Re: [RFC PATCH 5/6] x86/mm/tlb: Flush remote and local TLBs concurrently
From: Nadav Amit <hidden>
Date: 2019-05-27 17:49:11
Also in:
kvm, lkml, xen-devel
On May 27, 2019, at 2:47 AM, Peter Zijlstra [off-list ref] wrote: On Sat, May 25, 2019 at 10:54:50AM +0200, Juergen Gross wrote:quoted
On 25/05/2019 10:22, Nadav Amit wrote:quoted
quoted
diff --git a/arch/x86/include/asm/paravirt_types.h b/arch/x86/include/asm/paravirt_types.h index 946f8f1f1efc..3a156e63c57d 100644 --- a/arch/x86/include/asm/paravirt_types.h +++ b/arch/x86/include/asm/paravirt_types.h@@ -211,6 +211,12 @@ struct pv_mmu_ops {void (*flush_tlb_user)(void); void (*flush_tlb_kernel)(void); void (*flush_tlb_one_user)(unsigned long addr); + /* + * flush_tlb_multi() is the preferred interface. When it is used, + * flush_tlb_others() should return false.This comment does not make sense. flush_tlb_others() return type is void.I suspect that is an artifact from before the static_key; an attempt to make the pv interface less awkward.
Yes, remainders that should have been removed - I will remove them for the next version.
quoted hunk ↗ jump to hunk
Something like the below would work for KVM I suspect, the others (Hyper-V and Xen are more 'interesting'). ------ a/arch/x86/kernel/kvm.c +++ b/arch/x86/kernel/kvm.c@@ -580,7 +580,7 @@ static void __init kvm_apf_trap_init(voistatic DEFINE_PER_CPU(cpumask_var_t, __pv_tlb_mask); -static void kvm_flush_tlb_others(const struct cpumask *cpumask, +static void kvm_flush_tlb_multi(const struct cpumask *cpumask, const struct flush_tlb_info *info) { u8 state;@@ -594,6 +594,9 @@ static void kvm_flush_tlb_others(const s* queue flush_on_enter for pre-empted vCPUs */ for_each_cpu(cpu, flushmask) { + if (cpu == smp_processor_id()) + continue; + src = &per_cpu(steal_time, cpu); state = READ_ONCE(src->preempted); if ((state & KVM_VCPU_PREEMPTED)) {@@ -603,7 +606,7 @@ static void kvm_flush_tlb_others(const s} } - native_flush_tlb_others(flushmask, info); + native_flush_tlb_multi(flushmask, info); } static void __init kvm_guest_init(void)@@ -628,9 +631,8 @@ static void __init kvm_guest_init(void)if (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) && !kvm_para_has_hint(KVM_HINTS_REALTIME) && kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) { - pv_ops.mmu.flush_tlb_others = kvm_flush_tlb_others; + pv_ops.mmu.flush_tlb_multi = kvm_flush_tlb_multi; pv_ops.mmu.tlb_remove_table = tlb_remove_table; - static_key_disable(&flush_tlb_multi_enabled.key); } if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
That’s what I have as well ;-). As you mentioned (in another email), specifically hyper-v code seems convoluted to me. In general, I prefer not to touch KVM/Xen/hyper-v, but you twist my arm, I will send a compile-tested version for Xen and hyper-v.