vfio: selftests: Add iommufd mode

Add a new IOMMU mode for using iommufd directly. In this mode userspace
opens /dev/iommu and binds it to a device FD acquired through
/dev/vfio/devices/vfioX.

Acked-by: Shuah Khan <[e-mail]>
Signed-off-by: David Matlack <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Alex Williamson <[e-mail]>

tools/testing/selftests/vfio/lib/include/vfio_util.h

···
 FIXTURE_VARIANT_ADD_IOMMU_MODE(vfio_type1_iommu, ##__VA_ARGS__); \
 FIXTURE_VARIANT_ADD_IOMMU_MODE(vfio_type1v2_iommu, ##__VA_ARGS__); \
 FIXTURE_VARIANT_ADD_IOMMU_MODE(iommufd_compat_type1, ##__VA_ARGS__); \
-FIXTURE_VARIANT_ADD_IOMMU_MODE(iommufd_compat_type1v2, ##__VA_ARGS__)
+FIXTURE_VARIANT_ADD_IOMMU_MODE(iommufd_compat_type1v2, ##__VA_ARGS__); \
+FIXTURE_VARIANT_ADD_IOMMU_MODE(iommufd, ##__VA_ARGS__)
 
 struct vfio_pci_bar {
 	struct vfio_region_info info;
···
 	const struct vfio_iommu_mode *iommu_mode;
 	int group_fd;
 	int container_fd;
+
+	int iommufd;
+	u32 ioas_id;
 
 	struct vfio_device_info info;
 	struct vfio_region_info config_space;

tools/testing/selftests/vfio/lib/vfio_pci_device.c

···
 #include <sys/ioctl.h>
 #include <sys/mman.h>
 
+#include <uapi/linux/types.h>
 #include <linux/limits.h>
 #include <linux/mman.h>
 #include <linux/types.h>
 #include <linux/vfio.h>
+#include <linux/iommufd.h>
 
 #include "../../../kselftest.h"
 #include <vfio_util.h>
···
 	ioctl_assert(device->fd, VFIO_DEVICE_GET_IRQ_INFO, irq_info);
 }
 
-void vfio_pci_dma_map(struct vfio_pci_device *device,
-		      struct vfio_dma_region *region)
+static void vfio_iommu_dma_map(struct vfio_pci_device *device,
+			       struct vfio_dma_region *region)
 {
-	struct vfio_iommu_type1_dma_map map = {
-		.argsz = sizeof(map),
+	struct vfio_iommu_type1_dma_map args = {
+		.argsz = sizeof(args),
 		.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE,
 		.vaddr = (u64)region->vaddr,
 		.iova = region->iova,
 		.size = region->size,
 	};
 
-	ioctl_assert(device->container_fd, VFIO_IOMMU_MAP_DMA, &map);
+	ioctl_assert(device->container_fd, VFIO_IOMMU_MAP_DMA, &args);
+}
+
+static void iommufd_dma_map(struct vfio_pci_device *device,
+			    struct vfio_dma_region *region)
+{
+	struct iommu_ioas_map args = {
+		.size = sizeof(args),
+		.flags = IOMMU_IOAS_MAP_READABLE |
+			 IOMMU_IOAS_MAP_WRITEABLE |
+			 IOMMU_IOAS_MAP_FIXED_IOVA,
+		.user_va = (u64)region->vaddr,
+		.iova = region->iova,
+		.length = region->size,
+		.ioas_id = device->ioas_id,
+	};
+
+	ioctl_assert(device->iommufd, IOMMU_IOAS_MAP, &args);
+}
+
+void vfio_pci_dma_map(struct vfio_pci_device *device,
+		      struct vfio_dma_region *region)
+{
+	if (device->iommufd)
+		iommufd_dma_map(device, region);
+	else
+		vfio_iommu_dma_map(device, region);
 
 	list_add(&region->link, &device->dma_regions);
+}
+
+static void vfio_iommu_dma_unmap(struct vfio_pci_device *device,
+				 struct vfio_dma_region *region)
+{
+	struct vfio_iommu_type1_dma_unmap args = {
+		.argsz = sizeof(args),
+		.iova = region->iova,
+		.size = region->size,
+	};
+
+	ioctl_assert(device->container_fd, VFIO_IOMMU_UNMAP_DMA, &args);
+}
+
+static void iommufd_dma_unmap(struct vfio_pci_device *device,
+			      struct vfio_dma_region *region)
+{
+	struct iommu_ioas_unmap args = {
+		.size = sizeof(args),
+		.iova = region->iova,
+		.length = region->size,
+		.ioas_id = device->ioas_id,
+	};
+
+	ioctl_assert(device->iommufd, IOMMU_IOAS_UNMAP, &args);
 }
 
 void vfio_pci_dma_unmap(struct vfio_pci_device *device,
 			struct vfio_dma_region *region)
 {
-	struct vfio_iommu_type1_dma_unmap unmap = {
-		.argsz = sizeof(unmap),
-		.iova = region->iova,
-		.size = region->size,
-	};
-
-	ioctl_assert(device->container_fd, VFIO_IOMMU_UNMAP_DMA, &unmap);
+	if (device->iommufd)
+		iommufd_dma_unmap(device, region);
+	else
+		vfio_iommu_dma_unmap(device, region);
 
 	list_del(&region->link);
 }
···
 	return group;
 }
 
-static void vfio_pci_container_setup(struct vfio_pci_device *device)
-{
-	const char *path = device->iommu_mode->container_path;
-	int version;
-
-	device->container_fd = open(path, O_RDWR);
-	VFIO_ASSERT_GE(device->container_fd, 0, "open(%s) failed\n", path);
-
-	version = ioctl(device->container_fd, VFIO_GET_API_VERSION);
-	VFIO_ASSERT_EQ(version, VFIO_API_VERSION);
-}
-
 static void vfio_pci_group_setup(struct vfio_pci_device *device, const char *bdf)
 {
 	struct vfio_group_status group_status = {
···
 	ioctl_assert(device->group_fd, VFIO_GROUP_SET_CONTAINER, &device->container_fd);
 }
 
-static void vfio_pci_iommu_setup(struct vfio_pci_device *device)
+static void vfio_pci_container_setup(struct vfio_pci_device *device, const char *bdf)
 {
 	unsigned long iommu_type = device->iommu_mode->iommu_type;
+	const char *path = device->iommu_mode->container_path;
+	int version;
 	int ret;
 
-	INIT_LIST_HEAD(&device->dma_regions);
+	device->container_fd = open(path, O_RDWR);
+	VFIO_ASSERT_GE(device->container_fd, 0, "open(%s) failed\n", path);
+
+	version = ioctl(device->container_fd, VFIO_GET_API_VERSION);
+	VFIO_ASSERT_EQ(version, VFIO_API_VERSION, "Unsupported version: %d\n", version);
+
+	vfio_pci_group_setup(device, bdf);
 
 	ret = ioctl(device->container_fd, VFIO_CHECK_EXTENSION, iommu_type);
 	VFIO_ASSERT_GT(ret, 0, "VFIO IOMMU type %lu not supported\n", iommu_type);
 
 	ioctl_assert(device->container_fd, VFIO_SET_IOMMU, (void *)iommu_type);
-}
-
-static void vfio_pci_device_setup(struct vfio_pci_device *device, const char *bdf)
-{
-	int i;
 
 	device->fd = ioctl(device->group_fd, VFIO_GROUP_GET_DEVICE_FD, bdf);
 	VFIO_ASSERT_GE(device->fd, 0);
+}
+
+static void vfio_pci_device_setup(struct vfio_pci_device *device)
+{
+	int i;
 
 	device->info.argsz = sizeof(device->info);
 	ioctl_assert(device->fd, VFIO_DEVICE_GET_INFO, &device->info);
···
 		.container_path = "/dev/iommu",
 		.iommu_type = VFIO_TYPE1v2_IOMMU,
 	},
+	{
+		.name = "iommufd",
+	},
 };
 
 const char *default_iommu_mode = "vfio_type1_iommu";
···
 	VFIO_FAIL("Unrecognized IOMMU mode: %s\n", iommu_mode);
 }
 
+static void vfio_device_bind_iommufd(int device_fd, int iommufd)
+{
+	struct vfio_device_bind_iommufd args = {
+		.argsz = sizeof(args),
+		.iommufd = iommufd,
+	};
+
+	ioctl_assert(device_fd, VFIO_DEVICE_BIND_IOMMUFD, &args);
+}
+
+static u32 iommufd_ioas_alloc(int iommufd)
+{
+	struct iommu_ioas_alloc args = {
+		.size = sizeof(args),
+	};
+
+	ioctl_assert(iommufd, IOMMU_IOAS_ALLOC, &args);
+	return args.out_ioas_id;
+}
+
+static void vfio_device_attach_iommufd_pt(int device_fd, u32 pt_id)
+{
+	struct vfio_device_attach_iommufd_pt args = {
+		.argsz = sizeof(args),
+		.pt_id = pt_id,
+	};
+
+	ioctl_assert(device_fd, VFIO_DEVICE_ATTACH_IOMMUFD_PT, &args);
+}
+
+static void vfio_pci_iommufd_setup(struct vfio_pci_device *device, const char *bdf)
+{
+	const char *cdev_path = vfio_pci_get_cdev_path(bdf);
+
+	device->fd = open(cdev_path, O_RDWR);
+	VFIO_ASSERT_GE(device->fd, 0);
+	free((void *)cdev_path);
+
+	/*
+	 * Require device->iommufd to be >0 so that a simple non-0 check can be
+	 * used to check if iommufd is enabled. In practice open() will never
+	 * return 0 unless stdin is closed.
+	 */
+	device->iommufd = open("/dev/iommu", O_RDWR);
+	VFIO_ASSERT_GT(device->iommufd, 0);
+
+	vfio_device_bind_iommufd(device->fd, device->iommufd);
+	device->ioas_id = iommufd_ioas_alloc(device->iommufd);
+	vfio_device_attach_iommufd_pt(device->fd, device->ioas_id);
+}
+
 struct vfio_pci_device *vfio_pci_device_init(const char *bdf, const char *iommu_mode)
 {
 	struct vfio_pci_device *device;
···
 	device = calloc(1, sizeof(*device));
 	VFIO_ASSERT_NOT_NULL(device);
 
+	INIT_LIST_HEAD(&device->dma_regions);
+
 	device->iommu_mode = lookup_iommu_mode(iommu_mode);
 
-	vfio_pci_container_setup(device);
-	vfio_pci_group_setup(device, bdf);
-	vfio_pci_iommu_setup(device);
-	vfio_pci_device_setup(device, bdf);
+	if (device->iommu_mode->container_path)
+		vfio_pci_container_setup(device, bdf);
+	else
+		vfio_pci_iommufd_setup(device, bdf);
 
+	vfio_pci_device_setup(device);
 	vfio_pci_driver_probe(device);
 
 	return device;
···
 		VFIO_ASSERT_EQ(close(device->msi_eventfds[i]), 0);
 	}
 
-	VFIO_ASSERT_EQ(close(device->group_fd), 0);
-	VFIO_ASSERT_EQ(close(device->container_fd), 0);
+	if (device->iommufd) {
+		VFIO_ASSERT_EQ(close(device->iommufd), 0);
+	} else {
+		VFIO_ASSERT_EQ(close(device->group_fd), 0);
+		VFIO_ASSERT_EQ(close(device->container_fd), 0);
+	}
 
 	free(device);
 }