linux-kernel - [PATCH 3/3] KVM: LAPIC: Optimize timer latency further

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <1557401361-3828-4-git-send-email-wanpengli@tencent.com>
Date:   Thu,  9 May 2019 19:29:21 +0800
From:   Wanpeng Li <kernellwp@...il.com>
To:     linux-kernel@...r.kernel.org, kvm@...r.kernel.org
Cc:     Paolo Bonzini <pbonzini@...hat.com>,
        Radim Krčmář <rkrcmar@...hat.com>,
        Sean Christopherson <sean.j.christopherson@...el.com>,
        Liran Alon <liran.alon@...cle.com>
Subject: [PATCH 3/3] KVM: LAPIC: Optimize timer latency further

From: Wanpeng Li <wanpengli@...cent.com>

Advance lapic timer tries to hidden the hypervisor overhead between host 
timer fires and the guest awares the timer is fired. However, it just hidden 
the time between apic_timer_fn/handle_preemption_timer -> wait_lapic_expire, 
instead of the real position of vmentry which is mentioned in the orignial 
commit d0659d946be0 ("KVM: x86: add option to advance tscdeadline hrtimer 
expiration"). There is 700+ cpu cycles between the end of wait_lapic_expire 
and before world switch on my haswell desktop, it will be 2400+ cycles if 
vmentry_l1d_flush is tuned to always. 

This patch tries to narrow the last gap, it measures the time between 
the end of wait_lapic_expire and before world switch, we take this 
time into consideration when busy waiting, otherwise, the guest still 
awares the latency between wait_lapic_expire and world switch, we also 
consider this when adaptively tuning the timer advancement. The patch 
can reduce 50% latency (~1600+ cycles to ~800+ cycles on a haswell 
desktop) for kvm-unit-tests/tscdeadline_latency when testing busy waits.

Cc: Paolo Bonzini <pbonzini@...hat.com>
Cc: Radim Krčmář <rkrcmar@...hat.com>
Cc: Sean Christopherson <sean.j.christopherson@...el.com>
Cc: Liran Alon <liran.alon@...cle.com>
Signed-off-by: Wanpeng Li <wanpengli@...cent.com>
---
 arch/x86/kvm/lapic.c   | 23 +++++++++++++++++++++--
 arch/x86/kvm/lapic.h   |  8 ++++++++
 arch/x86/kvm/vmx/vmx.c |  2 ++
 3 files changed, 31 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index e7a0660..01d3a87 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1545,13 +1545,19 @@ void wait_lapic_expire(struct kvm_vcpu *vcpu)
 
 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
 	apic->lapic_timer.expired_tscdeadline = 0;
-	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
+	guest_tsc = kvm_read_l1_tsc(vcpu, (apic->lapic_timer.measure_delay_done == 2) ?
+		rdtsc() + apic->lapic_timer.vmentry_delay : rdtsc());
 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
 
 	if (guest_tsc < tsc_deadline)
 		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
 
 	adaptive_tune_timer_advancement(vcpu, guest_tsc, tsc_deadline);
+
+	if (!apic->lapic_timer.measure_delay_done) {
+		apic->lapic_timer.measure_delay_done = 1;
+		apic->lapic_timer.vmentry_delay = rdtsc();
+	}
 }
 
 static void start_sw_tscdeadline(struct kvm_lapic *apic)
@@ -1837,6 +1843,18 @@ static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
 	}
 }
 
+void kvm_lapic_measure_vmentry_delay(struct kvm_vcpu *vcpu)
+{
+	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
+
+	if (ktimer->measure_delay_done == 1) {
+		ktimer->vmentry_delay = rdtsc() -
+			ktimer->vmentry_delay;
+		ktimer->measure_delay_done = 2;
+	}
+}
+EXPORT_SYMBOL_GPL(kvm_lapic_measure_vmentry_delay);
+
 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
 {
 	int ret = 0;
@@ -2318,7 +2336,8 @@ int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
 		apic->lapic_timer.timer_advance_ns = timer_advance_ns;
 		apic->lapic_timer.timer_advance_adjust_done = true;
 	}
-
+	apic->lapic_timer.vmentry_delay = 0;
+	apic->lapic_timer.measure_delay_done = 0;
 
 	/*
 	 * APIC is created enabled. This will prevent kvm_lapic_set_base from
diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
index d6d049b..f1d037b 100644
--- a/arch/x86/kvm/lapic.h
+++ b/arch/x86/kvm/lapic.h
@@ -35,6 +35,13 @@ struct kvm_timer {
 	atomic_t pending;			/* accumulated triggered timers */
 	bool hv_timer_in_use;
 	bool timer_advance_adjust_done;
+	/**
+	 * 0 unstart measure
+	 * 1 start record
+	 * 2 get delta
+	 */
+	u32 measure_delay_done;
+	u64 vmentry_delay;
 };
 
 struct kvm_lapic {
@@ -230,6 +237,7 @@ void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu);
 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu);
 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu);
 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu);
+void kvm_lapic_measure_vmentry_delay(struct kvm_vcpu *vcpu);
 
 static inline enum lapic_mode kvm_apic_mode(u64 apic_base)
 {
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 9663d41..a939bf5 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -6437,6 +6437,8 @@ static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
 	if (vcpu->arch.cr2 != read_cr2())
 		write_cr2(vcpu->arch.cr2);
 
+	kvm_lapic_measure_vmentry_delay(vcpu);
+
 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
 				   vmx->loaded_vmcs->launched);
 
-- 
2.7.4