -26,6 +26,7 @@
>>   #include "cpuid.h"
>>   #include "pmu.h"
>>   #include "hyperv.h"
>> +#include "lapic.h"
>>   
>>   #include <linux/clocksource.h>
>>   #include <linux/interrupt.h>
>> @@ -7163,6 +7164,22 @@ static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
>>   	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
>>   }
>>   
>> +void kvm_vcpu_activate_apicv(struct kvm_vcpu *vcpu)
>> +{
>> +	if (!lapic_in_kernel(vcpu)) {
>> +		WARN_ON_ONCE(!vcpu->arch.apicv_active);
>> +		return;
>> +	}
>> +	if (vcpu->arch.apicv_active)
>> +		return;
>> +
>> +	vcpu->arch.apicv_active = true;
>> +	kvm_apic_update_apicv(vcpu);
>> +
>> +	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
>> +}
>> +EXPORT_SYMBOL_GPL(kvm_vcpu_activate_apicv);
>> +
>>   void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
>>   {
>>   	if (!lapic_in_kernel(vcpu)) {
>> @@ -7173,8 +7190,11 @@ void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
>>   		return;
>>   
>>   	vcpu->arch.apicv_active = false;
>> +	kvm_apic_update_apicv(vcpu);
>> +
>>   	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
>>   }
>> +EXPORT_SYMBOL_GPL(kvm_vcpu_deactivate_apicv);
>>   
>>   int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
>>   {
>> @@ -7668,6 +7688,58 @@ void kvm_make_scan_ioapic_request(struct kvm *kvm)
>>   	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
>>   }
>>   
>> +void kvm_make_apicv_activate_request(struct kvm_vcpu *vcpu)
>> +{
>> +	int i;
>> +	struct kvm_vcpu *v;
>> +	struct kvm *kvm = vcpu->kvm;
>> +
>> +	mutex_lock(&kvm->arch.apicv_lock);
>> +	if (kvm->arch.apicv_state != APICV_DEACTIVATED) {
>> +		mutex_unlock(&kvm->arch.apicv_lock);
>> +		return;
>> +	}
>> +
>> +	kvm_for_each_vcpu(i, v, kvm)
>> +		kvm_clear_request(KVM_REQ_APICV_DEACTIVATE, v);
>> +
>> +	if (kvm_x86_ops->pre_update_apicv_exec_ctrl)
>> +		kvm_x86_ops->pre_update_apicv_exec_ctrl(vcpu, true);
>> +
>> +	kvm->arch.apicv_state = APICV_ACTIVATED;
>> +
>> +	kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_ACTIVATE);
>> +
>> +	mutex_unlock(&kvm->arch.apicv_lock);
>> +}
>> +EXPORT_SYMBOL_GPL(kvm_make_apicv_activate_request);
>> +
>> +void kvm_make_apicv_deactivate_request(struct kvm_vcpu *vcpu, bool disable)
>> +{
>> +	int i;
>> +	struct kvm_vcpu *v;
>> +	struct kvm *kvm = vcpu->kvm;
>> +
>> +	mutex_lock(&kvm->arch.apicv_lock);
>> +	if (kvm->arch.apicv_state == APICV_DEACTIVATED) {
>> +		mutex_unlock(&kvm->arch.apicv_lock);
>> +		return;
>> +	}
>> +
>> +	kvm_for_each_vcpu(i, v, kvm)
>> +		kvm_clear_request(KVM_REQ_APICV_ACTIVATE, v);
> 
> Could you please elaborate on when we need to eat the
> KVM_REQ_APICV_ACTIVATE request here (and KVM_REQ_APICV_DEACTIVATE in
> kvm_make_apicv_activate_request() respectively)? To me, this looks like
> a possible source of hard-to-debug problems in the future.

	CPU0					CPU1

kvm_make_apicv_activate_request()	...
Handle KVM_REQ_APICV_ACTIVATE		KVM_REQ_APICV_ACTIVATE==1
vcpu_run()				...
kvm_make_apicv_deactivate_request()	...

In that case CPU1 would have both activate and deactivate requests 
pending. You can see the same logic above in the activate() function, 
just reverse.

I agree that it probably needs at least a comment to explain why we have 
it. But when I looked at the code, I could not think of a nicer way to 
implement it either.


Alex



Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Ralf Herbrich
Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879