diff --git a/arch/s390/kvm/guestdbg.c b/arch/s390/kvm/guestdbg.c
index 1e0849e209650dbe8e5de156a6afa1d3bdf45bf3..31a05330d11c77fc0c34cd30ea54bc2075a2c3d9 100644
--- a/arch/s390/kvm/guestdbg.c
+++ b/arch/s390/kvm/guestdbg.c
@@ -439,6 +439,23 @@ static int debug_exit_required(struct kvm_vcpu *vcpu)
 #define guest_per_enabled(vcpu) \
 			     (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PER)
 
+int kvm_s390_handle_per_ifetch_icpt(struct kvm_vcpu *vcpu)
+{
+	const u8 ilen = kvm_s390_get_ilen(vcpu);
+	struct kvm_s390_pgm_info pgm_info = {
+		.code = PGM_PER,
+		.per_code = PER_EVENT_IFETCH >> 24,
+		.per_address = __rewind_psw(vcpu->arch.sie_block->gpsw, ilen),
+	};
+
+	/*
+	 * The PSW points to the next instruction, therefore the intercepted
+	 * instruction generated a PER i-fetch event. PER address therefore
+	 * points at the previous PSW address (could be an EXECUTE function).
+	 */
+	return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
+}
+
 static void filter_guest_per_event(struct kvm_vcpu *vcpu)
 {
 	u32 perc = vcpu->arch.sie_block->perc << 24;
diff --git a/arch/s390/kvm/intercept.c b/arch/s390/kvm/intercept.c
index 9359f65c8634ba3f6924b42ac863618e59be6e9f..850be47c4cc93fa4b7033a096b986a1670cb3cdd 100644
--- a/arch/s390/kvm/intercept.c
+++ b/arch/s390/kvm/intercept.c
@@ -364,6 +364,8 @@ static int handle_operexc(struct kvm_vcpu *vcpu)
 
 int kvm_handle_sie_intercept(struct kvm_vcpu *vcpu)
 {
+	int rc, per_rc = 0;
+
 	if (kvm_is_ucontrol(vcpu->kvm))
 		return -EOPNOTSUPP;
 
@@ -372,7 +374,8 @@ int kvm_handle_sie_intercept(struct kvm_vcpu *vcpu)
 	case 0x18:
 		return handle_noop(vcpu);
 	case 0x04:
-		return handle_instruction(vcpu);
+		rc = handle_instruction(vcpu);
+		break;
 	case 0x08:
 		return handle_prog(vcpu);
 	case 0x14:
@@ -384,10 +387,18 @@ int kvm_handle_sie_intercept(struct kvm_vcpu *vcpu)
 	case 0x28:
 		return handle_stop(vcpu);
 	case 0x2c:
-		return handle_operexc(vcpu);
+		rc = handle_operexc(vcpu);
+		break;
 	case 0x38:
-		return handle_partial_execution(vcpu);
+		rc = handle_partial_execution(vcpu);
+		break;
 	default:
 		return -EOPNOTSUPP;
 	}
+
+	/* process PER, also if the instrution is processed in user space */
+	if (vcpu->arch.sie_block->icptstatus & 0x02 &&
+	    (!rc || rc == -EOPNOTSUPP))
+		per_rc = kvm_s390_handle_per_ifetch_icpt(vcpu);
+	return per_rc ? per_rc : rc;
 }
diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
index 031f451bb2cf79a2550b14eb6bd2701a0e29d765..b8432862a81715761fbe274071d0988021b61a54 100644
--- a/arch/s390/kvm/kvm-s390.h
+++ b/arch/s390/kvm/kvm-s390.h
@@ -238,6 +238,8 @@ static inline void kvm_s390_forward_psw(struct kvm_vcpu *vcpu, int ilen)
 }
 static inline void kvm_s390_retry_instr(struct kvm_vcpu *vcpu)
 {
+	/* don't inject PER events if we re-execute the instruction */
+	vcpu->arch.sie_block->icptstatus &= ~0x02;
 	kvm_s390_rewind_psw(vcpu, kvm_s390_get_ilen(vcpu));
 }
 
@@ -377,6 +379,7 @@ int kvm_s390_import_bp_data(struct kvm_vcpu *vcpu,
 			    struct kvm_guest_debug *dbg);
 void kvm_s390_clear_bp_data(struct kvm_vcpu *vcpu);
 void kvm_s390_prepare_debug_exit(struct kvm_vcpu *vcpu);
+int kvm_s390_handle_per_ifetch_icpt(struct kvm_vcpu *vcpu);
 void kvm_s390_handle_per_event(struct kvm_vcpu *vcpu);
 
 /* support for Basic/Extended SCA handling */