[RFC V3 5/6] KVM: X86: add lazy tscdeadline support to reduce vm-exit of msr-write

From: Wang Jianchao
Date: Sun Jul 16 2023 - 22:36:20 EST


This patch adds the main logic of lazy_tscdeadline of host side.
There are 3 operations:
- UPDATE, when the guest update msr of tsc deadline, we need to
update the value of 'armed' field of kvm_lazy_tscdeadline
- KICK, when the hv or sw timer is fired, we need to check the
'pending' field to decide whether to re-arm timer or inject
local timer vector. The sw timer is not in vcpu context, so a
new kvm req is added to handle the kick in vcpu context.
- CLEAR, this is a bit tricky. We need to clear the 'armed' field
properly otherwise the guestOS can be hung.

The scenerios need to do CLEAR:
- convert between period & onshot and tscdeadline
- mask the lapic timer
- tscdeadline value has expired before we arm the timer

Here is the test result of netperf TCP_RR on loopback,
Close Open
--------------------------------------------------------
VM-Exit
sum 12617503 5815737
intr 0% 37023 0% 33002
cpuid 0% 1 0% 0
halt 19% 2503932 47% 2780683
msr-write 79% 10046340 51% 2966824
pause 0% 90 0% 84
ept-violation 0% 584 0% 336
ept-misconfig 0% 0 0% 2
preemption-timer 0% 29518 0% 34800
-------------------------------------------------------
MSR-Write
sum 10046455 2966864
apic-icr 25% 2533498 93% 2781235
tsc-deadline 74% 7512945 6% 185629

The vm-exit caused by writing msr of tsc-deadline is reduced by 70%

Signed-off-by: Li Shujin <arkinjob@xxxxxxxxxxx>
Signed-off-by: Wang Jianchao <jianchwa@xxxxxxxxxxx>
---
arch/x86/include/asm/kvm_host.h | 1 +
arch/x86/kvm/lapic.c | 93 +++++++++++++++++++++++++++++++++++++----
arch/x86/kvm/lapic.h | 3 +-
arch/x86/kvm/x86.c | 3 ++
4 files changed, 90 insertions(+), 10 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index b036874..b217ae7 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -113,6 +113,7 @@
KVM_ARCH_REQ_FLAGS(31, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
#define KVM_REQ_HV_TLB_FLUSH \
KVM_ARCH_REQ_FLAGS(32, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
+#define KVM_REQ_LAZY_TSCDEADLINE KVM_ARCH_REQ(33)

#define CR0_RESERVED_BITS \
(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 71da41e..781516f 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1720,6 +1720,54 @@ void kvm_lazy_tscdeadline_exit(struct kvm_vcpu *vcpu)
vcpu->arch.lazy_tscdeadline.guest = NULL;
}

+static void kvm_lazy_tscdeadline_update(struct kvm_vcpu *vcpu, u64 tsc)
+{
+ struct kvm_host_lazy_tscdeadline *hlt = &vcpu->arch.lazy_tscdeadline;
+
+ if (!(hlt->msr_val & KVM_MSR_ENABLED) ||
+ !hlt->guest)
+ return;
+
+ hlt->guest->armed = tsc;
+ hlt->cached_armed = tsc;
+}
+
+bool kvm_lazy_tscdeadline_kick(struct kvm_vcpu *vcpu)
+{
+ struct kvm_host_lazy_tscdeadline *hlt = &vcpu->arch.lazy_tscdeadline;
+ u64 next;
+ bool ret = false;
+
+ if (!hlt->cached_armed ||
+ !(hlt->msr_val & KVM_MSR_ENABLED) ||
+ !hlt->guest)
+ return ret;
+
+ next = hlt->guest->pending;
+ if (next && next > hlt->guest->armed) {
+ kvm_set_lapic_tscdeadline_msr(vcpu, next);
+ ret = true;
+ } else {
+ hlt->guest->armed = 0;
+ hlt->cached_armed = 0;
+ }
+
+ return ret;
+}
+
+void kvm_lazy_tscdeadline_clear(struct kvm_vcpu *vcpu)
+{
+ struct kvm_host_lazy_tscdeadline *hlt = &vcpu->arch.lazy_tscdeadline;
+
+ if (!hlt->cached_armed ||
+ !(hlt->msr_val & KVM_MSR_ENABLED) ||
+ !hlt->guest)
+ return;
+
+ hlt->guest->armed = 0;
+ hlt->cached_armed = 0;
+}
+
static void update_divide_count(struct kvm_lapic *apic)
{
u32 tmp1, tmp2, tdcr;
@@ -1765,8 +1813,12 @@ static void cancel_apic_timer(struct kvm_lapic *apic)

static void apic_update_lvtt(struct kvm_lapic *apic)
{
- u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
- apic->lapic_timer.timer_mode_mask;
+ u32 reg, timer_mode;
+ bool clear;
+
+ reg = kvm_lapic_get_reg(apic, APIC_LVTT);
+ clear = !!(reg & APIC_LVT_MASKED);
+ timer_mode = reg & apic->lapic_timer.timer_mode_mask;

if (apic->lapic_timer.timer_mode != timer_mode) {
if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
@@ -1775,10 +1827,14 @@ static void apic_update_lvtt(struct kvm_lapic *apic)
kvm_lapic_set_reg(apic, APIC_TMICT, 0);
apic->lapic_timer.period = 0;
apic->lapic_timer.tscdeadline = 0;
+ clear = true;
}
apic->lapic_timer.timer_mode = timer_mode;
limit_periodic_timer_frequency(apic);
}
+
+ if (clear)
+ kvm_lazy_tscdeadline_clear(apic->vcpu);
}

/*
@@ -1966,8 +2022,15 @@ static void start_sw_tscdeadline(struct kvm_lapic *apic)
expire = ktime_add_ns(now, ns);
expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
- } else
+ } else {
apic_timer_expired(apic, false);
+ /*
+ * If the current pending tscdeadline has been expired, we need
+ * to clear the armed_tscddl otherwise guest will skip following
+ * msr wtite and clock event hangs
+ */
+ kvm_lazy_tscdeadline_clear(vcpu);
+ }

local_irq_restore(flags);
}
@@ -2145,6 +2208,9 @@ static bool start_hv_timer(struct kvm_lapic *apic)
}
}

+ if (apic_lvtt_tscdeadline(apic) && expired)
+ kvm_lazy_tscdeadline_clear(vcpu);
+
trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);

return true;
@@ -2189,8 +2255,12 @@ void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
if (!apic->lapic_timer.hv_timer_in_use)
goto out;
WARN_ON(kvm_vcpu_is_blocking(vcpu));
- apic_timer_expired(apic, false);
- cancel_hv_timer(apic);
+
+ if (!apic_lvtt_tscdeadline(apic) ||
+ !kvm_lazy_tscdeadline_kick(vcpu)) {
+ apic_timer_expired(apic, false);
+ cancel_hv_timer(apic);
+ }

if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
advance_periodic_target_expiration(apic);
@@ -2522,6 +2592,7 @@ void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)

hrtimer_cancel(&apic->lapic_timer.timer);
apic->lapic_timer.tscdeadline = data;
+ kvm_lazy_tscdeadline_update(vcpu, data);
start_apic_timer(apic);
}

@@ -2802,15 +2873,19 @@ static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
{
struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
+ enum hrtimer_restart ret = HRTIMER_NORESTART;

apic_timer_expired(apic, true);

- if (lapic_is_periodic(apic)) {
+ if (apic_lvtt_tscdeadline(apic)) {
+ kvm_make_request(KVM_REQ_LAZY_TSCDEADLINE, apic->vcpu);
+ } else if (lapic_is_periodic(apic)) {
advance_periodic_target_expiration(apic);
hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
- return HRTIMER_RESTART;
- } else
- return HRTIMER_NORESTART;
+ ret = HRTIMER_RESTART;
+ }
+
+ return ret;
}

int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
index 51b9d5b..0387a02 100644
--- a/arch/x86/kvm/lapic.h
+++ b/arch/x86/kvm/lapic.h
@@ -280,5 +280,6 @@ static inline u8 kvm_xapic_id(struct kvm_lapic *apic)

int kvm_lazy_tscdeadline_init(struct kvm_vcpu *vcpu);
void kvm_lazy_tscdeadline_exit(struct kvm_vcpu *vcpu);
-
+void kvm_lazy_tscdeadline_clear(struct kvm_vcpu *vcpu);
+bool kvm_lazy_tscdeadline_kick(struct kvm_vcpu *vcpu);
#endif
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 7225fc9..26f0ef3 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3879,6 +3879,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
return 1;

if (!(data & KVM_MSR_ENABLED)) {
+ kvm_lazy_tscdeadline_clear(vcpu);
kvm_lazy_tscdeadline_exit(vcpu);
} else {
kvm_lazy_tscdeadline_exit(vcpu);
@@ -10584,6 +10585,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
}
if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
record_steal_time(vcpu);
+ if (kvm_check_request(KVM_REQ_LAZY_TSCDEADLINE, vcpu))
+ kvm_lazy_tscdeadline_kick(vcpu);
#ifdef CONFIG_KVM_SMM
if (kvm_check_request(KVM_REQ_SMI, vcpu))
process_smi(vcpu);
--
2.7.4