Using hypercall to send IPIs by one vmexit instead of one by one for xAPIC/x2APIC physical mode and one vmexit per-cluster for x2APIC cluster mode.
Even if enable qemu interrupt remapping and PV TLB Shootdown, I can still observe ~14% performance boost by ebizzy benchmark for 64 vCPUs VM, the total msr-induced vmexits reduce ~70%. The patchset implements the PV IPIs for vCPUs <= 128 VM, this is really common in cloud environment, after this patchset is applied, I can continue to add > 64 vCPUs VM support and that implementation has to introduce more complex logic. Cc: Paolo Bonzini <[email protected]> Cc: Radim Krčmář <[email protected]> Cc: Vitaly Kuznetsov <[email protected]> v1 -> v2: * sparse apic id > 128, or any other errors, fallback to original apic hooks * have two bitmask arguments so that one hypercall handles 128 vCPUs * fix KVM_FEATURE_PV_SEND_IPI doc * document hypercall * fix NMI selftest fails * fix build errors reported by 0day Wanpeng Li (2): KVM: X86: Implement PV IPI in linux guest KVM: X86: Implement PV send IPI support Documentation/virtual/kvm/cpuid.txt | 4 ++ Documentation/virtual/kvm/hypercalls.txt | 6 ++ arch/x86/include/uapi/asm/kvm_para.h | 1 + arch/x86/kernel/kvm.c | 99 ++++++++++++++++++++++++++++++++ arch/x86/kvm/cpuid.c | 3 +- arch/x86/kvm/x86.c | 42 ++++++++++++++ include/uapi/linux/kvm_para.h | 1 + 7 files changed, 155 insertions(+), 1 deletion(-) -- 2.7.4
