On Fri, May 08, 2026 at 08:14:08PM +0000, David Matlack wrote:
On 2026-05-05 10:14 PM, Samiullah Khawaja wrote:
Add a test for iommufd to verify creation of multiple iommus using an
already setup iommufd and switch between them.
Signed-off-by: Samiullah Khawaja <[email protected]>
---
tools/testing/selftests/vfio/Makefile | 1 +
.../vfio/vfio_iommufd_multi_iommu_test.c | 203 ++++++++++++++++++
2 files changed, 204 insertions(+)
create mode 100644 tools/testing/selftests/vfio/vfio_iommufd_multi_iommu_test.c
diff --git a/tools/testing/selftests/vfio/Makefile
b/tools/testing/selftests/vfio/Makefile
index 0684932d91bf..24bdeaad590f 100644
--- a/tools/testing/selftests/vfio/Makefile
+++ b/tools/testing/selftests/vfio/Makefile
@@ -8,6 +8,7 @@ else
CFLAGS = $(KHDR_INCLUDES)
TEST_GEN_PROGS += vfio_dma_mapping_test
TEST_GEN_PROGS += vfio_dma_mapping_mmio_test
+TEST_GEN_PROGS += vfio_iommufd_multi_iommu_test
TEST_GEN_PROGS += vfio_iommufd_setup_test
TEST_GEN_PROGS += vfio_pci_device_test
TEST_GEN_PROGS += vfio_pci_device_init_perf_test
diff --git a/tools/testing/selftests/vfio/vfio_iommufd_multi_iommu_test.c
b/tools/testing/selftests/vfio/vfio_iommufd_multi_iommu_test.c
new file mode 100644
index 000000000000..7199c662637c
--- /dev/null
+++ b/tools/testing/selftests/vfio/vfio_iommufd_multi_iommu_test.c
@@ -0,0 +1,203 @@
+// SPDX-License-Identifier: GPL-2.0-only
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+
+#include <linux/sizes.h>
+#include <linux/vfio.h>
+
+#include <libvfio.h>
+
+#include "kselftest_harness.h"
+
+static const char *device_bdf;
+
+#define IOMMU_DEFAULT 0
+#define IOMMU_WITH_PT 1
+#define IOMMU_WITHOUT_PT 2
+
+static void region_setup(struct iommu *iommu,
+ struct iova_allocator *iova_allocator,
+ struct dma_region *region, u64 size)
+{
+ const int flags = MAP_SHARED | MAP_ANONYMOUS;
+ const int prot = PROT_READ | PROT_WRITE;
+ void *vaddr;
+
+ vaddr = mmap(NULL, size, prot, flags, -1, 0);
+ VFIO_ASSERT_NE(vaddr, MAP_FAILED);
+
+ region->vaddr = vaddr;
+ region->iova = iova_allocator_alloc(iova_allocator, size);
+ region->size = size;
+
+ iommu_map(iommu, region);
+}
+
+static void region_teardown(struct iommu *iommu, struct dma_region *region)
+{
+ iommu_unmap(iommu, region);
+ VFIO_ASSERT_EQ(munmap(region->vaddr, region->size), 0);
+}
Can you create library helpers in the library to share with
vfio_pci_driver_test.c?
dma_region_setup()
dma_region_teardown()
I thought it is ok to duplicate this as each test (in future) might have
different mapping requirements? we might have memfd, guest_memfd,
dma_buf with various mapping strategies for different usecases?
Or you are thinking of a basic helper that can be used by tests if they
are not doing anything fancy?
+
+FIXTURE(vfio_iommufd_multi_iommu_test) {
+ struct iommu *iommu;
+ struct vfio_pci_device *device;
+ struct iova_allocator *iova_allocator;
+ struct dma_region memcpy_region;
+ void *vaddr;
+
+ u64 size;
+ void *src;
+ void *dst;
+ iova_t src_iova;
+ iova_t dst_iova;
+};
+
+FIXTURE_SETUP(vfio_iommufd_multi_iommu_test)
+{
+ struct vfio_pci_driver *driver;
+
+ self->iommu = iommu_init("iommufd");
MODE_IOMMUFD
Will update.
+ self->device = vfio_pci_device_init(device_bdf, self->iommu);
+ self->iova_allocator = iova_allocator_init(self->iommu);
+
+ driver = &self->device->driver;
+
+ region_setup(self->iommu, self->iova_allocator, &self->memcpy_region,
SZ_1G);
+ region_setup(self->iommu, self->iova_allocator, &driver->region, SZ_2M);
+
+ if (driver->ops)
+ vfio_pci_driver_init(self->device);
+
+ self->size = self->memcpy_region.size / 2;
+ self->src = self->memcpy_region.vaddr;
+ self->dst = self->src + self->size;
+
+ self->src_iova = to_iova(self->device, self->src);
+ self->dst_iova = to_iova(self->device, self->dst);
+}
+
[snip]
+static void test_memcpy(struct vfio_pci_device *device, void *src, void *dst,
+ iova_t src_iova, iova_t dst_iova, u64 size, char val)
+{
+ if (!device->driver.ops)
+ return;
+
+ memset(src, val, size);
+ memset(dst, 0, size);
+
+ vfio_pci_driver_memcpy_start(device, src_iova, dst_iova, size, 100);
Why do 100 copies?
Will remove in next revision.
+ VFIO_ASSERT_EQ(vfio_pci_driver_memcpy_wait(device), 0);
+ VFIO_ASSERT_EQ(memcmp(src, dst, size), 0);
+}
+
+TEST_F(vfio_iommufd_multi_iommu_test, memcpy)
+{
+ struct dma_region memcpy_region1, driver_region1;
+ struct dma_region memcpy_region2, driver_region2;
+ struct iommu *iommu1;
+ struct iommu *iommu2;
+
+ iommu1 = setup_variant_iommu(self->iommu, self->device->dev_id,
+ variant->start_iommu);
+ if (iommu1 != self->iommu) {
+ memcpy_region1 = self->memcpy_region;
+ driver_region1 = self->device->driver.region;
+
+ iommu_map(iommu1, &memcpy_region1);
Can you make a helper to copy dma_region into a new iommu? It should set
up the new struct dma_region based on the old one, re-initialize the
link field, and then call iommu_map().
Can you please clarify, I didn't get it completely. so basically
something that can be called like following?
setup_copy_dma_regions(self, iommu1, &memcpy_region1, &driver_region1);
I think I get that you are concerned about link init. Maybe I can do?
int clone_dma_region(struct dma_region *source_region,
struct dma_region *dest_region);
We can maybe add it to the library?
+ iommu_map(iommu1, &driver_region1);
+
+ vfio_pci_device_attach_iommu(self->device, iommu1);
+ }
+
+ test_memcpy(self->device, self->src, self->dst, self->src_iova,
+ self->dst_iova, self->size, 'x');
nit: You can pass in self to avoid unpacking it in every call to
test_memcpy(). The type would be:
struct _test_data_vfio_iommufd_multi_iommu_test *self
Agreed. Will do.
+
+ iommu2 = setup_variant_iommu(self->iommu, self->device->dev_id,
+ variant->end_iommu);
+ if (iommu2 != self->iommu) {
+ memcpy_region2 = self->memcpy_region;
+ driver_region2 = self->device->driver.region;
+
+ iommu_map(iommu2, &memcpy_region2);
+ iommu_map(iommu2, &driver_region2);
+ }
+
+ vfio_pci_device_attach_iommu(self->device, iommu2);
+ test_memcpy(self->device, self->src, self->dst, self->src_iova,
+ self->dst_iova, self->size, 'a');
test_memcpy() zeroes self->dst every time so do you need to use a
different character for each call?
I will update this.
+
+ vfio_pci_device_attach_iommu(self->device, iommu1);
+ test_memcpy(self->device, self->src, self->dst, self->src_iova,
+ self->dst_iova, self->size, '1');
+
+ if (iommu1 != self->iommu) {
+ /* attach back to default iommu for cleanup */
+ vfio_pci_device_attach_iommu(self->device, self->iommu);
+ iommu_unmap(iommu1, &memcpy_region1);
+ iommu_unmap(iommu1, &driver_region1);
+ iommu_cleanup(iommu1);
+ }
+
+ if (iommu2 != self->iommu) {
+ iommu_unmap(iommu2, &memcpy_region2);
+ iommu_unmap(iommu2, &driver_region2);
nit: You don't need to unmap before cleanup. Is this just to exercise
that unmap works on the variant iommus?
Not really. I thought it is needed. But I think I will keep it to
exercise that. Will add a comment to clarify.
+ iommu_cleanup(iommu2);
+ }
+}
+
+int main(int argc, char *argv[])
+{
+ device_bdf = vfio_selftests_get_bdf(&argc, argv);
+
+ return test_harness_run(argc, argv);
+}
--
2.54.0.545.g6539524ca2-goog
Thanks,
Sami
