On Tue, Nov 28, 2017 at 09:34:56AM +0800, Longpeng (Mike) wrote:
>
>
> On 2017/11/28 4:55, Michael S. Tsirkin wrote:
>
> > On Mon, Nov 27, 2017 at 09:51:27PM +0100, Paolo Bonzini wrote:
> >> On 27/11/2017 21:45, Michael S. Tsirkin wrote:
> >>> On Sat, Nov 25, 2017 at 02:09:32PM +0100, Jan H. Schönherr wrote:
> >>>> If host CPUs are dedicated to a VM, we can avoid VM exits on HLT,
> >>>> reducing the wake-up latency on posted interrupts.
> >>>>
> >>>> This reintroduces a feature that has been there at some point --
> >>>> see Linux 3.4 commit 10166744b80a ("KVM: VMX: remove yield_on_hlt")
> >>>> for the removal -- but with the additional ability to enable it only
> >>>> for selected VMs (and supporting SVM as well).
> >>>>
> >>>> Signed-off-by: Jan H. Schönherr <jschoenh@xxxxxxxxx>
> >>>
> >>>
> >>> If you are going to do this, why not expose mwait
> >>> in the cpuid thus making guests use mwait to halt?
> >>>
> >>> What are the advantages of doing this using halt specifically?
> >>
> >> Not all guests use MWAIT, I suppose.
> >>
> >> Paolo
> >
> > In that case, it would be nice to document which guests of interest
> > don't. E.g. I don't think there are still supported versions of RHEL
> > that don't use MWAIT.
> >
>
>
> Some old kernels, E.g. my kernel is 3.10.0-514 based on RHEL 7.3, don't use
> MWAIT if the idle-driver is not supported (we can see
> "/sys/devices/system/cpu/cpuidle/current_driver" in guest is "none").
This is exactly what I was saying. mwait is supported by this guest -
the reason idle driver is not used is because QEMU does not enable the
mwait flag in the CPUID.
You just need a QEMU patch to expose CPUID.
However, if mwait is disabled to avoid guest entering low PM states,
then maybe it makes sense to allow a non-exiting HLT instead.
Pls add that in the commit log as a motivation.
> So the
> idle routine will use the kernel's default routine.
>
> The default idle routine is selected when starting,
>
> old kernel:
> '''
> select_idle_routine()
> if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
> x86_idle = amd_e400_idle;
> } else
> x86_idle = default_idle;
> '''
>
> newer kernel:
> '''
> select_idle_routine()
> if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
> pr_info("using AMD E400 aware idle routine\n");
> x86_idle = amd_e400_idle;
> } else if (prefer_mwait_c1_over_halt(c)) {
> pr_info("using mwait in idle threads\n");
> x86_idle = mwait_idle;
> } else
> x86_idle = default_idle;
> '''
>
> So, some old guests don't use MWAIT as default idle routine.
>
> >
> >
> >>>
> >>>> ---
> >>>> Note: AMD code paths are only compile tested
> >>>> ---
> >>>> Documentation/virtual/kvm/api.txt | 12 +++++++++++-
> >>>> arch/x86/include/asm/kvm_host.h | 1 +
> >>>> arch/x86/kvm/svm.c | 3 ++-
> >>>> arch/x86/kvm/vmx.c | 33 +++++++++++++++++++++++++++------
> >>>> arch/x86/kvm/x86.c | 5 +++++
> >>>> arch/x86/kvm/x86.h | 5 +++++
> >>>> include/uapi/linux/kvm.h | 1 +
> >>>> 7 files changed, 52 insertions(+), 8 deletions(-)
> >>>>
> >>>> diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt
> >>>> index 0ee812c..c06bb41 100644
> >>>> --- a/Documentation/virtual/kvm/api.txt
> >>>> +++ b/Documentation/virtual/kvm/api.txt
> >>>> @@ -4172,7 +4172,17 @@ Returns: 0 on success
> >>>> This capability indicates that a guest using memory monitoring instructions
> >>>> (MWAIT/MWAITX) to stop a virtual CPU will not cause a VM exit. As such, time
> >>>> spent while a virtual CPU is halted in this way will then be accounted for as
> >>>> -guest running time on the host (as opposed to e.g. HLT).
> >>>> +guest running time on the host.
> >>>> +
> >>>> +7.14 KVM_CAP_X86_GUEST_HLT
> >>>> +
> >>>> +Architectures: x86
> >>>> +Parameters: none
> >>>> +Returns: 0 on success
> >>>> +
> >>>> +This capability indicates that a guest using HLT to stop a virtual CPU will not
> >>>> +cause a VM exit. As such, time spent while a virtual CPU is halted in this way
> >>>> +will then be accounted for as guest running time on the host.
> >>>>
> >>>> 8. Other capabilities.
> >>>> ----------------------
> >>>> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> >>>> index f7bcfaa..3197c2d 100644
> >>>> --- a/arch/x86/include/asm/kvm_host.h
> >>>> +++ b/arch/x86/include/asm/kvm_host.h
> >>>> @@ -781,6 +781,7 @@ struct kvm_arch {
> >>>>
> >>>> gpa_t wall_clock;
> >>>>
> >>>> + bool hlt_in_guest;
> >>>> bool mwait_in_guest;
> >>>>
> >>>> bool ept_identity_pagetable_done;
> >>>> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
> >>>> index ef1b320..c135b98 100644
> >>>> --- a/arch/x86/kvm/svm.c
> >>>> +++ b/arch/x86/kvm/svm.c
> >>>> @@ -1236,7 +1236,6 @@ static void init_vmcb(struct vcpu_svm *svm)
> >>>> set_intercept(svm, INTERCEPT_RDPMC);
> >>>> set_intercept(svm, INTERCEPT_CPUID);
> >>>> set_intercept(svm, INTERCEPT_INVD);
> >>>> - set_intercept(svm, INTERCEPT_HLT);
> >>>> set_intercept(svm, INTERCEPT_INVLPG);
> >>>> set_intercept(svm, INTERCEPT_INVLPGA);
> >>>> set_intercept(svm, INTERCEPT_IOIO_PROT);
> >>>> @@ -1257,6 +1256,8 @@ static void init_vmcb(struct vcpu_svm *svm)
> >>>> set_intercept(svm, INTERCEPT_MONITOR);
> >>>> set_intercept(svm, INTERCEPT_MWAIT);
> >>>> }
> >>>> + if (!kvm_hlt_in_guest(svm->vcpu.kvm))
> >>>> + set_intercept(svm, INTERCEPT_HLT);
> >>>>
> >>>> control->iopm_base_pa = __sme_set(iopm_base);
> >>>> control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
> >>>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> >>>> index a067735..1b67433 100644
> >>>> --- a/arch/x86/kvm/vmx.c
> >>>> +++ b/arch/x86/kvm/vmx.c
> >>>> @@ -2446,6 +2446,25 @@ static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
> >>>> vmx_set_interrupt_shadow(vcpu, 0);
> >>>> }
> >>>>
> >>>> +static void vmx_set_intr_info(struct kvm_vcpu *vcpu, u32 intr)
> >>>> +{
> >>>> + vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
> >>>> +
> >>>> + /*
> >>>> + * Ensure that we clear the HLT state in the VMCS. We don't need to
> >>>> + * explicitly skip the instruction because if the HLT state is set, then
> >>>> + * the instruction is already executing and RIP has already been
> >>>> + * advanced.
> >>>> + */
> >>>> + if (!kvm_hlt_in_guest(vcpu->kvm) || !(intr & INTR_INFO_VALID_MASK))
> >>>> + return;
> >>>> + if (is_external_interrupt(intr) || is_nmi(intr))
> >>>> + return;
> >>>> + if (vmcs_read32(GUEST_ACTIVITY_STATE) != GUEST_ACTIVITY_HLT)
> >>>> + return;
> >>>> + vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
> >>>> +}
> >>>> +
> >>>> static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
> >>>> unsigned long exit_qual)
> >>>> {
> >>>> @@ -2540,7 +2559,7 @@ static void vmx_queue_exception(struct kvm_vcpu *vcpu)
> >>>> } else
> >>>> intr_info |= INTR_TYPE_HARD_EXCEPTION;
> >>>>
> >>>> - vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
> >>>> + vmx_set_intr_info(vcpu, intr_info);
> >>>> }
> >>>>
> >>>> static bool vmx_rdtscp_supported(void)
> >>>> @@ -5298,6 +5317,8 @@ static u32 vmx_exec_control(struct vcpu_vmx *vmx)
> >>>> if (kvm_mwait_in_guest(vmx->vcpu.kvm))
> >>>> exec_control &= ~(CPU_BASED_MWAIT_EXITING |
> >>>> CPU_BASED_MONITOR_EXITING);
> >>>> + if (kvm_hlt_in_guest(vmx->vcpu.kvm))
> >>>> + exec_control &= ~CPU_BASED_HLT_EXITING;
> >>>> return exec_control;
> >>>> }
> >>>>
> >>>> @@ -5635,7 +5656,7 @@ static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
> >>>>
> >>>> setup_msrs(vmx);
> >>>>
> >>>> - vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
> >>>> + vmx_set_intr_info(vcpu, 0); /* 22.2.1 */
> >>>>
> >>>> if (cpu_has_vmx_tpr_shadow() && !init_event) {
> >>>> vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
> >>>> @@ -5729,7 +5750,7 @@ static void vmx_inject_irq(struct kvm_vcpu *vcpu)
> >>>> vmx->vcpu.arch.event_exit_inst_len);
> >>>> } else
> >>>> intr |= INTR_TYPE_EXT_INTR;
> >>>> - vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
> >>>> + vmx_set_intr_info(vcpu, intr);
> >>>> }
> >>>>
> >>>> static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
> >>>> @@ -5758,8 +5779,8 @@ static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
> >>>> return;
> >>>> }
> >>>>
> >>>> - vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
> >>>> - INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
> >>>> + vmx_set_intr_info(vcpu, INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK |
> >>>> + NMI_VECTOR);
> >>>> }
> >>>>
> >>>> static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
> >>>> @@ -9301,7 +9322,7 @@ static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
> >>>> VM_ENTRY_INSTRUCTION_LEN,
> >>>> VM_ENTRY_EXCEPTION_ERROR_CODE);
> >>>>
> >>>> - vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
> >>>> + vmx_set_intr_info(vcpu, 0);
> >>>> }
> >>>>
> >>>> static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
> >>>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> >>>> index fe6627a..f17c520 100644
> >>>> --- a/arch/x86/kvm/x86.c
> >>>> +++ b/arch/x86/kvm/x86.c
> >>>> @@ -2755,6 +2755,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
> >>>> case KVM_CAP_SET_BOOT_CPU_ID:
> >>>> case KVM_CAP_SPLIT_IRQCHIP:
> >>>> case KVM_CAP_IMMEDIATE_EXIT:
> >>>> + case KVM_CAP_X86_GUEST_HLT:
> >>>> r = 1;
> >>>> break;
> >>>> case KVM_CAP_ADJUST_CLOCK:
> >>>> @@ -4068,6 +4069,10 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
> >>>> r = 0;
> >>>> }
> >>>> break;
> >>>> + case KVM_CAP_X86_GUEST_HLT:
> >>>> + kvm->arch.hlt_in_guest = true;
> >>>> + r = 0;
> >>>> + break;
> >>>> default:
> >>>> r = -EINVAL;
> >>>> break;
> >>>> diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
> >>>> index ed8e150..b2066aa 100644
> >>>> --- a/arch/x86/kvm/x86.h
> >>>> +++ b/arch/x86/kvm/x86.h
> >>>> @@ -266,4 +266,9 @@ static inline bool kvm_mwait_in_guest(struct kvm *kvm)
> >>>> return kvm->arch.mwait_in_guest;
> >>>> }
> >>>>
> >>>> +static inline bool kvm_hlt_in_guest(struct kvm *kvm)
> >>>> +{
> >>>> + return kvm->arch.hlt_in_guest;
> >>>> +}
> >>>> +
> >>>> #endif
> >>>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> >>>> index 282d7613..ff8f266 100644
> >>>> --- a/include/uapi/linux/kvm.h
> >>>> +++ b/include/uapi/linux/kvm.h
> >>>> @@ -932,6 +932,7 @@ struct kvm_ppc_resize_hpt {
> >>>> #define KVM_CAP_HYPERV_SYNIC2 148
> >>>> #define KVM_CAP_HYPERV_VP_INDEX 149
> >>>> #define KVM_CAP_S390_AIS_MIGRATION 150
> >>>> +#define KVM_CAP_X86_GUEST_HLT 151
> >>>>
> >>>> #ifdef KVM_CAP_IRQ_ROUTING
> >>>>
> >>>> --
> >>>> 2.3.1.dirty
> >
> > .
> >
>
>
> --
> Regards,
> Longpeng(Mike)
