diff options
| author | Wanpeng Li <wanpengli@tencent.com> | 2020-03-26 10:20:02 +0800 |
|---|---|---|
| committer | Paolo Bonzini <pbonzini@redhat.com> | 2020-03-26 05:58:26 -0400 |
| commit | d5361678e63c8a5e72d75cee6d15b840c44306f2 (patch) | |
| tree | f7ff872c969cbe77c3c7e324302781097c0e3d5e /arch/x86/kvm/x86.c | |
| parent | 8a1038de11a5536c76054061837b11648bec5b46 (diff) | |
| download | linux-d5361678e63c8a5e72d75cee6d15b840c44306f2.tar.gz linux-d5361678e63c8a5e72d75cee6d15b840c44306f2.tar.bz2 linux-d5361678e63c8a5e72d75cee6d15b840c44306f2.zip | |
KVM: X86: Micro-optimize IPI fastpath delay
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>
Message-Id: <1585189202-1708-4-git-send-email-wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Diffstat (limited to 'arch/x86/kvm/x86.c')
| -rw-r--r-- | arch/x86/kvm/x86.c | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 62d614586402..6fa014ccd253 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -1585,8 +1585,12 @@ 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; |