Everytime guest write msr of MSR_IA32_TSC_DEADLINE, a vm-exit occurs
and host side handle it. However, a lot of the vm-exit is unnecessary
because the timer is often over-written before it expires. This patch
add a new pv feature, namely, lazy tscdeadline, as following,
v : write to msr of tsc deadline
| : timer armed by tsc deadline
v v v v v | | | | |
---------------------------------------> Time
The timer armed by msr write is over-written before expires
v v v v v | |
---------------------------------------> Time
'- arm -'
The 1st timer is responsible for arming the next timer.
When an armed timer is expired, it will check pending and arm a new timer.
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 msr-write is reduced by 70% at least
Signed-off-by: Li Shujin <arkinjob@xxxxxxxxxxx>
Signed-off-by: Wang Jianchao <jianchwa@xxxxxxxxxxx>
---
arch/x86/include/asm/kvm_host.h | 12 +++-
arch/x86/kernel/apic/apic.c | 47 ++++++++++++++-
arch/x86/kvm/cpuid.c | 1 +
arch/x86/kvm/lapic.c | 128 +++++++++++++++++++++++++++++++++++++---
arch/x86/kvm/lapic.h | 4 ++
arch/x86/kvm/x86.c | 17 +++++-
6 files changed, 195 insertions(+), 14 deletions(-)
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 6edb1ac..6b20609 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 \
@@ -713,6 +714,13 @@ struct kvm_queued_exception {
bool has_payload;
};
+struct kvm_host_lazy_tscdeadline {
+ u64 msr_val;
+ u64 cached_armed;
+ struct page *page;
+ struct kvm_lazy_tscdeadline *guest;
+};
+
struct kvm_vcpu_arch {
/*
* rip and regs accesses must go through
@@ -944,9 +952,7 @@ struct kvm_vcpu_arch {
struct gfn_to_hva_cache data;
} pv_eoi;
- struct {
- u64 msr_val;
- } lazy_tscdeadline;
+ struct kvm_host_lazy_tscdeadline lazy_tscdeadline;
u64 msr_kvm_poll_control;
diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
index 7705571..0fe1215 100644
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -62,7 +62,10 @@
#include <asm/intel-family.h>
#include <asm/irq_regs.h>
#include <asm/cpu.h>
+#include <linux/kvm_para.h>
+DECLARE_PER_CPU_DECRYPTED(struct kvm_lazy_tscdeadline, kvm_lazy_tscdeadline);
+
unsigned int num_processors;
unsigned disabled_cpus;
@@ -492,6 +495,43 @@ static int lapic_next_deadline(unsigned long delta,
return 0;
}
+static int kvm_lapic_next_deadline(unsigned long delta,
+ struct clock_event_device *evt)
+{
+ struct kvm_lazy_tscdeadline *lazy_tscddl = this_cpu_ptr(&kvm_lazy_tscdeadline);
+ u64 tsc;
+
+ /* This MSR is special and need a special fence: */
+ weak_wrmsr_fence();
+
+ tsc = rdtsc() + (((u64) delta) * TSC_DIVISOR);
+
+ /*
+ * v : write to msr of tsc deadline
+ * | : timer armed by tsc deadline
+ *
+ * v v v v v | | | | |
+ * ---------------------------------------> Time
+ *
+ * The timer armed by msr write is over-written before expires
+ *
+ * v v v v v | |
+ * ---------------------------------------> Time
+ * '- arm -'
+ *
+ * The 1st timer is responsible for arming the next timer.
+ * - armed, the time has been armed on host side by msr-write
+ * - pending, the next expires time set by guest side
+ * When timer of 'armed' is expired, it will check pending and
+ * arm a new timer.
+ */
+ lazy_tscddl->pending = tsc;
+ if (!lazy_tscddl->armed || tsc < lazy_tscddl->armed)
+ wrmsrl(MSR_IA32_TSC_DEADLINE, tsc);
+
+ return 0;
+}
+
static int lapic_timer_shutdown(struct clock_event_device *evt)
{
unsigned int v;
@@ -636,7 +676,12 @@ static void setup_APIC_timer(void)
levt->name = "lapic-deadline";
levt->features &= ~(CLOCK_EVT_FEAT_PERIODIC |
CLOCK_EVT_FEAT_DUMMY);
- levt->set_next_event = lapic_next_deadline;
+ if (kvm_para_available() &&
+ kvm_para_has_feature(KVM_FEATURE_LAZY_TSCDEADLINE)) {
+ levt->set_next_event = kvm_lapic_next_deadline;
+ } else {
+ levt->set_next_event = lapic_next_deadline;
+ }
clockevents_config_and_register(levt,
tsc_khz * (1000 / TSC_DIVISOR),
0xF, ~0UL);
diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
index 0c9660a..5a12601 100644
--- a/arch/x86/kvm/cpuid.c
+++ b/arch/x86/kvm/cpuid.c
@@ -1118,6 +1118,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
(1 << KVM_FEATURE_PV_SEND_IPI) |
(1 << KVM_FEATURE_POLL_CONTROL) |
(1 << KVM_FEATURE_PV_SCHED_YIELD) |
+ (1 << KVM_FEATURE_LAZY_TSCDEADLINE) |
(1 << KVM_FEATURE_ASYNC_PF_INT);
if (sched_info_on())
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 3c300a1..9d137c8 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1681,6 +1681,94 @@ static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
return 0;
}
+int kvm_lazy_tscdeadline_init(struct kvm_vcpu *vcpu)
+{
+ struct kvm_memory_slot *slot;
+ struct kvm_memslots *slots;
+ struct page *page;
+ kvm_pfn_t pfn;
+ gfn_t gfn;
+ void *hva;
+ u64 msr;
+
+ slots = kvm_memslots(vcpu->kvm);
+ msr = vcpu->arch.lazy_tscdeadline.msr_val;
+ gfn = msr >> PAGE_SHIFT;
+ slot = __gfn_to_memslot(slots, gfn);
+ pfn = gfn_to_pfn_memslot(slot, gfn);
+
+ if (!pfn_valid(pfn))
+ return -EINVAL;
+
+ page = pfn_to_page(pfn);
+ hva = kmap(page);
+ if (!hva)
+ return -EFAULT;
+
+ hva += offset_in_page(msr - KVM_MSR_ENABLED);
+ get_page(page);
+ vcpu->arch.lazy_tscdeadline.page = page;
+ vcpu->arch.lazy_tscdeadline.guest = hva;
+
+ return 0;
+}
+
+void kvm_lazy_tscdeadline_exit(struct kvm_vcpu *vcpu)
+{
+ if (vcpu->arch.lazy_tscdeadline.page) {
+ kunmap((void *)vcpu->arch.lazy_tscdeadline.guest);
+ put_page(vcpu->arch.lazy_tscdeadline.page);
+ }
+
+ vcpu->arch.lazy_tscdeadline.cached_armed = 0;
+ vcpu->arch.lazy_tscdeadline.page = NULL;
+ vcpu->arch.lazy_tscdeadline.guest = NULL;
+}
+
+static void kvm_lazy_tscdeadline_update(struct kvm_vcpu *vcpu, u64 tsc)
+{
+ struct kvm_host_lazy_tscdeadline *host = &vcpu->arch.lazy_tscdeadline;
+
+ if (!(host->msr_val & KVM_MSR_ENABLED) ||
+ !host->guest)
+ return;
+
+ host->guest->armed = tsc;
+ host->cached_armed = tsc;
+}
+
+void kvm_lazy_tscdeadline_kick(struct kvm_vcpu *vcpu)
+{
+ struct kvm_host_lazy_tscdeadline *host = &vcpu->arch.lazy_tscdeadline;
+ u64 next;
+
+ if (!host->cached_armed ||
+ !(host->msr_val & KVM_MSR_ENABLED) ||
+ !host->guest)
+ return;
+
+ next = host->guest->pending;
+ if (next && next > host->guest->armed) {
+ kvm_set_lapic_tscdeadline_msr(vcpu, next);
+ } else {
+ host->guest->armed = 0;
+ host->cached_armed = 0;
+ }
+}
+
+void kvm_lazy_tscdeadline_clear(struct kvm_vcpu *vcpu)
+{
+ struct kvm_host_lazy_tscdeadline *host = &vcpu->arch.lazy_tscdeadline;
+
+ if (!host->cached_armed ||
+ !(host->msr_val & KVM_MSR_ENABLED) ||
+ !host->guest)
+ return;
+
+ host->guest->armed = 0;
+ host->cached_armed = 0;
+}
+
static void update_divide_count(struct kvm_lapic *apic)
{
u32 tmp1, tmp2, tdcr;
@@ -1726,8 +1814,12 @@ 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 ==
@@ -1736,10 +1828,14 @@ 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);
}
/*
@@ -1927,8 +2023,15 @@ 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);
}
@@ -2100,9 +2203,11 @@ static bool start_hv_timer(struct kvm_lapic *apic)
*/
if (atomic_read(&ktimer->pending)) {
cancel_hv_timer(apic);
+ kvm_lazy_tscdeadline_clear(vcpu);
} else if (expired) {
apic_timer_expired(apic, false);
cancel_hv_timer(apic);
+ kvm_lazy_tscdeadline_clear(vcpu);
}
}
@@ -2153,7 +2258,9 @@ void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
apic_timer_expired(apic, false);
cancel_hv_timer(apic);
- if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
+ if (apic_lvtt_tscdeadline(apic)) {
+ kvm_lazy_tscdeadline_kick(vcpu);
+ } else if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
advance_periodic_target_expiration(apic);
restart_apic_timer(apic);
}
@@ -2483,6 +2590,7 @@ 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);
}
@@ -2762,15 +2870,19 @@ 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 0a0ea4b..30a8cf1 100644
--- a/arch/x86/kvm/lapic.h
+++ b/arch/x86/kvm/lapic.h
@@ -278,4 +278,8 @@ static inline u8 kvm_xapic_id(struct kvm_lapic *apic)
return kvm_lapic_get_reg(apic, APIC_ID) >> 24;
}
+void kvm_lazy_tscdeadline_clear(struct kvm_vcpu *vcpu);
+void kvm_lazy_tscdeadline_kick(struct kvm_vcpu *vcpu);
+int kvm_lazy_tscdeadline_init(struct kvm_vcpu *vcpu);
+void kvm_lazy_tscdeadline_exit(struct kvm_vcpu *vcpu);
#endif
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 15c265a..8270bd5 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3873,8 +3873,17 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
case MSR_KVM_LAZY_TSCDEADLINE:
if (!guest_pv_has(vcpu, KVM_FEATURE_LAZY_TSCDEADLINE))
return 1;
-
- vcpu->arch.lazy_tscdeadline.msr_val = data;
+
+ if (!(data & KVM_MSR_ENABLED)) {
+ kvm_lazy_tscdeadline_clear(vcpu);
+ kvm_lazy_tscdeadline_exit(vcpu);
+ } else {
+ vcpu->arch.lazy_tscdeadline.msr_val = data;
+ if (kvm_lazy_tscdeadline_init(vcpu)) {
+ vcpu->arch.lazy_tscdeadline.msr_val = 0;
+ return 1;
+ }
+ }
break;
case MSR_IA32_MCG_CTL:
@@ -10576,6 +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);
@@ -11965,6 +11976,8 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
kvmclock_reset(vcpu);
+ kvm_lazy_tscdeadline_exit(vcpu);
+
static_call(kvm_x86_vcpu_free)(vcpu);
kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
--
2.7.4
