Re: [PATCH v2 5/6] kvm: allocate vcpu pointer array separately

[Marc Zyngier]

On Fri, 03 Sep 2021 14:08:06 +0100,
Juergen Gross <jgross@xxxxxxxx> wrote:
> 
> Prepare support of very large vcpu numbers per guest by moving the
> vcpu pointer array out of struct kvm.
> 
> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
> ---
> V2:
> - rebase to new kvm_arch_free_vm() implementation
> ---
>  arch/arm64/kvm/arm.c            | 21 +++++++++++++++++++--
>  arch/x86/include/asm/kvm_host.h |  5 +----
>  arch/x86/kvm/x86.c              | 18 ++++++++++++++++++
>  include/linux/kvm_host.h        | 17 +++++++++++++++--
>  4 files changed, 53 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index 38fff5963d9f..8bb5caeba007 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -293,10 +293,27 @@ long kvm_arch_dev_ioctl(struct file *filp,
>  
>  struct kvm *kvm_arch_alloc_vm(void)
>  {
> +	struct kvm *kvm;
> +
> +	if (!has_vhe())
> +		kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
> +	else
> +		kvm = vzalloc(sizeof(struct kvm));
> +
> +	if (!kvm)
> +		return NULL;
> +
>  	if (!has_vhe())
> -		return kzalloc(sizeof(struct kvm), GFP_KERNEL);
> +		kvm->vcpus = kcalloc(KVM_MAX_VCPUS, sizeof(void *), GFP_KERNEL);
> +	else
> +		kvm->vcpus = vzalloc(KVM_MAX_VCPUS * sizeof(void *));
> +
> +	if (!kvm->vcpus) {
> +		kvm_arch_free_vm(kvm);
> +		kvm = NULL;
> +	}
>  
> -	return vzalloc(sizeof(struct kvm));
> +	return kvm;
>  }
>  
>  int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index f16fadfc030a..6c28d0800208 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1517,10 +1517,7 @@ static inline void kvm_ops_static_call_update(void)
>  }
>  
>  #define __KVM_HAVE_ARCH_VM_ALLOC
> -static inline struct kvm *kvm_arch_alloc_vm(void)
> -{
> -	return __vmalloc(kvm_x86_ops.vm_size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> -}
> +struct kvm *kvm_arch_alloc_vm(void);
>  
>  #define __KVM_HAVE_ARCH_VM_FREE
>  void kvm_arch_free_vm(struct kvm *kvm);
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index cc552763f0e4..ff142b6dd00c 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -11126,6 +11126,24 @@ void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
>  	static_call(kvm_x86_sched_in)(vcpu, cpu);
>  }
>  
> +struct kvm *kvm_arch_alloc_vm(void)
> +{
> +	struct kvm *kvm;
> +
> +	kvm = __vmalloc(kvm_x86_ops.vm_size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +	if (!kvm)
> +		return NULL;
> +
> +	kvm->vcpus = __vmalloc(KVM_MAX_VCPUS * sizeof(void *),
> +			       GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +	if (!kvm->vcpus) {
> +		vfree(kvm);
> +		kvm = NULL;
> +	}
> +
> +	return kvm;
> +}
> +
>  void kvm_arch_free_vm(struct kvm *kvm)
>  {
>  	kfree(to_kvm_hv(kvm)->hv_pa_pg);
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index d75e9c2a00b1..9e2a5f1c6f54 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -536,7 +536,7 @@ struct kvm {
>  	struct mutex slots_arch_lock;
>  	struct mm_struct *mm; /* userspace tied to this vm */
>  	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
> -	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
> +	struct kvm_vcpu **vcpus;

At this stage, I really wonder why we are not using an xarray instead.

I wrote this [1] a while ago, and nothing caught fire. It was also a
net deletion of code...

	M.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/log/?h=kvm-arm64/vcpu-xarray

-- 
Without deviation from the norm, progress is not possible.