Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Wanpeng Li <kernellwp@gmail.com>
To:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org
Cc:     Paolo Bonzini <pbonzini@redhat.com>,
        Sean Christopherson <sean.j.christopherson@intel.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>
Subject: [PATCH 3/3] KVM: X86: Micro-optimize IPI fastpath delay
Date:   Thu, 26 Mar 2020 10:20:02 +0800
Message-Id: <1585189202-1708-4-git-send-email-wanpengli@tencent.com>
In-Reply-To: <1585189202-1708-1-git-send-email-wanpengli@tencent.com>
References: <1585189202-1708-1-git-send-email-wanpengli@tencent.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: Wanpeng Li <wanpengli@tencent.com>

This patch optimizes the virtual IPI fastpath emulation sequence:

write ICR2                          send virtual IPI
read ICR2                           write ICR2
send virtual IPI         ==>        write ICR
write ICR

We can observe ~0.67% performance improvement for IPI microbenchmark
(https://lore.kernel.org/kvm/20171219085010.4081-1-ynorov@caviumnetworks.com/) 
on Skylake server.

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
---
 arch/x86/kvm/lapic.c | 4 ++--
 arch/x86/kvm/lapic.h | 1 +
 arch/x86/kvm/x86.c   | 6 +++++-
 3 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index e3099c6..338de38 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1226,7 +1226,7 @@ void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
 }
 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
 
-static void apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
+void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
 {
 	struct kvm_lapic_irq irq;
 
@@ -1940,7 +1940,7 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
 	case APIC_ICR:
 		/* No delay here, so we always clear the pending bit */
 		val &= ~(1 << 12);
-		apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
+		kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
 		kvm_lapic_set_reg(apic, APIC_ICR, val);
 		break;
 
diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
index ec6fbfe..bc76860 100644
--- a/arch/x86/kvm/lapic.h
+++ b/arch/x86/kvm/lapic.h
@@ -95,6 +95,7 @@ void kvm_apic_update_apicv(struct kvm_vcpu *vcpu);
 
 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map);
+void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high);
 
 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu);
 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info);
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 50ef1c5..c4bb7d8 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1561,8 +1561,12 @@ static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data
 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
 		((data & APIC_MODE_MASK) == APIC_DM_FIXED)) {
 
+		data &= ~(1 << 12);
+		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
-		return kvm_lapic_reg_write(vcpu->arch.apic, APIC_ICR, (u32)data);
+		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
+		trace_kvm_apic_write(APIC_ICR, (u32)data);
+		return 0;
 	}
 
 	return 1;
-- 
2.7.4