Kernel KVM virtualization development
 help / color / mirror / Atom feed
From: David Matlack <dmatlack@google.com>
To: Aaron Lewis <aaronlewis@google.com>
Cc: kvm@vger.kernel.org, alex@shazbot.org, jgg@nvidia.com
Subject: Re: [PATCH 2/4] vfio: selftests: Add memfd test to vfio_dma_mapping_perf_test
Date: Tue, 14 Jul 2026 16:53:07 +0000	[thread overview]
Message-ID: <alZpc7oSwtw7hExk@google.com> (raw)
In-Reply-To: <20260701203311.326798-3-aaronlewis@google.com>

On 2026-07-01 08:33 PM, Aaron Lewis wrote:
> Add a second test to "vfio_dma_mapping_perf_test" to evaluate memfd
> latencies. A key motivator for adding this test is to demonstrate that
> the IOMMU_IOAS_MAP_FILE ioctl is significantly faster than other methods
> of mapping DMA regions. While this performance difference is not fully
> apparent with the test's current capabilities, it will become evident
> as more features are introduced later in the series.
> 
> To support this, add IOMMU_IOAS_MAP_FILE ioctl support to the VFIO
> selftest library via iommufd_map_file().
> 
> Signed-off-by: Aaron Lewis <aaronlewis@google.com>
> ---
>  .../vfio/lib/include/libvfio/iommu.h          |   7 ++
>  tools/testing/selftests/vfio/lib/iommu.c      |  24 ++++
>  .../vfio/vfio_dma_mapping_perf_test.c         | 118 ++++++++++++++++++
>  3 files changed, 149 insertions(+)
> 
> diff --git a/tools/testing/selftests/vfio/lib/include/libvfio/iommu.h b/tools/testing/selftests/vfio/lib/include/libvfio/iommu.h
> index e9a3386a4719..eeef5990292c 100644
> --- a/tools/testing/selftests/vfio/lib/include/libvfio/iommu.h
> +++ b/tools/testing/selftests/vfio/lib/include/libvfio/iommu.h
> @@ -35,6 +35,13 @@ struct iommu {
>  struct iommu *iommu_init(const char *iommu_mode);
>  void iommu_cleanup(struct iommu *iommu);
>  
> +int __iommufd_map_file(struct iommu *iommu, struct dma_region *region, int fd);
> +
> +static inline void iommufd_map_file(struct iommu *iommu, struct dma_region *region, int fd)
> +{
> +	VFIO_ASSERT_EQ(__iommufd_map_file(iommu, region, fd), 0);
> +}
> +
>  int __iommu_map(struct iommu *iommu, struct dma_region *region);
>  
>  static inline void iommu_map(struct iommu *iommu, struct dma_region *region)
> diff --git a/tools/testing/selftests/vfio/lib/iommu.c b/tools/testing/selftests/vfio/lib/iommu.c
> index 035dac069d60..6730b39060ca 100644
> --- a/tools/testing/selftests/vfio/lib/iommu.c
> +++ b/tools/testing/selftests/vfio/lib/iommu.c
> @@ -97,6 +97,30 @@ iova_t iommu_hva2iova(struct iommu *iommu, void *vaddr)
>  	return iova;
>  }
>  
> +int __iommufd_map_file(struct iommu *iommu, struct dma_region *region, int fd)
> +{
> +	VFIO_ASSERT_TRUE(iommu->iommufd, "IOMMU_IOAS_MAP_FILE is an IOMMUFD IOCTL.");
> +
> +	struct iommu_ioas_map_file args = {
> +		.size = sizeof(args),
> +		.flags = IOMMU_IOAS_MAP_READABLE |
> +			IOMMU_IOAS_MAP_WRITEABLE |
> +			IOMMU_IOAS_MAP_FIXED_IOVA,
> +		.ioas_id = iommu->ioas_id,
> +		.fd = fd,
> +		.start = 0,
> +		.iova = region->iova,
> +		.length = region->size,
> +	};
> +
> +	if (ioctl(iommu->iommufd, IOMMU_IOAS_MAP_FILE, &args))
> +		return -errno;
> +
> +	list_add(&region->link, &iommu->dma_regions);
> +
> +	return 0;
> +}
> +
>  static int vfio_iommu_map(struct iommu *iommu, struct dma_region *region)
>  {
>  	struct vfio_iommu_type1_dma_map args = {
> diff --git a/tools/testing/selftests/vfio/vfio_dma_mapping_perf_test.c b/tools/testing/selftests/vfio/vfio_dma_mapping_perf_test.c
> index 87d84450fc47..4639bda7ebaa 100644
> --- a/tools/testing/selftests/vfio/vfio_dma_mapping_perf_test.c
> +++ b/tools/testing/selftests/vfio/vfio_dma_mapping_perf_test.c
> @@ -117,6 +117,124 @@ TEST_F(vfio_dma_mapping_perf_test, dma_map_unmap)
>  	printf("Completed munmap() in %.2lfms\n", timer_elapsed_ms(start));
>  }
>  
> +FIXTURE(vfio_dma_mapping_perf_memfd_test) {
> +	struct iommu *iommu;
> +	struct vfio_pci_device *device;
> +	struct iova_allocator *iova_allocator;
> +};
> +
> +FIXTURE_VARIANT(vfio_dma_mapping_perf_memfd_test) {
> +	const char *iommu_mode;
> +	int mmap_flags;
> +	int memfd_flags;
> +};
> +
> +#define FIXTURE_VARIANT_ADD_IOMMU_MODE(_iommu_mode, _name, _mmap_flags, _memfd_flags) \
> +FIXTURE_VARIANT_ADD(vfio_dma_mapping_perf_memfd_test, _iommu_mode ## _ ## _name) {    \
> +	.iommu_mode = #_iommu_mode,						      \
> +	.mmap_flags = MAP_SHARED | (_mmap_flags),				      \
> +	.memfd_flags = MAP_SHARED | (_memfd_flags),				      \
> +}
> +
> +FIXTURE_VARIANT_ADD_ALL_IOMMU_MODES(memfd, 0, 0);
> +FIXTURE_VARIANT_ADD_ALL_IOMMU_MODES(memfd_hugetlb_2mb,
> +				    MAP_HUGETLB | MAP_HUGE_2MB,
> +				    MFD_HUGETLB | MFD_HUGE_2MB);
> +FIXTURE_VARIANT_ADD_ALL_IOMMU_MODES(memfd_hugetlb_1gb,
> +				    MAP_HUGETLB | MAP_HUGE_1GB,
> +				    MFD_HUGETLB | MFD_HUGE_1GB);

Only iommufd supports file-based mapping so not need to replicate across
all other IOMMU modes. Just run with iommufd.

And then you can drop the special case below where you fallback to
HVA-based mapping in the file-based test, which is largely duplicating
the other test.

If you want to also have test coverage for HVA-based mapping of
file-backed memory then you should add that separately, and include
iommufd in it.

> +
> +#undef FIXTURE_VARIANT_ADD_IOMMU_MODE
> +
> +FIXTURE_SETUP(vfio_dma_mapping_perf_memfd_test)
> +{
> +	self->iommu = iommu_init(variant->iommu_mode);
> +	self->device = vfio_pci_device_init(device_bdf, self->iommu);
> +	self->iova_allocator = iova_allocator_init(self->iommu);
> +}
> +
> +FIXTURE_TEARDOWN(vfio_dma_mapping_perf_memfd_test)
> +{
> +	iova_allocator_cleanup(self->iova_allocator);
> +	vfio_pci_device_cleanup(self->device);
> +	iommu_cleanup(self->iommu);
> +}
> +
> +static void *setup_memfd(int *fd, u64 size, int mmap_flags, int mfd_flags)
> +{
> +	void *buf = MAP_FAILED;
> +	struct timespec start;
> +
> +	timer_start(&start);
> +	*fd = memfd_create("vfio_dma_mapping_perf_memfd_test", mfd_flags);
> +	printf("Completed memfd_create() in %.2lfms\n", timer_elapsed_ms(start));
> +	if (*fd <= 0)
> +		return MAP_FAILED;
> +
> +	if (ftruncate(*fd, size))
> +		goto out;
> +
> +	timer_start(&start);
> +	buf = mmap(NULL, size, PROT_READ | PROT_WRITE, mmap_flags, *fd, 0);
> +	printf("Completed mmap() for memfd in %.2lfms\n", timer_elapsed_ms(start));
> +
> +out:
> +	if (buf == MAP_FAILED)
> +		close(*fd);
> +
> +	return buf;
> +}
> +
> +static void teardown_memfd(int fd, u64 size, void *vaddr)
> +{
> +	struct timespec start;
> +
> +	if (vaddr != MAP_FAILED) {
> +		timer_start(&start);
> +		munmap(vaddr, size);
> +		printf("Completed munmap() in %.2lfms\n", timer_elapsed_ms(start));
> +	}
> +
> +	if (fd != -1) {
> +		timer_start(&start);
> +		close(fd);
> +		printf("Completed close() in %.2lfms\n", timer_elapsed_ms(start));
> +	}
> +}
> +
> +TEST_F(vfio_dma_mapping_perf_memfd_test, dma_map_unmap_from_file)
> +{
> +	const u64 size = SZ_1G;
> +	struct dma_region region;
> +	struct timespec start;
> +	u64 unmapped;
> +	int rc, fd;
> +
> +	region.vaddr = setup_memfd(&fd, size, variant->mmap_flags, variant->memfd_flags);
> +	ASSERT_NE(region.vaddr, MAP_FAILED);
> +
> +	region.iova = iova_allocator_alloc(self->iova_allocator, size);
> +	region.size = size;
> +
> +	timer_start(&start);
> +	if (strcmp(variant->iommu_mode, MODE_IOMMUFD) == 0) {
> +		iommufd_map_file(self->iommu, &region, fd);
> +	} else {
> +		iommu_map(self->iommu, &region);
> +	}
> +	printf("Mapped HVA %p (size %luG) at IOVA 0x%lx in %.2lfms\n",
> +	       region.vaddr, size / SZ_1G, region.iova, timer_elapsed_ms(start));
> +	ASSERT_EQ(region.iova, to_iova(self->device, region.vaddr));
> +
> +	timer_start(&start);
> +	rc = __iommu_unmap(self->iommu, &region, &unmapped);
> +	printf("Unmapped IOVA 0x%lx in %.2lfms\n", region.iova, timer_elapsed_ms(start));
> +	ASSERT_EQ(rc, 0);
> +	ASSERT_EQ(unmapped, region.size);

Maybe we should add this assert to iommu_unmap() and you can just use
iommu_unmap() here and in the other test?

> +
> +	teardown_memfd(fd, size, region.vaddr);
> +}
> +
>  int main(int argc, char *argv[])
>  {
>  	device_bdf = vfio_selftests_get_bdf(&argc, argv);
> -- 
> 2.55.0.rc0.799.gd6f94ed593-goog
> 

  parent reply	other threads:[~2026-07-14 16:53 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-07-01 20:33 [PATCH 0/4] Introduce vfio_dma_mapping_perf_test Aaron Lewis
2026-07-01 20:33 ` [PATCH 1/4] vfio: selftests: " Aaron Lewis
2026-07-01 20:44   ` sashiko-bot
2026-07-14 16:44   ` David Matlack
2026-07-01 20:33 ` [PATCH 2/4] vfio: selftests: Add memfd test to vfio_dma_mapping_perf_test Aaron Lewis
2026-07-01 20:43   ` sashiko-bot
2026-07-14 16:53   ` David Matlack [this message]
2026-07-01 20:33 ` [PATCH 3/4] vfio: selftests: Allow a size for vfio_dma_mapping_perf_test Aaron Lewis
2026-07-01 20:44   ` sashiko-bot
2026-07-14 17:49   ` David Matlack
2026-07-01 20:33 ` [PATCH 4/4] vfio: selftests: Allow the flag MAP_POPULATE to be set on the cmdline Aaron Lewis
2026-07-07 20:30   ` David Matlack
2026-07-07 20:32     ` David Matlack
2026-07-14 16:27 ` [PATCH 0/4] Introduce vfio_dma_mapping_perf_test David Matlack

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=alZpc7oSwtw7hExk@google.com \
    --to=dmatlack@google.com \
    --cc=aaronlewis@google.com \
    --cc=alex@shazbot.org \
    --cc=jgg@nvidia.com \
    --cc=kvm@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox