Re: [PATCH 2/2] vfio: selftests: Add iommufd multi iommu test

[David Matlack <dmatlack@google.com>]

On 2026-05-11 08:53 PM, Samiullah Khawaja wrote:
> 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 <skhawaja@google.com>
> > > ---
> > >  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?

We have 2 tests now that need the same code to setup their regions so we
might as well share it. As new and more complex use-cases arise we can
figure out how to support them (per-test logic or fancier library
support).

> > > +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 was thinking something like this exposed by lib/iommu.c:

void iommu_copy_mapping(struct iommu *iommu,
			struct dma_region *new_region,
			struct dma_region *existing_region)
{
	*new_region = *existing_region;
	INIT_LIST_HEAD(&new_region->link);
	iommu_map(iommu, new_region);
}

And then in the test you just do:

if (iommu1 != self->iommu)
	iommu_copy_mapping(iommu1, memcpy_region1, self->memcpy_region);