On Thu, Jan 09, 2025, James Houghton wrote:
> James Houghton (13):
> KVM: Add KVM_MEM_USERFAULT memslot flag and bitmap
> KVM: Add KVM_MEMORY_EXIT_FLAG_USERFAULT
> KVM: Allow late setting of KVM_MEM_USERFAULT on guest_memfd memslot
> KVM: Advertise KVM_CAP_USERFAULT in KVM_CHECK_EXTENSION
Starting with some series-wide feedback, the granularity of these first few
patches is too fine. I normally like to split things up, but honestly, this is
such a small feature that I don't see much point in separating the uAPI from the
infrastructure.
To avoid cyclical dependencies between common KVM and arch code, we can do all
the prep, but not fully enable+advertise support on any architecture until all
targeted architectures are fully ready.
In other words, I think we should squish these into one patch, minus this bit at
the very end of the series (spoiler alert):
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index ce7bf5de6d72..0106d6d461a3 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -1545,6 +1545,9 @@ static int check_memory_region_flags(struct kvm *kvm,
!(mem->flags & KVM_MEM_GUEST_MEMFD))
valid_flags |= KVM_MEM_READONLY;
+ if (IS_ENABLED(CONFIG_KVM_GENERIC_PAGE_FAULT))
+ valid_flags |= KVM_MEM_USERFAULT;
+
if (mem->flags & ~valid_flags)
return -EINVAL;
@@ -4824,6 +4827,9 @@ static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
case KVM_CAP_CHECK_EXTENSION_VM:
case KVM_CAP_ENABLE_CAP_VM:
case KVM_CAP_HALT_POLL:
+#ifdef CONFIG_KVM_GENERIC_PAGE_FAULT
+ case KVM_CAP_USERFAULT:
+#endif
return 1;
#ifdef CONFIG_KVM_MMIO
case KVM_CAP_COALESCED_MMIO:
