[PATCH v2 0/3] Provides userspace with per-VM capability to improve latency

[PATCH v2 0/3] Provides userspace with per-VM capability to improve latency

[Wanpeng Li]

Provides userspace with per-VM capability(KVM_CAP_X86_DISABLE_EXITS) to 
not intercept MWAIT/HLT/PAUSE in order that to improve latency in some 
workloads.

The patchset implements the original proposal from Radim.
https://www.spinics.net/lists/kvm/msg146879.html

In addition, thanks to Jan H. Schönherr's attempt last year.

v1 -> v2:
 * remove blinding setting KVM_ENABLE_CAP statement in doc
 * move PV_UNHALT associated statement to 2/3
 * rename kvm_mwait_can_in_guest to kvm_can_mwait_in_guest
 * remove unconditionally set INTERCEPT HLT in svm 
 * call vmx_clear_hlt() from pre_enter_smm()
 * add a check to kvm_update_cpuid() that forbits KVM_FEATURE_PV_UNHALT
   when halt exits are disabld

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Jan H. Schönherr <jschoenh@amazon.de>

Wanpeng Li (3):
  KVM: X86: Provides userspace with a capability to not intercept MWAIT
  KVM: X86: Provides userspace with a capability to not intercept HLT
  KVM: X86: Provides userspace with a capability to not intercept PAUSE

 Documentation/virtual/kvm/api.txt | 24 ++++++++++++-------
 arch/x86/include/asm/kvm_host.h   |  4 ++++
 arch/x86/kvm/cpuid.c              |  5 ++++
 arch/x86/kvm/svm.c                |  9 ++++---
 arch/x86/kvm/vmx.c                | 50 ++++++++++++++++++++++++++++++++-------
 arch/x86/kvm/x86.c                | 29 +++++++++++++++++++----
 arch/x86/kvm/x86.h                | 24 +++++++++++++++----
 include/uapi/linux/kvm.h          |  2 +-
 tools/include/uapi/linux/kvm.h    |  2 +-
 9 files changed, 118 insertions(+), 31 deletions(-)

-- 
2.7.4

[PATCH v2 1/3] KVM: X86: Provides userspace with a capability to not intercept MWAIT

[Wanpeng Li]

From: Wanpeng Li <wanpengli@tencent.com>

Allowing a guest to execute MWAIT without interception enables a guest
to put a (physical) CPU into a power saving state, where it takes
longer to return from than what may be desired by the host.

Don't give a guest that power over a host by default. (Especially,
since nothing prevents a guest from using MWAIT even when it is not
advertised via CPUID.)

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Jan H. Schönherr <jschoenh@amazon.de>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
---
 Documentation/virtual/kvm/api.txt | 23 ++++++++++++++---------
 arch/x86/include/asm/kvm_host.h   |  2 ++
 arch/x86/kvm/svm.c                |  2 +-
 arch/x86/kvm/vmx.c                |  9 +++++----
 arch/x86/kvm/x86.c                | 24 ++++++++++++++++++++----
 arch/x86/kvm/x86.h                | 10 +++++-----
 include/uapi/linux/kvm.h          |  2 +-
 tools/include/uapi/linux/kvm.h    |  2 +-
 8 files changed, 49 insertions(+), 25 deletions(-)

diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt
index 98de506..76e5a15 100644
--- a/Documentation/virtual/kvm/api.txt
+++ b/Documentation/virtual/kvm/api.txt
@@ -4358,6 +4358,20 @@ enables QEMU to build error log and branch to guest kernel registered
 machine check handling routine. Without this capability KVM will
 branch to guests' 0x200 interrupt vector.
 
+7.13 KVM_CAP_X86_DISABLE_EXITS
+
+Architectures: x86
+Parameters: args[0] defines which exits are disabled
+Returns: 0 on success, -EINVAL when args[0] contains invalid exits
+
+Valid exits in args[0] are
+
+#define KVM_X86_DISABLE_EXITS_MWAIT            (1 << 0)
+
+Enabling this capability on a VM provides userspace with a way to no
+longer intercepts some instructions for improved latency in some
+workloads.
+
 8. Other capabilities.
 ----------------------
 
@@ -4470,15 +4484,6 @@ reserved.
     Both registers and addresses are 64-bits wide.
     It will be possible to run 64-bit or 32-bit guest code.
 
-8.8 KVM_CAP_X86_GUEST_MWAIT
-
-Architectures: x86
-
-This capability indicates that guest using memory monotoring instructions
-(MWAIT/MWAITX) to stop the virtual CPU will not cause a VM exit.  As such time
-spent while 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).
-
 8.9 KVM_CAP_ARM_USER_IRQ
 
 Architectures: arm, arm64
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 0395c35..e107171 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -811,6 +811,8 @@ struct kvm_arch {
 
 	gpa_t wall_clock;
 
+	bool mwait_in_guest;
+
 	bool ept_identity_pagetable_done;
 	gpa_t ept_identity_map_addr;
 
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index be9c839..321b3fd 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1390,7 +1390,7 @@ static void init_vmcb(struct vcpu_svm *svm)
 	set_intercept(svm, INTERCEPT_XSETBV);
 	set_intercept(svm, INTERCEPT_RSM);
 
-	if (!kvm_mwait_in_guest()) {
+	if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
 		set_intercept(svm, INTERCEPT_MONITOR);
 		set_intercept(svm, INTERCEPT_MWAIT);
 	}
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 6cefd7b..2302ae2 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -3733,13 +3733,11 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
 	      CPU_BASED_UNCOND_IO_EXITING |
 	      CPU_BASED_MOV_DR_EXITING |
 	      CPU_BASED_USE_TSC_OFFSETING |
+	      CPU_BASED_MWAIT_EXITING |
+	      CPU_BASED_MONITOR_EXITING |
 	      CPU_BASED_INVLPG_EXITING |
 	      CPU_BASED_RDPMC_EXITING;
 
-	if (!kvm_mwait_in_guest())
-		min |= CPU_BASED_MWAIT_EXITING |
-			CPU_BASED_MONITOR_EXITING;
-
 	opt = CPU_BASED_TPR_SHADOW |
 	      CPU_BASED_USE_MSR_BITMAPS |
 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
@@ -5531,6 +5529,9 @@ static u32 vmx_exec_control(struct vcpu_vmx *vmx)
 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
 				CPU_BASED_CR3_LOAD_EXITING  |
 				CPU_BASED_INVLPG_EXITING;
+	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
+		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
+				CPU_BASED_MONITOR_EXITING);
 	return exec_control;
 }
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 36ef3d8..5fae476 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2809,9 +2809,15 @@ static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
 	return r;
 }
 
+static inline bool kvm_can_mwait_in_guest(void)
+{
+	return boot_cpu_has(X86_FEATURE_MWAIT) &&
+		!boot_cpu_has_bug(X86_BUG_MONITOR);
+}
+
 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 {
-	int r;
+	int r = 0;
 
 	switch (ext) {
 	case KVM_CAP_IRQCHIP:
@@ -2867,8 +2873,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_ADJUST_CLOCK:
 		r = KVM_CLOCK_TSC_STABLE;
 		break;
-	case KVM_CAP_X86_GUEST_MWAIT:
-		r = kvm_mwait_in_guest();
+	case KVM_CAP_X86_DISABLE_EXITS:
+		if(kvm_can_mwait_in_guest())
+			r |= KVM_X86_DISABLE_EXITS_MWAIT;
 		break;
 	case KVM_CAP_X86_SMM:
 		/* SMBASE is usually relocated above 1M on modern chipsets,
@@ -2909,7 +2916,6 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 		r = KVM_X2APIC_API_VALID_FLAGS;
 		break;
 	default:
-		r = 0;
 		break;
 	}
 	return r;
@@ -4214,6 +4220,16 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 
 		r = 0;
 		break;
+	case KVM_CAP_X86_DISABLE_EXITS:
+		r = -EINVAL;
+		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
+			break;
+
+		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
+			kvm_can_mwait_in_guest())
+			kvm->arch.mwait_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 b91215d..cd1215e 100644
--- a/arch/x86/kvm/x86.h
+++ b/arch/x86/kvm/x86.h
@@ -2,8 +2,6 @@
 #ifndef ARCH_X86_KVM_X86_H
 #define ARCH_X86_KVM_X86_H
 
-#include <asm/processor.h>
-#include <asm/mwait.h>
 #include <linux/kvm_host.h>
 #include <asm/pvclock.h>
 #include "kvm_cache_regs.h"
@@ -264,10 +262,12 @@ static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
 	    __rem;						\
 	 })
 
-static inline bool kvm_mwait_in_guest(void)
+#define KVM_X86_DISABLE_EXITS_MWAIT          (1 << 0)
+#define KVM_X86_DISABLE_VALID_EXITS          (KVM_X86_DISABLE_EXITS_MWAIT)
+
+static inline bool kvm_mwait_in_guest(struct kvm *kvm)
 {
-	return boot_cpu_has(X86_FEATURE_MWAIT) &&
-		!boot_cpu_has_bug(X86_BUG_MONITOR);
+	return kvm->arch.mwait_in_guest;
 }
 
 #endif
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index 088c2c9..1065006 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -929,7 +929,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_S390_GS 140
 #define KVM_CAP_S390_AIS 141
 #define KVM_CAP_SPAPR_TCE_VFIO 142
-#define KVM_CAP_X86_GUEST_MWAIT 143
+#define KVM_CAP_X86_DISABLE_EXITS 143
 #define KVM_CAP_ARM_USER_IRQ 144
 #define KVM_CAP_S390_CMMA_MIGRATION 145
 #define KVM_CAP_PPC_FWNMI 146
diff --git a/tools/include/uapi/linux/kvm.h b/tools/include/uapi/linux/kvm.h
index 0fb5ef9..b13c257 100644
--- a/tools/include/uapi/linux/kvm.h
+++ b/tools/include/uapi/linux/kvm.h
@@ -924,7 +924,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_S390_GS 140
 #define KVM_CAP_S390_AIS 141
 #define KVM_CAP_SPAPR_TCE_VFIO 142
-#define KVM_CAP_X86_GUEST_MWAIT 143
+#define KVM_CAP_X86_DISABLE_EXITS 143
 #define KVM_CAP_ARM_USER_IRQ 144
 #define KVM_CAP_S390_CMMA_MIGRATION 145
 #define KVM_CAP_PPC_FWNMI 146
-- 
2.7.4

[PATCH v2 2/3] KVM: X86: Provides userspace with a capability to not intercept HLT

[Wanpeng Li]

From: Wanpeng Li <wanpengli@tencent.com>

If host CPUs are dedicated to a VM, we can avoid VM exits on HLT.
This patch adds the per-VM non-HLT-exiting capability.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Jan H. Schönherr <jschoenh@amazon.de>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
---
 Documentation/virtual/kvm/api.txt |  3 ++-
 arch/x86/include/asm/kvm_host.h   |  1 +
 arch/x86/kvm/cpuid.c              |  5 +++++
 arch/x86/kvm/svm.c                |  4 +++-
 arch/x86/kvm/vmx.c                | 24 ++++++++++++++++++++++++
 arch/x86/kvm/x86.c                |  3 +++
 arch/x86/kvm/x86.h                |  9 ++++++++-
 7 files changed, 46 insertions(+), 3 deletions(-)

diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt
index 76e5a15..b46494d 100644
--- a/Documentation/virtual/kvm/api.txt
+++ b/Documentation/virtual/kvm/api.txt
@@ -4367,10 +4367,11 @@ Returns: 0 on success, -EINVAL when args[0] contains invalid exits
 Valid exits in args[0] are
 
 #define KVM_X86_DISABLE_EXITS_MWAIT            (1 << 0)
+#define KVM_X86_DISABLE_EXITS_HLT              (1 << 1)
 
 Enabling this capability on a VM provides userspace with a way to no
 longer intercepts some instructions for improved latency in some
-workloads.
+workloads. Do not enable KVM_FEATURE_PV_UNHALT if you disable HLT exits.
 
 8. Other capabilities.
 ----------------------
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index e107171..1a79065 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -812,6 +812,7 @@ struct kvm_arch {
 	gpa_t wall_clock;
 
 	bool mwait_in_guest;
+	bool hlt_in_guest;
 
 	bool ept_identity_pagetable_done;
 	gpa_t ept_identity_map_addr;
diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
index e2d3050..82055b9 100644
--- a/arch/x86/kvm/cpuid.c
+++ b/arch/x86/kvm/cpuid.c
@@ -135,6 +135,11 @@ int kvm_update_cpuid(struct kvm_vcpu *vcpu)
 			return -EINVAL;
 	}
 
+	best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
+	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
+		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
+		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
+
 	/* Update physical-address width */
 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 	kvm_mmu_reset_context(vcpu);
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 321b3fd..0b2e7af 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1372,7 +1372,6 @@ 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);
@@ -1395,6 +1394,9 @@ static void init_vmcb(struct vcpu_svm *svm)
 		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));
 	control->int_ctl = V_INTR_MASKING_MASK;
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 2302ae2..fa0c5e1 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -2543,6 +2543,19 @@ static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit
 	return 0;
 }
 
+static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
+{
+	/*
+	 * 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) &&
+			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
+		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
+}
+
 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
@@ -2573,6 +2586,8 @@ static void vmx_queue_exception(struct kvm_vcpu *vcpu)
 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
 
 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
+
+	vmx_clear_hlt(vcpu);
 }
 
 static bool vmx_rdtscp_supported(void)
@@ -5532,6 +5547,8 @@ 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;
 }
 
@@ -5893,6 +5910,8 @@ static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 	update_exception_bitmap(vcpu);
 
 	vpid_sync_context(vmx->vpid);
+	if (init_event)
+		vmx_clear_hlt(vcpu);
 }
 
 /*
@@ -5963,6 +5982,8 @@ static void vmx_inject_irq(struct kvm_vcpu *vcpu)
 	} else
 		intr |= INTR_TYPE_EXT_INTR;
 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
+
+	vmx_clear_hlt(vcpu);
 }
 
 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
@@ -5993,6 +6014,8 @@ static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
 
 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
+
+	vmx_clear_hlt(vcpu);
 }
 
 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
@@ -12314,6 +12337,7 @@ static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
 
 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
 	vmx->nested.vmxon = false;
+	vmx_clear_hlt(vcpu);
 	return 0;
 }
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 5fae476..73255e6 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2874,6 +2874,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 		r = KVM_CLOCK_TSC_STABLE;
 		break;
 	case KVM_CAP_X86_DISABLE_EXITS:
+		r |=  KVM_X86_DISABLE_EXITS_HTL;
 		if(kvm_can_mwait_in_guest())
 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
 		break;
@@ -4228,6 +4229,8 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
 			kvm_can_mwait_in_guest())
 			kvm->arch.mwait_in_guest = true;
+		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
+			kvm->arch.hlt_in_guest = true;
 		r = 0;
 		break;
 	default:
diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
index cd1215e..d4ddb00 100644
--- a/arch/x86/kvm/x86.h
+++ b/arch/x86/kvm/x86.h
@@ -263,11 +263,18 @@ static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
 	 })
 
 #define KVM_X86_DISABLE_EXITS_MWAIT          (1 << 0)
-#define KVM_X86_DISABLE_VALID_EXITS          (KVM_X86_DISABLE_EXITS_MWAIT)
+#define KVM_X86_DISABLE_EXITS_HTL            (1 << 1)
+#define KVM_X86_DISABLE_VALID_EXITS          (KVM_X86_DISABLE_EXITS_MWAIT | \
+                                              KVM_X86_DISABLE_EXITS_HTL)
 
 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
-- 
2.7.4

[PATCH v2 3/3] KVM: X86: Provides userspace with a capability to not intercept PAUSE

[Wanpeng Li]

From: Wanpeng Li <wanpengli@tencent.com>

Allow to disable pause loop exit/pause filtering on a per VM basis.

If some VMs have dedicated host CPUs, they won't be negatively affected
due to needlessly intercepted PAUSE instructions.

Thanks to Jan H. Schönherr's initial patch.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Jan H. Schönherr <jschoenh@amazon.de>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
---
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/svm.c              |  3 ++-
 arch/x86/kvm/vmx.c              | 17 +++++++++++++----
 arch/x86/kvm/x86.c              |  4 +++-
 arch/x86/kvm/x86.h              |  9 ++++++++-
 5 files changed, 27 insertions(+), 7 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 1a79065..d73ca26 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -813,6 +813,7 @@ struct kvm_arch {
 
 	bool mwait_in_guest;
 	bool hlt_in_guest;
+	bool pause_in_guest;
 
 	bool ept_identity_pagetable_done;
 	gpa_t ept_identity_map_addr;
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 0b2e7af..ddc705c 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1452,7 +1452,8 @@ static void init_vmcb(struct vcpu_svm *svm)
 	svm->nested.vmcb = 0;
 	svm->vcpu.arch.hflags = 0;
 
-	if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
+	if (boot_cpu_has(X86_FEATURE_PAUSEFILTER) &&
+	    !kvm_pause_in_guest(svm->vcpu.kvm)) {
 		control->pause_filter_count = 3000;
 		set_intercept(svm, INTERCEPT_PAUSE);
 	}
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index fa0c5e1..400f9d1 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -5582,7 +5582,7 @@ static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
 	}
 	if (!enable_unrestricted_guest)
 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
-	if (!ple_gap)
+	if (kvm_pause_in_guest(vmx->vcpu.kvm))
 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
 	if (!kvm_vcpu_apicv_active(vcpu))
 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
@@ -5745,7 +5745,7 @@ static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
 	}
 
-	if (ple_gap) {
+	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
 		vmcs_write32(PLE_GAP, ple_gap);
 		vmx->ple_window = ple_window;
 		vmx->ple_window_dirty = true;
@@ -7182,7 +7182,7 @@ static __exit void hardware_unsetup(void)
  */
 static int handle_pause(struct kvm_vcpu *vcpu)
 {
-	if (ple_gap)
+	if (!kvm_pause_in_guest(vcpu->kvm))
 		grow_ple_window(vcpu);
 
 	/*
@@ -9878,6 +9878,13 @@ static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
 	return ERR_PTR(err);
 }
 
+static int vmx_vm_init(struct kvm *kvm)
+{
+	if (!ple_gap)
+		kvm->arch.pause_in_guest = true;
+	return 0;
+}
+
 static void __init vmx_check_processor_compat(void *rtn)
 {
 	struct vmcs_config vmcs_conf;
@@ -12019,7 +12026,7 @@ static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
 
 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
 {
-	if (ple_gap)
+	if (!kvm_pause_in_guest(vcpu->kvm))
 		shrink_ple_window(vcpu);
 }
 
@@ -12379,6 +12386,8 @@ static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
 	.cpu_has_accelerated_tpr = report_flexpriority,
 	.cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
 
+	.vm_init = vmx_vm_init,
+
 	.vcpu_create = vmx_create_vcpu,
 	.vcpu_free = vmx_free_vcpu,
 	.vcpu_reset = vmx_vcpu_reset,
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 73255e6..8060f27 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2874,7 +2874,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 		r = KVM_CLOCK_TSC_STABLE;
 		break;
 	case KVM_CAP_X86_DISABLE_EXITS:
-		r |=  KVM_X86_DISABLE_EXITS_HTL;
+		r |=  KVM_X86_DISABLE_EXITS_HTL | KVM_X86_DISABLE_EXITS_PAUSE;
 		if(kvm_can_mwait_in_guest())
 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
 		break;
@@ -4231,6 +4231,8 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 			kvm->arch.mwait_in_guest = true;
 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
 			kvm->arch.hlt_in_guest = true;
+		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
+			kvm->arch.pause_in_guest = true;
 		r = 0;
 		break;
 	default:
diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
index d4ddb00..658ea9a 100644
--- a/arch/x86/kvm/x86.h
+++ b/arch/x86/kvm/x86.h
@@ -264,8 +264,10 @@ static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
 
 #define KVM_X86_DISABLE_EXITS_MWAIT          (1 << 0)
 #define KVM_X86_DISABLE_EXITS_HTL            (1 << 1)
+#define KVM_X86_DISABLE_EXITS_PAUSE          (1 << 2)
 #define KVM_X86_DISABLE_VALID_EXITS          (KVM_X86_DISABLE_EXITS_MWAIT | \
-                                              KVM_X86_DISABLE_EXITS_HTL)
+                                              KVM_X86_DISABLE_EXITS_HTL | \
+                                              KVM_X86_DISABLE_EXITS_PAUSE)
 
 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
 {
@@ -277,4 +279,9 @@ static inline bool kvm_hlt_in_guest(struct kvm *kvm)
 	return kvm->arch.hlt_in_guest;
 }
 
+static inline bool kvm_pause_in_guest(struct kvm *kvm)
+{
+	return kvm->arch.pause_in_guest;
+}
+
 #endif
-- 
2.7.4

Re: [PATCH v2 1/3] KVM: X86: Provides userspace with a capability to not intercept MWAIT

[Jim Mattson]

Is there a need for a new API for yielding MONITOR/MWAIT to the guest?
Why not just tie this to the guest CPUID.01H:ECX[MWAIT] being set?

On Mon, Mar 12, 2018 at 4:53 AM, Wanpeng Li <kernellwp@gmail.com> wrote:
> From: Wanpeng Li <wanpengli@tencent.com>
>
> Allowing a guest to execute MWAIT without interception enables a guest
> to put a (physical) CPU into a power saving state, where it takes
> longer to return from than what may be desired by the host.
>
> Don't give a guest that power over a host by default. (Especially,
> since nothing prevents a guest from using MWAIT even when it is not
> advertised via CPUID.)
>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Radim Krčmář <rkrcmar@redhat.com>
> Cc: Jan H. Schönherr <jschoenh@amazon.de>
> Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> ---
>  Documentation/virtual/kvm/api.txt | 23 ++++++++++++++---------
>  arch/x86/include/asm/kvm_host.h   |  2 ++
>  arch/x86/kvm/svm.c                |  2 +-
>  arch/x86/kvm/vmx.c                |  9 +++++----
>  arch/x86/kvm/x86.c                | 24 ++++++++++++++++++++----
>  arch/x86/kvm/x86.h                | 10 +++++-----
>  include/uapi/linux/kvm.h          |  2 +-
>  tools/include/uapi/linux/kvm.h    |  2 +-
>  8 files changed, 49 insertions(+), 25 deletions(-)
>
> diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt
> index 98de506..76e5a15 100644
> --- a/Documentation/virtual/kvm/api.txt
> +++ b/Documentation/virtual/kvm/api.txt
> @@ -4358,6 +4358,20 @@ enables QEMU to build error log and branch to guest kernel registered
>  machine check handling routine. Without this capability KVM will
>  branch to guests' 0x200 interrupt vector.
>
> +7.13 KVM_CAP_X86_DISABLE_EXITS
> +
> +Architectures: x86
> +Parameters: args[0] defines which exits are disabled
> +Returns: 0 on success, -EINVAL when args[0] contains invalid exits
> +
> +Valid exits in args[0] are
> +
> +#define KVM_X86_DISABLE_EXITS_MWAIT            (1 << 0)
> +
> +Enabling this capability on a VM provides userspace with a way to no
> +longer intercepts some instructions for improved latency in some
> +workloads.
> +
>  8. Other capabilities.
>  ----------------------
>
> @@ -4470,15 +4484,6 @@ reserved.
>      Both registers and addresses are 64-bits wide.
>      It will be possible to run 64-bit or 32-bit guest code.
>
> -8.8 KVM_CAP_X86_GUEST_MWAIT
> -
> -Architectures: x86
> -
> -This capability indicates that guest using memory monotoring instructions
> -(MWAIT/MWAITX) to stop the virtual CPU will not cause a VM exit.  As such time
> -spent while 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).
> -
>  8.9 KVM_CAP_ARM_USER_IRQ
>
>  Architectures: arm, arm64
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 0395c35..e107171 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -811,6 +811,8 @@ struct kvm_arch {
>
>         gpa_t wall_clock;
>
> +       bool mwait_in_guest;
> +
>         bool ept_identity_pagetable_done;
>         gpa_t ept_identity_map_addr;
>
> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
> index be9c839..321b3fd 100644
> --- a/arch/x86/kvm/svm.c
> +++ b/arch/x86/kvm/svm.c
> @@ -1390,7 +1390,7 @@ static void init_vmcb(struct vcpu_svm *svm)
>         set_intercept(svm, INTERCEPT_XSETBV);
>         set_intercept(svm, INTERCEPT_RSM);
>
> -       if (!kvm_mwait_in_guest()) {
> +       if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
>                 set_intercept(svm, INTERCEPT_MONITOR);
>                 set_intercept(svm, INTERCEPT_MWAIT);
>         }
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 6cefd7b..2302ae2 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -3733,13 +3733,11 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
>               CPU_BASED_UNCOND_IO_EXITING |
>               CPU_BASED_MOV_DR_EXITING |
>               CPU_BASED_USE_TSC_OFFSETING |
> +             CPU_BASED_MWAIT_EXITING |
> +             CPU_BASED_MONITOR_EXITING |
>               CPU_BASED_INVLPG_EXITING |
>               CPU_BASED_RDPMC_EXITING;
>
> -       if (!kvm_mwait_in_guest())
> -               min |= CPU_BASED_MWAIT_EXITING |
> -                       CPU_BASED_MONITOR_EXITING;
> -
>         opt = CPU_BASED_TPR_SHADOW |
>               CPU_BASED_USE_MSR_BITMAPS |
>               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
> @@ -5531,6 +5529,9 @@ static u32 vmx_exec_control(struct vcpu_vmx *vmx)
>                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
>                                 CPU_BASED_CR3_LOAD_EXITING  |
>                                 CPU_BASED_INVLPG_EXITING;
> +       if (kvm_mwait_in_guest(vmx->vcpu.kvm))
> +               exec_control &= ~(CPU_BASED_MWAIT_EXITING |
> +                               CPU_BASED_MONITOR_EXITING);
>         return exec_control;
>  }
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 36ef3d8..5fae476 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2809,9 +2809,15 @@ static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
>         return r;
>  }
>
> +static inline bool kvm_can_mwait_in_guest(void)
> +{
> +       return boot_cpu_has(X86_FEATURE_MWAIT) &&
> +               !boot_cpu_has_bug(X86_BUG_MONITOR);
> +}
> +
>  int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>  {
> -       int r;
> +       int r = 0;
>
>         switch (ext) {
>         case KVM_CAP_IRQCHIP:
> @@ -2867,8 +2873,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>         case KVM_CAP_ADJUST_CLOCK:
>                 r = KVM_CLOCK_TSC_STABLE;
>                 break;
> -       case KVM_CAP_X86_GUEST_MWAIT:
> -               r = kvm_mwait_in_guest();
> +       case KVM_CAP_X86_DISABLE_EXITS:
> +               if(kvm_can_mwait_in_guest())
> +                       r |= KVM_X86_DISABLE_EXITS_MWAIT;
>                 break;
>         case KVM_CAP_X86_SMM:
>                 /* SMBASE is usually relocated above 1M on modern chipsets,
> @@ -2909,7 +2916,6 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>                 r = KVM_X2APIC_API_VALID_FLAGS;
>                 break;
>         default:
> -               r = 0;
>                 break;
>         }
>         return r;
> @@ -4214,6 +4220,16 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
>
>                 r = 0;
>                 break;
> +       case KVM_CAP_X86_DISABLE_EXITS:
> +               r = -EINVAL;
> +               if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
> +                       break;
> +
> +               if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
> +                       kvm_can_mwait_in_guest())
> +                       kvm->arch.mwait_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 b91215d..cd1215e 100644
> --- a/arch/x86/kvm/x86.h
> +++ b/arch/x86/kvm/x86.h
> @@ -2,8 +2,6 @@
>  #ifndef ARCH_X86_KVM_X86_H
>  #define ARCH_X86_KVM_X86_H
>
> -#include <asm/processor.h>
> -#include <asm/mwait.h>
>  #include <linux/kvm_host.h>
>  #include <asm/pvclock.h>
>  #include "kvm_cache_regs.h"
> @@ -264,10 +262,12 @@ static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
>             __rem;                                              \
>          })
>
> -static inline bool kvm_mwait_in_guest(void)
> +#define KVM_X86_DISABLE_EXITS_MWAIT          (1 << 0)
> +#define KVM_X86_DISABLE_VALID_EXITS          (KVM_X86_DISABLE_EXITS_MWAIT)
> +
> +static inline bool kvm_mwait_in_guest(struct kvm *kvm)
>  {
> -       return boot_cpu_has(X86_FEATURE_MWAIT) &&
> -               !boot_cpu_has_bug(X86_BUG_MONITOR);
> +       return kvm->arch.mwait_in_guest;
>  }
>
>  #endif
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index 088c2c9..1065006 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -929,7 +929,7 @@ struct kvm_ppc_resize_hpt {
>  #define KVM_CAP_S390_GS 140
>  #define KVM_CAP_S390_AIS 141
>  #define KVM_CAP_SPAPR_TCE_VFIO 142
> -#define KVM_CAP_X86_GUEST_MWAIT 143
> +#define KVM_CAP_X86_DISABLE_EXITS 143
>  #define KVM_CAP_ARM_USER_IRQ 144
>  #define KVM_CAP_S390_CMMA_MIGRATION 145
>  #define KVM_CAP_PPC_FWNMI 146
> diff --git a/tools/include/uapi/linux/kvm.h b/tools/include/uapi/linux/kvm.h
> index 0fb5ef9..b13c257 100644
> --- a/tools/include/uapi/linux/kvm.h
> +++ b/tools/include/uapi/linux/kvm.h
> @@ -924,7 +924,7 @@ struct kvm_ppc_resize_hpt {
>  #define KVM_CAP_S390_GS 140
>  #define KVM_CAP_S390_AIS 141
>  #define KVM_CAP_SPAPR_TCE_VFIO 142
> -#define KVM_CAP_X86_GUEST_MWAIT 143
> +#define KVM_CAP_X86_DISABLE_EXITS 143
>  #define KVM_CAP_ARM_USER_IRQ 144
>  #define KVM_CAP_S390_CMMA_MIGRATION 145
>  #define KVM_CAP_PPC_FWNMI 146
> --
> 2.7.4
>

Re: [PATCH v2 1/3] KVM: X86: Provides userspace with a capability to not intercept MWAIT

[Wanpeng Li]

Hi Jim,
2018-03-14 2:21 GMT+08:00 Jim Mattson <jmattson@google.com>:
> Is there a need for a new API for yielding MONITOR/MWAIT to the guest?
> Why not just tie this to the guest CPUID.01H:ECX[MWAIT] being set?

The API also will be used by HLT/PAUSE. Please refer to Paolo's
original proposal though I didn't find a link which is replied by
Paolo direclty. https://marc.info/?l=kvm&m=151182818103804&w=2

Regards,
Wanpeng Li

>
> On Mon, Mar 12, 2018 at 4:53 AM, Wanpeng Li <kernellwp@gmail.com> wrote:
>> From: Wanpeng Li <wanpengli@tencent.com>
>>
>> Allowing a guest to execute MWAIT without interception enables a guest
>> to put a (physical) CPU into a power saving state, where it takes
>> longer to return from than what may be desired by the host.
>>
>> Don't give a guest that power over a host by default. (Especially,
>> since nothing prevents a guest from using MWAIT even when it is not
>> advertised via CPUID.)
>>
>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>> Cc: Radim Krčmář <rkrcmar@redhat.com>
>> Cc: Jan H. Schönherr <jschoenh@amazon.de>
>> Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
>> ---
>>  Documentation/virtual/kvm/api.txt | 23 ++++++++++++++---------
>>  arch/x86/include/asm/kvm_host.h   |  2 ++
>>  arch/x86/kvm/svm.c                |  2 +-
>>  arch/x86/kvm/vmx.c                |  9 +++++----
>>  arch/x86/kvm/x86.c                | 24 ++++++++++++++++++++----
>>  arch/x86/kvm/x86.h                | 10 +++++-----
>>  include/uapi/linux/kvm.h          |  2 +-
>>  tools/include/uapi/linux/kvm.h    |  2 +-
>>  8 files changed, 49 insertions(+), 25 deletions(-)
>>
>> diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt
>> index 98de506..76e5a15 100644
>> --- a/Documentation/virtual/kvm/api.txt
>> +++ b/Documentation/virtual/kvm/api.txt
>> @@ -4358,6 +4358,20 @@ enables QEMU to build error log and branch to guest kernel registered
>>  machine check handling routine. Without this capability KVM will
>>  branch to guests' 0x200 interrupt vector.
>>
>> +7.13 KVM_CAP_X86_DISABLE_EXITS
>> +
>> +Architectures: x86
>> +Parameters: args[0] defines which exits are disabled
>> +Returns: 0 on success, -EINVAL when args[0] contains invalid exits
>> +
>> +Valid exits in args[0] are
>> +
>> +#define KVM_X86_DISABLE_EXITS_MWAIT            (1 << 0)
>> +
>> +Enabling this capability on a VM provides userspace with a way to no
>> +longer intercepts some instructions for improved latency in some
>> +workloads.
>> +
>>  8. Other capabilities.
>>  ----------------------
>>
>> @@ -4470,15 +4484,6 @@ reserved.
>>      Both registers and addresses are 64-bits wide.
>>      It will be possible to run 64-bit or 32-bit guest code.
>>
>> -8.8 KVM_CAP_X86_GUEST_MWAIT
>> -
>> -Architectures: x86
>> -
>> -This capability indicates that guest using memory monotoring instructions
>> -(MWAIT/MWAITX) to stop the virtual CPU will not cause a VM exit.  As such time
>> -spent while 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).
>> -
>>  8.9 KVM_CAP_ARM_USER_IRQ
>>
>>  Architectures: arm, arm64
>> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
>> index 0395c35..e107171 100644
>> --- a/arch/x86/include/asm/kvm_host.h
>> +++ b/arch/x86/include/asm/kvm_host.h
>> @@ -811,6 +811,8 @@ struct kvm_arch {
>>
>>         gpa_t wall_clock;
>>
>> +       bool mwait_in_guest;
>> +
>>         bool ept_identity_pagetable_done;
>>         gpa_t ept_identity_map_addr;
>>
>> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
>> index be9c839..321b3fd 100644
>> --- a/arch/x86/kvm/svm.c
>> +++ b/arch/x86/kvm/svm.c
>> @@ -1390,7 +1390,7 @@ static void init_vmcb(struct vcpu_svm *svm)
>>         set_intercept(svm, INTERCEPT_XSETBV);
>>         set_intercept(svm, INTERCEPT_RSM);
>>
>> -       if (!kvm_mwait_in_guest()) {
>> +       if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
>>                 set_intercept(svm, INTERCEPT_MONITOR);
>>                 set_intercept(svm, INTERCEPT_MWAIT);
>>         }
>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>> index 6cefd7b..2302ae2 100644
>> --- a/arch/x86/kvm/vmx.c
>> +++ b/arch/x86/kvm/vmx.c
>> @@ -3733,13 +3733,11 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
>>               CPU_BASED_UNCOND_IO_EXITING |
>>               CPU_BASED_MOV_DR_EXITING |
>>               CPU_BASED_USE_TSC_OFFSETING |
>> +             CPU_BASED_MWAIT_EXITING |
>> +             CPU_BASED_MONITOR_EXITING |
>>               CPU_BASED_INVLPG_EXITING |
>>               CPU_BASED_RDPMC_EXITING;
>>
>> -       if (!kvm_mwait_in_guest())
>> -               min |= CPU_BASED_MWAIT_EXITING |
>> -                       CPU_BASED_MONITOR_EXITING;
>> -
>>         opt = CPU_BASED_TPR_SHADOW |
>>               CPU_BASED_USE_MSR_BITMAPS |
>>               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
>> @@ -5531,6 +5529,9 @@ static u32 vmx_exec_control(struct vcpu_vmx *vmx)
>>                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
>>                                 CPU_BASED_CR3_LOAD_EXITING  |
>>                                 CPU_BASED_INVLPG_EXITING;
>> +       if (kvm_mwait_in_guest(vmx->vcpu.kvm))
>> +               exec_control &= ~(CPU_BASED_MWAIT_EXITING |
>> +                               CPU_BASED_MONITOR_EXITING);
>>         return exec_control;
>>  }
>>
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index 36ef3d8..5fae476 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -2809,9 +2809,15 @@ static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
>>         return r;
>>  }
>>
>> +static inline bool kvm_can_mwait_in_guest(void)
>> +{
>> +       return boot_cpu_has(X86_FEATURE_MWAIT) &&
>> +               !boot_cpu_has_bug(X86_BUG_MONITOR);
>> +}
>> +
>>  int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>>  {
>> -       int r;
>> +       int r = 0;
>>
>>         switch (ext) {
>>         case KVM_CAP_IRQCHIP:
>> @@ -2867,8 +2873,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>>         case KVM_CAP_ADJUST_CLOCK:
>>                 r = KVM_CLOCK_TSC_STABLE;
>>                 break;
>> -       case KVM_CAP_X86_GUEST_MWAIT:
>> -               r = kvm_mwait_in_guest();
>> +       case KVM_CAP_X86_DISABLE_EXITS:
>> +               if(kvm_can_mwait_in_guest())
>> +                       r |= KVM_X86_DISABLE_EXITS_MWAIT;
>>                 break;
>>         case KVM_CAP_X86_SMM:
>>                 /* SMBASE is usually relocated above 1M on modern chipsets,
>> @@ -2909,7 +2916,6 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>>                 r = KVM_X2APIC_API_VALID_FLAGS;
>>                 break;
>>         default:
>> -               r = 0;
>>                 break;
>>         }
>>         return r;
>> @@ -4214,6 +4220,16 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
>>
>>                 r = 0;
>>                 break;
>> +       case KVM_CAP_X86_DISABLE_EXITS:
>> +               r = -EINVAL;
>> +               if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
>> +                       break;
>> +
>> +               if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
>> +                       kvm_can_mwait_in_guest())
>> +                       kvm->arch.mwait_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 b91215d..cd1215e 100644
>> --- a/arch/x86/kvm/x86.h
>> +++ b/arch/x86/kvm/x86.h
>> @@ -2,8 +2,6 @@
>>  #ifndef ARCH_X86_KVM_X86_H
>>  #define ARCH_X86_KVM_X86_H
>>
>> -#include <asm/processor.h>
>> -#include <asm/mwait.h>
>>  #include <linux/kvm_host.h>
>>  #include <asm/pvclock.h>
>>  #include "kvm_cache_regs.h"
>> @@ -264,10 +262,12 @@ static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
>>             __rem;                                              \
>>          })
>>
>> -static inline bool kvm_mwait_in_guest(void)
>> +#define KVM_X86_DISABLE_EXITS_MWAIT          (1 << 0)
>> +#define KVM_X86_DISABLE_VALID_EXITS          (KVM_X86_DISABLE_EXITS_MWAIT)
>> +
>> +static inline bool kvm_mwait_in_guest(struct kvm *kvm)
>>  {
>> -       return boot_cpu_has(X86_FEATURE_MWAIT) &&
>> -               !boot_cpu_has_bug(X86_BUG_MONITOR);
>> +       return kvm->arch.mwait_in_guest;
>>  }
>>
>>  #endif
>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
>> index 088c2c9..1065006 100644
>> --- a/include/uapi/linux/kvm.h
>> +++ b/include/uapi/linux/kvm.h
>> @@ -929,7 +929,7 @@ struct kvm_ppc_resize_hpt {
>>  #define KVM_CAP_S390_GS 140
>>  #define KVM_CAP_S390_AIS 141
>>  #define KVM_CAP_SPAPR_TCE_VFIO 142
>> -#define KVM_CAP_X86_GUEST_MWAIT 143
>> +#define KVM_CAP_X86_DISABLE_EXITS 143
>>  #define KVM_CAP_ARM_USER_IRQ 144
>>  #define KVM_CAP_S390_CMMA_MIGRATION 145
>>  #define KVM_CAP_PPC_FWNMI 146
>> diff --git a/tools/include/uapi/linux/kvm.h b/tools/include/uapi/linux/kvm.h
>> index 0fb5ef9..b13c257 100644
>> --- a/tools/include/uapi/linux/kvm.h
>> +++ b/tools/include/uapi/linux/kvm.h
>> @@ -924,7 +924,7 @@ struct kvm_ppc_resize_hpt {
>>  #define KVM_CAP_S390_GS 140
>>  #define KVM_CAP_S390_AIS 141
>>  #define KVM_CAP_SPAPR_TCE_VFIO 142
>> -#define KVM_CAP_X86_GUEST_MWAIT 143
>> +#define KVM_CAP_X86_DISABLE_EXITS 143
>>  #define KVM_CAP_ARM_USER_IRQ 144
>>  #define KVM_CAP_S390_CMMA_MIGRATION 145
>>  #define KVM_CAP_PPC_FWNMI 146
>> --
>> 2.7.4
>>

Re: [PATCH v2 0/3] Provides userspace with per-VM capability to improve latency

[Wanpeng Li]

2018-03-15 0:11 GMT+08:00 Paolo Bonzini <pbonzini@redhat.com>:
> On 12/03/2018 12:53, Wanpeng Li wrote:
>> Provides userspace with per-VM capability(KVM_CAP_X86_DISABLE_EXITS) to
>> not intercept MWAIT/HLT/PAUSE in order that to improve latency in some
>> workloads.
>>
>> The patchset implements the original proposal from Radim.
>> https://www.spinics.net/lists/kvm/msg146879.html
>>
>> In addition, thanks to Jan H. Schönherr's attempt last year.
>>
>> v1 -> v2:
>>  * remove blinding setting KVM_ENABLE_CAP statement in doc
>>  * move PV_UNHALT associated statement to 2/3
>>  * rename kvm_mwait_can_in_guest to kvm_can_mwait_in_guest
>>  * remove unconditionally set INTERCEPT HLT in svm
>>  * call vmx_clear_hlt() from pre_enter_smm()
>>  * add a check to kvm_update_cpuid() that forbits KVM_FEATURE_PV_UNHALT
>>    when halt exits are disabld
>>
>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>> Cc: Radim Krčmář <rkrcmar@redhat.com>
>> Cc: Jan H. Schönherr <jschoenh@amazon.de>
>>
>> Wanpeng Li (3):
>>   KVM: X86: Provides userspace with a capability to not intercept MWAIT
>>   KVM: X86: Provides userspace with a capability to not intercept HLT
>>   KVM: X86: Provides userspace with a capability to not intercept PAUSE
>>
>>  Documentation/virtual/kvm/api.txt | 24 ++++++++++++-------
>>  arch/x86/include/asm/kvm_host.h   |  4 ++++
>>  arch/x86/kvm/cpuid.c              |  5 ++++
>>  arch/x86/kvm/svm.c                |  9 ++++---
>>  arch/x86/kvm/vmx.c                | 50 ++++++++++++++++++++++++++++++++-------
>>  arch/x86/kvm/x86.c                | 29 +++++++++++++++++++----
>>  arch/x86/kvm/x86.h                | 24 +++++++++++++++----
>>  include/uapi/linux/kvm.h          |  2 +-
>>  tools/include/uapi/linux/kvm.h    |  2 +-
>>  9 files changed, 118 insertions(+), 31 deletions(-)
>>
>
> Queued, thanks.

Thanks Paolo. :)

>Do you have QEMU patches to automatically enable this
> together with HINTS_DEDICATED?

I will cook the patches. :)

Regards,
Wanpeng Li

Re: [PATCH v2 0/3] Provides userspace with per-VM capability to improve latency

[Wanpeng Li]

2018-03-12 19:53 GMT+08:00 Wanpeng Li <kernellwp@gmail.com>:
> Provides userspace with per-VM capability(KVM_CAP_X86_DISABLE_EXITS) to
> not intercept MWAIT/HLT/PAUSE in order that to improve latency in some
> workloads.

When running cyclictest in the guest w/ vCPU pin on host and
cyclictest pin in guest, the avg latency can be reduced 40%.

Regards,
Wanpeng Li

>
> The patchset implements the original proposal from Radim.
> https://www.spinics.net/lists/kvm/msg146879.html
>
> In addition, thanks to Jan H. Schönherr's attempt last year.
>
> v1 -> v2:
>  * remove blinding setting KVM_ENABLE_CAP statement in doc
>  * move PV_UNHALT associated statement to 2/3
>  * rename kvm_mwait_can_in_guest to kvm_can_mwait_in_guest
>  * remove unconditionally set INTERCEPT HLT in svm
>  * call vmx_clear_hlt() from pre_enter_smm()
>  * add a check to kvm_update_cpuid() that forbits KVM_FEATURE_PV_UNHALT
>    when halt exits are disabld
>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Radim Krčmář <rkrcmar@redhat.com>
> Cc: Jan H. Schönherr <jschoenh@amazon.de>
>
> Wanpeng Li (3):
>   KVM: X86: Provides userspace with a capability to not intercept MWAIT
>   KVM: X86: Provides userspace with a capability to not intercept HLT
>   KVM: X86: Provides userspace with a capability to not intercept PAUSE
>
>  Documentation/virtual/kvm/api.txt | 24 ++++++++++++-------
>  arch/x86/include/asm/kvm_host.h   |  4 ++++
>  arch/x86/kvm/cpuid.c              |  5 ++++
>  arch/x86/kvm/svm.c                |  9 ++++---
>  arch/x86/kvm/vmx.c                | 50 ++++++++++++++++++++++++++++++++-------
>  arch/x86/kvm/x86.c                | 29 +++++++++++++++++++----
>  arch/x86/kvm/x86.h                | 24 +++++++++++++++----
>  include/uapi/linux/kvm.h          |  2 +-
>  tools/include/uapi/linux/kvm.h    |  2 +-
>  9 files changed, 118 insertions(+), 31 deletions(-)
>
> --
> 2.7.4
>