ceph/src/seastar/dpdk/lib/librte_eal/common/include/rte_vfio.h

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2017 6WIND S.A.
   3  */
   4
   5 #ifndef _RTE_VFIO_H_
   6 #define _RTE_VFIO_H_
   7
   8 /**
   9  * @file
  10  * RTE VFIO. This library provides various VFIO related utility functions.
  11  */
  12
  13 #ifdef __cplusplus
  14 extern "C" {
  15 #endif
  16
  17 #include <stdint.h>
  18
  19 /*
  20  * determine if VFIO is present on the system
  21  */
  22 #if !defined(VFIO_PRESENT) && defined(RTE_EAL_VFIO)
  23 #include <linux/version.h>
  24 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 6, 0)
  25 #define VFIO_PRESENT
  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 */
  31
  32 #ifdef VFIO_PRESENT
  33
  34 #include <linux/vfio.h>
  35
  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"
  44
  45 /* NOIOMMU is defined from kernel version 4.5 onwards */
  46 #ifdef VFIO_NOIOMMU_IOMMU
  47 #define RTE_VFIO_NOIOMMU VFIO_NOIOMMU_IOMMU
  48 #else
  49 #define RTE_VFIO_NOIOMMU 8
  50 #endif
  51
  52 /*
  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.
  55  */
  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)
  59 #else
  60 #define RTE_VFIO_INFO_FLAG_CAPS (1 << 3)
  61 #define VFIO_CAP_OFFSET(x) (x->resv)
  62 struct vfio_info_cap_header {
  63         uint16_t id;
  64         uint16_t version;
  65         uint32_t next;
  66 };
  67 #endif
  68
  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
  72 #else
  73 #define RTE_VFIO_CAP_MSIX_MAPPABLE 3
  74 #endif
  75
  76 #else /* not VFIO_PRESENT */
  77
  78 /* we don't need an actual definition, only pointer is used */
  79 struct vfio_device_info;
  80
  81 #endif /* VFIO_PRESENT */
  82
  83 #define RTE_VFIO_DEFAULT_CONTAINER_FD (-1)
  84
  85 /**
  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.
  89  *
  90  * This function is only relevant to linux and will return
  91  * an error on BSD.
  92  *
  93  * @param sysfs_base
  94  *   sysfs path prefix.
  95  *
  96  * @param dev_addr
  97  *   device location.
  98  *
  99  * @param vfio_dev_fd
 100  *   VFIO fd.
 101  *
 102  * @param device_info
 103  *   Device information.
 104  *
 105  * @return
 106  *   0 on success.
 107  *   <0 on failure.
 108  *   >1 if the device cannot be managed this way.
 109  */
 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);
 112
 113 /**
 114  * Release a device mapped to a VFIO-managed I/O MMU group.
 115  *
 116  * This function is only relevant to linux and will return
 117  * an error on BSD.
 118  *
 119  * @param sysfs_base
 120  *   sysfs path prefix.
 121  *
 122  * @param dev_addr
 123  *   device location.
 124  *
 125  * @param fd
 126  *   VFIO fd.
 127  *
 128  * @return
 129  *   0 on success.
 130  *   <0 on failure.
 131  */
 132 int rte_vfio_release_device(const char *sysfs_base, const char *dev_addr, int fd);
 133
 134 /**
 135  * Enable a VFIO-related kmod.
 136  *
 137  * This function is only relevant to linux and will return
 138  * an error on BSD.
 139  *
 140  * @param modname
 141  *   kernel module name.
 142  *
 143  * @return
 144  *   0 on success.
 145  *   <0 on failure.
 146  */
 147 int rte_vfio_enable(const char *modname);
 148
 149 /**
 150  * Check whether a VFIO-related kmod is enabled.
 151  *
 152  * This function is only relevant to linux and will return
 153  * an error on BSD.
 154  *
 155  * @param modname
 156  *   kernel module name.
 157  *
 158  * @return
 159  *   !0 if true.
 160  *   0 otherwise.
 161  */
 162 int rte_vfio_is_enabled(const char *modname);
 163
 164 /**
 165  * Whether VFIO NOIOMMU mode is enabled.
 166  *
 167  * This function is only relevant to linux and will return
 168  * an error on BSD.
 169  *
 170  * @return
 171  *   !0 if true.
 172  *   0 otherwise.
 173  */
 174 int rte_vfio_noiommu_is_enabled(void);
 175
 176 /**
 177  * Remove group fd from internal VFIO group fd array/
 178  *
 179  * This function is only relevant to linux and will return
 180  * an error on BSD.
 181  *
 182  * @param vfio_group_fd
 183  *   VFIO Group FD.
 184  *
 185  * @return
 186  *   0 on success.
 187  *   <0 on failure.
 188  */
 189 int
 190 rte_vfio_clear_group(int vfio_group_fd);
 191
 192 /**
 193  * Map memory region for use with VFIO.
 194  *
 195  * @note Require at least one device to be attached at the time of
 196  *       mapping. DMA maps done via this API will only apply to default
 197  *       container and will not apply to any of the containers created
 198  *       via rte_vfio_container_create().
 199  *
 200  * @param vaddr
 201  *   Starting virtual address of memory to be mapped.
 202  *
 203  * @param iova
 204  *   Starting IOVA address of memory to be mapped.
 205  *
 206  * @param len
 207  *   Length of memory segment being mapped.
 208  *
 209  * @return
 210  *   0 if success.
 211  *   -1 on error.
 212  */
 213 int
 214 rte_vfio_dma_map(uint64_t vaddr, uint64_t iova, uint64_t len);
 215
 216
 217 /**
 218  * Unmap memory region from VFIO.
 219  *
 220  * @param vaddr
 221  *   Starting virtual address of memory to be unmapped.
 222  *
 223  * @param iova
 224  *   Starting IOVA address of memory to be unmapped.
 225  *
 226  * @param len
 227  *   Length of memory segment being unmapped.
 228  *
 229  * @return
 230  *   0 if success.
 231  *   -1 on error.
 232  */
 233
 234 int
 235 rte_vfio_dma_unmap(uint64_t vaddr, uint64_t iova, uint64_t len);
 236 /**
 237  * Parse IOMMU group number for a device
 238  *
 239  * This function is only relevant to linux and will return
 240  * an error on BSD.
 241  *
 242  * @param sysfs_base
 243  *   sysfs path prefix.
 244  *
 245  * @param dev_addr
 246  *   device location.
 247  *
 248  * @param iommu_group_num
 249  *   iommu group number
 250  *
 251  * @return
 252  *  >0 on success
 253  *   0 for non-existent group or VFIO
 254  *  <0 for errors
 255  */
 256 int
 257 rte_vfio_get_group_num(const char *sysfs_base,
 258                       const char *dev_addr, int *iommu_group_num);
 259
 260 /**
 261  * Open a new VFIO container fd
 262  *
 263  * This function is only relevant to linux and will return
 264  * an error on BSD.
 265  *
 266  * @return
 267  *  > 0 container fd
 268  *  < 0 for errors
 269  */
 270 int
 271 rte_vfio_get_container_fd(void);
 272
 273 /**
 274  * Open VFIO group fd or get an existing one
 275  *
 276  * This function is only relevant to linux and will return
 277  * an error on BSD.
 278  *
 279  * @param iommu_group_num
 280  *   iommu group number
 281  *
 282  * @return
 283  *  > 0 group fd
 284  *  < 0 for errors
 285  */
 286 int
 287 rte_vfio_get_group_fd(int iommu_group_num);
 288
 289 /**
 290  * Create a new container for device binding.
 291  *
 292  * @note Any newly allocated DPDK memory will not be mapped into these
 293  *       containers by default, user needs to manage DMA mappings for
 294  *       any container created by this API.
 295  *
 296  * @note When creating containers using this API, the container will only be
 297  *       available in the process that has created it. Sharing containers and
 298  *       devices between multiple processes is not supported.
 299  *
 300  * @return
 301  *   the container fd if successful
 302  *   <0 if failed
 303  */
 304 int
 305 rte_vfio_container_create(void);
 306
 307 /**
 308  * Destroy the container, unbind all vfio groups within it.
 309  *
 310  * @param container_fd
 311  *   the container fd to destroy
 312  *
 313  * @return
 314  *    0 if successful
 315  *   <0 if failed
 316  */
 317 int
 318 rte_vfio_container_destroy(int container_fd);
 319
 320 /**
 321  * Bind a IOMMU group to a container.
 322  *
 323  * @param container_fd
 324  *   the container's fd
 325  *
 326  * @param iommu_group_num
 327  *   the iommu group number to bind to container
 328  *
 329  * @return
 330  *   group fd if successful
 331  *   <0 if failed
 332  */
 333 int
 334 rte_vfio_container_group_bind(int container_fd, int iommu_group_num);
 335
 336 /**
 337  * Unbind a IOMMU group from a container.
 338  *
 339  * @param container_fd
 340  *   the container fd of container
 341  *
 342  * @param iommu_group_num
 343  *   the iommu group number to delete from container
 344  *
 345  * @return
 346  *    0 if successful
 347  *   <0 if failed
 348  */
 349 int
 350 rte_vfio_container_group_unbind(int container_fd, int iommu_group_num);
 351
 352 /**
 353  * Perform DMA mapping for devices in a container.
 354  *
 355  * @param container_fd
 356  *   the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to
 357  *   use the default container.
 358  *
 359  * @param vaddr
 360  *   Starting virtual address of memory to be mapped.
 361  *
 362  * @param iova
 363  *   Starting IOVA address of memory to be mapped.
 364  *
 365  * @param len
 366  *   Length of memory segment being mapped.
 367  *
 368  * @return
 369  *    0 if successful
 370  *   <0 if failed
 371  */
 372 int
 373 rte_vfio_container_dma_map(int container_fd, uint64_t vaddr,
 374                 uint64_t iova, uint64_t len);
 375
 376 /**
 377  * Perform DMA unmapping for devices in a container.
 378  *
 379  * @param container_fd
 380  *   the specified container fd. Use RTE_VFIO_DEFAULT_CONTAINER_FD to
 381  *   use the default container.
 382  *
 383  * @param vaddr
 384  *   Starting virtual address of memory to be unmapped.
 385  *
 386  * @param iova
 387  *   Starting IOVA address of memory to be unmapped.
 388  *
 389  * @param len
 390  *   Length of memory segment being unmapped.
 391  *
 392  * @return
 393  *    0 if successful
 394  *   <0 if failed
 395  */
 396 int
 397 rte_vfio_container_dma_unmap(int container_fd, uint64_t vaddr,
 398                 uint64_t iova, uint64_t len);
 399
 400 #ifdef __cplusplus
 401 }
 402 #endif
 403
 404 #endif /* _RTE_VFIO_H_ */