linux-headers/linux/iommufd.h

   1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
   2 /* Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES.
   3  */
   4 #ifndef _IOMMUFD_H
   5 #define _IOMMUFD_H
   6
   7 #include <linux/types.h>
   8 #include <linux/ioctl.h>
   9
  10 #define IOMMUFD_TYPE (';')
  11
  12 /**
  13  * DOC: General ioctl format
  14  *
  15  * The ioctl interface follows a general format to allow for extensibility. Each
  16  * ioctl is passed in a structure pointer as the argument providing the size of
  17  * the structure in the first u32. The kernel checks that any structure space
  18  * beyond what it understands is 0. This allows userspace to use the backward
  19  * compatible portion while consistently using the newer, larger, structures.
  20  *
  21  * ioctls use a standard meaning for common errnos:
  22  *
  23  *  - ENOTTY: The IOCTL number itself is not supported at all
  24  *  - E2BIG: The IOCTL number is supported, but the provided structure has
  25  *    non-zero in a part the kernel does not understand.
  26  *  - EOPNOTSUPP: The IOCTL number is supported, and the structure is
  27  *    understood, however a known field has a value the kernel does not
  28  *    understand or support.
  29  *  - EINVAL: Everything about the IOCTL was understood, but a field is not
  30  *    correct.
  31  *  - ENOENT: An ID or IOVA provided does not exist.
  32  *  - ENOMEM: Out of memory.
  33  *  - EOVERFLOW: Mathematics overflowed.
  34  *
  35  * As well as additional errnos, within specific ioctls.
  36  */
  37 enum {
  38         IOMMUFD_CMD_BASE = 0x80,
  39         IOMMUFD_CMD_DESTROY = IOMMUFD_CMD_BASE,
  40         IOMMUFD_CMD_IOAS_ALLOC,
  41         IOMMUFD_CMD_IOAS_ALLOW_IOVAS,
  42         IOMMUFD_CMD_IOAS_COPY,
  43         IOMMUFD_CMD_IOAS_IOVA_RANGES,
  44         IOMMUFD_CMD_IOAS_MAP,
  45         IOMMUFD_CMD_IOAS_UNMAP,
  46         IOMMUFD_CMD_OPTION,
  47         IOMMUFD_CMD_VFIO_IOAS,
  48         IOMMUFD_CMD_HWPT_ALLOC,
  49         IOMMUFD_CMD_GET_HW_INFO,
  50 };
  51
  52 /**
  53  * struct iommu_destroy - ioctl(IOMMU_DESTROY)
  54  * @size: sizeof(struct iommu_destroy)
  55  * @id: iommufd object ID to destroy. Can be any destroyable object type.
  56  *
  57  * Destroy any object held within iommufd.
  58  */
  59 struct iommu_destroy {
  60         __u32 size;
  61         __u32 id;
  62 };
  63 #define IOMMU_DESTROY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DESTROY)
  64
  65 /**
  66  * struct iommu_ioas_alloc - ioctl(IOMMU_IOAS_ALLOC)
  67  * @size: sizeof(struct iommu_ioas_alloc)
  68  * @flags: Must be 0
  69  * @out_ioas_id: Output IOAS ID for the allocated object
  70  *
  71  * Allocate an IO Address Space (IOAS) which holds an IO Virtual Address (IOVA)
  72  * to memory mapping.
  73  */
  74 struct iommu_ioas_alloc {
  75         __u32 size;
  76         __u32 flags;
  77         __u32 out_ioas_id;
  78 };
  79 #define IOMMU_IOAS_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOC)
  80
  81 /**
  82  * struct iommu_iova_range - ioctl(IOMMU_IOVA_RANGE)
  83  * @start: First IOVA
  84  * @last: Inclusive last IOVA
  85  *
  86  * An interval in IOVA space.
  87  */
  88 struct iommu_iova_range {
  89         __aligned_u64 start;
  90         __aligned_u64 last;
  91 };
  92
  93 /**
  94  * struct iommu_ioas_iova_ranges - ioctl(IOMMU_IOAS_IOVA_RANGES)
  95  * @size: sizeof(struct iommu_ioas_iova_ranges)
  96  * @ioas_id: IOAS ID to read ranges from
  97  * @num_iovas: Input/Output total number of ranges in the IOAS
  98  * @__reserved: Must be 0
  99  * @allowed_iovas: Pointer to the output array of struct iommu_iova_range
 100  * @out_iova_alignment: Minimum alignment required for mapping IOVA
 101  *
 102  * Query an IOAS for ranges of allowed IOVAs. Mapping IOVA outside these ranges
 103  * is not allowed. num_iovas will be set to the total number of iovas and
 104  * the allowed_iovas[] will be filled in as space permits.
 105  *
 106  * The allowed ranges are dependent on the HW path the DMA operation takes, and
 107  * can change during the lifetime of the IOAS. A fresh empty IOAS will have a
 108  * full range, and each attached device will narrow the ranges based on that
 109  * device's HW restrictions. Detaching a device can widen the ranges. Userspace
 110  * should query ranges after every attach/detach to know what IOVAs are valid
 111  * for mapping.
 112  *
 113  * On input num_iovas is the length of the allowed_iovas array. On output it is
 114  * the total number of iovas filled in. The ioctl will return -EMSGSIZE and set
 115  * num_iovas to the required value if num_iovas is too small. In this case the
 116  * caller should allocate a larger output array and re-issue the ioctl.
 117  *
 118  * out_iova_alignment returns the minimum IOVA alignment that can be given
 119  * to IOMMU_IOAS_MAP/COPY. IOVA's must satisfy::
 120  *
 121  *   starting_iova % out_iova_alignment == 0
 122  *   (starting_iova + length) % out_iova_alignment == 0
 123  *
 124  * out_iova_alignment can be 1 indicating any IOVA is allowed. It cannot
 125  * be higher than the system PAGE_SIZE.
 126  */
 127 struct iommu_ioas_iova_ranges {
 128         __u32 size;
 129         __u32 ioas_id;
 130         __u32 num_iovas;
 131         __u32 __reserved;
 132         __aligned_u64 allowed_iovas;
 133         __aligned_u64 out_iova_alignment;
 134 };
 135 #define IOMMU_IOAS_IOVA_RANGES _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_IOVA_RANGES)
 136
 137 /**
 138  * struct iommu_ioas_allow_iovas - ioctl(IOMMU_IOAS_ALLOW_IOVAS)
 139  * @size: sizeof(struct iommu_ioas_allow_iovas)
 140  * @ioas_id: IOAS ID to allow IOVAs from
 141  * @num_iovas: Input/Output total number of ranges in the IOAS
 142  * @__reserved: Must be 0
 143  * @allowed_iovas: Pointer to array of struct iommu_iova_range
 144  *
 145  * Ensure a range of IOVAs are always available for allocation. If this call
 146  * succeeds then IOMMU_IOAS_IOVA_RANGES will never return a list of IOVA ranges
 147  * that are narrower than the ranges provided here. This call will fail if
 148  * IOMMU_IOAS_IOVA_RANGES is currently narrower than the given ranges.
 149  *
 150  * When an IOAS is first created the IOVA_RANGES will be maximally sized, and as
 151  * devices are attached the IOVA will narrow based on the device restrictions.
 152  * When an allowed range is specified any narrowing will be refused, ie device
 153  * attachment can fail if the device requires limiting within the allowed range.
 154  *
 155  * Automatic IOVA allocation is also impacted by this call. MAP will only
 156  * allocate within the allowed IOVAs if they are present.
 157  *
 158  * This call replaces the entire allowed list with the given list.
 159  */
 160 struct iommu_ioas_allow_iovas {
 161         __u32 size;
 162         __u32 ioas_id;
 163         __u32 num_iovas;
 164         __u32 __reserved;
 165         __aligned_u64 allowed_iovas;
 166 };
 167 #define IOMMU_IOAS_ALLOW_IOVAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOW_IOVAS)
 168
 169 /**
 170  * enum iommufd_ioas_map_flags - Flags for map and copy
 171  * @IOMMU_IOAS_MAP_FIXED_IOVA: If clear the kernel will compute an appropriate
 172  *                             IOVA to place the mapping at
 173  * @IOMMU_IOAS_MAP_WRITEABLE: DMA is allowed to write to this mapping
 174  * @IOMMU_IOAS_MAP_READABLE: DMA is allowed to read from this mapping
 175  */
 176 enum iommufd_ioas_map_flags {
 177         IOMMU_IOAS_MAP_FIXED_IOVA = 1 << 0,
 178         IOMMU_IOAS_MAP_WRITEABLE = 1 << 1,
 179         IOMMU_IOAS_MAP_READABLE = 1 << 2,
 180 };
 181
 182 /**
 183  * struct iommu_ioas_map - ioctl(IOMMU_IOAS_MAP)
 184  * @size: sizeof(struct iommu_ioas_map)
 185  * @flags: Combination of enum iommufd_ioas_map_flags
 186  * @ioas_id: IOAS ID to change the mapping of
 187  * @__reserved: Must be 0
 188  * @user_va: Userspace pointer to start mapping from
 189  * @length: Number of bytes to map
 190  * @iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is set
 191  *        then this must be provided as input.
 192  *
 193  * Set an IOVA mapping from a user pointer. If FIXED_IOVA is specified then the
 194  * mapping will be established at iova, otherwise a suitable location based on
 195  * the reserved and allowed lists will be automatically selected and returned in
 196  * iova.
 197  *
 198  * If IOMMU_IOAS_MAP_FIXED_IOVA is specified then the iova range must currently
 199  * be unused, existing IOVA cannot be replaced.
 200  */
 201 struct iommu_ioas_map {
 202         __u32 size;
 203         __u32 flags;
 204         __u32 ioas_id;
 205         __u32 __reserved;
 206         __aligned_u64 user_va;
 207         __aligned_u64 length;
 208         __aligned_u64 iova;
 209 };
 210 #define IOMMU_IOAS_MAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP)
 211
 212 /**
 213  * struct iommu_ioas_copy - ioctl(IOMMU_IOAS_COPY)
 214  * @size: sizeof(struct iommu_ioas_copy)
 215  * @flags: Combination of enum iommufd_ioas_map_flags
 216  * @dst_ioas_id: IOAS ID to change the mapping of
 217  * @src_ioas_id: IOAS ID to copy from
 218  * @length: Number of bytes to copy and map
 219  * @dst_iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is
 220  *            set then this must be provided as input.
 221  * @src_iova: IOVA to start the copy
 222  *
 223  * Copy an already existing mapping from src_ioas_id and establish it in
 224  * dst_ioas_id. The src iova/length must exactly match a range used with
 225  * IOMMU_IOAS_MAP.
 226  *
 227  * This may be used to efficiently clone a subset of an IOAS to another, or as a
 228  * kind of 'cache' to speed up mapping. Copy has an efficiency advantage over
 229  * establishing equivalent new mappings, as internal resources are shared, and
 230  * the kernel will pin the user memory only once.
 231  */
 232 struct iommu_ioas_copy {
 233         __u32 size;
 234         __u32 flags;
 235         __u32 dst_ioas_id;
 236         __u32 src_ioas_id;
 237         __aligned_u64 length;
 238         __aligned_u64 dst_iova;
 239         __aligned_u64 src_iova;
 240 };
 241 #define IOMMU_IOAS_COPY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_COPY)
 242
 243 /**
 244  * struct iommu_ioas_unmap - ioctl(IOMMU_IOAS_UNMAP)
 245  * @size: sizeof(struct iommu_ioas_unmap)
 246  * @ioas_id: IOAS ID to change the mapping of
 247  * @iova: IOVA to start the unmapping at
 248  * @length: Number of bytes to unmap, and return back the bytes unmapped
 249  *
 250  * Unmap an IOVA range. The iova/length must be a superset of a previously
 251  * mapped range used with IOMMU_IOAS_MAP or IOMMU_IOAS_COPY. Splitting or
 252  * truncating ranges is not allowed. The values 0 to U64_MAX will unmap
 253  * everything.
 254  */
 255 struct iommu_ioas_unmap {
 256         __u32 size;
 257         __u32 ioas_id;
 258         __aligned_u64 iova;
 259         __aligned_u64 length;
 260 };
 261 #define IOMMU_IOAS_UNMAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_UNMAP)
 262
 263 /**
 264  * enum iommufd_option - ioctl(IOMMU_OPTION_RLIMIT_MODE) and
 265  *                       ioctl(IOMMU_OPTION_HUGE_PAGES)
 266  * @IOMMU_OPTION_RLIMIT_MODE:
 267  *    Change how RLIMIT_MEMLOCK accounting works. The caller must have privilege
 268  *    to invoke this. Value 0 (default) is user based accouting, 1 uses process
 269  *    based accounting. Global option, object_id must be 0
 270  * @IOMMU_OPTION_HUGE_PAGES:
 271  *    Value 1 (default) allows contiguous pages to be combined when generating
 272  *    iommu mappings. Value 0 disables combining, everything is mapped to
 273  *    PAGE_SIZE. This can be useful for benchmarking.  This is a per-IOAS
 274  *    option, the object_id must be the IOAS ID.
 275  */
 276 enum iommufd_option {
 277         IOMMU_OPTION_RLIMIT_MODE = 0,
 278         IOMMU_OPTION_HUGE_PAGES = 1,
 279 };
 280
 281 /**
 282  * enum iommufd_option_ops - ioctl(IOMMU_OPTION_OP_SET) and
 283  *                           ioctl(IOMMU_OPTION_OP_GET)
 284  * @IOMMU_OPTION_OP_SET: Set the option's value
 285  * @IOMMU_OPTION_OP_GET: Get the option's value
 286  */
 287 enum iommufd_option_ops {
 288         IOMMU_OPTION_OP_SET = 0,
 289         IOMMU_OPTION_OP_GET = 1,
 290 };
 291
 292 /**
 293  * struct iommu_option - iommu option multiplexer
 294  * @size: sizeof(struct iommu_option)
 295  * @option_id: One of enum iommufd_option
 296  * @op: One of enum iommufd_option_ops
 297  * @__reserved: Must be 0
 298  * @object_id: ID of the object if required
 299  * @val64: Option value to set or value returned on get
 300  *
 301  * Change a simple option value. This multiplexor allows controlling options
 302  * on objects. IOMMU_OPTION_OP_SET will load an option and IOMMU_OPTION_OP_GET
 303  * will return the current value.
 304  */
 305 struct iommu_option {
 306         __u32 size;
 307         __u32 option_id;
 308         __u16 op;
 309         __u16 __reserved;
 310         __u32 object_id;
 311         __aligned_u64 val64;
 312 };
 313 #define IOMMU_OPTION _IO(IOMMUFD_TYPE, IOMMUFD_CMD_OPTION)
 314
 315 /**
 316  * enum iommufd_vfio_ioas_op - IOMMU_VFIO_IOAS_* ioctls
 317  * @IOMMU_VFIO_IOAS_GET: Get the current compatibility IOAS
 318  * @IOMMU_VFIO_IOAS_SET: Change the current compatibility IOAS
 319  * @IOMMU_VFIO_IOAS_CLEAR: Disable VFIO compatibility
 320  */
 321 enum iommufd_vfio_ioas_op {
 322         IOMMU_VFIO_IOAS_GET = 0,
 323         IOMMU_VFIO_IOAS_SET = 1,
 324         IOMMU_VFIO_IOAS_CLEAR = 2,
 325 };
 326
 327 /**
 328  * struct iommu_vfio_ioas - ioctl(IOMMU_VFIO_IOAS)
 329  * @size: sizeof(struct iommu_vfio_ioas)
 330  * @ioas_id: For IOMMU_VFIO_IOAS_SET the input IOAS ID to set
 331  *           For IOMMU_VFIO_IOAS_GET will output the IOAS ID
 332  * @op: One of enum iommufd_vfio_ioas_op
 333  * @__reserved: Must be 0
 334  *
 335  * The VFIO compatibility support uses a single ioas because VFIO APIs do not
 336  * support the ID field. Set or Get the IOAS that VFIO compatibility will use.
 337  * When VFIO_GROUP_SET_CONTAINER is used on an iommufd it will get the
 338  * compatibility ioas, either by taking what is already set, or auto creating
 339  * one. From then on VFIO will continue to use that ioas and is not effected by
 340  * this ioctl. SET or CLEAR does not destroy any auto-created IOAS.
 341  */
 342 struct iommu_vfio_ioas {
 343         __u32 size;
 344         __u32 ioas_id;
 345         __u16 op;
 346         __u16 __reserved;
 347 };
 348 #define IOMMU_VFIO_IOAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VFIO_IOAS)
 349
 350 /**
 351  * struct iommu_hwpt_alloc - ioctl(IOMMU_HWPT_ALLOC)
 352  * @size: sizeof(struct iommu_hwpt_alloc)
 353  * @flags: Must be 0
 354  * @dev_id: The device to allocate this HWPT for
 355  * @pt_id: The IOAS to connect this HWPT to
 356  * @out_hwpt_id: The ID of the new HWPT
 357  * @__reserved: Must be 0
 358  *
 359  * Explicitly allocate a hardware page table object. This is the same object
 360  * type that is returned by iommufd_device_attach() and represents the
 361  * underlying iommu driver's iommu_domain kernel object.
 362  *
 363  * A HWPT will be created with the IOVA mappings from the given IOAS.
 364  */
 365 struct iommu_hwpt_alloc {
 366         __u32 size;
 367         __u32 flags;
 368         __u32 dev_id;
 369         __u32 pt_id;
 370         __u32 out_hwpt_id;
 371         __u32 __reserved;
 372 };
 373 #define IOMMU_HWPT_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_ALLOC)
 374
 375 /**
 376  * struct iommu_hw_info_vtd - Intel VT-d hardware information
 377  *
 378  * @flags: Must be 0
 379  * @__reserved: Must be 0
 380  *
 381  * @cap_reg: Value of Intel VT-d capability register defined in VT-d spec
 382  *           section 11.4.2 Capability Register.
 383  * @ecap_reg: Value of Intel VT-d capability register defined in VT-d spec
 384  *            section 11.4.3 Extended Capability Register.
 385  *
 386  * User needs to understand the Intel VT-d specification to decode the
 387  * register value.
 388  */
 389 struct iommu_hw_info_vtd {
 390         __u32 flags;
 391         __u32 __reserved;
 392         __aligned_u64 cap_reg;
 393         __aligned_u64 ecap_reg;
 394 };
 395
 396 /**
 397  * enum iommu_hw_info_type - IOMMU Hardware Info Types
 398  * @IOMMU_HW_INFO_TYPE_NONE: Used by the drivers that do not report hardware
 399  *                           info
 400  * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type
 401  */
 402 enum iommu_hw_info_type {
 403         IOMMU_HW_INFO_TYPE_NONE,
 404         IOMMU_HW_INFO_TYPE_INTEL_VTD,
 405 };
 406
 407 /**
 408  * struct iommu_hw_info - ioctl(IOMMU_GET_HW_INFO)
 409  * @size: sizeof(struct iommu_hw_info)
 410  * @flags: Must be 0
 411  * @dev_id: The device bound to the iommufd
 412  * @data_len: Input the length of a user buffer in bytes. Output the length of
 413  *            data that kernel supports
 414  * @data_uptr: User pointer to a user-space buffer used by the kernel to fill
 415  *             the iommu type specific hardware information data
 416  * @out_data_type: Output the iommu hardware info type as defined in the enum
 417  *                 iommu_hw_info_type.
 418  * @__reserved: Must be 0
 419  *
 420  * Query an iommu type specific hardware information data from an iommu behind
 421  * a given device that has been bound to iommufd. This hardware info data will
 422  * be used to sync capabilities between the virtual iommu and the physical
 423  * iommu, e.g. a nested translation setup needs to check the hardware info, so
 424  * a guest stage-1 page table can be compatible with the physical iommu.
 425  *
 426  * To capture an iommu type specific hardware information data, @data_uptr and
 427  * its length @data_len must be provided. Trailing bytes will be zeroed if the
 428  * user buffer is larger than the data that kernel has. Otherwise, kernel only
 429  * fills the buffer using the given length in @data_len. If the ioctl succeeds,
 430  * @data_len will be updated to the length that kernel actually supports,
 431  * @out_data_type will be filled to decode the data filled in the buffer
 432  * pointed by @data_uptr. Input @data_len == zero is allowed.
 433  */
 434 struct iommu_hw_info {
 435         __u32 size;
 436         __u32 flags;
 437         __u32 dev_id;
 438         __u32 data_len;
 439         __aligned_u64 data_uptr;
 440         __u32 out_data_type;
 441         __u32 __reserved;
 442 };
 443 #define IOMMU_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_GET_HW_INFO)
 444 #endif