Allow up to 16 VM planes, it's a nice round number.

FIXME: online_vcpus is used by x86 code that deals with TSC synchronization.
Maybe kvmclock should be moved to planex.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch/x86/include/asm/kvm_host.h | 3 +++
 arch/x86/kvm/x86.c              | 6 ++++++
 2 files changed, 9 insertions(+)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 0344e8bed319..d0cb177b6f52 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -2339,6 +2339,8 @@ enum {
 # define kvm_memslots_for_spte_role(kvm, role) __kvm_memslots(kvm, 0)
 #endif
 
+#define KVM_MAX_VCPU_PLANES	16
+
 int kvm_cpu_has_injectable_intr(struct kvm_vcpu *v);
 int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu);
 int kvm_cpu_has_extint(struct kvm_vcpu *v);
@@ -2455,6 +2457,7 @@ int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages);
  */
 #define KVM_EXIT_HYPERCALL_MBZ		GENMASK_ULL(31, 1)
 
+int kvm_arch_nr_vcpu_planes(struct kvm *kvm);
 bool kvm_arch_planes_share_fpu(struct kvm *kvm);
 
 #endif /* _ASM_X86_KVM_HOST_H */
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 4546d1049f43..86d1a567f62e 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -13989,6 +13989,12 @@ int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
 }
 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
 
+int kvm_arch_nr_vcpu_planes(struct kvm *kvm)
+{
+	/* TODO: use kvm_x86_ops so that SNP can use planes for VTPLs.  */
+	return kvm->arch.has_protected_state ? 1 : KVM_MAX_VCPU_PLANES;
+}
+
 bool kvm_arch_planes_share_fpu(struct kvm *kvm)
 {
 	return !kvm || kvm->arch.planes_share_fpu;
-- 
2.49.0

On Tue, Apr 01, 2025, Paolo Bonzini wrote:
> Allow up to 16 VM planes, it's a nice round number.
> 
> FIXME: online_vcpus is used by x86 code that deals with TSC synchronization.
> Maybe kvmclock should be moved to planex.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  arch/x86/include/asm/kvm_host.h | 3 +++
>  arch/x86/kvm/x86.c              | 6 ++++++
>  2 files changed, 9 insertions(+)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 0344e8bed319..d0cb177b6f52 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -2339,6 +2339,8 @@ enum {
>  # define kvm_memslots_for_spte_role(kvm, role) __kvm_memslots(kvm, 0)
>  #endif
>  
> +#define KVM_MAX_VCPU_PLANES	16

I'm pretty sure x86 can't support 16 planes.  "union kvm_mmu_page_role" needs
to incorporate the plane, otherwise per-plane memory attributes will won't work.
And adding four bits to the plane would theoretically put us in danger of
overflowing gfn_write_track (in practice, I highly, highly doubt that can happen).

Why not start with 4 planes?  Or even 2?  Expanding the number of planes should
be much easier than contracting.  Based on the VTL and VMPL roadmaps, 4 planes
will probably be enough for many years to come.