[PATCH v3 12/16] KVM: selftests: Verify non-postable IRQ remapping
From: Josh Hilke
Date: Tue Apr 21 2026 - 19:22:56 EST
From: David Matlack <dmatlack@xxxxxxxxxx>
Add the -n flag to tools/testing/selftests/kvm/irq_test.c to route a
portion of device interrupts as NMIs (Non-Maskable Interrupts) into the
guest using an alternating pattern of 4 NMIs followed by 4 regular
interrupts.
While this adds coverage for NMI injection, the primary goal is to
validate KVM's handling of non-postable interrupt delivery—specifically
the transitions between posted and remapped modes. NMIs are used for
this purpose because they are a reliable, architectural way to force
these code paths.
Co-developed-by: Josh Hilke <jrhilke@xxxxxxxxxx>
Signed-off-by: Josh Hilke <jrhilke@xxxxxxxxxx>
Signed-off-by: David Matlack <dmatlack@xxxxxxxxxx>
---
tools/testing/selftests/kvm/irq_test.c | 44 ++++++++++++++++++++------
1 file changed, 34 insertions(+), 10 deletions(-)
diff --git a/tools/testing/selftests/kvm/irq_test.c b/tools/testing/selftests/kvm/irq_test.c
index 9efb7e655623..03b88b7585b3 100644
--- a/tools/testing/selftests/kvm/irq_test.c
+++ b/tools/testing/selftests/kvm/irq_test.c
@@ -17,6 +17,7 @@
static u64 timeout_ns = 2ULL * 1000 * 1000 * 1000;
static bool guest_ready_for_irqs[KVM_MAX_VCPUS];
static bool guest_received_irq[KVM_MAX_VCPUS];
+static bool guest_received_nmi[KVM_MAX_VCPUS];
static bool irq_affinity;
static bool block_vcpus;
static bool done;
@@ -33,6 +34,11 @@ static void guest_irq_handler(struct ex_regs *regs)
x2apic_write_reg(APIC_EOI, 0);
}
+static void guest_nmi_handler(struct ex_regs *regs)
+{
+ WRITE_ONCE(guest_received_nmi[guest_get_vcpu_id()], true);
+}
+
static void guest_code(void)
{
x2apic_enable();
@@ -93,7 +99,7 @@ static void trigger_interrupt(struct vfio_pci_device *device, int eventfd)
static void kvm_route_msi(struct kvm_vm *vm, u32 gsi, struct kvm_vcpu *vcpu,
- u8 vector)
+ u8 vector, bool use_nmi)
{
struct {
struct kvm_irq_routing head;
@@ -106,7 +112,7 @@ static void kvm_route_msi(struct kvm_vm *vm, u32 gsi, struct kvm_vcpu *vcpu,
routes->entries[0].gsi = gsi;
routes->entries[0].type = KVM_IRQ_ROUTING_MSI;
routes->entries[0].u.msi.address_lo = 0xFEE00000 | (vcpu->id << 12);
- routes->entries[0].u.msi.data = vector;
+ routes->entries[0].u.msi.data = use_nmi ? NMI_VECTOR | (4 << 8) : vector;
vm_ioctl(vm, KVM_SET_GSI_ROUTING, routes);
}
@@ -120,7 +126,7 @@ static void kvm_clear_gsi_routes(struct kvm_vm *vm)
static void help(const char *name)
{
- printf("Usage: %s [-a] [-b] [-c] [-d <segment:bus:device.function>] [-h] [-i nr_irqs]\n", name);
+ printf("Usage: %s [-a] [-b] [-c] [-d <segment:bus:device.function>] [-h] [-i nr_irqs] [-n]\n", name);
printf("\n");
printf("Tests KVM IRQ injection via irqfd using an emulated eventfd.\n");
printf("-a Randomly affinitize the device's host IRQ to different physical CPUs throughout the test\n");
@@ -128,6 +134,7 @@ static void help(const char *name)
printf("-c Destroy and recreate KVM's GSI routing table in between some interrupts\n");
printf("-d Use a VFIO device to send MSI-X interrupts instead of using an emulated eventfd\n");
printf("-i The number of IRQs to generate during the test\n");
+ printf("-n Deliver 50 percent of IRQs as non-maskable interrupts\n");
printf("\n");
exit(KSFT_FAIL);
}
@@ -149,6 +156,7 @@ int main(int argc, char **argv)
u32 gsi = kvm_random_u64_in_range(&kvm_rng, 24, KVM_MAX_IRQ_ROUTES - 1);
u8 vector = kvm_random_u64_in_range(&kvm_rng, 32, UINT8_MAX);
+ bool clear_routes = false, use_nmi = false;
int i, j, c, msi, irq, eventfd, irq_cpu;
struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
pthread_t vcpu_threads[KVM_MAX_VCPUS];
@@ -156,11 +164,10 @@ int main(int argc, char **argv)
int nr_irqs = 1000, nr_vcpus = 1;
const char *device_bdf = NULL;
FILE *irq_affinity_fp = NULL;
- bool clear_routes = false;
struct iommu *iommu;
struct kvm_vm *vm;
- while ((c = getopt(argc, argv, "abcd:hi:")) != -1) {
+ while ((c = getopt(argc, argv, "abcd:hi:n")) != -1) {
switch (c) {
case 'a':
irq_affinity = true;
@@ -177,6 +184,9 @@ int main(int argc, char **argv)
case 'i':
nr_irqs = atoi_positive("Number of IRQs", optarg);
break;
+ case 'n':
+ use_nmi = true;
+ break;
case 'h':
default:
help(argv[0]);
@@ -187,6 +197,7 @@ int main(int argc, char **argv)
vm = vm_create_with_vcpus(nr_vcpus, guest_code, vcpus);
vm_install_exception_handler(vm, vector, guest_irq_handler);
+ vm_install_exception_handler(vm, NMI_VECTOR, guest_nmi_handler);
if (device_bdf) {
iommu = iommu_init(default_iommu_mode);
@@ -224,35 +235,45 @@ int main(int argc, char **argv)
for (i = 0; i < nr_irqs; i++) {
const bool do_clear_routes = clear_routes && (i & BIT(3));
+ const bool do_use_nmi = use_nmi && (i & BIT(2));
struct kvm_vcpu *vcpu = vcpus[i % nr_vcpus];
struct timespec start;
if (do_clear_routes)
kvm_clear_gsi_routes(vm);
- kvm_route_msi(vm, gsi, vcpu, vector);
+ kvm_route_msi(vm, gsi, vcpu, vector, do_use_nmi);
if (irq_affinity && vcpu->id == 0) {
irq_cpu = kvm_random_u64(&kvm_rng) % get_nprocs();
write_proc_irq_affinity(irq_affinity_fp, irq, irq_cpu);
}
- for (j = 0; j < nr_vcpus; j++)
+ for (j = 0; j < nr_vcpus; j++) {
TEST_ASSERT(
!SYNC_FROM_GUEST_AND_READ(vm, guest_received_irq[vcpus[j]->id]),
"IRQ flag for vCPU %d not clear prior to test",
vcpus[j]->id);
+ TEST_ASSERT(
+ !SYNC_FROM_GUEST_AND_READ(vm, guest_received_nmi[vcpus[j]->id]),
+ "NMI flag for vCPU %d not clear prior to test",
+ vcpus[j]->id);
+ }
trigger_interrupt(device, eventfd);
clock_gettime(CLOCK_MONOTONIC, &start);
for (;;) {
- if (SYNC_FROM_GUEST_AND_READ(vm, guest_received_irq[vcpu->id]))
+ if (!do_use_nmi && SYNC_FROM_GUEST_AND_READ(vm, guest_received_irq[vcpu->id]))
+ break;
+
+ if (do_use_nmi && SYNC_FROM_GUEST_AND_READ(vm, guest_received_nmi[vcpu->id]))
break;
if (timespec_to_ns(timespec_elapsed(start)) > timeout_ns) {
printf("Timeout waiting for interrupt!\n");
printf(" vCPU: %d\n", vcpu->id);
+ printf(" is interrupt NMI: %s\n", do_use_nmi ? "true" : "false");
if (irq_affinity)
printf(" irq_cpu: %d\n", irq_cpu);
@@ -261,7 +282,10 @@ int main(int argc, char **argv)
}
}
- WRITE_AND_SYNC_TO_GUEST(vm, guest_received_irq[vcpu->id], false);
+ if (do_use_nmi)
+ WRITE_AND_SYNC_TO_GUEST(vm, guest_received_nmi[vcpu->id], false);
+ else
+ WRITE_AND_SYNC_TO_GUEST(vm, guest_received_irq[vcpu->id], false);
}
WRITE_AND_SYNC_TO_GUEST(vm, done, true);
@@ -273,7 +297,7 @@ int main(int argc, char **argv)
* below, will hang.
*/
if (block_vcpus) {
- kvm_route_msi(vm, gsi, vcpus[i], vector);
+ kvm_route_msi(vm, gsi, vcpus[i], vector, false);
trigger_interrupt(device, eventfd);
}
--
2.54.0.rc2.533.g4f5dca5207-goog