Messages in this thread |  | | From | "Pratik R. Sampat" <> | | Subject | [RFC PATCH 0/4] SEV-SNP Unaccepted Memory Hotplug | | Date | Tue, 25 Nov 2025 11:57:49 -0600 |
| |
Guest memory hot-plug/remove via the QEMU monitor is used by virtual machines to dynamically scale the memory capacity of a system with virtually zero downtime to the guest. For confidential VMs, memory has to be first accepted before it can be used.
The unaccepted memory feature provides a mechanism to accept memory either up-front or right before it is needed. The unaccepted table that tracks this information is allocated and memory block reserved at boot time. For memory hotplug, this means the table cannot be updated to track additional regions and accept them as the guest physical memory grows.
This proof-of-concept series extends the unaccepted memory infrastructure to support memory hotplug and hot-unplug on the SNP platform. On a high-level, it does so by decoupling the memory bitmap from the unaccepted table so that kernel can manage bitmap when memory is added. For hot-remove, it reverts the page states so that the hypervisor can reuse that memory. Hot-remove also presents a unique scenario where the memory we attempt to share can already be in a shared state set externally which can cause pvalidation on the platform to fail since no updates were made to the validated bit. Handle this case by tracking the state of hotplugged memory within the guest and disallow pvalidate operations on the same state.
Usage (for SNP guests) ---------------------- Step1: Spawn a QEMU SNP guest with the additional parameter of slots and maximum possible memory, along with the initial memory as below: "-m X,slots=Y,maxmem=Z".
Use the "accept_memory=[eager|lazy]" kernel command-line parameter to specify whether hotplugged memory should be accepted immediately upon addition or only when first accessed. By default, lazy acceptance is used.
Step2: Once the guest is booted, launch the qemu monitor and hotplug the memory as follows: (qemu) object_add memory-backend-memfd,id=mem1,size=1G (qemu) device_add pc-dimm,id=dimm1,memdev=mem1
Step3: If using auto-onlining by either: a) echo online > /sys/devices/system/memory/auto_online_blocks, OR b) enable CONFIG_MHP_DEFAULT_ONLINE_TYPE_* while compiling kernel Memory should show up automatically.
Otherwise, memory can also be onlined by echoing 1 to the newly added blocks in: /sys/devices/system/memory/memoryXX/online
Step4: If accept_memory is set to eager, all memory is accepted immediately. Otherwise, memory is accepted on access. For the latter, acceptance can be triggered by simply running a program such as stress-ng that requests enough memory to cover the newly allocated hotplugged regions.
$ stress-ng --vm 1 --vm-bytes={X}G -t {T}s
Step5: memory can be hot-removed using the qemu monitor using: (qemu) device_remove dimm1 (qemu) object_remove mem1
Tip: Enable the kvm_convert_memory event in QEMU to observe memory conversions between private and shared during hotplug/remove.
The series is based on git.kernel.org/pub/scm/virt/kvm/kvm.git next
Comments and feedback appreciated!
Pratik R. Sampat (4): efi/libstub: Decouple memory bitmap from the unaccepted table mm: Add support for unaccepted memory hotplug x86/sev: Introduce hotplug-aware SNP page state validation mm: Add support for unaccepted memory hot-remove
arch/x86/boot/compressed/efi.h | 3 +- arch/x86/coco/sev/core.c | 127 +++++++++++++++- arch/x86/include/asm/sev.h | 34 +++++ arch/x86/include/asm/unaccepted_memory.h | 31 ++++ .../firmware/efi/libstub/unaccepted_memory.c | 12 +- drivers/firmware/efi/unaccepted_memory.c | 139 +++++++++++++++++- include/linux/efi.h | 3 +- include/linux/mm.h | 18 +++ mm/memory_hotplug.c | 9 ++ mm/page_alloc.c | 2 + 10 files changed, 363 insertions(+), 15 deletions(-)
-- 2.51.1
|  |