Re: [PATCH v2 2/5] vfio: selftests: Skip tests when the IOMMU mode is not available

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] vfio: selftests: iommu_mode_is_available() misses CONFIG_VFIO_GROUP check
--

commit 254a1695e07cd0eb0e8a8f399f3d939f79af60d2
Author: David Matlack <dmatlack@google.com>

vfio: selftests: Skip tests when the IOMMU mode is not available

This commit adds iommu_mode_is_available(), iommu_mode_unavailable_msg(),
and SKIP_IF_IOMMU_MODE_UNAVAILABLE() to skip VFIO selftest variants when
the requested IOMMU mode is unavailable on the system.

> diff --git a/tools/testing/selftests/vfio/lib/iommu.c b/tools/testing/selftests/vfio/lib/iommu.c
> index 0393536370e79..526bb21f27666 100644
> --- a/tools/testing/selftests/vfio/lib/iommu.c
> +++ b/tools/testing/selftests/vfio/lib/iommu.c
> @@ -111,6 +111,26 @@ static const struct iommu_mode *iommu_mode_lookup(const char *iommu_mode)
>  	VFIO_FAIL("Unrecognized IOMMU mode: %s\n", iommu_mode);
>  }
>  
> +bool iommu_mode_is_available(const char *iommu_mode)
> +{
> +	if (!iommu_mode)
> +		return iommu_mode_detect();
> +
> +	return iommu_mode_probe(iommu_mode_lookup(iommu_mode));
> +}

[Severity: Medium]
Can iommu_mode_is_available() return true for iommufd_compat_type1,
iommufd_compat_type1v2, and NULL when CONFIG_VFIO_GROUP is disabled (for
example on a cdev-only VFIO kernel with CONFIG_IOMMUFD=y and
CONFIG_VFIO_DEVICE_CDEV=y) or when the vfio module is not loaded?

In tools/testing/selftests/vfio/lib/iommu.c, iommu_mode_probe() only checks
for /dev/vfio/devices when !mode->container_path:

	/* iommufd mode requires VFIO device cdevs (CONFIG_VFIO_DEVICE_CDEV). */
	if (!mode->container_path && access("/dev/vfio/devices", F_OK))
		return false;

	fd = open(path, O_RDWR);
	if (fd < 0)
		return false;

	available = !mode->container_path ||
		    ioctl(fd, VFIO_CHECK_EXTENSION, mode->iommu_type) > 0;

For iommufd_compat_type1 and iommufd_compat_type1v2, mode->container_path
is "/dev/iommu". Because drivers/iommu/iommufd/vfio_compat.c is compiled
into iommufd unconditionally regardless of CONFIG_VFIO_GROUP, opening
/dev/iommu and calling VFIO_CHECK_EXTENSION succeeds whenever iommufd is
present.

However, vfio_pci_device_init() and device_init() in
vfio_pci_sriov_uapi_test.c route all modes with a non-NULL
mode->container_path through vfio_pci_container_setup() ->
vfio_pci_group_setup(), which opens /dev/vfio/<group_id> and asserts on
failure when CONFIG_VFIO_GROUP is disabled.

Should iommu_mode_probe() also verify that VFIO group support is available
(such as checking access("/sys/class/vfio", F_OK) == 0) when
mode->container_path is non-NULL?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261103002205.3348519-1-dmatlack@google.com?part=2