Re: [RFC 39/55] KVM: arm/arm64: Add mmu context for the nesting

From: Christoffer Dall
Date: Wed Feb 22 2017 - 08:35:15 EST


On Mon, Jan 09, 2017 at 01:24:35AM -0500, Jintack Lim wrote:
> Add the shadow stage-2 MMU context to be used for the nesting, but don't
> do anything with it yet.
>
> The host hypervisor maintains mmu structures for each nested VM. When
> entering a nested VM, the host hypervisor searches for the nested VM's
> mmu using vmid as a key. Note that this vmid is from the guest
> hypervisor's point of view.

I feel like I'm missing some overall design description or rationale of
why this is needed. Can you expand on this commit message a bit?

>
> Signed-off-by: Jintack Lim <jintack@xxxxxxxxxxxxxxx>
> ---
> arch/arm/include/asm/kvm_host.h | 3 ++
> arch/arm/kvm/arm.c | 1 +
> arch/arm64/include/asm/kvm_emulate.h | 13 ++++-----
> arch/arm64/include/asm/kvm_host.h | 19 +++++++++++++
> arch/arm64/include/asm/kvm_mmu.h | 31 ++++++++++++++++++++
> arch/arm64/kvm/Makefile | 1 +
> arch/arm64/kvm/context.c | 2 +-
> arch/arm64/kvm/mmu-nested.c | 55 ++++++++++++++++++++++++++++++++++++
> 8 files changed, 116 insertions(+), 9 deletions(-)
> create mode 100644 arch/arm64/kvm/mmu-nested.c
>
> diff --git a/arch/arm/include/asm/kvm_host.h b/arch/arm/include/asm/kvm_host.h
> index da45394..fbde48d 100644
> --- a/arch/arm/include/asm/kvm_host.h
> +++ b/arch/arm/include/asm/kvm_host.h
> @@ -82,6 +82,9 @@ struct kvm_arch {
> * here.
> */
>
> + /* Never used on arm but added to be compatible with arm64 */
> + struct list_head nested_mmu_list;
> +
> /* Interrupt controller */
> struct vgic_dist vgic;
> int max_vcpus;
> diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
> index 371b38e7..147df97 100644
> --- a/arch/arm/kvm/arm.c
> +++ b/arch/arm/kvm/arm.c
> @@ -146,6 +146,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
> /* Mark the initial VMID generation invalid */
> kvm->arch.mmu.vmid.vmid_gen = 0;
> kvm->arch.mmu.el2_vmid.vmid_gen = 0;
> + INIT_LIST_HEAD(&kvm->arch.nested_mmu_list);
>
> /* The maximum number of VCPUs is limited by the host's GIC model */
> kvm->arch.max_vcpus = vgic_present ?
> diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
> index 94068e7..abad676 100644
> --- a/arch/arm64/include/asm/kvm_emulate.h
> +++ b/arch/arm64/include/asm/kvm_emulate.h
> @@ -183,6 +183,11 @@ static inline bool vcpu_el2_imo_is_set(const struct kvm_vcpu *vcpu)
> return (vcpu_el2_reg(vcpu, HCR_EL2) & HCR_IMO);
> }
>
> +static inline bool vcpu_nested_stage2_enabled(const struct kvm_vcpu *vcpu)
> +{
> + return (vcpu_el2_reg(vcpu, HCR_EL2) & HCR_VM);
> +}
> +
> static inline u32 kvm_vcpu_get_hsr(const struct kvm_vcpu *vcpu)
> {
> return vcpu->arch.fault.esr_el2;
> @@ -363,12 +368,4 @@ static inline unsigned long vcpu_data_host_to_guest(struct kvm_vcpu *vcpu,
> return data; /* Leave LE untouched */
> }
>
> -static inline struct kvm_s2_vmid *vcpu_get_active_vmid(struct kvm_vcpu *vcpu)
> -{
> - if (unlikely(vcpu_mode_el2(vcpu)))
> - return &vcpu->kvm->arch.mmu.el2_vmid;
> -
> - return &vcpu->kvm->arch.mmu.vmid;
> -}
> -
> #endif /* __ARM64_KVM_EMULATE_H__ */
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index b33d35d..23e2267 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -65,6 +65,22 @@ struct kvm_s2_mmu {
> pgd_t *pgd;
> };
>
> +/* Per nested VM mmu structure */
> +struct kvm_nested_s2_mmu {
> + struct kvm_s2_mmu mmu;
> +
> + /*
> + * The vttbr value set by the guest hypervisor for this nested VM.
> + * vmid field is used as a key to search for this mmu structure among
> + * all nested VM mmu structures by the host hypervisor.
> + * baddr field is used to determine if we need to unmap stage 2
> + * shadow page tables.
> + */

I don't really understand this comment in isolation - especially not the
baddr part.

> + u64 virtual_vttbr;
> +
> + struct list_head list;
> +};
> +
> struct kvm_arch {
> /* Stage 2 paging state for the VM */
> struct kvm_s2_mmu mmu;
> @@ -80,6 +96,9 @@ struct kvm_arch {
>
> /* Timer */
> struct arch_timer_kvm timer;
> +
> + /* Stage 2 shadow paging contexts for nested L2 VM */
> + struct list_head nested_mmu_list;
> };
>
> #define KVM_NR_MEM_OBJS 40
> diff --git a/arch/arm64/include/asm/kvm_mmu.h b/arch/arm64/include/asm/kvm_mmu.h
> index a504162..d1ef650 100644
> --- a/arch/arm64/include/asm/kvm_mmu.h
> +++ b/arch/arm64/include/asm/kvm_mmu.h
> @@ -112,6 +112,7 @@
> #include <asm/cacheflush.h>
> #include <asm/mmu_context.h>
> #include <asm/pgtable.h>
> +#include <asm/kvm_emulate.h>
>
> static inline unsigned long __kern_hyp_va(unsigned long v)
> {
> @@ -323,6 +324,21 @@ static inline unsigned int kvm_get_vmid_bits(void)
> return (cpuid_feature_extract_unsigned_field(reg, ID_AA64MMFR1_VMIDBITS_SHIFT) == 2) ? 16 : 8;
> }
>
> +#ifdef CONFIG_KVM_ARM_NESTED_HYP
> +struct kvm_nested_s2_mmu *get_nested_mmu(struct kvm_vcpu *vcpu, u64 vttbr);
> +struct kvm_s2_mmu *vcpu_get_active_s2_mmu(struct kvm_vcpu *vcpu);
> +#else
> +static inline struct kvm_nested_s2_mmu *get_nested_mmu(struct kvm_vcpu *vcpu,
> + u64 vttbr)
> +{
> + return NULL;
> +}
> +static inline struct kvm_s2_mmu *vcpu_get_active_s2_mmu(struct kvm_vcpu *vcpu)
> +{
> + return &vcpu->kvm->arch.mmu;
> +}
> +#endif
> +
> static inline u64 kvm_get_vttbr(struct kvm_s2_vmid *vmid,
> struct kvm_s2_mmu *mmu)
> {
> @@ -334,5 +350,20 @@ static inline u64 kvm_get_vttbr(struct kvm_s2_vmid *vmid,
> return baddr | vmid_field;
> }
>
> +static inline u64 get_vmid(u64 vttbr)
> +{
> + return (vttbr & VTTBR_VMID_MASK(get_kvm_vmid_bits()))>>VTTBR_VMID_SHIFT;

whitespacealertbetweentheshiftmarker.

> +}
> +
> +static inline struct kvm_s2_vmid *vcpu_get_active_vmid(struct kvm_vcpu *vcpu)
> +{
> + struct kvm_s2_mmu *mmu = vcpu_get_active_s2_mmu(vcpu);
> +
> + if (unlikely(vcpu_mode_el2(vcpu)))
> + return &mmu->el2_vmid;
> + else
> + return &mmu->vmid;
> +}
> +
> #endif /* __ASSEMBLY__ */
> #endif /* __ARM64_KVM_MMU_H__ */
> diff --git a/arch/arm64/kvm/Makefile b/arch/arm64/kvm/Makefile
> index 8573faf..b0b1074 100644
> --- a/arch/arm64/kvm/Makefile
> +++ b/arch/arm64/kvm/Makefile
> @@ -36,5 +36,6 @@ kvm-$(CONFIG_KVM_ARM_HOST) += $(KVM)/arm/arch_timer.o
> kvm-$(CONFIG_KVM_ARM_PMU) += $(KVM)/arm/pmu.o
>
> kvm-$(CONFIG_KVM_ARM_NESTED_HYP) += handle_exit_nested.o
> +kvm-$(CONFIG_KVM_ARM_NESTED_HYP) += mmu-nested.o
> kvm-$(CONFIG_KVM_ARM_NESTED_HYP) += emulate-nested.o
> kvm-$(CONFIG_KVM_ARM_NESTED_HYP) += $(KVM)/arm/vgic/vgic-v2-nested.o
> diff --git a/arch/arm64/kvm/context.c b/arch/arm64/kvm/context.c
> index b2c0220..9ebc38f 100644
> --- a/arch/arm64/kvm/context.c
> +++ b/arch/arm64/kvm/context.c
> @@ -91,7 +91,7 @@ static void create_shadow_el1_sysregs(struct kvm_vcpu *vcpu)
>
> static void setup_s2_mmu(struct kvm_vcpu *vcpu)
> {
> - struct kvm_s2_mmu *mmu = &vcpu->kvm->arch.mmu;
> + struct kvm_s2_mmu *mmu = vcpu_get_active_s2_mmu(vcpu);
> struct kvm_s2_vmid *vmid = vcpu_get_active_vmid(vcpu);
>
> vcpu->arch.hw_vttbr = kvm_get_vttbr(vmid, mmu);
> diff --git a/arch/arm64/kvm/mmu-nested.c b/arch/arm64/kvm/mmu-nested.c
> new file mode 100644
> index 0000000..d52078f
> --- /dev/null
> +++ b/arch/arm64/kvm/mmu-nested.c
> @@ -0,0 +1,55 @@
> +/*
> + * Copyright (C) 2016 - Columbia University
> + * Author: Jintack Lim <jintack@xxxxxxxxxxxxxxx>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program. If not, see <http://www.gnu.org/licenses/>.
> + */
> +
> +#include <linux/kvm_host.h>
> +
> +#include <asm/kvm_arm.h>
> +#include <asm/kvm_emulate.h>
> +#include <asm/kvm_mmu.h>
> +#include <asm/kvm_nested.h>
> +
> +struct kvm_nested_s2_mmu *get_nested_mmu(struct kvm_vcpu *vcpu, u64 vttbr)
> +{
> + struct kvm_nested_s2_mmu *mmu;
> + u64 target_vmid = get_vmid(vttbr);
> + struct list_head *nested_mmu_list = &vcpu->kvm->arch.nested_mmu_list;
> +
> + list_for_each_entry_rcu(mmu, nested_mmu_list, list) {
> + u64 vmid = get_vmid(mmu->virtual_vttbr);
> +
> + if (target_vmid == vmid)

why is it sufficient to just look at the VMID and not having to consider
the baddr?

> + return mmu;
> + }
> + return NULL;
> +}
> +
> +struct kvm_s2_mmu *vcpu_get_active_s2_mmu(struct kvm_vcpu *vcpu)
> +{
> + struct kvm_nested_s2_mmu *nested_mmu;
> +
> + /* If we are NOT entering the nested VM, return mmu in kvm_arch */

this comment doesn't add any info not clear in the code

> + if (vcpu_mode_el2(vcpu) || !vcpu_nested_stage2_enabled(vcpu))
> + return &vcpu->kvm->arch.mmu;
> +
> + /* Otherwise, search for nested_mmu in the list */
> + nested_mmu = get_nested_mmu(vcpu, vcpu_el2_reg(vcpu, VTTBR_EL2));
> +
> + /* When this function is called, nested_mmu should be in the list */
> + BUG_ON(!nested_mmu);

can you provide a slightly stronger rationale behind why this BUG_ON
should never fire - I don't feel convinced right now.

> +
> + return &nested_mmu->mmu;
> +}
> --
> 1.9.1
>
>

Thanks,
-Christoffer