Re: [PATCH v12 23/45] KVM: Let userspace disable per-VM mem attributes, enable per-gmem attributes

On Wed, Sep 02, 2026, Binbin Wu wrote:

On 9/2/2026 2:26 AM, Sean Christopherson wrote:

quoted

On Tue, Sep 01, 2026, Fuad Tabba wrote:

quoted

Hi Sean, Ackerley,

On Mon, 31 Aug 2026 at 01:25, Ackerley Tng via B4 Relay
[off-list ref] wrote:
...

quoted

+       kvm.gmem_in_place_conversion=
+                       [KVM] Controls whether KVM enables in-place conversion
+                       support for guest_memfd and tracks the private/shared
+                       state of memory per guest_memfd instead of per VM.
+
+                       If enabled, KVM enables the KVM_SET_MEMORY_ATTRIBUTES2
+                       ioctl on guest_memfd file descriptors and disables the
+                       legacy VM-scoped KVM_SET_MEMORY_ATTRIBUTES ioctl for
+                       private memory state tracking. Only the
+                       KVM_MEMORY_ATTRIBUTE_PRIVATE attribute moves to
+                       per-guest_memfd tracking; other attributes remain
+                       per-VM.
+
+                       This parameter toggles KVM's in-place conversion
+                       capability support. Whether a VMM uses separate backends
+                       or out-of-place memory management is determined by
+                       userspace VMM design.
+
+                       Note, this parameter is only available when
+                       CONFIG_KVM_VM_MEMORY_ATTRIBUTES=y. When
+                       CONFIG_KVM_VM_MEMORY_ATTRIBUTES is not set, in-place
+                       conversion is unconditionally enabled.
+
+                       Default is Y (on).

The parameter is only settable when CONFIG_KVM_VM_MEMORY_ATTRIBUTES=y, and
in that config gmem_in_place_conversion initialises to
!IS_ENABLED(CONFIG_KVM_VM_MEMORY_ATTRIBUTES), i.e. N, not Y. Is the doc line
wrong, or is N-by-default not what was intended here?

The doc is wrong.  The intent is to maintain backwards compatibility with existing
deployments if KVM_VM_MEMORY_ATTRIBUTES=y.

Will it result in the situation that kernel builds with CONFIG_KVM_VM_MEMORY_ATTRIBUTES
enabled by default for backwards compatibility, which requires users to explicitly enable 
gmem_in_place_conversion by the kernel option? Then it effectively disables
gmem_in_place_conversion by default.

Yes.  My intent/expectation is that the only reason to build with
CONFIG_KVM_VM_MEMORY_ATTRIBUTES=y is for backwards compatibility, and so it makes
sense to have the existing behavior be the default.  If a deployment only wants
the new behavior, then it can simply build with CONFIG_KVM_VM_MEMORY_ATTRIBUTES=n.

Folks that want only the new behavior but are using a distro/pre-built kernel will
need to manually enable in-place conversion, but that's not too onerous, and it's
simply not feasible to provide defaults that work for every possible scenario.

But I guess it's OK for CSPs since they know what they want.

Yep.