It's not a lockless walk of the memslots. The walk of memslots is already
"lockless" in that the memslots are protected by SRCU, not by mmu_lock.
On Tue, Feb 04, 2025, James Houghton wrote:
> It is possible to correctly do aging without taking the KVM MMU lock;
> this option allows such architectures to do so. Architectures that
> select CONFIG_KVM_MMU_NOTIFIER_AGING_LOCKLESS are responsible for
> correctness.
>
> Suggested-by: Yu Zhao <yuzhao@xxxxxxxxxx>
> Signed-off-by: James Houghton <jthoughton@xxxxxxxxxx>
> Reviewed-by: David Matlack <dmatlack@xxxxxxxxxx>
> ---
> include/linux/kvm_host.h | 1 +
> virt/kvm/Kconfig | 2 ++
> virt/kvm/kvm_main.c | 24 +++++++++++++++++-------
> 3 files changed, 20 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index f34f4cfaa513..c28a6aa1f2ed 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -267,6 +267,7 @@ struct kvm_gfn_range {
> union kvm_mmu_notifier_arg arg;
> enum kvm_gfn_range_filter attr_filter;
> bool may_block;
> + bool lockless;
> };
> bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range);
> bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range);
> diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
> index 54e959e7d68f..9356f4e4e255 100644
> --- a/virt/kvm/Kconfig
> +++ b/virt/kvm/Kconfig
> @@ -102,6 +102,8 @@ config KVM_GENERIC_MMU_NOTIFIER
>
> config KVM_ELIDE_TLB_FLUSH_IF_YOUNG
> depends on KVM_GENERIC_MMU_NOTIFIER
> +
> +config KVM_MMU_NOTIFIER_AGING_LOCKLESS
> bool
As noted by Stephen[*], this steals the "bool" from KVM_ELIDE_TLB_FLUSH_IF_YOUNG.
Looking at it with fresh eyes, it also fails to take a depenency on
KVM_GENERIC_MMU_NOTIFIER.
Lastly, the name is unnecessarily long. The "NOTIFIER" part is superfluous and
can be dropped, as it's a property of the architecture's MMU, not of KVM's
mmu_notifier implementation. E.g. if KVM ever did aging outside of the notifier,
then this Kconfig would be relevant for that flow as well. The dependency on
KVM_GENERIC_MMU_NOTIFIER is what communicates that its currently used only by
mmu_notifier aging.
Actually, I take "Lastly" back. IMO, it reads much better as LOCKLESS_AGING,
because LOCKLESS is an adverb that describes the AGING process.
[*] https://lore.kernel.org/all/20250214181401.4e7dd91d@xxxxxxxxxxxxxxxx
TL;DR: I'm squashing this:
diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig
index f0a60e59c884..fe8ea8c097de 100644
--- a/arch/x86/kvm/Kconfig
+++ b/arch/x86/kvm/Kconfig
@@ -22,7 +22,7 @@ config KVM_X86
select KVM_COMMON
select KVM_GENERIC_MMU_NOTIFIER
select KVM_ELIDE_TLB_FLUSH_IF_YOUNG
- select KVM_MMU_NOTIFIER_AGING_LOCKLESS
+ select KVM_MMU_LOCKLESS_AGING
select HAVE_KVM_IRQCHIP
select HAVE_KVM_PFNCACHE
select HAVE_KVM_DIRTY_RING_TSO
diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
index 9356f4e4e255..746e1f466aa6 100644
--- a/virt/kvm/Kconfig
+++ b/virt/kvm/Kconfig
@@ -102,8 +102,10 @@ config KVM_GENERIC_MMU_NOTIFIER
config KVM_ELIDE_TLB_FLUSH_IF_YOUNG
depends on KVM_GENERIC_MMU_NOTIFIER
+ bool
-config KVM_MMU_NOTIFIER_AGING_LOCKLESS
+config KVM_MMU_LOCKLESS_AGING
+ depends on KVM_GENERIC_MMU_NOTIFIER
bool
config KVM_GENERIC_MEMORY_ATTRIBUTES
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index e514e3db1b31..201c14ff476f 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -655,8 +655,7 @@ static __always_inline int kvm_age_hva_range(struct mmu_notifier *mn,
.on_lock = (void *)kvm_null_fn,
.flush_on_ret = flush_on_ret,
.may_block = false,
- .lockless =
- IS_ENABLED(CONFIG_KVM_MMU_NOTIFIER_AGING_LOCKLESS),
+ .lockless = IS_ENABLED(CONFIG_KVM_MMU_LOCKLESS_AGING),
};
return kvm_handle_hva_range(kvm, &range).ret;
