Gavin Shan <gshan@xxxxxxxxxx> writes:
> Hi Vitaly,
>
> On 4/29/20 7:36 PM, Vitaly Kuznetsov wrote:
>> Currently, APF mechanism relies on the #PF abuse where the token is being
>> passed through CR2. If we switch to using interrupts to deliver page-ready
>> notifications we need a different way to pass the data. Extent the existing
>> 'struct kvm_vcpu_pv_apf_data' with token information.
>>
>> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
>> ---
>> arch/x86/include/uapi/asm/kvm_para.h | 3 ++-
>> arch/x86/kvm/x86.c | 10 ++++++----
>> 2 files changed, 8 insertions(+), 5 deletions(-)
>>
>> diff --git a/arch/x86/include/uapi/asm/kvm_para.h b/arch/x86/include/uapi/asm/kvm_para.h
>> index 2a8e0b6b9805..df2ba34037a2 100644
>> --- a/arch/x86/include/uapi/asm/kvm_para.h
>> +++ b/arch/x86/include/uapi/asm/kvm_para.h
>> @@ -113,7 +113,8 @@ struct kvm_mmu_op_release_pt {
>>
>> struct kvm_vcpu_pv_apf_data {
>> __u32 reason;
>> - __u8 pad[60];
>> + __u32 token;
>> + __u8 pad[56];
>> __u32 enabled;
>> };
>>
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index b93133ee07ba..7c21c0cf0a33 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -2662,7 +2662,7 @@ static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
>> }
>>
>> if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
>> - sizeof(u32)))
>> + sizeof(u64)))
>> return 1;
>>
>> vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
>> @@ -10352,8 +10352,9 @@ static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
>> }
>> }
>>
>> -static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
>> +static int apf_put_user(struct kvm_vcpu *vcpu, u32 reason, u32 token)
>> {
>> + u64 val = (u64)token << 32 | reason;
>>
>> return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
>> sizeof(val));
>> @@ -10405,7 +10406,8 @@ void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
>> kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
>>
>> if (kvm_can_deliver_async_pf(vcpu) &&
>> - !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
>> + !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT,
>> + work->arch.token)) {
>> fault.vector = PF_VECTOR;
>> fault.error_code_valid = true;
>> fault.error_code = 0;
>> @@ -10438,7 +10440,7 @@ void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
>> trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
>>
>> if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
>> - !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
>> + !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY, work->arch.token)) {
>> fault.vector = PF_VECTOR;
>> fault.error_code_valid = true;
>> fault.error_code = 0;
>>
>
> It would be as below based on two facts: (1) token is more important than reason;
> (2) token will be put into high word of @val. I think apf_{get,put}_user() might
> be worthy to be inline. However, it's not a big deal.
This is to be changed in v1 as we agreed to drop page-ready delivery via
#PF completely.
> static inline int apf_put_user(struct kvm_vcpu *vcpu, u32 token, u32 reason)
>
Yes, it makes sense to inline these. Thanks!
--
Vitaly
