2017-11-10 16:24 GMT+08:00 Paolo Bonzini <pbonzini@xxxxxxxxxx>:
> On 10/11/2017 08:04, Wanpeng Li wrote:
>> From: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
>>
>> Remote flushing api's does a busy wait which is fine in bare-metal
>> scenario. But with-in the guest, the vcpus might have been pre-empted
>> or blocked. In this scenario, the initator vcpu would end up
>> busy-waiting for a long amount of time.
>>
>> This patch set implements para-virt flush tlbs making sure that it
>> does not wait for vcpus that are sleeping. And all the sleeping vcpus
>> flush the tlb on guest enter.
>>
>> The best result is achieved when we're overcommiting the host by running
>> multiple vCPUs on each pCPU. In this case PV tlb flush avoids touching
>> vCPUs which are not scheduled and avoid the wait on the main CPU.
>>
>> Test on a Haswell i7 desktop 4 cores (2HT), so 8 pCPUs, running ebizzy in
>> one linux guest.
>>
>> ebizzy -M
>> vanilla optimized boost
>> 8 vCPUs 10152 10083 -0.68%
>> 16 vCPUs 1224 4866 297.5%
>> 24 vCPUs 1109 3871 249%
>> 32 vCPUs 1025 3375 229.3%
>>
>> Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
>> Cc: Radim Krčmář <rkrcmar@xxxxxxxxxx>
>> Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
>> ---
>> Documentation/virtual/kvm/cpuid.txt | 4 ++++
>> arch/x86/include/uapi/asm/kvm_para.h | 2 ++
>> arch/x86/kernel/kvm.c | 31 +++++++++++++++++++++++++++++++
>> 3 files changed, 37 insertions(+)
>>
>> diff --git a/Documentation/virtual/kvm/cpuid.txt b/Documentation/virtual/kvm/cpuid.txt
>> index 117066a..9693fcc 100644
>> --- a/Documentation/virtual/kvm/cpuid.txt
>> +++ b/Documentation/virtual/kvm/cpuid.txt
>> @@ -60,6 +60,10 @@ KVM_FEATURE_PV_DEDICATED || 8 || guest checks this feature bit
>> || || mizations such as usage of
>> || || qspinlocks.
>> ------------------------------------------------------------------------------
>> +KVM_FEATURE_PV_TLB_FLUSH || 9 || guest checks this feature bit
>> + || || before enabling paravirtualized
>> + || || tlb flush.
>> +------------------------------------------------------------------------------
>> KVM_FEATURE_CLOCKSOURCE_STABLE_BIT || 24 || host will warn if no guest-side
>> || || per-cpu warps are expected in
>> || || kvmclock.
>> diff --git a/arch/x86/include/uapi/asm/kvm_para.h b/arch/x86/include/uapi/asm/kvm_para.h
>> index 9ead1ed..a028479 100644
>> --- a/arch/x86/include/uapi/asm/kvm_para.h
>> +++ b/arch/x86/include/uapi/asm/kvm_para.h
>> @@ -25,6 +25,7 @@
>> #define KVM_FEATURE_PV_EOI 6
>> #define KVM_FEATURE_PV_UNHALT 7
>> #define KVM_FEATURE_PV_DEDICATED 8
>> +#define KVM_FEATURE_PV_TLB_FLUSH 9
>>
>> /* The last 8 bits are used to indicate how to interpret the flags field
>> * in pvclock structure. If no bits are set, all flags are ignored.
>> @@ -53,6 +54,7 @@ struct kvm_steal_time {
>>
>> #define KVM_VCPU_NOT_PREEMPTED (0 << 0)
>> #define KVM_VCPU_PREEMPTED (1 << 0)
>> +#define KVM_VCPU_SHOULD_FLUSH (1 << 1)
>>
>> #define KVM_CLOCK_PAIRING_WALLCLOCK 0
>> struct kvm_clock_pairing {
>> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
>> index 66ed3bc..50f4b6a 100644
>> --- a/arch/x86/kernel/kvm.c
>> +++ b/arch/x86/kernel/kvm.c
>> @@ -465,6 +465,33 @@ static void __init kvm_apf_trap_init(void)
>> update_intr_gate(X86_TRAP_PF, async_page_fault);
>> }
>>
>> +static cpumask_t flushmask;
>
> Hi Wanpeng,
>
> are you going to send v3 with a percpu variable?
Hi Peterz,
I found big performance difference as I discuss with you several days ago.
ebizzy -M
vanilla static/local cpumask per-cpu cpumask
8 vCPUs 10152 10083 10117
16 vCPUs 1224 4866 10008
24 vCPUs 1109 3871 9928
32 vCPUs 1025 3375 9811
In addition, I can observe ~50% perf top time is occupied by
smp_call_function_many(), ~30% perf top time is occupied by
call_function_interrupt() in the guest when running ebizzy for
static/local cpumask variable. However, I almost can't observe these
IPI stuffs after changing to per-cpu variable. Any opinions?
Regards,
Wanpeng Li
>
> Paolo
>
>> +static void kvm_flush_tlb_others(const struct cpumask *cpumask,
>> + const struct flush_tlb_info *info)
>> +{
>> + u8 state;
>> + int cpu;
>> + struct kvm_steal_time *src;
>> +
>> + cpumask_copy(&flushmask, cpumask);
>> + /*
>> + * We have to call flush only on online vCPUs. And
>> + * queue flush_on_enter for pre-empted vCPUs
>> + */
>> + for_each_cpu(cpu, cpumask) {
>> + src = &per_cpu(steal_time, cpu);
>> + state = src->preempted;
>> + if ((state & KVM_VCPU_PREEMPTED)) {
>> + if (cmpxchg(&src->preempted, state, state |
>> + KVM_VCPU_SHOULD_FLUSH) == state)
>> + __cpumask_clear_cpu(cpu, &flushmask);
>> + }
>> + }
>> +
>> + native_flush_tlb_others(&flushmask, info);
>> +}
>> +
>> void __init kvm_guest_init(void)
>> {
>> int i;
>> @@ -484,6 +511,10 @@ void __init kvm_guest_init(void)
>> pv_time_ops.steal_clock = kvm_steal_clock;
>> }
>>
>> + if (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) &&
>> + !kvm_para_has_feature(KVM_FEATURE_PV_DEDICATED))
>> + pv_mmu_ops.flush_tlb_others = kvm_flush_tlb_others;
>> +
>> if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
>> apic_set_eoi_write(kvm_guest_apic_eoi_write);
>>
>>
>
