Alexander Graf <g...@amazon.com> writes:
> On 08.11.23 12:17, Nicolas Saenz Julienne wrote:
>> Prepare infrastructure to be able to return data through the XMM
>> registers when Hyper-V hypercalls are issues in fast mode. The XMM
>> registers are exposed to user-space through KVM_EXIT_HYPERV_HCALL and
>> restored on successful hypercall completion.
>>
>> Signed-off-by: Nicolas Saenz Julienne <nsa...@amazon.com>
>> ---
>> arch/x86/include/asm/hyperv-tlfs.h | 2 +-
>> arch/x86/kvm/hyperv.c | 33 +++++++++++++++++++++++++++++-
>> include/uapi/linux/kvm.h | 6 ++++++
>> 3 files changed, 39 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/x86/include/asm/hyperv-tlfs.h
>> b/arch/x86/include/asm/hyperv-tlfs.h
>> index 2ff26f53cd62..af594aa65307 100644
>> --- a/arch/x86/include/asm/hyperv-tlfs.h
>> +++ b/arch/x86/include/asm/hyperv-tlfs.h
>> @@ -49,7 +49,7 @@
>> /* Support for physical CPU dynamic partitioning events is available*/
>> #define HV_X64_CPU_DYNAMIC_PARTITIONING_AVAILABLE BIT(3)
>> /*
>> - * Support for passing hypercall input parameter block via XMM
>> + * Support for passing hypercall input and output parameter block via XMM
>> * registers is available
>> */
>> #define HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE BIT(4)
>> diff --git a/arch/x86/kvm/hyperv.c b/arch/x86/kvm/hyperv.c
>> index 238afd7335e4..e1bc861ab3b0 100644
>> --- a/arch/x86/kvm/hyperv.c
>> +++ b/arch/x86/kvm/hyperv.c
>> @@ -1815,6 +1815,7 @@ struct kvm_hv_hcall {
>> u16 rep_idx;
>> bool fast;
>> bool rep;
>> + bool xmm_dirty;
>> sse128_t xmm[HV_HYPERCALL_MAX_XMM_REGISTERS];
>>
>> /*
>> @@ -2346,9 +2347,33 @@ static int kvm_hv_hypercall_complete(struct kvm_vcpu
>> *vcpu, u64 result)
>> return ret;
>> }
>>
>> +static void kvm_hv_write_xmm(struct kvm_hyperv_xmm_reg *xmm)
>> +{
>> + int reg;
>> +
>> + kvm_fpu_get();
>> + for (reg = 0; reg < HV_HYPERCALL_MAX_XMM_REGISTERS; reg++) {
>> + const sse128_t data = sse128(xmm[reg].low, xmm[reg].high);
>> + _kvm_write_sse_reg(reg, &data);
>> + }
>> + kvm_fpu_put();
>> +}
>> +
>> +static bool kvm_hv_is_xmm_output_hcall(u16 code)
>> +{
>> + return false;
>> +}
>> +
>> static int kvm_hv_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
>> {
>> - return kvm_hv_hypercall_complete(vcpu,
>> vcpu->run->hyperv.u.hcall.result);
>> + bool fast = !!(vcpu->run->hyperv.u.hcall.input & HV_HYPERCALL_FAST_BIT);
>> + u16 code = vcpu->run->hyperv.u.hcall.input & 0xffff;
>> + u64 result = vcpu->run->hyperv.u.hcall.result;
>> +
>> + if (kvm_hv_is_xmm_output_hcall(code) && hv_result_success(result) &&
>> fast)
>> + kvm_hv_write_xmm(vcpu->run->hyperv.u.hcall.xmm);
>> +
>> + return kvm_hv_hypercall_complete(vcpu, result);
>> }
>>
>> static u16 kvm_hvcall_signal_event(struct kvm_vcpu *vcpu, struct
>> kvm_hv_hcall *hc)
>> @@ -2623,6 +2648,9 @@ int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
>> break;
>> }
>>
>> + if ((ret & HV_HYPERCALL_RESULT_MASK) == HV_STATUS_SUCCESS &&
>> hc.xmm_dirty)
>> + kvm_hv_write_xmm((struct kvm_hyperv_xmm_reg*)hc.xmm);
>> +
>> hypercall_complete:
>> return kvm_hv_hypercall_complete(vcpu, ret);
>>
>> @@ -2632,6 +2660,8 @@ int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
>> vcpu->run->hyperv.u.hcall.input = hc.param;
>> vcpu->run->hyperv.u.hcall.params[0] = hc.ingpa;
>> vcpu->run->hyperv.u.hcall.params[1] = hc.outgpa;
>> + if (hc.fast)
>> + memcpy(vcpu->run->hyperv.u.hcall.xmm, hc.xmm, sizeof(hc.xmm));
>> vcpu->arch.complete_userspace_io = kvm_hv_hypercall_complete_userspace;
>> return 0;
>> }
>> @@ -2780,6 +2810,7 @@ int kvm_get_hv_cpuid(struct kvm_vcpu *vcpu, struct
>> kvm_cpuid2 *cpuid,
>> ent->ebx |= HV_ENABLE_EXTENDED_HYPERCALLS;
>>
>> ent->edx |= HV_X64_HYPERCALL_XMM_INPUT_AVAILABLE;
>> + ent->edx |= HV_X64_HYPERCALL_XMM_OUTPUT_AVAILABLE;
>
>
> Shouldn't this be guarded by an ENABLE_CAP to make sure old user space
> that doesn't know about xmm outputs is still able to run with newer kernels?
>
No, we don't do CAPs for new Hyper-V features anymore since we have
KVM_GET_SUPPORTED_HV_CPUID. Userspace is not supposed to simply copy
its output into guest visible CPUIDs, it must only enable features it
knows. Even 'hv_passthrough' option in QEMU doesn't pass unknown
features through.
>
>> ent->edx |= HV_FEATURE_FREQUENCY_MSRS_AVAILABLE;
>> ent->edx |= HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE;
>>
>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
>> index d7a01766bf21..5ce06a1eee2b 100644
>> --- a/include/uapi/linux/kvm.h
>> +++ b/include/uapi/linux/kvm.h
>> @@ -192,6 +192,11 @@ struct kvm_s390_cmma_log {
>> __u64 values;
>> };
>>
>> +struct kvm_hyperv_xmm_reg {
>> + __u64 low;
>> + __u64 high;
>> +};
>> +
>> struct kvm_hyperv_exit {
>> #define KVM_EXIT_HYPERV_SYNIC 1
>> #define KVM_EXIT_HYPERV_HCALL 2
>> @@ -210,6 +215,7 @@ struct kvm_hyperv_exit {
>> __u64 input;
>> __u64 result;
>> __u64 params[2];
>> + struct kvm_hyperv_xmm_reg xmm[6];
>
>
> Would this change the size of struct kvm_hyperv_exit? And if so,
> wouldn't that potentially be a UABI breakage?
>
Yes. 'struct kvm_hyperv_exit' has 'type' field which determines which
particular type of the union (synic/hcall/syndbg) is used. The easiest
would probably be to introduce a new type (hcall_with_xmm or something
like that).
>
> Alex
>
>
>
>
> Amazon Development Center Germany GmbH
> Krausenstr. 38
> 10117 Berlin
> Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss
> Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
> Sitz: Berlin
> Ust-ID: DE 289 237 879
>
>
--
Vitaly
