Re: [PATCH v5 7/7] KVM: SVM: hyper-v: Direct Virtual Flush support

[Vitaly Kuznetsov <vkuznets@redhat.com>]

Vineeth Pillai <viremana@linux.microsoft.com> writes:

> From Hyper-V TLFS:
>  "The hypervisor exposes hypercalls (HvFlushVirtualAddressSpace,
>   HvFlushVirtualAddressSpaceEx, HvFlushVirtualAddressList, and
>   HvFlushVirtualAddressListEx) that allow operating systems to more
>   efficiently manage the virtual TLB. The L1 hypervisor can choose to
>   allow its guest to use those hypercalls and delegate the responsibility
>   to handle them to the L0 hypervisor. This requires the use of a
>   partition assist page."
>
> Add the Direct Virtual Flush support for SVM.
>
> Related VMX changes:
> commit 6f6a657c9998 ("KVM/Hyper-V/VMX: Add direct tlb flush support")
>
> Signed-off-by: Vineeth Pillai <viremana@linux.microsoft.com>
> ---
>  arch/x86/kvm/Makefile           |  4 ++++
>  arch/x86/kvm/svm/svm.c          |  2 ++
>  arch/x86/kvm/svm/svm_onhyperv.c | 41 +++++++++++++++++++++++++++++++++
>  arch/x86/kvm/svm/svm_onhyperv.h | 36 +++++++++++++++++++++++++++++
>  4 files changed, 83 insertions(+)
>  create mode 100644 arch/x86/kvm/svm/svm_onhyperv.c
>
> diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
> index a06745c2fef1..83331376b779 100644
> --- a/arch/x86/kvm/Makefile
> +++ b/arch/x86/kvm/Makefile
> @@ -32,6 +32,10 @@ kvm-intel-$(CONFIG_X86_SGX_KVM)	+= vmx/sgx.o
>  
>  kvm-amd-y		+= svm/svm.o svm/vmenter.o svm/pmu.o svm/nested.o svm/avic.o svm/sev.o
>  
> +ifdef CONFIG_HYPERV
> +kvm-amd-y		+= svm/svm_onhyperv.o
> +endif
> +
>  obj-$(CONFIG_KVM)	+= kvm.o
>  obj-$(CONFIG_KVM_INTEL)	+= kvm-intel.o
>  obj-$(CONFIG_KVM_AMD)	+= kvm-amd.o
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index d2a625411059..5139cb6baadc 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -3779,6 +3779,8 @@ static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
>  	}
>  	svm->vmcb->save.cr2 = vcpu->arch.cr2;
>  
> +	svm_hv_update_vp_id(svm->vmcb, vcpu);
> +
>  	/*
>  	 * Run with all-zero DR6 unless needed, so that we can get the exact cause
>  	 * of a #DB.
> diff --git a/arch/x86/kvm/svm/svm_onhyperv.c b/arch/x86/kvm/svm/svm_onhyperv.c
> new file mode 100644
> index 000000000000..3281856ebd94
> --- /dev/null
> +++ b/arch/x86/kvm/svm/svm_onhyperv.c
> @@ -0,0 +1,41 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * KVM L1 hypervisor optimizations on Hyper-V for SVM.
> + */
> +
> +#include <linux/kvm_host.h>
> +#include "kvm_cache_regs.h"
> +
> +#include <asm/mshyperv.h>
> +
> +#include "svm.h"
> +#include "svm_ops.h"
> +
> +#include "hyperv.h"
> +#include "kvm_onhyperv.h"
> +#include "svm_onhyperv.h"
> +
> +int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
> +{
> +	struct hv_enlightenments *hve;
> +	struct hv_partition_assist_pg **p_hv_pa_pg =
> +			&to_kvm_hv(vcpu->kvm)->hv_pa_pg;
> +
> +	if (!*p_hv_pa_pg)
> +		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
> +
> +	if (!*p_hv_pa_pg)
> +		return -ENOMEM;
> +
> +	hve = (struct hv_enlightenments *)to_svm(vcpu)->vmcb->control.reserved_sw;
> +
> +	hve->partition_assist_page = __pa(*p_hv_pa_pg);
> +	hve->hv_vm_id = (unsigned long)vcpu->kvm;
> +	if (!hve->hv_enlightenments_control.nested_flush_hypercall) {
> +		hve->hv_enlightenments_control.nested_flush_hypercall = 1;
> +		vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_HV_NESTED_ENLIGHTENMENTS);
> +	}
> +
> +	return 0;
> +}
> +
> diff --git a/arch/x86/kvm/svm/svm_onhyperv.h b/arch/x86/kvm/svm/svm_onhyperv.h
> index 0f262460b2e6..7487052fcef8 100644
> --- a/arch/x86/kvm/svm/svm_onhyperv.h
> +++ b/arch/x86/kvm/svm/svm_onhyperv.h
> @@ -36,6 +36,8 @@ struct hv_enlightenments {
>   */
>  #define VMCB_HV_NESTED_ENLIGHTENMENTS VMCB_SW
>  
> +int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu);
> +
>  static inline void svm_hv_init_vmcb(struct vmcb *vmcb)
>  {
>  	struct hv_enlightenments *hve =
> @@ -55,6 +57,23 @@ static inline void svm_hv_hardware_setup(void)
>  		svm_x86_ops.tlb_remote_flush_with_range =
>  				hv_remote_flush_tlb_with_range;
>  	}
> +
> +	if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH) {
> +		int cpu;
> +
> +		pr_info("kvm: Hyper-V Direct TLB Flush enabled\n");
> +		for_each_online_cpu(cpu) {
> +			struct hv_vp_assist_page *vp_ap =
> +				hv_get_vp_assist_page(cpu);
> +
> +			if (!vp_ap)
> +				continue;
> +
> +			vp_ap->nested_control.features.directhypercall = 1;
> +		}
> +		svm_x86_ops.enable_direct_tlbflush =
> +				hv_enable_direct_tlbflush;
> +	}
>  }
>  
>  static inline void svm_hv_vmcb_dirty_nested_enlightenments(
> @@ -74,6 +93,18 @@ static inline void svm_hv_vmcb_dirty_nested_enlightenments(
>  	    hve->hv_enlightenments_control.msr_bitmap)
>  		vmcb_mark_dirty(vmcb, VMCB_HV_NESTED_ENLIGHTENMENTS);
>  }
> +
> +static inline void svm_hv_update_vp_id(struct vmcb *vmcb,
> +		struct kvm_vcpu *vcpu)
> +{
> +	struct hv_enlightenments *hve =
> +		(struct hv_enlightenments *)vmcb->control.reserved_sw;
> +
> +	if (hve->hv_vp_id != to_hv_vcpu(vcpu)->vp_index) {
> +		hve->hv_vp_id = to_hv_vcpu(vcpu)->vp_index;
> +		vmcb_mark_dirty(vmcb, VMCB_HV_NESTED_ENLIGHTENMENTS);
> +	}

This blows up in testing when no Hyper-V context was created on a vCPU,
e.g. when running KVM selftests (to_hv_vcpu(vcpu) is NULL when no
Hyper-V emulation features were requested on a vCPU but
svm_hv_update_vp_id() is called unconditionally by svm_vcpu_run()).

I'll be sending a patch to fix the immediate issue but I was wondering
why we need to call svm_hv_update_vp_id() from svm_vcpu_run() as VP
index is unlikely to change; we can probably just call it from
kvm_hv_set_msr() instead.


> +}
>  #else
>  
>  static inline void svm_hv_init_vmcb(struct vmcb *vmcb)
> @@ -88,6 +119,11 @@ static inline void svm_hv_vmcb_dirty_nested_enlightenments(
>  		struct kvm_vcpu *vcpu)
>  {
>  }
> +
> +static inline void svm_hv_update_vp_id(struct vmcb *vmcb,
> +		struct kvm_vcpu *vcpu)
> +{
> +}
>  #endif /* CONFIG_HYPERV */
>  
>  #endif /* __ARCH_X86_KVM_SVM_ONHYPERV_H__ */

-- 
Vitaly