path: root/Documentation/powerpc
diff options
authorHollis Blanchard <hollisb@us.ibm.com>2008-04-16 23:28:09 -0500
committerAvi Kivity <avi@qumranet.com>2008-04-27 18:21:39 +0300
commitbbf45ba57eaec56569918a8bab96ab653bd45ec1 (patch)
tree63c53b1c1d93ec6559c7695c16b2345238e270f5 /Documentation/powerpc
parent513014b717203d1d689652d0fda86eee959a6a8a (diff)
KVM: ppc: PowerPC 440 KVM implementation
This functionality is definitely experimental, but is capable of running unmodified PowerPC 440 Linux kernels as guests on a PowerPC 440 host. (Only tested with 440EP "Bamboo" guests so far, but with appropriate userspace support other SoC/board combinations should work.) See Documentation/powerpc/kvm_440.txt for technical details. [stephen: build fix] Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com> Acked-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Avi Kivity <avi@qumranet.com>
Diffstat (limited to 'Documentation/powerpc')
1 files changed, 41 insertions, 0 deletions
diff --git a/Documentation/powerpc/kvm_440.txt b/Documentation/powerpc/kvm_440.txt
new file mode 100644
index 00000000000..c02a003fa03
--- /dev/null
+++ b/Documentation/powerpc/kvm_440.txt
@@ -0,0 +1,41 @@
+Hollis Blanchard <hollisb@us.ibm.com>
+15 Apr 2008
+Various notes on the implementation of KVM for PowerPC 440:
+To enforce isolation, host userspace, guest kernel, and guest userspace all
+run at user privilege level. Only the host kernel runs in supervisor mode.
+Executing privileged instructions in the guest traps into KVM (in the host
+kernel), where we decode and emulate them. Through this technique, unmodified
+440 Linux kernels can be run (slowly) as guests. Future performance work will
+focus on reducing the overhead and frequency of these traps.
+The usual code flow is started from userspace invoking an "run" ioctl, which
+causes KVM to switch into guest context. We use IVPR to hijack the host
+interrupt vectors while running the guest, which allows us to direct all
+interrupts to kvmppc_handle_interrupt(). At this point, we could either
+- handle the interrupt completely (e.g. emulate "mtspr SPRG0"), or
+- let the host interrupt handler run (e.g. when the decrementer fires), or
+- return to host userspace (e.g. when the guest performs device MMIO)
+Address spaces: We take advantage of the fact that Linux doesn't use the AS=1
+address space (in host or guest), which gives us virtual address space to use
+for guest mappings. While the guest is running, the host kernel remains mapped
+in AS=0, but the guest can only use AS=1 mappings.
+TLB entries: The TLB entries covering the host linear mapping remain
+present while running the guest. This reduces the overhead of lightweight
+exits, which are handled by KVM running in the host kernel. We keep three
+copies of the TLB:
+ - guest TLB: contents of the TLB as the guest sees it
+ - shadow TLB: the TLB that is actually in hardware while guest is running
+ - host TLB: to restore TLB state when context switching guest -> host
+When a TLB miss occurs because a mapping was not present in the shadow TLB,
+but was present in the guest TLB, KVM handles the fault without invoking the
+guest. Large guest pages are backed by multiple 4KB shadow pages through this
+IO: MMIO and DCR accesses are emulated by userspace. We use virtio for network
+and block IO, so those drivers must be enabled in the guest. It's possible
+that some qemu device emulation (e.g. e1000 or rtl8139) may also work with
+little effort.