Live migration software in L0 needs to be able to capture an accurate state of the L1 guest in order to save it and later restore it. One of the complications of doing this when you have a nested/L2 guest is that the state from the L0 user space point of view is very indeterminstic. The state can be coming from the L1 or the L2 guest depending on when the vCPU was kicked out of non-root mode. There is also no way currently to extract the extra meta-state that L0 holds in order to run L2s. In order to fix this, always switch to L1 state before exiting to user space (Patch 1) and provide a new ioctl command to extra the extra meta-state (Patch 2). This is still an early RFC because: 1- To get feedback on the approach taken here. 2- I am still debugging a VMEntry failure for L2 that occassionaly happens when I run a pause+unpause loop against an L1 guest running L2s. Also there is some meta-state that I am not extracting yet (will update this soon). L1 seems to be very stable though. KarimAllah Ahmed (2): KVM/nVMX: Cleanly exit from L2 to L1 on user-space exit KVM/nVMX: Add support for saving/restoring L2 meta-state stored by L0 arch/x86/include/asm/kvm_host.h | 5 +++ arch/x86/include/uapi/asm/kvm.h | 7 ++++ arch/x86/kvm/vmx.c | 83 +++++++++++++++++++++++++++++++++++++++++ arch/x86/kvm/x86.c | 43 ++++++++++++++++++--- include/linux/kvm_host.h | 2 + include/uapi/linux/kvm.h | 3 ++ virt/kvm/kvm_main.c | 24 ++++++++++++ 7 files changed, 162 insertions(+), 5 deletions(-) Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx> Cc: Radim Krčmář <rkrcmar@xxxxxxxxxx> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx> Cc: Ingo Molnar <mingo@xxxxxxxxxx> Cc: H. Peter Anvin <hpa@xxxxxxxxx> Cc: x86@xxxxxxxxxx Cc: kvm@xxxxxxxxxxxxxxx Cc: linux-kernel@xxxxxxxxxxxxxxx Signed-off-by: KarimAllah Ahmed <karahmed@xxxxxxxxx> -- 2.7.4
