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         IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING,
  51         IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP,
  52 };
  53
  54 /**
  55  * struct iommu_destroy - ioctl(IOMMU_DESTROY)
  56  * @size: sizeof(struct iommu_destroy)
  57  * @id: iommufd object ID to destroy. Can be any destroyable object type.
  58  *
  59  * Destroy any object held within iommufd.
  60  */
  61 struct iommu_destroy {
  62         __u32 size;
  63         __u32 id;
  64 };
  65 #define IOMMU_DESTROY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DESTROY)
  66
  67 /**
  68  * struct iommu_ioas_alloc - ioctl(IOMMU_IOAS_ALLOC)
  69  * @size: sizeof(struct iommu_ioas_alloc)
  70  * @flags: Must be 0
  71  * @out_ioas_id: Output IOAS ID for the allocated object
  72  *
  73  * Allocate an IO Address Space (IOAS) which holds an IO Virtual Address (IOVA)
  74  * to memory mapping.
  75  */
  76 struct iommu_ioas_alloc {
  77         __u32 size;
  78         __u32 flags;
  79         __u32 out_ioas_id;
  80 };
  81 #define IOMMU_IOAS_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOC)
  82
  83 /**
  84  * struct iommu_iova_range - ioctl(IOMMU_IOVA_RANGE)
  85  * @start: First IOVA
  86  * @last: Inclusive last IOVA
  87  *
  88  * An interval in IOVA space.
  89  */
  90 struct iommu_iova_range {
  91         __aligned_u64 start;
  92         __aligned_u64 last;
  93 };
  94
  95 /**
  96  * struct iommu_ioas_iova_ranges - ioctl(IOMMU_IOAS_IOVA_RANGES)
  97  * @size: sizeof(struct iommu_ioas_iova_ranges)
  98  * @ioas_id: IOAS ID to read ranges from
  99  * @num_iovas: Input/Output total number of ranges in the IOAS
 100  * @__reserved: Must be 0
 101  * @allowed_iovas: Pointer to the output array of struct iommu_iova_range
 102  * @out_iova_alignment: Minimum alignment required for mapping IOVA
 103  *
 104  * Query an IOAS for ranges of allowed IOVAs. Mapping IOVA outside these ranges
 105  * is not allowed. num_iovas will be set to the total number of iovas and
 106  * the allowed_iovas[] will be filled in as space permits.
 107  *
 108  * The allowed ranges are dependent on the HW path the DMA operation takes, and
 109  * can change during the lifetime of the IOAS. A fresh empty IOAS will have a
 110  * full range, and each attached device will narrow the ranges based on that
 111  * device's HW restrictions. Detaching a device can widen the ranges. Userspace
 112  * should query ranges after every attach/detach to know what IOVAs are valid
 113  * for mapping.
 114  *
 115  * On input num_iovas is the length of the allowed_iovas array. On output it is
 116  * the total number of iovas filled in. The ioctl will return -EMSGSIZE and set
 117  * num_iovas to the required value if num_iovas is too small. In this case the
 118  * caller should allocate a larger output array and re-issue the ioctl.
 119  *
 120  * out_iova_alignment returns the minimum IOVA alignment that can be given
 121  * to IOMMU_IOAS_MAP/COPY. IOVA's must satisfy::
 122  *
 123  *   starting_iova % out_iova_alignment == 0
 124  *   (starting_iova + length) % out_iova_alignment == 0
 125  *
 126  * out_iova_alignment can be 1 indicating any IOVA is allowed. It cannot
 127  * be higher than the system PAGE_SIZE.
 128  */
 129 struct iommu_ioas_iova_ranges {
 130         __u32 size;
 131         __u32 ioas_id;
 132         __u32 num_iovas;
 133         __u32 __reserved;
 134         __aligned_u64 allowed_iovas;
 135         __aligned_u64 out_iova_alignment;
 136 };
 137 #define IOMMU_IOAS_IOVA_RANGES _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_IOVA_RANGES)
 138
 139 /**
 140  * struct iommu_ioas_allow_iovas - ioctl(IOMMU_IOAS_ALLOW_IOVAS)
 141  * @size: sizeof(struct iommu_ioas_allow_iovas)
 142  * @ioas_id: IOAS ID to allow IOVAs from
 143  * @num_iovas: Input/Output total number of ranges in the IOAS
 144  * @__reserved: Must be 0
 145  * @allowed_iovas: Pointer to array of struct iommu_iova_range
 146  *
 147  * Ensure a range of IOVAs are always available for allocation. If this call
 148  * succeeds then IOMMU_IOAS_IOVA_RANGES will never return a list of IOVA ranges
 149  * that are narrower than the ranges provided here. This call will fail if
 150  * IOMMU_IOAS_IOVA_RANGES is currently narrower than the given ranges.
 151  *
 152  * When an IOAS is first created the IOVA_RANGES will be maximally sized, and as
 153  * devices are attached the IOVA will narrow based on the device restrictions.
 154  * When an allowed range is specified any narrowing will be refused, ie device
 155  * attachment can fail if the device requires limiting within the allowed range.
 156  *
 157  * Automatic IOVA allocation is also impacted by this call. MAP will only
 158  * allocate within the allowed IOVAs if they are present.
 159  *
 160  * This call replaces the entire allowed list with the given list.
 161  */
 162 struct iommu_ioas_allow_iovas {
 163         __u32 size;
 164         __u32 ioas_id;
 165         __u32 num_iovas;
 166         __u32 __reserved;
 167         __aligned_u64 allowed_iovas;
 168 };
 169 #define IOMMU_IOAS_ALLOW_IOVAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOW_IOVAS)
 170
 171 /**
 172  * enum iommufd_ioas_map_flags - Flags for map and copy
 173  * @IOMMU_IOAS_MAP_FIXED_IOVA: If clear the kernel will compute an appropriate
 174  *                             IOVA to place the mapping at
 175  * @IOMMU_IOAS_MAP_WRITEABLE: DMA is allowed to write to this mapping
 176  * @IOMMU_IOAS_MAP_READABLE: DMA is allowed to read from this mapping
 177  */
 178 enum iommufd_ioas_map_flags {
 179         IOMMU_IOAS_MAP_FIXED_IOVA = 1 << 0,
 180         IOMMU_IOAS_MAP_WRITEABLE = 1 << 1,
 181         IOMMU_IOAS_MAP_READABLE = 1 << 2,
 182 };
 183
 184 /**
 185  * struct iommu_ioas_map - ioctl(IOMMU_IOAS_MAP)
 186  * @size: sizeof(struct iommu_ioas_map)
 187  * @flags: Combination of enum iommufd_ioas_map_flags
 188  * @ioas_id: IOAS ID to change the mapping of
 189  * @__reserved: Must be 0
 190  * @user_va: Userspace pointer to start mapping from
 191  * @length: Number of bytes to map
 192  * @iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is set
 193  *        then this must be provided as input.
 194  *
 195  * Set an IOVA mapping from a user pointer. If FIXED_IOVA is specified then the
 196  * mapping will be established at iova, otherwise a suitable location based on
 197  * the reserved and allowed lists will be automatically selected and returned in
 198  * iova.
 199  *
 200  * If IOMMU_IOAS_MAP_FIXED_IOVA is specified then the iova range must currently
 201  * be unused, existing IOVA cannot be replaced.
 202  */
 203 struct iommu_ioas_map {
 204         __u32 size;
 205         __u32 flags;
 206         __u32 ioas_id;
 207         __u32 __reserved;
 208         __aligned_u64 user_va;
 209         __aligned_u64 length;
 210         __aligned_u64 iova;
 211 };
 212 #define IOMMU_IOAS_MAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP)
 213
 214 /**
 215  * struct iommu_ioas_copy - ioctl(IOMMU_IOAS_COPY)
 216  * @size: sizeof(struct iommu_ioas_copy)
 217  * @flags: Combination of enum iommufd_ioas_map_flags
 218  * @dst_ioas_id: IOAS ID to change the mapping of
 219  * @src_ioas_id: IOAS ID to copy from
 220  * @length: Number of bytes to copy and map
 221  * @dst_iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is
 222  *            set then this must be provided as input.
 223  * @src_iova: IOVA to start the copy
 224  *
 225  * Copy an already existing mapping from src_ioas_id and establish it in
 226  * dst_ioas_id. The src iova/length must exactly match a range used with
 227  * IOMMU_IOAS_MAP.
 228  *
 229  * This may be used to efficiently clone a subset of an IOAS to another, or as a
 230  * kind of 'cache' to speed up mapping. Copy has an efficiency advantage over
 231  * establishing equivalent new mappings, as internal resources are shared, and
 232  * the kernel will pin the user memory only once.
 233  */
 234 struct iommu_ioas_copy {
 235         __u32 size;
 236         __u32 flags;
 237         __u32 dst_ioas_id;
 238         __u32 src_ioas_id;
 239         __aligned_u64 length;
 240         __aligned_u64 dst_iova;
 241         __aligned_u64 src_iova;
 242 };
 243 #define IOMMU_IOAS_COPY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_COPY)
 244
 245 /**
 246  * struct iommu_ioas_unmap - ioctl(IOMMU_IOAS_UNMAP)
 247  * @size: sizeof(struct iommu_ioas_unmap)
 248  * @ioas_id: IOAS ID to change the mapping of
 249  * @iova: IOVA to start the unmapping at
 250  * @length: Number of bytes to unmap, and return back the bytes unmapped
 251  *
 252  * Unmap an IOVA range. The iova/length must be a superset of a previously
 253  * mapped range used with IOMMU_IOAS_MAP or IOMMU_IOAS_COPY. Splitting or
 254  * truncating ranges is not allowed. The values 0 to U64_MAX will unmap
 255  * everything.
 256  */
 257 struct iommu_ioas_unmap {
 258         __u32 size;
 259         __u32 ioas_id;
 260         __aligned_u64 iova;
 261         __aligned_u64 length;
 262 };
 263 #define IOMMU_IOAS_UNMAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_UNMAP)
 264
 265 /**
 266  * enum iommufd_option - ioctl(IOMMU_OPTION_RLIMIT_MODE) and
 267  *                       ioctl(IOMMU_OPTION_HUGE_PAGES)
 268  * @IOMMU_OPTION_RLIMIT_MODE:
 269  *    Change how RLIMIT_MEMLOCK accounting works. The caller must have privilege
 270  *    to invoke this. Value 0 (default) is user based accouting, 1 uses process
 271  *    based accounting. Global option, object_id must be 0
 272  * @IOMMU_OPTION_HUGE_PAGES:
 273  *    Value 1 (default) allows contiguous pages to be combined when generating
 274  *    iommu mappings. Value 0 disables combining, everything is mapped to
 275  *    PAGE_SIZE. This can be useful for benchmarking.  This is a per-IOAS
 276  *    option, the object_id must be the IOAS ID.
 277  */
 278 enum iommufd_option {
 279         IOMMU_OPTION_RLIMIT_MODE = 0,
 280         IOMMU_OPTION_HUGE_PAGES = 1,
 281 };
 282
 283 /**
 284  * enum iommufd_option_ops - ioctl(IOMMU_OPTION_OP_SET) and
 285  *                           ioctl(IOMMU_OPTION_OP_GET)
 286  * @IOMMU_OPTION_OP_SET: Set the option's value
 287  * @IOMMU_OPTION_OP_GET: Get the option's value
 288  */
 289 enum iommufd_option_ops {
 290         IOMMU_OPTION_OP_SET = 0,
 291         IOMMU_OPTION_OP_GET = 1,
 292 };
 293
 294 /**
 295  * struct iommu_option - iommu option multiplexer
 296  * @size: sizeof(struct iommu_option)
 297  * @option_id: One of enum iommufd_option
 298  * @op: One of enum iommufd_option_ops
 299  * @__reserved: Must be 0
 300  * @object_id: ID of the object if required
 301  * @val64: Option value to set or value returned on get
 302  *
 303  * Change a simple option value. This multiplexor allows controlling options
 304  * on objects. IOMMU_OPTION_OP_SET will load an option and IOMMU_OPTION_OP_GET
 305  * will return the current value.
 306  */
 307 struct iommu_option {
 308         __u32 size;
 309         __u32 option_id;
 310         __u16 op;
 311         __u16 __reserved;
 312         __u32 object_id;
 313         __aligned_u64 val64;
 314 };
 315 #define IOMMU_OPTION _IO(IOMMUFD_TYPE, IOMMUFD_CMD_OPTION)
 316
 317 /**
 318  * enum iommufd_vfio_ioas_op - IOMMU_VFIO_IOAS_* ioctls
 319  * @IOMMU_VFIO_IOAS_GET: Get the current compatibility IOAS
 320  * @IOMMU_VFIO_IOAS_SET: Change the current compatibility IOAS
 321  * @IOMMU_VFIO_IOAS_CLEAR: Disable VFIO compatibility
 322  */
 323 enum iommufd_vfio_ioas_op {
 324         IOMMU_VFIO_IOAS_GET = 0,
 325         IOMMU_VFIO_IOAS_SET = 1,
 326         IOMMU_VFIO_IOAS_CLEAR = 2,
 327 };
 328
 329 /**
 330  * struct iommu_vfio_ioas - ioctl(IOMMU_VFIO_IOAS)
 331  * @size: sizeof(struct iommu_vfio_ioas)
 332  * @ioas_id: For IOMMU_VFIO_IOAS_SET the input IOAS ID to set
 333  *           For IOMMU_VFIO_IOAS_GET will output the IOAS ID
 334  * @op: One of enum iommufd_vfio_ioas_op
 335  * @__reserved: Must be 0
 336  *
 337  * The VFIO compatibility support uses a single ioas because VFIO APIs do not
 338  * support the ID field. Set or Get the IOAS that VFIO compatibility will use.
 339  * When VFIO_GROUP_SET_CONTAINER is used on an iommufd it will get the
 340  * compatibility ioas, either by taking what is already set, or auto creating
 341  * one. From then on VFIO will continue to use that ioas and is not effected by
 342  * this ioctl. SET or CLEAR does not destroy any auto-created IOAS.
 343  */
 344 struct iommu_vfio_ioas {
 345         __u32 size;
 346         __u32 ioas_id;
 347         __u16 op;
 348         __u16 __reserved;
 349 };
 350 #define IOMMU_VFIO_IOAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VFIO_IOAS)
 351
 352 /**
 353  * enum iommufd_hwpt_alloc_flags - Flags for HWPT allocation
 354  * @IOMMU_HWPT_ALLOC_NEST_PARENT: If set, allocate a HWPT that can serve as
 355  *                                the parent HWPT in a nesting configuration.
 356  * @IOMMU_HWPT_ALLOC_DIRTY_TRACKING: Dirty tracking support for device IOMMU is
 357  *                                   enforced on device attachment
 358  */
 359 enum iommufd_hwpt_alloc_flags {
 360         IOMMU_HWPT_ALLOC_NEST_PARENT = 1 << 0,
 361         IOMMU_HWPT_ALLOC_DIRTY_TRACKING = 1 << 1,
 362 };
 363
 364 /**
 365  * enum iommu_hwpt_vtd_s1_flags - Intel VT-d stage-1 page table
 366  *                                entry attributes
 367  * @IOMMU_VTD_S1_SRE: Supervisor request
 368  * @IOMMU_VTD_S1_EAFE: Extended access enable
 369  * @IOMMU_VTD_S1_WPE: Write protect enable
 370  */
 371 enum iommu_hwpt_vtd_s1_flags {
 372         IOMMU_VTD_S1_SRE = 1 << 0,
 373         IOMMU_VTD_S1_EAFE = 1 << 1,
 374         IOMMU_VTD_S1_WPE = 1 << 2,
 375 };
 376
 377 /**
 378  * struct iommu_hwpt_vtd_s1 - Intel VT-d stage-1 page table
 379  *                            info (IOMMU_HWPT_DATA_VTD_S1)
 380  * @flags: Combination of enum iommu_hwpt_vtd_s1_flags
 381  * @pgtbl_addr: The base address of the stage-1 page table.
 382  * @addr_width: The address width of the stage-1 page table
 383  * @__reserved: Must be 0
 384  */
 385 struct iommu_hwpt_vtd_s1 {
 386         __aligned_u64 flags;
 387         __aligned_u64 pgtbl_addr;
 388         __u32 addr_width;
 389         __u32 __reserved;
 390 };
 391
 392 /**
 393  * enum iommu_hwpt_data_type - IOMMU HWPT Data Type
 394  * @IOMMU_HWPT_DATA_NONE: no data
 395  * @IOMMU_HWPT_DATA_VTD_S1: Intel VT-d stage-1 page table
 396  */
 397 enum iommu_hwpt_data_type {
 398         IOMMU_HWPT_DATA_NONE,
 399         IOMMU_HWPT_DATA_VTD_S1,
 400 };
 401
 402 /**
 403  * struct iommu_hwpt_alloc - ioctl(IOMMU_HWPT_ALLOC)
 404  * @size: sizeof(struct iommu_hwpt_alloc)
 405  * @flags: Combination of enum iommufd_hwpt_alloc_flags
 406  * @dev_id: The device to allocate this HWPT for
 407  * @pt_id: The IOAS or HWPT to connect this HWPT to
 408  * @out_hwpt_id: The ID of the new HWPT
 409  * @__reserved: Must be 0
 410  * @data_type: One of enum iommu_hwpt_data_type
 411  * @data_len: Length of the type specific data
 412  * @data_uptr: User pointer to the type specific data
 413  *
 414  * Explicitly allocate a hardware page table object. This is the same object
 415  * type that is returned by iommufd_device_attach() and represents the
 416  * underlying iommu driver's iommu_domain kernel object.
 417  *
 418  * A kernel-managed HWPT will be created with the mappings from the given
 419  * IOAS via the @pt_id. The @data_type for this allocation must be set to
 420  * IOMMU_HWPT_DATA_NONE. The HWPT can be allocated as a parent HWPT for a
 421  * nesting configuration by passing IOMMU_HWPT_ALLOC_NEST_PARENT via @flags.
 422  *
 423  * A user-managed nested HWPT will be created from a given parent HWPT via
 424  * @pt_id, in which the parent HWPT must be allocated previously via the
 425  * same ioctl from a given IOAS (@pt_id). In this case, the @data_type
 426  * must be set to a pre-defined type corresponding to an I/O page table
 427  * type supported by the underlying IOMMU hardware.
 428  *
 429  * If the @data_type is set to IOMMU_HWPT_DATA_NONE, @data_len and
 430  * @data_uptr should be zero. Otherwise, both @data_len and @data_uptr
 431  * must be given.
 432  */
 433 struct iommu_hwpt_alloc {
 434         __u32 size;
 435         __u32 flags;
 436         __u32 dev_id;
 437         __u32 pt_id;
 438         __u32 out_hwpt_id;
 439         __u32 __reserved;
 440         __u32 data_type;
 441         __u32 data_len;
 442         __aligned_u64 data_uptr;
 443 };
 444 #define IOMMU_HWPT_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_ALLOC)
 445
 446 /**
 447  * enum iommu_hw_info_vtd_flags - Flags for VT-d hw_info
 448  * @IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17: If set, disallow read-only mappings
 449  *                                         on a nested_parent domain.
 450  *                                         https://www.intel.com/content/www/us/en/content-details/772415/content-details.html
 451  */
 452 enum iommu_hw_info_vtd_flags {
 453         IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17 = 1 << 0,
 454 };
 455
 456 /**
 457  * struct iommu_hw_info_vtd - Intel VT-d hardware information
 458  *
 459  * @flags: Combination of enum iommu_hw_info_vtd_flags
 460  * @__reserved: Must be 0
 461  *
 462  * @cap_reg: Value of Intel VT-d capability register defined in VT-d spec
 463  *           section 11.4.2 Capability Register.
 464  * @ecap_reg: Value of Intel VT-d capability register defined in VT-d spec
 465  *            section 11.4.3 Extended Capability Register.
 466  *
 467  * User needs to understand the Intel VT-d specification to decode the
 468  * register value.
 469  */
 470 struct iommu_hw_info_vtd {
 471         __u32 flags;
 472         __u32 __reserved;
 473         __aligned_u64 cap_reg;
 474         __aligned_u64 ecap_reg;
 475 };
 476
 477 /**
 478  * enum iommu_hw_info_type - IOMMU Hardware Info Types
 479  * @IOMMU_HW_INFO_TYPE_NONE: Used by the drivers that do not report hardware
 480  *                           info
 481  * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type
 482  */
 483 enum iommu_hw_info_type {
 484         IOMMU_HW_INFO_TYPE_NONE,
 485         IOMMU_HW_INFO_TYPE_INTEL_VTD,
 486 };
 487
 488 /**
 489  * enum iommufd_hw_capabilities
 490  * @IOMMU_HW_CAP_DIRTY_TRACKING: IOMMU hardware support for dirty tracking
 491  *                               If available, it means the following APIs
 492  *                               are supported:
 493  *
 494  *                                   IOMMU_HWPT_GET_DIRTY_BITMAP
 495  *                                   IOMMU_HWPT_SET_DIRTY_TRACKING
 496  *
 497  */
 498 enum iommufd_hw_capabilities {
 499         IOMMU_HW_CAP_DIRTY_TRACKING = 1 << 0,
 500 };
 501
 502 /**
 503  * struct iommu_hw_info - ioctl(IOMMU_GET_HW_INFO)
 504  * @size: sizeof(struct iommu_hw_info)
 505  * @flags: Must be 0
 506  * @dev_id: The device bound to the iommufd
 507  * @data_len: Input the length of a user buffer in bytes. Output the length of
 508  *            data that kernel supports
 509  * @data_uptr: User pointer to a user-space buffer used by the kernel to fill
 510  *             the iommu type specific hardware information data
 511  * @out_data_type: Output the iommu hardware info type as defined in the enum
 512  *                 iommu_hw_info_type.
 513  * @out_capabilities: Output the generic iommu capability info type as defined
 514  *                    in the enum iommu_hw_capabilities.
 515  * @__reserved: Must be 0
 516  *
 517  * Query an iommu type specific hardware information data from an iommu behind
 518  * a given device that has been bound to iommufd. This hardware info data will
 519  * be used to sync capabilities between the virtual iommu and the physical
 520  * iommu, e.g. a nested translation setup needs to check the hardware info, so
 521  * a guest stage-1 page table can be compatible with the physical iommu.
 522  *
 523  * To capture an iommu type specific hardware information data, @data_uptr and
 524  * its length @data_len must be provided. Trailing bytes will be zeroed if the
 525  * user buffer is larger than the data that kernel has. Otherwise, kernel only
 526  * fills the buffer using the given length in @data_len. If the ioctl succeeds,
 527  * @data_len will be updated to the length that kernel actually supports,
 528  * @out_data_type will be filled to decode the data filled in the buffer
 529  * pointed by @data_uptr. Input @data_len == zero is allowed.
 530  */
 531 struct iommu_hw_info {
 532         __u32 size;
 533         __u32 flags;
 534         __u32 dev_id;
 535         __u32 data_len;
 536         __aligned_u64 data_uptr;
 537         __u32 out_data_type;
 538         __u32 __reserved;
 539         __aligned_u64 out_capabilities;
 540 };
 541 #define IOMMU_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_GET_HW_INFO)
 542
 543 /*
 544  * enum iommufd_hwpt_set_dirty_tracking_flags - Flags for steering dirty
 545  *                                              tracking
 546  * @IOMMU_HWPT_DIRTY_TRACKING_ENABLE: Enable dirty tracking
 547  */
 548 enum iommufd_hwpt_set_dirty_tracking_flags {
 549         IOMMU_HWPT_DIRTY_TRACKING_ENABLE = 1,
 550 };
 551
 552 /**
 553  * struct iommu_hwpt_set_dirty_tracking - ioctl(IOMMU_HWPT_SET_DIRTY_TRACKING)
 554  * @size: sizeof(struct iommu_hwpt_set_dirty_tracking)
 555  * @flags: Combination of enum iommufd_hwpt_set_dirty_tracking_flags
 556  * @hwpt_id: HW pagetable ID that represents the IOMMU domain
 557  * @__reserved: Must be 0
 558  *
 559  * Toggle dirty tracking on an HW pagetable.
 560  */
 561 struct iommu_hwpt_set_dirty_tracking {
 562         __u32 size;
 563         __u32 flags;
 564         __u32 hwpt_id;
 565         __u32 __reserved;
 566 };
 567 #define IOMMU_HWPT_SET_DIRTY_TRACKING _IO(IOMMUFD_TYPE, \
 568                                           IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING)
 569
 570 /**
 571  * enum iommufd_hwpt_get_dirty_bitmap_flags - Flags for getting dirty bits
 572  * @IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR: Just read the PTEs without clearing
 573  *                                        any dirty bits metadata. This flag
 574  *                                        can be passed in the expectation
 575  *                                        where the next operation is an unmap
 576  *                                        of the same IOVA range.
 577  *
 578  */
 579 enum iommufd_hwpt_get_dirty_bitmap_flags {
 580         IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR = 1,
 581 };
 582
 583 /**
 584  * struct iommu_hwpt_get_dirty_bitmap - ioctl(IOMMU_HWPT_GET_DIRTY_BITMAP)
 585  * @size: sizeof(struct iommu_hwpt_get_dirty_bitmap)
 586  * @hwpt_id: HW pagetable ID that represents the IOMMU domain
 587  * @flags: Combination of enum iommufd_hwpt_get_dirty_bitmap_flags
 588  * @__reserved: Must be 0
 589  * @iova: base IOVA of the bitmap first bit
 590  * @length: IOVA range size
 591  * @page_size: page size granularity of each bit in the bitmap
 592  * @data: bitmap where to set the dirty bits. The bitmap bits each
 593  *        represent a page_size which you deviate from an arbitrary iova.
 594  *
 595  * Checking a given IOVA is dirty:
 596  *
 597  *  data[(iova / page_size) / 64] & (1ULL << ((iova / page_size) % 64))
 598  *
 599  * Walk the IOMMU pagetables for a given IOVA range to return a bitmap
 600  * with the dirty IOVAs. In doing so it will also by default clear any
 601  * dirty bit metadata set in the IOPTE.
 602  */
 603 struct iommu_hwpt_get_dirty_bitmap {
 604         __u32 size;
 605         __u32 hwpt_id;
 606         __u32 flags;
 607         __u32 __reserved;
 608         __aligned_u64 iova;
 609         __aligned_u64 length;
 610         __aligned_u64 page_size;
 611         __aligned_u64 data;
 612 };
 613 #define IOMMU_HWPT_GET_DIRTY_BITMAP _IO(IOMMUFD_TYPE, \
 614                                         IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP)
 615
 616 #endif