[PATCH v2 50/60] KVM: PPC: Book3S HV P9: Avoid tlbsync sequence on radix guest exit

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



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@xxxxxxxxx>
---
 arch/powerpc/kvm/book3s_hv.c          | 31 +++++++++++++++++++++++----
 arch/powerpc/kvm/book3s_hv_p9_entry.c |  9 --------
 2 files changed, 27 insertions(+), 13 deletions(-)

diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 3983c5fa065a..7d08b826d355 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -3028,6 +3028,25 @@ 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;
@@ -3055,10 +3074,14 @@ 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
diff --git a/arch/powerpc/kvm/book3s_hv_p9_entry.c b/arch/powerpc/kvm/book3s_hv_p9_entry.c
index 183d5884e362..94b15294a388 100644
--- a/arch/powerpc/kvm/book3s_hv_p9_entry.c
+++ b/arch/powerpc/kvm/book3s_hv_p9_entry.c
@@ -1045,15 +1045,6 @@ 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.
-- 
2.23.0




[Index of Archives]     [KVM Development]     [KVM ARM]     [KVM ia64]     [Linux Virtualization]     [Linux USB Devel]     [Linux Video]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Big List of Linux Books]

  Powered by Linux