Re: [RFC PATCH v2 1/7] KVM: Add generic infrastructure to support pinned VMIDs

[Sean Christopherson <seanjc@google.com>]

On Thu, Feb 08, 2024, Shameer Kolothum wrote:
> Provide generic helper functions to get/put pinned VMIDs if the arch
> supports it.

IMO, this has no business being in generic KVM.  Multiple architectures have
constructs that are _similar_ ARM's VMID, but AFAICT the exact semantics are very
ARM specific.  I.e. odds are very good that the only thing that can actually use
kvm_pinned_vmid_get() is the SMMU, because the concept won't fit any other
architecture.

The other issue is that other architectures support building KVM as a module,
which makes it much more difficult to guarantee the safety of these hooks.

> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@huawei.com>
> ---
>  include/linux/kvm_host.h | 18 ++++++++++++++++++
>  virt/kvm/Kconfig         |  3 +++
>  virt/kvm/kvm_main.c      | 23 +++++++++++++++++++++++
>  3 files changed, 44 insertions(+)
> 
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index 7e7fd25b09b3..610e239bea46 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -2311,6 +2311,24 @@ static inline void kvm_handle_signal_exit(struct kvm_vcpu *vcpu)
>  }
>  #endif /* CONFIG_KVM_XFER_TO_GUEST_WORK */
>  
> +#ifdef CONFIG_HAVE_KVM_PINNED_VMID
> +int kvm_arch_pinned_vmid_get(struct kvm *kvm);
> +void kvm_arch_pinned_vmid_put(struct kvm *kvm);
> +#endif
> +
> +#ifdef CONFIG_HAVE_KVM_PINNED_VMID
> +int kvm_pinned_vmid_get(struct kvm *kvm);
> +void kvm_pinned_vmid_put(struct kvm *kvm);
> +#else
> +static inline int kvm_pinned_vmid_get(struct kvm *kvm)
> +{
> +	return -EINVAL;
> +}
> +
> +static inline void kvm_pinned_vmid_put(struct kvm *kvm)
> +{
> +}
> +#endif
>  /*
>   * If more than one page is being (un)accounted, @virt must be the address of
>   * the first page of a block of pages what were allocated together (i.e
> diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
> index 184dab4ee871..a3052c8e3ac4 100644
> --- a/virt/kvm/Kconfig
> +++ b/virt/kvm/Kconfig
> @@ -108,3 +108,6 @@ config KVM_GENERIC_PRIVATE_MEM
>         select KVM_GENERIC_MEMORY_ATTRIBUTES
>         select KVM_PRIVATE_MEM
>         bool
> +
> +config HAVE_KVM_PINNED_VMID
> +       bool
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 10bfc88a69f7..f84d6da5f464 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -3918,6 +3918,29 @@ bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
>  }
>  EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
>  
> +#ifdef CONFIG_HAVE_KVM_PINNED_VMID
> +int kvm_pinned_vmid_get(struct kvm *kvm)
> +{
> +	int ret;
> +
> +	if (!kvm_get_kvm_safe(kvm))
> +		return -ENOENT;
> +	ret  = kvm_arch_pinned_vmid_get(kvm);
> +	if (ret < 0)
> +		kvm_put_kvm(kvm);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(kvm_pinned_vmid_get);
> +
> +void kvm_pinned_vmid_put(struct kvm *kvm)
> +{
> +	kvm_arch_pinned_vmid_put(kvm);
> +	kvm_put_kvm(kvm);
> +}
> +EXPORT_SYMBOL_GPL(kvm_pinned_vmid_put);
> +#endif
> +
>  #ifndef CONFIG_S390
>  /*
>   * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
> -- 
> 2.34.1
>