@@ -3025,6 +3025,25 @@ static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
smp_call_function_single(i, do_nothing, NULL, 1);
}
+static void do_migrate_away_vcpu(void *arg)
+{
+ struct kvm_vcpu *vcpu = arg;
+ struct kvm *kvm = vcpu->kvm;
+
+ /*
+ * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync;
+ * ptesync sequence on the old CPU before migrating to a new one, in
+ * case we interrupted the guest between a tlbie ; eieio ;
+ * tlbsync; ptesync sequence.
+ *
+ * Otherwise, ptesync is sufficient.
+ */
+ if (kvm->arch.lpcr & LPCR_GTSE)
+ asm volatile("eieio; tlbsync; ptesync");
+ else
+ asm volatile("ptesync");
+}
+
static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
{
struct kvm_nested_guest *nested = vcpu->arch.nested;
@@ -3052,10 +3071,14 @@ static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
* so we use a single bit in .need_tlb_flush for all 4 threads.
*/
if (prev_cpu != pcpu) {
- if (prev_cpu >= 0 &&
- cpu_first_tlb_thread_sibling(prev_cpu) !=
- cpu_first_tlb_thread_sibling(pcpu))
- radix_flush_cpu(kvm, prev_cpu, vcpu);
+ if (prev_cpu >= 0) {
+ if (cpu_first_tlb_thread_sibling(prev_cpu) !=
+ cpu_first_tlb_thread_sibling(pcpu))
+ radix_flush_cpu(kvm, prev_cpu, vcpu);
+
+ smp_call_function_single(prev_cpu,
+ do_migrate_away_vcpu, vcpu, 1);
+ }
if (nested)
nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
else
@@ -1049,15 +1049,6 @@ int kvmhv_vcpu_entry_p9(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpc
local_paca->kvm_hstate.in_guest = KVM_GUEST_MODE_NONE;
- if (kvm_is_radix(kvm)) {
- /*
- * Since this is radix, do a eieio; tlbsync; ptesync sequence
- * in case we interrupted the guest between a tlbie and a
- * ptesync.
- */
- asm volatile("eieio; tlbsync; ptesync");
- }
-
/*
* cp_abort is required if the processor supports local copy-paste
* to clear the copy buffer that was under control of the guest.
Use the existing TLB flushing logic to IPI the previous CPU and run the necessary barriers before running a guest vCPU on a new physical CPU, to do the necessary radix GTSE barriers for handling the case of an interrupted guest tlbie sequence. This results in more IPIs than the TLB flush logic requires, but it's a significant win for common case scheduling when the vCPU remains on the same physical CPU. This saves about 520 cycles (nearly 10%) on a guest entry+exit micro benchmark on a POWER9. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> --- arch/powerpc/kvm/book3s_hv.c | 31 +++++++++++++++++++++++---- arch/powerpc/kvm/book3s_hv_p9_entry.c | 9 -------- 2 files changed, 27 insertions(+), 13 deletions(-)