Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
1
fork

Configure Feed

Select the types of activity you want to include in your feed.

at master 106 lines 4.0 kB view raw
1.. SPDX-License-Identifier: GPL-2.0 2 3==================== 4Protected KVM (pKVM) 5==================== 6 7**NOTE**: pKVM is currently an experimental, development feature and 8subject to breaking changes as new isolation features are implemented. 9Please reach out to the developers at kvmarm@lists.linux.dev if you have 10any questions. 11 12Overview 13======== 14 15Booting a host kernel with '``kvm-arm.mode=protected``' enables 16"Protected KVM" (pKVM). During boot, pKVM installs a stage-2 identity 17map page-table for the host and uses it to isolate the hypervisor 18running at EL2 from the rest of the host running at EL1/0. 19 20pKVM permits creation of protected virtual machines (pVMs) by passing 21the ``KVM_VM_TYPE_ARM_PROTECTED`` machine type identifier to the 22``KVM_CREATE_VM`` ioctl(). The hypervisor isolates pVMs from the host by 23unmapping pages from the stage-2 identity map as they are accessed by a 24pVM. Hypercalls are provided for a pVM to share specific regions of its 25IPA space back with the host, allowing for communication with the VMM. 26A Linux guest must be configured with ``CONFIG_ARM_PKVM_GUEST=y`` in 27order to issue these hypercalls. 28 29See hypercalls.rst for more details. 30 31Isolation mechanisms 32==================== 33 34pKVM relies on a number of mechanisms to isolate PVMs from the host: 35 36CPU memory isolation 37-------------------- 38 39Status: Isolation of anonymous memory and metadata pages. 40 41Metadata pages (e.g. page-table pages and '``struct kvm_vcpu``' pages) 42are donated from the host to the hypervisor during pVM creation and 43are consequently unmapped from the stage-2 identity map until the pVM is 44destroyed. 45 46Similarly to regular KVM, pages are lazily mapped into the guest in 47response to stage-2 page faults handled by the host. However, when 48running a pVM, these pages are first pinned and then unmapped from the 49stage-2 identity map as part of the donation procedure. This gives rise 50to some user-visible differences when compared to non-protected VMs, 51largely due to the lack of MMU notifiers: 52 53* Memslots cannot be moved or deleted once the pVM has started running. 54* Read-only memslots and dirty logging are not supported. 55* With the exception of swap, file-backed pages cannot be mapped into a 56 pVM. 57* Donated pages are accounted against ``RLIMIT_MLOCK`` and so the VMM 58 must have a sufficient resource limit or be granted ``CAP_IPC_LOCK``. 59 The lack of a runtime reclaim mechanism means that memory locked for 60 a pVM will remain locked until the pVM is destroyed. 61* Changes to the VMM address space (e.g. a ``MAP_FIXED`` mmap() over a 62 mapping associated with a memslot) are not reflected in the guest and 63 may lead to loss of coherency. 64* Accessing pVM memory that has not been shared back will result in the 65 delivery of a SIGSEGV. 66* If a system call accesses pVM memory that has not been shared back 67 then it will either return ``-EFAULT`` or forcefully reclaim the 68 memory pages. Reclaimed memory is zeroed by the hypervisor and a 69 subsequent attempt to access it in the pVM will return ``-EFAULT`` 70 from the ``VCPU_RUN`` ioctl(). 71 72CPU state isolation 73------------------- 74 75Status: **Unimplemented.** 76 77DMA isolation using an IOMMU 78---------------------------- 79 80Status: **Unimplemented.** 81 82Proxying of Trustzone services 83------------------------------ 84 85Status: FF-A and PSCI calls from the host are proxied by the pKVM 86hypervisor. 87 88The FF-A proxy ensures that the host cannot share pVM or hypervisor 89memory with Trustzone as part of a "confused deputy" attack. 90 91The PSCI proxy ensures that CPUs always have the stage-2 identity map 92installed when they are executing in the host. 93 94Protected VM firmware (pvmfw) 95----------------------------- 96 97Status: **Unimplemented.** 98 99Resources 100========= 101 102Quentin Perret's KVM Forum 2022 talk entitled "Protected KVM on arm64: A 103technical deep dive" remains a good resource for learning more about 104pKVM, despite some of the details having changed in the meantime: 105 106https://www.youtube.com/watch?v=9npebeVFbFw