2020-02-18 10:50:18

by Wanpeng Li

Subject: [PATCH] KVM: LAPIC: Recalculate apic map in batch

From: Wanpeng Li <[email protected]>

In the vCPU reset and set APIC_BASE MSR path, the apic map will be recalculated
several times, each time it will consume 10+ us observed by ftrace in my
non-overcommit environment since the expensive memory allocate/mutex/rcu etc
operations. This patch optimizes it by recaluating apic map in batch, I hope
this can benefit the serverless scenario which can frequently create/destroy
VMs.

Signed-off-by: Wanpeng Li <[email protected]>
---
arch/x86/kvm/lapic.c | 74 ++++++++++++++++++++++++++++++++++------------------
arch/x86/kvm/lapic.h | 3 ++-
arch/x86/kvm/x86.c | 3 ++-
3 files changed, 53 insertions(+), 27 deletions(-)

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index afcd30d..71f843a 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -164,7 +164,7 @@ static void kvm_apic_map_free(struct rcu_head *rcu)
kvfree(map);
}

-static void recalculate_apic_map(struct kvm *kvm)
+void kvm_recalculate_apic_map(struct kvm *kvm)
{
struct kvm_apic_map *new, *old = NULL;
struct kvm_vcpu *vcpu;
@@ -244,9 +244,10 @@ static void recalculate_apic_map(struct kvm *kvm)
kvm_make_scan_ioapic_request(kvm);
}

-static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
+static inline bool apic_set_spiv(struct kvm_lapic *apic, u32 val)
{
bool enabled = val & APIC_SPIV_APIC_ENABLED;
+ bool need_recal = false;

kvm_lapic_set_reg(apic, APIC_SPIV, val);

@@ -257,20 +258,20 @@ static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
else
static_key_slow_inc(&apic_sw_disabled.key);

- recalculate_apic_map(apic->vcpu->kvm);
+ need_recal = true;
}
+
+ return need_recal;
}

static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
{
kvm_lapic_set_reg(apic, APIC_ID, id << 24);
- recalculate_apic_map(apic->vcpu->kvm);
}

static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
{
kvm_lapic_set_reg(apic, APIC_LDR, id);
- recalculate_apic_map(apic->vcpu->kvm);
}

static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
@@ -286,7 +287,6 @@ static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)

kvm_lapic_set_reg(apic, APIC_ID, id);
kvm_lapic_set_reg(apic, APIC_LDR, ldr);
- recalculate_apic_map(apic->vcpu->kvm);
}

static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
@@ -1882,14 +1882,16 @@ static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
{
int ret = 0;
+ bool need_recal = false;

trace_kvm_apic_write(reg, val);

switch (reg) {
case APIC_ID: /* Local APIC ID */
- if (!apic_x2apic_mode(apic))
+ if (!apic_x2apic_mode(apic)) {
kvm_apic_set_xapic_id(apic, val >> 24);
- else
+ need_recal = true;
+ } else
ret = 1;
break;

@@ -1903,16 +1905,17 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
break;

case APIC_LDR:
- if (!apic_x2apic_mode(apic))
+ if (!apic_x2apic_mode(apic)) {
kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
- else
+ need_recal = true;
+ } else
ret = 1;
break;

case APIC_DFR:
if (!apic_x2apic_mode(apic)) {
kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
- recalculate_apic_map(apic->vcpu->kvm);
+ need_recal = true;
} else
ret = 1;
break;
@@ -1921,7 +1924,8 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
u32 mask = 0x3ff;
if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
mask |= APIC_SPIV_DIRECTED_EOI;
- apic_set_spiv(apic, val & mask);
+ if (apic_set_spiv(apic, val & mask))
+ need_recal = true;
if (!(val & APIC_SPIV_APIC_ENABLED)) {
int i;
u32 lvt_val;
@@ -2018,6 +2022,9 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
break;
}

+ if (need_recal)
+ kvm_recalculate_apic_map(apic->vcpu->kvm);
+
return ret;
}
EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
@@ -2143,10 +2150,11 @@ u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
return (tpr & 0xf0) >> 4;
}

-void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
+bool kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
{
u64 old_value = vcpu->arch.apic_base;
struct kvm_lapic *apic = vcpu->arch.apic;
+ bool need_recal = false;

if (!apic)
value |= MSR_IA32_APICBASE_BSP;
@@ -2157,7 +2165,7 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
kvm_update_cpuid(vcpu);

if (!apic)
- return;
+ return need_recal;

/* update jump label if enable bit changes */
if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
@@ -2166,12 +2174,14 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
static_key_slow_dec_deferred(&apic_hw_disabled);
} else {
static_key_slow_inc(&apic_hw_disabled.key);
- recalculate_apic_map(vcpu->kvm);
}
+ need_recal = true;
}

- if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
+ if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE)) {
kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
+ need_recal = true;
+ }

if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
kvm_x86_ops->set_virtual_apic_mode(vcpu);
@@ -2182,6 +2192,8 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
if ((value & MSR_IA32_APICBASE_ENABLE) &&
apic->base_address != APIC_DEFAULT_PHYS_BASE)
pr_warn_once("APIC base relocation is unsupported by KVM");
+
+ return need_recal;
}

void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
@@ -2203,6 +2215,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
{
struct kvm_lapic *apic = vcpu->arch.apic;
int i;
+ bool need_recal = false;

if (!apic)
return;
@@ -2214,6 +2227,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
MSR_IA32_APICBASE_ENABLE);
kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
+ need_recal = true;
}
kvm_apic_set_version(apic->vcpu);

@@ -2227,10 +2241,13 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));

kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
- apic_set_spiv(apic, 0xff);
+ if (apic_set_spiv(apic, 0xff))
+ need_recal = true;
kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
- if (!apic_x2apic_mode(apic))
+ if (!apic_x2apic_mode(apic)) {
kvm_apic_set_ldr(apic, 0);
+ need_recal = true;
+ }
kvm_lapic_set_reg(apic, APIC_ESR, 0);
kvm_lapic_set_reg(apic, APIC_ICR, 0);
kvm_lapic_set_reg(apic, APIC_ICR2, 0);
@@ -2246,8 +2263,8 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
update_divide_count(apic);
atomic_set(&apic->lapic_timer.pending, 0);
if (kvm_vcpu_is_bsp(vcpu))
- kvm_lapic_set_base(vcpu,
- vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
+ need_recal = kvm_lapic_set_base(vcpu,
+ vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
vcpu->arch.pv_eoi.msr_val = 0;
apic_update_ppr(apic);
if (vcpu->arch.apicv_active) {
@@ -2258,6 +2275,9 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)

vcpu->arch.apic_arb_prio = 0;
vcpu->arch.apic_attention = 0;
+
+ if (need_recal)
+ kvm_recalculate_apic_map(vcpu->kvm);
}

/*
@@ -2478,18 +2498,22 @@ int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
{
struct kvm_lapic *apic = vcpu->arch.apic;
int r;
+ bool need_recal = false;

-
- kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
+ need_recal = kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
/* set SPIV separately to get count of SW disabled APICs right */
- apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
+ if (apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV))))
+ need_recal = true;

r = kvm_apic_state_fixup(vcpu, s, true);
- if (r)
+ if (r) {
+ if (need_recal)
+ kvm_recalculate_apic_map(vcpu->kvm);
return r;
+ }
memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));

- recalculate_apic_map(vcpu->kvm);
+ kvm_recalculate_apic_map(vcpu->kvm);
kvm_apic_set_version(vcpu);

apic_update_ppr(apic);
diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
index ec6fbfe..58c07ad 100644
--- a/arch/x86/kvm/lapic.h
+++ b/arch/x86/kvm/lapic.h
@@ -76,8 +76,9 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event);
u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu);
void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8);
void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu);
-void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
+bool kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu);
+void kvm_recalculate_apic_map(struct kvm *kvm);
void kvm_apic_set_version(struct kvm_vcpu *vcpu);
int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val);
int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 79bc995..e961e65 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -349,7 +349,8 @@ int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
return 1;
}

- kvm_lapic_set_base(vcpu, msr_info->data);
+ if (kvm_lapic_set_base(vcpu, msr_info->data))
+ kvm_recalculate_apic_map(vcpu->kvm);
return 0;
}
EXPORT_SYMBOL_GPL(kvm_set_apic_base);
--
2.7.4

2020-02-18 12:25:18

by Vitaly Kuznetsov

Subject: Re: [PATCH] KVM: LAPIC: Recalculate apic map in batch

Wanpeng Li <[email protected]> writes:

> From: Wanpeng Li <[email protected]>
>
> In the vCPU reset and set APIC_BASE MSR path, the apic map will be recalculated
> several times, each time it will consume 10+ us observed by ftrace in my
> non-overcommit environment since the expensive memory allocate/mutex/rcu etc
> operations. This patch optimizes it by recaluating apic map in batch, I hope
> this can benefit the serverless scenario which can frequently create/destroy
> VMs.
>
> Signed-off-by: Wanpeng Li <[email protected]>

An alternative idea: instead of making every caller return bool and
every call site handle the result (once) just add a
KVM_REQ_APIC_MAP_RECALC flag or a boolean flag to struct kvm. I
understand it may not be that easy as it sounds as we may be conunting
on valid mapping somewhere before we actually get to handiling
KVM_REQ_APIC_MAP_RECALC but we may preserve *some*
recalculate_apic_map() calls (and make it reset KVM_REQ_APIC_MAP_RECALC).

> ---
> arch/x86/kvm/lapic.c | 74 ++++++++++++++++++++++++++++++++++------------------
> arch/x86/kvm/lapic.h | 3 ++-
> arch/x86/kvm/x86.c | 3 ++-
> 3 files changed, 53 insertions(+), 27 deletions(-)
>
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index afcd30d..71f843a 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -164,7 +164,7 @@ static void kvm_apic_map_free(struct rcu_head *rcu)
> kvfree(map);
> }
>
> -static void recalculate_apic_map(struct kvm *kvm)
> +void kvm_recalculate_apic_map(struct kvm *kvm)
> {
> struct kvm_apic_map *new, *old = NULL;
> struct kvm_vcpu *vcpu;
> @@ -244,9 +244,10 @@ static void recalculate_apic_map(struct kvm *kvm)
> kvm_make_scan_ioapic_request(kvm);
> }
>
> -static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
> +static inline bool apic_set_spiv(struct kvm_lapic *apic, u32 val)
> {
> bool enabled = val & APIC_SPIV_APIC_ENABLED;
> + bool need_recal = false;
>
> kvm_lapic_set_reg(apic, APIC_SPIV, val);
>
> @@ -257,20 +258,20 @@ static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
> else
> static_key_slow_inc(&apic_sw_disabled.key);
>
> - recalculate_apic_map(apic->vcpu->kvm);
> + need_recal = true;
> }
> +
> + return need_recal;
> }
>
> static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
> {
> kvm_lapic_set_reg(apic, APIC_ID, id << 24);
> - recalculate_apic_map(apic->vcpu->kvm);
> }
>
> static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
> {
> kvm_lapic_set_reg(apic, APIC_LDR, id);
> - recalculate_apic_map(apic->vcpu->kvm);
> }
>
> static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
> @@ -286,7 +287,6 @@ static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
>
> kvm_lapic_set_reg(apic, APIC_ID, id);
> kvm_lapic_set_reg(apic, APIC_LDR, ldr);
> - recalculate_apic_map(apic->vcpu->kvm);
> }
>
> static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
> @@ -1882,14 +1882,16 @@ static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
> int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> {
> int ret = 0;
> + bool need_recal = false;
>
> trace_kvm_apic_write(reg, val);
>
> switch (reg) {
> case APIC_ID: /* Local APIC ID */
> - if (!apic_x2apic_mode(apic))
> + if (!apic_x2apic_mode(apic)) {
> kvm_apic_set_xapic_id(apic, val >> 24);
> - else
> + need_recal = true;
> + } else
> ret = 1;
> break;
>
> @@ -1903,16 +1905,17 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> break;
>
> case APIC_LDR:
> - if (!apic_x2apic_mode(apic))
> + if (!apic_x2apic_mode(apic)) {
> kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
> - else
> + need_recal = true;
> + } else
> ret = 1;
> break;
>
> case APIC_DFR:
> if (!apic_x2apic_mode(apic)) {
> kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
> - recalculate_apic_map(apic->vcpu->kvm);
> + need_recal = true;
> } else
> ret = 1;
> break;
> @@ -1921,7 +1924,8 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> u32 mask = 0x3ff;
> if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
> mask |= APIC_SPIV_DIRECTED_EOI;
> - apic_set_spiv(apic, val & mask);
> + if (apic_set_spiv(apic, val & mask))
> + need_recal = true;
> if (!(val & APIC_SPIV_APIC_ENABLED)) {
> int i;
> u32 lvt_val;
> @@ -2018,6 +2022,9 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> break;
> }
>
> + if (need_recal)
> + kvm_recalculate_apic_map(apic->vcpu->kvm);
> +
> return ret;
> }
> EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
> @@ -2143,10 +2150,11 @@ u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
> return (tpr & 0xf0) >> 4;
> }
>
> -void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> +bool kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> {
> u64 old_value = vcpu->arch.apic_base;
> struct kvm_lapic *apic = vcpu->arch.apic;
> + bool need_recal = false;
>
> if (!apic)
> value |= MSR_IA32_APICBASE_BSP;
> @@ -2157,7 +2165,7 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> kvm_update_cpuid(vcpu);
>
> if (!apic)
> - return;
> + return need_recal;

Just 'return false;' maybe?

>
> /* update jump label if enable bit changes */
> if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
> @@ -2166,12 +2174,14 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> static_key_slow_dec_deferred(&apic_hw_disabled);
> } else {
> static_key_slow_inc(&apic_hw_disabled.key);
> - recalculate_apic_map(vcpu->kvm);
> }
> + need_recal = true;
> }
>
> - if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
> + if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE)) {
> kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
> + need_recal = true;
> + }
>
> if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
> kvm_x86_ops->set_virtual_apic_mode(vcpu);
> @@ -2182,6 +2192,8 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> if ((value & MSR_IA32_APICBASE_ENABLE) &&
> apic->base_address != APIC_DEFAULT_PHYS_BASE)
> pr_warn_once("APIC base relocation is unsupported by KVM");
> +
> + return need_recal;
> }
>
> void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
> @@ -2203,6 +2215,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> {
> struct kvm_lapic *apic = vcpu->arch.apic;
> int i;
> + bool need_recal = false;
>
> if (!apic)
> return;
> @@ -2214,6 +2227,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
> MSR_IA32_APICBASE_ENABLE);
> kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
> + need_recal = true;
> }
> kvm_apic_set_version(apic->vcpu);
>
> @@ -2227,10 +2241,13 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
>
> kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
> - apic_set_spiv(apic, 0xff);
> + if (apic_set_spiv(apic, 0xff))
> + need_recal = true;
> kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
> - if (!apic_x2apic_mode(apic))
> + if (!apic_x2apic_mode(apic)) {
> kvm_apic_set_ldr(apic, 0);
> + need_recal = true;
> + }
> kvm_lapic_set_reg(apic, APIC_ESR, 0);
> kvm_lapic_set_reg(apic, APIC_ICR, 0);
> kvm_lapic_set_reg(apic, APIC_ICR2, 0);
> @@ -2246,8 +2263,8 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> update_divide_count(apic);
> atomic_set(&apic->lapic_timer.pending, 0);
> if (kvm_vcpu_is_bsp(vcpu))
> - kvm_lapic_set_base(vcpu,
> - vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
> + need_recal = kvm_lapic_set_base(vcpu,
> + vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
> vcpu->arch.pv_eoi.msr_val = 0;
> apic_update_ppr(apic);
> if (vcpu->arch.apicv_active) {
> @@ -2258,6 +2275,9 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
>
> vcpu->arch.apic_arb_prio = 0;
> vcpu->arch.apic_attention = 0;
> +
> + if (need_recal)
> + kvm_recalculate_apic_map(vcpu->kvm);
> }
>
> /*
> @@ -2478,18 +2498,22 @@ int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
> {
> struct kvm_lapic *apic = vcpu->arch.apic;
> int r;
> + bool need_recal = false;
>
> -
> - kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
> + need_recal = kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
> /* set SPIV separately to get count of SW disabled APICs right */
> - apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
> + if (apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV))))
> + need_recal = true;
>
> r = kvm_apic_state_fixup(vcpu, s, true);
> - if (r)
> + if (r) {
> + if (need_recal)
> + kvm_recalculate_apic_map(vcpu->kvm);

This is an error condition, I'd suggest to drop need_recal check here
and do it unconditionally.

> return r;
> + }
> memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
>
> - recalculate_apic_map(vcpu->kvm);
> + kvm_recalculate_apic_map(vcpu->kvm);
> kvm_apic_set_version(vcpu);
>
> apic_update_ppr(apic);
> diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
> index ec6fbfe..58c07ad 100644
> --- a/arch/x86/kvm/lapic.h
> +++ b/arch/x86/kvm/lapic.h
> @@ -76,8 +76,9 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event);
> u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu);
> void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8);
> void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu);
> -void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
> +bool kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
> u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu);
> +void kvm_recalculate_apic_map(struct kvm *kvm);
> void kvm_apic_set_version(struct kvm_vcpu *vcpu);
> int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val);
> int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 79bc995..e961e65 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -349,7 +349,8 @@ int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> return 1;
> }
>
> - kvm_lapic_set_base(vcpu, msr_info->data);
> + if (kvm_lapic_set_base(vcpu, msr_info->data))
> + kvm_recalculate_apic_map(vcpu->kvm);
> return 0;
> }
> EXPORT_SYMBOL_GPL(kvm_set_apic_base);

--
Vitaly

2020-02-19 00:49:38

by Wanpeng Li

Subject: Re: [PATCH] KVM: LAPIC: Recalculate apic map in batch

On Tue, 18 Feb 2020 at 20:24, Vitaly Kuznetsov <[email protected]> wrote:
>
> Wanpeng Li <[email protected]> writes:
>
> > From: Wanpeng Li <[email protected]>
> >
> > In the vCPU reset and set APIC_BASE MSR path, the apic map will be recalculated
> > several times, each time it will consume 10+ us observed by ftrace in my
> > non-overcommit environment since the expensive memory allocate/mutex/rcu etc
> > operations. This patch optimizes it by recaluating apic map in batch, I hope
> > this can benefit the serverless scenario which can frequently create/destroy
> > VMs.
> >
> > Signed-off-by: Wanpeng Li <[email protected]>
>
> An alternative idea: instead of making every caller return bool and
> every call site handle the result (once) just add a
> KVM_REQ_APIC_MAP_RECALC flag or a boolean flag to struct kvm. I
> understand it may not be that easy as it sounds as we may be conunting
> on valid mapping somewhere before we actually get to handiling

Yes.

> KVM_REQ_APIC_MAP_RECALC but we may preserve *some*
> recalculate_apic_map() calls (and make it reset KVM_REQ_APIC_MAP_RECALC).

Paolo, keep the caller return bool or add a booleen flag to struct
kvm, what do you think?

Wanpeng

2020-02-25 07:56:24

by Wanpeng Li

Subject: Re: [PATCH] KVM: LAPIC: Recalculate apic map in batch

ping,
On Tue, 18 Feb 2020 at 18:49, Wanpeng Li <[email protected]> wrote:
>
> From: Wanpeng Li <[email protected]>
>
> In the vCPU reset and set APIC_BASE MSR path, the apic map will be recalculated
> several times, each time it will consume 10+ us observed by ftrace in my
> non-overcommit environment since the expensive memory allocate/mutex/rcu etc
> operations. This patch optimizes it by recaluating apic map in batch, I hope
> this can benefit the serverless scenario which can frequently create/destroy
> VMs.
>
> Signed-off-by: Wanpeng Li <[email protected]>
> ---
> arch/x86/kvm/lapic.c | 74 ++++++++++++++++++++++++++++++++++------------------
> arch/x86/kvm/lapic.h | 3 ++-
> arch/x86/kvm/x86.c | 3 ++-
> 3 files changed, 53 insertions(+), 27 deletions(-)
>
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index afcd30d..71f843a 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -164,7 +164,7 @@ static void kvm_apic_map_free(struct rcu_head *rcu)
> kvfree(map);
> }
>
> -static void recalculate_apic_map(struct kvm *kvm)
> +void kvm_recalculate_apic_map(struct kvm *kvm)
> {
> struct kvm_apic_map *new, *old = NULL;
> struct kvm_vcpu *vcpu;
> @@ -244,9 +244,10 @@ static void recalculate_apic_map(struct kvm *kvm)
> kvm_make_scan_ioapic_request(kvm);
> }
>
> -static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
> +static inline bool apic_set_spiv(struct kvm_lapic *apic, u32 val)
> {
> bool enabled = val & APIC_SPIV_APIC_ENABLED;
> + bool need_recal = false;
>
> kvm_lapic_set_reg(apic, APIC_SPIV, val);
>
> @@ -257,20 +258,20 @@ static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
> else
> static_key_slow_inc(&apic_sw_disabled.key);
>
> - recalculate_apic_map(apic->vcpu->kvm);
> + need_recal = true;
> }
> +
> + return need_recal;
> }
>
> static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
> {
> kvm_lapic_set_reg(apic, APIC_ID, id << 24);
> - recalculate_apic_map(apic->vcpu->kvm);
> }
>
> static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
> {
> kvm_lapic_set_reg(apic, APIC_LDR, id);
> - recalculate_apic_map(apic->vcpu->kvm);
> }
>
> static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
> @@ -286,7 +287,6 @@ static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
>
> kvm_lapic_set_reg(apic, APIC_ID, id);
> kvm_lapic_set_reg(apic, APIC_LDR, ldr);
> - recalculate_apic_map(apic->vcpu->kvm);
> }
>
> static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
> @@ -1882,14 +1882,16 @@ static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
> int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> {
> int ret = 0;
> + bool need_recal = false;
>
> trace_kvm_apic_write(reg, val);
>
> switch (reg) {
> case APIC_ID: /* Local APIC ID */
> - if (!apic_x2apic_mode(apic))
> + if (!apic_x2apic_mode(apic)) {
> kvm_apic_set_xapic_id(apic, val >> 24);
> - else
> + need_recal = true;
> + } else
> ret = 1;
> break;
>
> @@ -1903,16 +1905,17 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> break;
>
> case APIC_LDR:
> - if (!apic_x2apic_mode(apic))
> + if (!apic_x2apic_mode(apic)) {
> kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
> - else
> + need_recal = true;
> + } else
> ret = 1;
> break;
>
> case APIC_DFR:
> if (!apic_x2apic_mode(apic)) {
> kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
> - recalculate_apic_map(apic->vcpu->kvm);
> + need_recal = true;
> } else
> ret = 1;
> break;
> @@ -1921,7 +1924,8 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> u32 mask = 0x3ff;
> if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
> mask |= APIC_SPIV_DIRECTED_EOI;
> - apic_set_spiv(apic, val & mask);
> + if (apic_set_spiv(apic, val & mask))
> + need_recal = true;
> if (!(val & APIC_SPIV_APIC_ENABLED)) {
> int i;
> u32 lvt_val;
> @@ -2018,6 +2022,9 @@ int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
> break;
> }
>
> + if (need_recal)
> + kvm_recalculate_apic_map(apic->vcpu->kvm);
> +
> return ret;
> }
> EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
> @@ -2143,10 +2150,11 @@ u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
> return (tpr & 0xf0) >> 4;
> }
>
> -void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> +bool kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> {
> u64 old_value = vcpu->arch.apic_base;
> struct kvm_lapic *apic = vcpu->arch.apic;
> + bool need_recal = false;
>
> if (!apic)
> value |= MSR_IA32_APICBASE_BSP;
> @@ -2157,7 +2165,7 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> kvm_update_cpuid(vcpu);
>
> if (!apic)
> - return;
> + return need_recal;
>
> /* update jump label if enable bit changes */
> if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
> @@ -2166,12 +2174,14 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> static_key_slow_dec_deferred(&apic_hw_disabled);
> } else {
> static_key_slow_inc(&apic_hw_disabled.key);
> - recalculate_apic_map(vcpu->kvm);
> }
> + need_recal = true;
> }
>
> - if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
> + if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE)) {
> kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
> + need_recal = true;
> + }
>
> if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
> kvm_x86_ops->set_virtual_apic_mode(vcpu);
> @@ -2182,6 +2192,8 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
> if ((value & MSR_IA32_APICBASE_ENABLE) &&
> apic->base_address != APIC_DEFAULT_PHYS_BASE)
> pr_warn_once("APIC base relocation is unsupported by KVM");
> +
> + return need_recal;
> }
>
> void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
> @@ -2203,6 +2215,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> {
> struct kvm_lapic *apic = vcpu->arch.apic;
> int i;
> + bool need_recal = false;
>
> if (!apic)
> return;
> @@ -2214,6 +2227,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
> MSR_IA32_APICBASE_ENABLE);
> kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
> + need_recal = true;
> }
> kvm_apic_set_version(apic->vcpu);
>
> @@ -2227,10 +2241,13 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
>
> kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
> - apic_set_spiv(apic, 0xff);
> + if (apic_set_spiv(apic, 0xff))
> + need_recal = true;
> kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
> - if (!apic_x2apic_mode(apic))
> + if (!apic_x2apic_mode(apic)) {
> kvm_apic_set_ldr(apic, 0);
> + need_recal = true;
> + }
> kvm_lapic_set_reg(apic, APIC_ESR, 0);
> kvm_lapic_set_reg(apic, APIC_ICR, 0);
> kvm_lapic_set_reg(apic, APIC_ICR2, 0);
> @@ -2246,8 +2263,8 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
> update_divide_count(apic);
> atomic_set(&apic->lapic_timer.pending, 0);
> if (kvm_vcpu_is_bsp(vcpu))
> - kvm_lapic_set_base(vcpu,
> - vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
> + need_recal = kvm_lapic_set_base(vcpu,
> + vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
> vcpu->arch.pv_eoi.msr_val = 0;
> apic_update_ppr(apic);
> if (vcpu->arch.apicv_active) {
> @@ -2258,6 +2275,9 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
>
> vcpu->arch.apic_arb_prio = 0;
> vcpu->arch.apic_attention = 0;
> +
> + if (need_recal)
> + kvm_recalculate_apic_map(vcpu->kvm);
> }
>
> /*
> @@ -2478,18 +2498,22 @@ int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
> {
> struct kvm_lapic *apic = vcpu->arch.apic;
> int r;
> + bool need_recal = false;
>
> -
> - kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
> + need_recal = kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
> /* set SPIV separately to get count of SW disabled APICs right */
> - apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
> + if (apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV))))
> + need_recal = true;
>
> r = kvm_apic_state_fixup(vcpu, s, true);
> - if (r)
> + if (r) {
> + if (need_recal)
> + kvm_recalculate_apic_map(vcpu->kvm);
> return r;
> + }
> memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
>
> - recalculate_apic_map(vcpu->kvm);
> + kvm_recalculate_apic_map(vcpu->kvm);
> kvm_apic_set_version(vcpu);
>
> apic_update_ppr(apic);
> diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
> index ec6fbfe..58c07ad 100644
> --- a/arch/x86/kvm/lapic.h
> +++ b/arch/x86/kvm/lapic.h
> @@ -76,8 +76,9 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event);
> u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu);
> void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8);
> void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu);
> -void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
> +bool kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
> u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu);
> +void kvm_recalculate_apic_map(struct kvm *kvm);
> void kvm_apic_set_version(struct kvm_vcpu *vcpu);
> int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val);
> int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 79bc995..e961e65 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -349,7 +349,8 @@ int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> return 1;
> }
>
> - kvm_lapic_set_base(vcpu, msr_info->data);
> + if (kvm_lapic_set_base(vcpu, msr_info->data))
> + kvm_recalculate_apic_map(vcpu->kvm);
> return 0;
> }
> EXPORT_SYMBOL_GPL(kvm_set_apic_base);
> --
> 2.7.4
>

2020-02-25 08:07:55

by Paolo Bonzini

Subject: Re: [PATCH] KVM: LAPIC: Recalculate apic map in batch

On 19/02/20 01:47, Wanpeng Li wrote:
>> An alternative idea: instead of making every caller return bool and
>> every call site handle the result (once) just add a
>> KVM_REQ_APIC_MAP_RECALC flag or a boolean flag to struct kvm. I
>> understand it may not be that easy as it sounds as we may be conunting
>> on valid mapping somewhere before we actually get to handiling
> Yes.
>
>> KVM_REQ_APIC_MAP_RECALC but we may preserve *some*
>> recalculate_apic_map() calls (and make it reset KVM_REQ_APIC_MAP_RECALC).
> Paolo, keep the caller return bool or add a booleen flag to struct
> kvm, what do you think?

A third possibility: add an apic_map field to struct kvm_lapic, so that
you don't have to add bool return values everywhere.

Paolo

2020-02-25 08:50:59

by Wanpeng Li

Subject: Re: [PATCH] KVM: LAPIC: Recalculate apic map in batch

On Tue, 25 Feb 2020 at 16:07, Paolo Bonzini <[email protected]> wrote:
>
> On 19/02/20 01:47, Wanpeng Li wrote:
> >> An alternative idea: instead of making every caller return bool and
> >> every call site handle the result (once) just add a
> >> KVM_REQ_APIC_MAP_RECALC flag or a boolean flag to struct kvm. I
> >> understand it may not be that easy as it sounds as we may be conunting
> >> on valid mapping somewhere before we actually get to handiling
> > Yes.
> >
> >> KVM_REQ_APIC_MAP_RECALC but we may preserve *some*
> >> recalculate_apic_map() calls (and make it reset KVM_REQ_APIC_MAP_RECALC).
> > Paolo, keep the caller return bool or add a booleen flag to struct
> > kvm, what do you think?
>
> A third possibility: add an apic_map field to struct kvm_lapic, so that
> you don't have to add bool return values everywhere.

This apic_map field is boolean, right?

Wanpeng

2020-02-25 08:56:35

by Paolo Bonzini

Subject: Re: [PATCH] KVM: LAPIC: Recalculate apic map in batch

On 25/02/20 09:31, Wanpeng Li wrote:
> On Tue, 25 Feb 2020 at 16:07, Paolo Bonzini <[email protected]> wrote:
>>
>> On 19/02/20 01:47, Wanpeng Li wrote:
>>>> An alternative idea: instead of making every caller return bool and
>>>> every call site handle the result (once) just add a
>>>> KVM_REQ_APIC_MAP_RECALC flag or a boolean flag to struct kvm. I
>>>> understand it may not be that easy as it sounds as we may be conunting
>>>> on valid mapping somewhere before we actually get to handiling
>>> Yes.
>>>
>>>> KVM_REQ_APIC_MAP_RECALC but we may preserve *some*
>>>> recalculate_apic_map() calls (and make it reset KVM_REQ_APIC_MAP_RECALC).
>>> Paolo, keep the caller return bool or add a booleen flag to struct
>>> kvm, what do you think?
>>
>> A third possibility: add an apic_map field to struct kvm_lapic, so that
>> you don't have to add bool return values everywhere.
>
> This apic_map field is boolean, right?

Right, and the name should really be apic_map_dirty.

Paolo