1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2017 6WIND S.A.
10 * RTE VFIO. This library provides various VFIO related utility functions.
20 * determine if VFIO is present on the system
22 #if !defined(VFIO_PRESENT) && defined(RTE_EAL_VFIO)
23 #include <linux/version.h>
24 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 6, 0)
26 #endif /* kernel version >= 3.6.0 */
27 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 0, 0)
28 #define HAVE_VFIO_DEV_REQ_INTERFACE
29 #endif /* kernel version >= 4.0.0 */
30 #endif /* RTE_EAL_VFIO */
34 #include <linux/vfio.h>
36 #define VFIO_DIR "/dev/vfio"
37 #define VFIO_CONTAINER_PATH "/dev/vfio/vfio"
38 #define VFIO_GROUP_FMT "/dev/vfio/%u"
39 #define VFIO_NOIOMMU_GROUP_FMT "/dev/vfio/noiommu-%u"
40 #define VFIO_GET_REGION_ADDR(x) ((uint64_t) x << 40ULL)
41 #define VFIO_GET_REGION_IDX(x) (x >> 40)
42 #define VFIO_NOIOMMU_MODE \
43 "/sys/module/vfio/parameters/enable_unsafe_noiommu_mode"
45 /* NOIOMMU is defined from kernel version 4.5 onwards */
46 #ifdef VFIO_NOIOMMU_IOMMU
47 #define RTE_VFIO_NOIOMMU VFIO_NOIOMMU_IOMMU
49 #define RTE_VFIO_NOIOMMU 8
53 * capabilities are only supported on kernel 4.6+. there were also some API
54 * changes as well, so add a macro to get cap offset.
56 #ifdef VFIO_REGION_INFO_FLAG_CAPS
57 #define RTE_VFIO_INFO_FLAG_CAPS VFIO_REGION_INFO_FLAG_CAPS
58 #define VFIO_CAP_OFFSET(x) (x->cap_offset)
60 #define RTE_VFIO_INFO_FLAG_CAPS (1 << 3)
61 #define VFIO_CAP_OFFSET(x) (x->resv)
62 struct vfio_info_cap_header {
69 /* kernels 4.16+ can map BAR containing MSI-X table */
70 #ifdef VFIO_REGION_INFO_CAP_MSIX_MAPPABLE
71 #define RTE_VFIO_CAP_MSIX_MAPPABLE VFIO_REGION_INFO_CAP_MSIX_MAPPABLE
73 #define RTE_VFIO_CAP_MSIX_MAPPABLE 3
76 #else /* not VFIO_PRESENT */
78 /* we don't need an actual definition, only pointer is used */
79 struct vfio_device_info;
81 #endif /* VFIO_PRESENT */
83 #define RTE_VFIO_DEFAULT_CONTAINER_FD (-1)
86 * Setup vfio_cfg for the device identified by its address.
87 * It discovers the configured I/O MMU groups or sets a new one for the device.
88 * If a new groups is assigned, the DMA mapping is performed.
90 * This function is only relevant to linux and will return
103 * Device information.
108 * >1 if the device cannot be managed this way.
110 int rte_vfio_setup_device(const char *sysfs_base, const char *dev_addr,
111 int *vfio_dev_fd, struct vfio_device_info *device_info);
114 * Release a device mapped to a VFIO-managed I/O MMU group.
116 * This function is only relevant to linux and will return
132 int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd);
135 * Enable a VFIO-related kmod.
137 * This function is only relevant to linux and will return
141 * kernel module name.
147 int rte_vfio_enable(const char *modname);
150 * Check whether a VFIO-related kmod is enabled.
152 * This function is only relevant to linux and will return
156 * kernel module name.
162 int rte_vfio_is_enabled(const char *modname);
165 * Whether VFIO NOIOMMU mode is enabled.
167 * This function is only relevant to linux and will return
175 int rte_vfio_noiommu_is_enabled(void);
178 * Remove group fd from internal VFIO group fd array/
180 * This function is only relevant to linux and will return
183 * @param vfio_group_fd
191 rte_vfio_clear_group(int vfio_group_fd);
194 * Parse IOMMU group number for a device
196 * This function is only relevant to linux and will return
205 * @param iommu_group_num
210 * 0 for non-existent group or VFIO
214 rte_vfio_get_group_num(const char *sysfs_base,
215 const char *dev_addr, int *iommu_group_num);
218 * Open a new VFIO container fd
220 * This function is only relevant to linux and will return
228 rte_vfio_get_container_fd(void);
231 * Open VFIO group fd or get an existing one
233 * This function is only relevant to linux and will return
236 * @param iommu_group_num
244 rte_vfio_get_group_fd(int iommu_group_num);
247 * Create a new container for device binding.
249 * @note Any newly allocated DPDK memory will not be mapped into these
250 * containers by default, user needs to manage DMA mappings for
251 * any container created by this API.
253 * @note When creating containers using this API, the container will only be
254 * available in the process that has created it. Sharing containers and
255 * devices between multiple processes is not supported.
258 * the container fd if successful
262 rte_vfio_container_create(void);
265 * Destroy the container, unbind all vfio groups within it.
267 * @param container_fd
268 * the container fd to destroy
275 rte_vfio_container_destroy(int container_fd);
278 * Bind a IOMMU group to a container.
280 * @param container_fd
283 * @param iommu_group_num
284 * the iommu group number to bind to container
287 * group fd if successful
291 rte_vfio_container_group_bind(int container_fd, int iommu_group_num);
294 * Unbind a IOMMU group from a container.
296 * @param container_fd
297 * the container fd of container
299 * @param iommu_group_num
300 * the iommu group number to delete from container
307 rte_vfio_container_group_unbind(int container_fd, int iommu_group_num);
310 * Perform DMA mapping for devices in a container.
312 * @param container_fd
313 * the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to
314 * use the default container.
317 * Starting virtual address of memory to be mapped.
320 * Starting IOVA address of memory to be mapped.
323 * Length of memory segment being mapped.
330 rte_vfio_container_dma_map(int container_fd, uint64_t vaddr,
331 uint64_t iova, uint64_t len);
334 * Perform DMA unmapping for devices in a container.
336 * @param container_fd
337 * the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to
338 * use the default container.
341 * Starting virtual address of memory to be unmapped.
344 * Starting IOVA address of memory to be unmapped.
347 * Length of memory segment being unmapped.
354 rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr,
355 uint64_t iova, uint64_t len);
361 #endif /* _RTE_VFIO_H_ */