1/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2/* Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES.
3 */
4#ifndef _UAPI_IOMMUFD_H
5#define _UAPI_IOMMUFD_H
6
7#include <linux/ioctl.h>
8#include <linux/types.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 */
37enum {
38 IOMMUFD_CMD_BASE = 0x80,
39 IOMMUFD_CMD_DESTROY = IOMMUFD_CMD_BASE,
40 IOMMUFD_CMD_IOAS_ALLOC = 0x81,
41 IOMMUFD_CMD_IOAS_ALLOW_IOVAS = 0x82,
42 IOMMUFD_CMD_IOAS_COPY = 0x83,
43 IOMMUFD_CMD_IOAS_IOVA_RANGES = 0x84,
44 IOMMUFD_CMD_IOAS_MAP = 0x85,
45 IOMMUFD_CMD_IOAS_UNMAP = 0x86,
46 IOMMUFD_CMD_OPTION = 0x87,
47 IOMMUFD_CMD_VFIO_IOAS = 0x88,
48 IOMMUFD_CMD_HWPT_ALLOC = 0x89,
49 IOMMUFD_CMD_GET_HW_INFO = 0x8a,
50 IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING = 0x8b,
51 IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP = 0x8c,
52 IOMMUFD_CMD_HWPT_INVALIDATE = 0x8d,
53 IOMMUFD_CMD_FAULT_QUEUE_ALLOC = 0x8e,
54 IOMMUFD_CMD_IOAS_MAP_FILE = 0x8f,
55 IOMMUFD_CMD_VIOMMU_ALLOC = 0x90,
56 IOMMUFD_CMD_VDEVICE_ALLOC = 0x91,
57 IOMMUFD_CMD_IOAS_CHANGE_PROCESS = 0x92,
58 IOMMUFD_CMD_VEVENTQ_ALLOC = 0x93,
59 IOMMUFD_CMD_HW_QUEUE_ALLOC = 0x94,
60};
61
62/**
63 * struct iommu_destroy - ioctl(IOMMU_DESTROY)
64 * @size: sizeof(struct iommu_destroy)
65 * @id: iommufd object ID to destroy. Can be any destroyable object type.
66 *
67 * Destroy any object held within iommufd.
68 */
69struct iommu_destroy {
70 __u32 size;
71 __u32 id;
72};
73#define IOMMU_DESTROY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DESTROY)
74
75/**
76 * struct iommu_ioas_alloc - ioctl(IOMMU_IOAS_ALLOC)
77 * @size: sizeof(struct iommu_ioas_alloc)
78 * @flags: Must be 0
79 * @out_ioas_id: Output IOAS ID for the allocated object
80 *
81 * Allocate an IO Address Space (IOAS) which holds an IO Virtual Address (IOVA)
82 * to memory mapping.
83 */
84struct iommu_ioas_alloc {
85 __u32 size;
86 __u32 flags;
87 __u32 out_ioas_id;
88};
89#define IOMMU_IOAS_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOC)
90
91/**
92 * struct iommu_iova_range - ioctl(IOMMU_IOVA_RANGE)
93 * @start: First IOVA
94 * @last: Inclusive last IOVA
95 *
96 * An interval in IOVA space.
97 */
98struct iommu_iova_range {
99 __aligned_u64 start;
100 __aligned_u64 last;
101};
102
103/**
104 * struct iommu_ioas_iova_ranges - ioctl(IOMMU_IOAS_IOVA_RANGES)
105 * @size: sizeof(struct iommu_ioas_iova_ranges)
106 * @ioas_id: IOAS ID to read ranges from
107 * @num_iovas: Input/Output total number of ranges in the IOAS
108 * @__reserved: Must be 0
109 * @allowed_iovas: Pointer to the output array of struct iommu_iova_range
110 * @out_iova_alignment: Minimum alignment required for mapping IOVA
111 *
112 * Query an IOAS for ranges of allowed IOVAs. Mapping IOVA outside these ranges
113 * is not allowed. num_iovas will be set to the total number of iovas and
114 * the allowed_iovas[] will be filled in as space permits.
115 *
116 * The allowed ranges are dependent on the HW path the DMA operation takes, and
117 * can change during the lifetime of the IOAS. A fresh empty IOAS will have a
118 * full range, and each attached device will narrow the ranges based on that
119 * device's HW restrictions. Detaching a device can widen the ranges. Userspace
120 * should query ranges after every attach/detach to know what IOVAs are valid
121 * for mapping.
122 *
123 * On input num_iovas is the length of the allowed_iovas array. On output it is
124 * the total number of iovas filled in. The ioctl will return -EMSGSIZE and set
125 * num_iovas to the required value if num_iovas is too small. In this case the
126 * caller should allocate a larger output array and re-issue the ioctl.
127 *
128 * out_iova_alignment returns the minimum IOVA alignment that can be given
129 * to IOMMU_IOAS_MAP/COPY. IOVA's must satisfy::
130 *
131 * starting_iova % out_iova_alignment == 0
132 * (starting_iova + length) % out_iova_alignment == 0
133 *
134 * out_iova_alignment can be 1 indicating any IOVA is allowed. It cannot
135 * be higher than the system PAGE_SIZE.
136 */
137struct iommu_ioas_iova_ranges {
138 __u32 size;
139 __u32 ioas_id;
140 __u32 num_iovas;
141 __u32 __reserved;
142 __aligned_u64 allowed_iovas;
143 __aligned_u64 out_iova_alignment;
144};
145#define IOMMU_IOAS_IOVA_RANGES _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_IOVA_RANGES)
146
147/**
148 * struct iommu_ioas_allow_iovas - ioctl(IOMMU_IOAS_ALLOW_IOVAS)
149 * @size: sizeof(struct iommu_ioas_allow_iovas)
150 * @ioas_id: IOAS ID to allow IOVAs from
151 * @num_iovas: Input/Output total number of ranges in the IOAS
152 * @__reserved: Must be 0
153 * @allowed_iovas: Pointer to array of struct iommu_iova_range
154 *
155 * Ensure a range of IOVAs are always available for allocation. If this call
156 * succeeds then IOMMU_IOAS_IOVA_RANGES will never return a list of IOVA ranges
157 * that are narrower than the ranges provided here. This call will fail if
158 * IOMMU_IOAS_IOVA_RANGES is currently narrower than the given ranges.
159 *
160 * When an IOAS is first created the IOVA_RANGES will be maximally sized, and as
161 * devices are attached the IOVA will narrow based on the device restrictions.
162 * When an allowed range is specified any narrowing will be refused, ie device
163 * attachment can fail if the device requires limiting within the allowed range.
164 *
165 * Automatic IOVA allocation is also impacted by this call. MAP will only
166 * allocate within the allowed IOVAs if they are present.
167 *
168 * This call replaces the entire allowed list with the given list.
169 */
170struct iommu_ioas_allow_iovas {
171 __u32 size;
172 __u32 ioas_id;
173 __u32 num_iovas;
174 __u32 __reserved;
175 __aligned_u64 allowed_iovas;
176};
177#define IOMMU_IOAS_ALLOW_IOVAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOW_IOVAS)
178
179/**
180 * enum iommufd_ioas_map_flags - Flags for map and copy
181 * @IOMMU_IOAS_MAP_FIXED_IOVA: If clear the kernel will compute an appropriate
182 * IOVA to place the mapping at
183 * @IOMMU_IOAS_MAP_WRITEABLE: DMA is allowed to write to this mapping
184 * @IOMMU_IOAS_MAP_READABLE: DMA is allowed to read from this mapping
185 */
186enum iommufd_ioas_map_flags {
187 IOMMU_IOAS_MAP_FIXED_IOVA = 1 << 0,
188 IOMMU_IOAS_MAP_WRITEABLE = 1 << 1,
189 IOMMU_IOAS_MAP_READABLE = 1 << 2,
190};
191
192/**
193 * struct iommu_ioas_map - ioctl(IOMMU_IOAS_MAP)
194 * @size: sizeof(struct iommu_ioas_map)
195 * @flags: Combination of enum iommufd_ioas_map_flags
196 * @ioas_id: IOAS ID to change the mapping of
197 * @__reserved: Must be 0
198 * @user_va: Userspace pointer to start mapping from
199 * @length: Number of bytes to map
200 * @iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is set
201 * then this must be provided as input.
202 *
203 * Set an IOVA mapping from a user pointer. If FIXED_IOVA is specified then the
204 * mapping will be established at iova, otherwise a suitable location based on
205 * the reserved and allowed lists will be automatically selected and returned in
206 * iova.
207 *
208 * If IOMMU_IOAS_MAP_FIXED_IOVA is specified then the iova range must currently
209 * be unused, existing IOVA cannot be replaced.
210 */
211struct iommu_ioas_map {
212 __u32 size;
213 __u32 flags;
214 __u32 ioas_id;
215 __u32 __reserved;
216 __aligned_u64 user_va;
217 __aligned_u64 length;
218 __aligned_u64 iova;
219};
220#define IOMMU_IOAS_MAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP)
221
222/**
223 * struct iommu_ioas_map_file - ioctl(IOMMU_IOAS_MAP_FILE)
224 * @size: sizeof(struct iommu_ioas_map_file)
225 * @flags: same as for iommu_ioas_map
226 * @ioas_id: same as for iommu_ioas_map
227 * @fd: the memfd to map
228 * @start: byte offset from start of file to map from
229 * @length: same as for iommu_ioas_map
230 * @iova: same as for iommu_ioas_map
231 *
232 * Set an IOVA mapping from a memfd file. All other arguments and semantics
233 * match those of IOMMU_IOAS_MAP.
234 */
235struct iommu_ioas_map_file {
236 __u32 size;
237 __u32 flags;
238 __u32 ioas_id;
239 __s32 fd;
240 __aligned_u64 start;
241 __aligned_u64 length;
242 __aligned_u64 iova;
243};
244#define IOMMU_IOAS_MAP_FILE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP_FILE)
245
246/**
247 * struct iommu_ioas_copy - ioctl(IOMMU_IOAS_COPY)
248 * @size: sizeof(struct iommu_ioas_copy)
249 * @flags: Combination of enum iommufd_ioas_map_flags
250 * @dst_ioas_id: IOAS ID to change the mapping of
251 * @src_ioas_id: IOAS ID to copy from
252 * @length: Number of bytes to copy and map
253 * @dst_iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is
254 * set then this must be provided as input.
255 * @src_iova: IOVA to start the copy
256 *
257 * Copy an already existing mapping from src_ioas_id and establish it in
258 * dst_ioas_id. The src iova/length must exactly match a range used with
259 * IOMMU_IOAS_MAP.
260 *
261 * This may be used to efficiently clone a subset of an IOAS to another, or as a
262 * kind of 'cache' to speed up mapping. Copy has an efficiency advantage over
263 * establishing equivalent new mappings, as internal resources are shared, and
264 * the kernel will pin the user memory only once.
265 */
266struct iommu_ioas_copy {
267 __u32 size;
268 __u32 flags;
269 __u32 dst_ioas_id;
270 __u32 src_ioas_id;
271 __aligned_u64 length;
272 __aligned_u64 dst_iova;
273 __aligned_u64 src_iova;
274};
275#define IOMMU_IOAS_COPY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_COPY)
276
277/**
278 * struct iommu_ioas_unmap - ioctl(IOMMU_IOAS_UNMAP)
279 * @size: sizeof(struct iommu_ioas_unmap)
280 * @ioas_id: IOAS ID to change the mapping of
281 * @iova: IOVA to start the unmapping at
282 * @length: Number of bytes to unmap, and return back the bytes unmapped
283 *
284 * Unmap an IOVA range. The iova/length must be a superset of a previously
285 * mapped range used with IOMMU_IOAS_MAP or IOMMU_IOAS_COPY. Splitting or
286 * truncating ranges is not allowed. The values 0 to U64_MAX will unmap
287 * everything.
288 */
289struct iommu_ioas_unmap {
290 __u32 size;
291 __u32 ioas_id;
292 __aligned_u64 iova;
293 __aligned_u64 length;
294};
295#define IOMMU_IOAS_UNMAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_UNMAP)
296
297/**
298 * enum iommufd_option - ioctl(IOMMU_OPTION_RLIMIT_MODE) and
299 * ioctl(IOMMU_OPTION_HUGE_PAGES)
300 * @IOMMU_OPTION_RLIMIT_MODE:
301 * Change how RLIMIT_MEMLOCK accounting works. The caller must have privilege
302 * to invoke this. Value 0 (default) is user based accounting, 1 uses process
303 * based accounting. Global option, object_id must be 0
304 * @IOMMU_OPTION_HUGE_PAGES:
305 * Value 1 (default) allows contiguous pages to be combined when generating
306 * iommu mappings. Value 0 disables combining, everything is mapped to
307 * PAGE_SIZE. This can be useful for benchmarking. This is a per-IOAS
308 * option, the object_id must be the IOAS ID.
309 */
310enum iommufd_option {
311 IOMMU_OPTION_RLIMIT_MODE = 0,
312 IOMMU_OPTION_HUGE_PAGES = 1,
313};
314
315/**
316 * enum iommufd_option_ops - ioctl(IOMMU_OPTION_OP_SET) and
317 * ioctl(IOMMU_OPTION_OP_GET)
318 * @IOMMU_OPTION_OP_SET: Set the option's value
319 * @IOMMU_OPTION_OP_GET: Get the option's value
320 */
321enum iommufd_option_ops {
322 IOMMU_OPTION_OP_SET = 0,
323 IOMMU_OPTION_OP_GET = 1,
324};
325
326/**
327 * struct iommu_option - iommu option multiplexer
328 * @size: sizeof(struct iommu_option)
329 * @option_id: One of enum iommufd_option
330 * @op: One of enum iommufd_option_ops
331 * @__reserved: Must be 0
332 * @object_id: ID of the object if required
333 * @val64: Option value to set or value returned on get
334 *
335 * Change a simple option value. This multiplexor allows controlling options
336 * on objects. IOMMU_OPTION_OP_SET will load an option and IOMMU_OPTION_OP_GET
337 * will return the current value.
338 */
339struct iommu_option {
340 __u32 size;
341 __u32 option_id;
342 __u16 op;
343 __u16 __reserved;
344 __u32 object_id;
345 __aligned_u64 val64;
346};
347#define IOMMU_OPTION _IO(IOMMUFD_TYPE, IOMMUFD_CMD_OPTION)
348
349/**
350 * enum iommufd_vfio_ioas_op - IOMMU_VFIO_IOAS_* ioctls
351 * @IOMMU_VFIO_IOAS_GET: Get the current compatibility IOAS
352 * @IOMMU_VFIO_IOAS_SET: Change the current compatibility IOAS
353 * @IOMMU_VFIO_IOAS_CLEAR: Disable VFIO compatibility
354 */
355enum iommufd_vfio_ioas_op {
356 IOMMU_VFIO_IOAS_GET = 0,
357 IOMMU_VFIO_IOAS_SET = 1,
358 IOMMU_VFIO_IOAS_CLEAR = 2,
359};
360
361/**
362 * struct iommu_vfio_ioas - ioctl(IOMMU_VFIO_IOAS)
363 * @size: sizeof(struct iommu_vfio_ioas)
364 * @ioas_id: For IOMMU_VFIO_IOAS_SET the input IOAS ID to set
365 * For IOMMU_VFIO_IOAS_GET will output the IOAS ID
366 * @op: One of enum iommufd_vfio_ioas_op
367 * @__reserved: Must be 0
368 *
369 * The VFIO compatibility support uses a single ioas because VFIO APIs do not
370 * support the ID field. Set or Get the IOAS that VFIO compatibility will use.
371 * When VFIO_GROUP_SET_CONTAINER is used on an iommufd it will get the
372 * compatibility ioas, either by taking what is already set, or auto creating
373 * one. From then on VFIO will continue to use that ioas and is not effected by
374 * this ioctl. SET or CLEAR does not destroy any auto-created IOAS.
375 */
376struct iommu_vfio_ioas {
377 __u32 size;
378 __u32 ioas_id;
379 __u16 op;
380 __u16 __reserved;
381};
382#define IOMMU_VFIO_IOAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VFIO_IOAS)
383
384/**
385 * enum iommufd_hwpt_alloc_flags - Flags for HWPT allocation
386 * @IOMMU_HWPT_ALLOC_NEST_PARENT: If set, allocate a HWPT that can serve as
387 * the parent HWPT in a nesting configuration.
388 * @IOMMU_HWPT_ALLOC_DIRTY_TRACKING: Dirty tracking support for device IOMMU is
389 * enforced on device attachment
390 * @IOMMU_HWPT_FAULT_ID_VALID: The fault_id field of hwpt allocation data is
391 * valid.
392 * @IOMMU_HWPT_ALLOC_PASID: Requests a domain that can be used with PASID. The
393 * domain can be attached to any PASID on the device.
394 * Any domain attached to the non-PASID part of the
395 * device must also be flagged, otherwise attaching a
396 * PASID will blocked.
397 * For the user that wants to attach PASID, ioas is
398 * not recommended for both the non-PASID part
399 * and PASID part of the device.
400 * If IOMMU does not support PASID it will return
401 * error (-EOPNOTSUPP).
402 */
403enum iommufd_hwpt_alloc_flags {
404 IOMMU_HWPT_ALLOC_NEST_PARENT = 1 << 0,
405 IOMMU_HWPT_ALLOC_DIRTY_TRACKING = 1 << 1,
406 IOMMU_HWPT_FAULT_ID_VALID = 1 << 2,
407 IOMMU_HWPT_ALLOC_PASID = 1 << 3,
408};
409
410/**
411 * enum iommu_hwpt_vtd_s1_flags - Intel VT-d stage-1 page table
412 * entry attributes
413 * @IOMMU_VTD_S1_SRE: Supervisor request
414 * @IOMMU_VTD_S1_EAFE: Extended access enable
415 * @IOMMU_VTD_S1_WPE: Write protect enable
416 */
417enum iommu_hwpt_vtd_s1_flags {
418 IOMMU_VTD_S1_SRE = 1 << 0,
419 IOMMU_VTD_S1_EAFE = 1 << 1,
420 IOMMU_VTD_S1_WPE = 1 << 2,
421};
422
423/**
424 * struct iommu_hwpt_vtd_s1 - Intel VT-d stage-1 page table
425 * info (IOMMU_HWPT_DATA_VTD_S1)
426 * @flags: Combination of enum iommu_hwpt_vtd_s1_flags
427 * @pgtbl_addr: The base address of the stage-1 page table.
428 * @addr_width: The address width of the stage-1 page table
429 * @__reserved: Must be 0
430 */
431struct iommu_hwpt_vtd_s1 {
432 __aligned_u64 flags;
433 __aligned_u64 pgtbl_addr;
434 __u32 addr_width;
435 __u32 __reserved;
436};
437
438/**
439 * struct iommu_hwpt_arm_smmuv3 - ARM SMMUv3 nested STE
440 * (IOMMU_HWPT_DATA_ARM_SMMUV3)
441 *
442 * @ste: The first two double words of the user space Stream Table Entry for
443 * the translation. Must be little-endian.
444 * Allowed fields: (Refer to "5.2 Stream Table Entry" in SMMUv3 HW Spec)
445 * - word-0: V, Cfg, S1Fmt, S1ContextPtr, S1CDMax
446 * - word-1: EATS, S1DSS, S1CIR, S1COR, S1CSH, S1STALLD
447 *
448 * -EIO will be returned if @ste is not legal or contains any non-allowed field.
449 * Cfg can be used to select a S1, Bypass or Abort configuration. A Bypass
450 * nested domain will translate the same as the nesting parent. The S1 will
451 * install a Context Descriptor Table pointing at userspace memory translated
452 * by the nesting parent.
453 */
454struct iommu_hwpt_arm_smmuv3 {
455 __aligned_le64 ste[2];
456};
457
458/**
459 * enum iommu_hwpt_data_type - IOMMU HWPT Data Type
460 * @IOMMU_HWPT_DATA_NONE: no data
461 * @IOMMU_HWPT_DATA_VTD_S1: Intel VT-d stage-1 page table
462 * @IOMMU_HWPT_DATA_ARM_SMMUV3: ARM SMMUv3 Context Descriptor Table
463 */
464enum iommu_hwpt_data_type {
465 IOMMU_HWPT_DATA_NONE = 0,
466 IOMMU_HWPT_DATA_VTD_S1 = 1,
467 IOMMU_HWPT_DATA_ARM_SMMUV3 = 2,
468};
469
470/**
471 * struct iommu_hwpt_alloc - ioctl(IOMMU_HWPT_ALLOC)
472 * @size: sizeof(struct iommu_hwpt_alloc)
473 * @flags: Combination of enum iommufd_hwpt_alloc_flags
474 * @dev_id: The device to allocate this HWPT for
475 * @pt_id: The IOAS or HWPT or vIOMMU to connect this HWPT to
476 * @out_hwpt_id: The ID of the new HWPT
477 * @__reserved: Must be 0
478 * @data_type: One of enum iommu_hwpt_data_type
479 * @data_len: Length of the type specific data
480 * @data_uptr: User pointer to the type specific data
481 * @fault_id: The ID of IOMMUFD_FAULT object. Valid only if flags field of
482 * IOMMU_HWPT_FAULT_ID_VALID is set.
483 * @__reserved2: Padding to 64-bit alignment. Must be 0.
484 *
485 * Explicitly allocate a hardware page table object. This is the same object
486 * type that is returned by iommufd_device_attach() and represents the
487 * underlying iommu driver's iommu_domain kernel object.
488 *
489 * A kernel-managed HWPT will be created with the mappings from the given
490 * IOAS via the @pt_id. The @data_type for this allocation must be set to
491 * IOMMU_HWPT_DATA_NONE. The HWPT can be allocated as a parent HWPT for a
492 * nesting configuration by passing IOMMU_HWPT_ALLOC_NEST_PARENT via @flags.
493 *
494 * A user-managed nested HWPT will be created from a given vIOMMU (wrapping a
495 * parent HWPT) or a parent HWPT via @pt_id, in which the parent HWPT must be
496 * allocated previously via the same ioctl from a given IOAS (@pt_id). In this
497 * case, the @data_type must be set to a pre-defined type corresponding to an
498 * I/O page table type supported by the underlying IOMMU hardware. The device
499 * via @dev_id and the vIOMMU via @pt_id must be associated to the same IOMMU
500 * instance.
501 *
502 * If the @data_type is set to IOMMU_HWPT_DATA_NONE, @data_len and
503 * @data_uptr should be zero. Otherwise, both @data_len and @data_uptr
504 * must be given.
505 */
506struct iommu_hwpt_alloc {
507 __u32 size;
508 __u32 flags;
509 __u32 dev_id;
510 __u32 pt_id;
511 __u32 out_hwpt_id;
512 __u32 __reserved;
513 __u32 data_type;
514 __u32 data_len;
515 __aligned_u64 data_uptr;
516 __u32 fault_id;
517 __u32 __reserved2;
518};
519#define IOMMU_HWPT_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_ALLOC)
520
521/**
522 * enum iommu_hw_info_vtd_flags - Flags for VT-d hw_info
523 * @IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17: If set, disallow read-only mappings
524 * on a nested_parent domain.
525 * https://www.intel.com/content/www/us/en/content-details/772415/content-details.html
526 */
527enum iommu_hw_info_vtd_flags {
528 IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17 = 1 << 0,
529};
530
531/**
532 * struct iommu_hw_info_vtd - Intel VT-d hardware information
533 *
534 * @flags: Combination of enum iommu_hw_info_vtd_flags
535 * @__reserved: Must be 0
536 *
537 * @cap_reg: Value of Intel VT-d capability register defined in VT-d spec
538 * section 11.4.2 Capability Register.
539 * @ecap_reg: Value of Intel VT-d capability register defined in VT-d spec
540 * section 11.4.3 Extended Capability Register.
541 *
542 * User needs to understand the Intel VT-d specification to decode the
543 * register value.
544 */
545struct iommu_hw_info_vtd {
546 __u32 flags;
547 __u32 __reserved;
548 __aligned_u64 cap_reg;
549 __aligned_u64 ecap_reg;
550};
551
552/**
553 * struct iommu_hw_info_arm_smmuv3 - ARM SMMUv3 hardware information
554 * (IOMMU_HW_INFO_TYPE_ARM_SMMUV3)
555 *
556 * @flags: Must be set to 0
557 * @__reserved: Must be 0
558 * @idr: Implemented features for ARM SMMU Non-secure programming interface
559 * @iidr: Information about the implementation and implementer of ARM SMMU,
560 * and architecture version supported
561 * @aidr: ARM SMMU architecture version
562 *
563 * For the details of @idr, @iidr and @aidr, please refer to the chapters
564 * from 6.3.1 to 6.3.6 in the SMMUv3 Spec.
565 *
566 * This reports the raw HW capability, and not all bits are meaningful to be
567 * read by userspace. Only the following fields should be used:
568 *
569 * idr[0]: ST_LEVEL, TERM_MODEL, STALL_MODEL, TTENDIAN , CD2L, ASID16, TTF
570 * idr[1]: SIDSIZE, SSIDSIZE
571 * idr[3]: BBML, RIL
572 * idr[5]: VAX, GRAN64K, GRAN16K, GRAN4K
573 *
574 * - S1P should be assumed to be true if a NESTED HWPT can be created
575 * - VFIO/iommufd only support platforms with COHACC, it should be assumed to be
576 * true.
577 * - ATS is a per-device property. If the VMM describes any devices as ATS
578 * capable in ACPI/DT it should set the corresponding idr.
579 *
580 * This list may expand in future (eg E0PD, AIE, PBHA, D128, DS etc). It is
581 * important that VMMs do not read bits outside the list to allow for
582 * compatibility with future kernels. Several features in the SMMUv3
583 * architecture are not currently supported by the kernel for nesting: HTTU,
584 * BTM, MPAM and others.
585 */
586struct iommu_hw_info_arm_smmuv3 {
587 __u32 flags;
588 __u32 __reserved;
589 __u32 idr[6];
590 __u32 iidr;
591 __u32 aidr;
592};
593
594/**
595 * struct iommu_hw_info_tegra241_cmdqv - NVIDIA Tegra241 CMDQV Hardware
596 * Information (IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV)
597 *
598 * @flags: Must be 0
599 * @version: Version number for the CMDQ-V HW for PARAM bits[03:00]
600 * @log2vcmdqs: Log2 of the total number of VCMDQs for PARAM bits[07:04]
601 * @log2vsids: Log2 of the total number of SID replacements for PARAM bits[15:12]
602 * @__reserved: Must be 0
603 *
604 * VMM can use these fields directly in its emulated global PARAM register. Note
605 * that only one Virtual Interface (VINTF) should be exposed to a VM, i.e. PARAM
606 * bits[11:08] should be set to 0 for log2 of the total number of VINTFs.
607 */
608struct iommu_hw_info_tegra241_cmdqv {
609 __u32 flags;
610 __u8 version;
611 __u8 log2vcmdqs;
612 __u8 log2vsids;
613 __u8 __reserved;
614};
615
616/**
617 * enum iommu_hw_info_type - IOMMU Hardware Info Types
618 * @IOMMU_HW_INFO_TYPE_NONE: Output by the drivers that do not report hardware
619 * info
620 * @IOMMU_HW_INFO_TYPE_DEFAULT: Input to request for a default type
621 * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type
622 * @IOMMU_HW_INFO_TYPE_ARM_SMMUV3: ARM SMMUv3 iommu info type
623 * @IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM
624 * SMMUv3) info type
625 */
626enum iommu_hw_info_type {
627 IOMMU_HW_INFO_TYPE_NONE = 0,
628 IOMMU_HW_INFO_TYPE_DEFAULT = 0,
629 IOMMU_HW_INFO_TYPE_INTEL_VTD = 1,
630 IOMMU_HW_INFO_TYPE_ARM_SMMUV3 = 2,
631 IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV = 3,
632};
633
634/**
635 * enum iommufd_hw_capabilities
636 * @IOMMU_HW_CAP_DIRTY_TRACKING: IOMMU hardware support for dirty tracking
637 * If available, it means the following APIs
638 * are supported:
639 *
640 * IOMMU_HWPT_GET_DIRTY_BITMAP
641 * IOMMU_HWPT_SET_DIRTY_TRACKING
642 *
643 * @IOMMU_HW_CAP_PCI_PASID_EXEC: Execute Permission Supported, user ignores it
644 * when the struct
645 * iommu_hw_info::out_max_pasid_log2 is zero.
646 * @IOMMU_HW_CAP_PCI_PASID_PRIV: Privileged Mode Supported, user ignores it
647 * when the struct
648 * iommu_hw_info::out_max_pasid_log2 is zero.
649 */
650enum iommufd_hw_capabilities {
651 IOMMU_HW_CAP_DIRTY_TRACKING = 1 << 0,
652 IOMMU_HW_CAP_PCI_PASID_EXEC = 1 << 1,
653 IOMMU_HW_CAP_PCI_PASID_PRIV = 1 << 2,
654};
655
656/**
657 * enum iommufd_hw_info_flags - Flags for iommu_hw_info
658 * @IOMMU_HW_INFO_FLAG_INPUT_TYPE: If set, @in_data_type carries an input type
659 * for user space to request for a specific info
660 */
661enum iommufd_hw_info_flags {
662 IOMMU_HW_INFO_FLAG_INPUT_TYPE = 1 << 0,
663};
664
665/**
666 * struct iommu_hw_info - ioctl(IOMMU_GET_HW_INFO)
667 * @size: sizeof(struct iommu_hw_info)
668 * @flags: Must be 0
669 * @dev_id: The device bound to the iommufd
670 * @data_len: Input the length of a user buffer in bytes. Output the length of
671 * data that kernel supports
672 * @data_uptr: User pointer to a user-space buffer used by the kernel to fill
673 * the iommu type specific hardware information data
674 * @in_data_type: This shares the same field with @out_data_type, making it be
675 * a bidirectional field. When IOMMU_HW_INFO_FLAG_INPUT_TYPE is
676 * set, an input type carried via this @in_data_type field will
677 * be valid, requesting for the info data to the given type. If
678 * IOMMU_HW_INFO_FLAG_INPUT_TYPE is unset, any input value will
679 * be seen as IOMMU_HW_INFO_TYPE_DEFAULT
680 * @out_data_type: Output the iommu hardware info type as defined in the enum
681 * iommu_hw_info_type.
682 * @out_capabilities: Output the generic iommu capability info type as defined
683 * in the enum iommu_hw_capabilities.
684 * @out_max_pasid_log2: Output the width of PASIDs. 0 means no PASID support.
685 * PCI devices turn to out_capabilities to check if the
686 * specific capabilities is supported or not.
687 * @__reserved: Must be 0
688 *
689 * Query an iommu type specific hardware information data from an iommu behind
690 * a given device that has been bound to iommufd. This hardware info data will
691 * be used to sync capabilities between the virtual iommu and the physical
692 * iommu, e.g. a nested translation setup needs to check the hardware info, so
693 * a guest stage-1 page table can be compatible with the physical iommu.
694 *
695 * To capture an iommu type specific hardware information data, @data_uptr and
696 * its length @data_len must be provided. Trailing bytes will be zeroed if the
697 * user buffer is larger than the data that kernel has. Otherwise, kernel only
698 * fills the buffer using the given length in @data_len. If the ioctl succeeds,
699 * @data_len will be updated to the length that kernel actually supports,
700 * @out_data_type will be filled to decode the data filled in the buffer
701 * pointed by @data_uptr. Input @data_len == zero is allowed.
702 */
703struct iommu_hw_info {
704 __u32 size;
705 __u32 flags;
706 __u32 dev_id;
707 __u32 data_len;
708 __aligned_u64 data_uptr;
709 union {
710 __u32 in_data_type;
711 __u32 out_data_type;
712 };
713 __u8 out_max_pasid_log2;
714 __u8 __reserved[3];
715 __aligned_u64 out_capabilities;
716};
717#define IOMMU_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_GET_HW_INFO)
718
719/*
720 * enum iommufd_hwpt_set_dirty_tracking_flags - Flags for steering dirty
721 * tracking
722 * @IOMMU_HWPT_DIRTY_TRACKING_ENABLE: Enable dirty tracking
723 */
724enum iommufd_hwpt_set_dirty_tracking_flags {
725 IOMMU_HWPT_DIRTY_TRACKING_ENABLE = 1,
726};
727
728/**
729 * struct iommu_hwpt_set_dirty_tracking - ioctl(IOMMU_HWPT_SET_DIRTY_TRACKING)
730 * @size: sizeof(struct iommu_hwpt_set_dirty_tracking)
731 * @flags: Combination of enum iommufd_hwpt_set_dirty_tracking_flags
732 * @hwpt_id: HW pagetable ID that represents the IOMMU domain
733 * @__reserved: Must be 0
734 *
735 * Toggle dirty tracking on an HW pagetable.
736 */
737struct iommu_hwpt_set_dirty_tracking {
738 __u32 size;
739 __u32 flags;
740 __u32 hwpt_id;
741 __u32 __reserved;
742};
743#define IOMMU_HWPT_SET_DIRTY_TRACKING _IO(IOMMUFD_TYPE, \
744 IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING)
745
746/**
747 * enum iommufd_hwpt_get_dirty_bitmap_flags - Flags for getting dirty bits
748 * @IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR: Just read the PTEs without clearing
749 * any dirty bits metadata. This flag
750 * can be passed in the expectation
751 * where the next operation is an unmap
752 * of the same IOVA range.
753 *
754 */
755enum iommufd_hwpt_get_dirty_bitmap_flags {
756 IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR = 1,
757};
758
759/**
760 * struct iommu_hwpt_get_dirty_bitmap - ioctl(IOMMU_HWPT_GET_DIRTY_BITMAP)
761 * @size: sizeof(struct iommu_hwpt_get_dirty_bitmap)
762 * @hwpt_id: HW pagetable ID that represents the IOMMU domain
763 * @flags: Combination of enum iommufd_hwpt_get_dirty_bitmap_flags
764 * @__reserved: Must be 0
765 * @iova: base IOVA of the bitmap first bit
766 * @length: IOVA range size
767 * @page_size: page size granularity of each bit in the bitmap
768 * @data: bitmap where to set the dirty bits. The bitmap bits each
769 * represent a page_size which you deviate from an arbitrary iova.
770 *
771 * Checking a given IOVA is dirty:
772 *
773 * data[(iova / page_size) / 64] & (1ULL << ((iova / page_size) % 64))
774 *
775 * Walk the IOMMU pagetables for a given IOVA range to return a bitmap
776 * with the dirty IOVAs. In doing so it will also by default clear any
777 * dirty bit metadata set in the IOPTE.
778 */
779struct iommu_hwpt_get_dirty_bitmap {
780 __u32 size;
781 __u32 hwpt_id;
782 __u32 flags;
783 __u32 __reserved;
784 __aligned_u64 iova;
785 __aligned_u64 length;
786 __aligned_u64 page_size;
787 __aligned_u64 data;
788};
789#define IOMMU_HWPT_GET_DIRTY_BITMAP _IO(IOMMUFD_TYPE, \
790 IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP)
791
792/**
793 * enum iommu_hwpt_invalidate_data_type - IOMMU HWPT Cache Invalidation
794 * Data Type
795 * @IOMMU_HWPT_INVALIDATE_DATA_VTD_S1: Invalidation data for VTD_S1
796 * @IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3: Invalidation data for ARM SMMUv3
797 */
798enum iommu_hwpt_invalidate_data_type {
799 IOMMU_HWPT_INVALIDATE_DATA_VTD_S1 = 0,
800 IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3 = 1,
801};
802
803/**
804 * enum iommu_hwpt_vtd_s1_invalidate_flags - Flags for Intel VT-d
805 * stage-1 cache invalidation
806 * @IOMMU_VTD_INV_FLAGS_LEAF: Indicates whether the invalidation applies
807 * to all-levels page structure cache or just
808 * the leaf PTE cache.
809 */
810enum iommu_hwpt_vtd_s1_invalidate_flags {
811 IOMMU_VTD_INV_FLAGS_LEAF = 1 << 0,
812};
813
814/**
815 * struct iommu_hwpt_vtd_s1_invalidate - Intel VT-d cache invalidation
816 * (IOMMU_HWPT_INVALIDATE_DATA_VTD_S1)
817 * @addr: The start address of the range to be invalidated. It needs to
818 * be 4KB aligned.
819 * @npages: Number of contiguous 4K pages to be invalidated.
820 * @flags: Combination of enum iommu_hwpt_vtd_s1_invalidate_flags
821 * @__reserved: Must be 0
822 *
823 * The Intel VT-d specific invalidation data for user-managed stage-1 cache
824 * invalidation in nested translation. Userspace uses this structure to
825 * tell the impacted cache scope after modifying the stage-1 page table.
826 *
827 * Invalidating all the caches related to the page table by setting @addr
828 * to be 0 and @npages to be U64_MAX.
829 *
830 * The device TLB will be invalidated automatically if ATS is enabled.
831 */
832struct iommu_hwpt_vtd_s1_invalidate {
833 __aligned_u64 addr;
834 __aligned_u64 npages;
835 __u32 flags;
836 __u32 __reserved;
837};
838
839/**
840 * struct iommu_viommu_arm_smmuv3_invalidate - ARM SMMUv3 cache invalidation
841 * (IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3)
842 * @cmd: 128-bit cache invalidation command that runs in SMMU CMDQ.
843 * Must be little-endian.
844 *
845 * Supported command list only when passing in a vIOMMU via @hwpt_id:
846 * CMDQ_OP_TLBI_NSNH_ALL
847 * CMDQ_OP_TLBI_NH_VA
848 * CMDQ_OP_TLBI_NH_VAA
849 * CMDQ_OP_TLBI_NH_ALL
850 * CMDQ_OP_TLBI_NH_ASID
851 * CMDQ_OP_ATC_INV
852 * CMDQ_OP_CFGI_CD
853 * CMDQ_OP_CFGI_CD_ALL
854 *
855 * -EIO will be returned if the command is not supported.
856 */
857struct iommu_viommu_arm_smmuv3_invalidate {
858 __aligned_le64 cmd[2];
859};
860
861/**
862 * struct iommu_hwpt_invalidate - ioctl(IOMMU_HWPT_INVALIDATE)
863 * @size: sizeof(struct iommu_hwpt_invalidate)
864 * @hwpt_id: ID of a nested HWPT or a vIOMMU, for cache invalidation
865 * @data_uptr: User pointer to an array of driver-specific cache invalidation
866 * data.
867 * @data_type: One of enum iommu_hwpt_invalidate_data_type, defining the data
868 * type of all the entries in the invalidation request array. It
869 * should be a type supported by the hwpt pointed by @hwpt_id.
870 * @entry_len: Length (in bytes) of a request entry in the request array
871 * @entry_num: Input the number of cache invalidation requests in the array.
872 * Output the number of requests successfully handled by kernel.
873 * @__reserved: Must be 0.
874 *
875 * Invalidate iommu cache for user-managed page table or vIOMMU. Modifications
876 * on a user-managed page table should be followed by this operation, if a HWPT
877 * is passed in via @hwpt_id. Other caches, such as device cache or descriptor
878 * cache can be flushed if a vIOMMU is passed in via the @hwpt_id field.
879 *
880 * Each ioctl can support one or more cache invalidation requests in the array
881 * that has a total size of @entry_len * @entry_num.
882 *
883 * An empty invalidation request array by setting @entry_num==0 is allowed, and
884 * @entry_len and @data_uptr would be ignored in this case. This can be used to
885 * check if the given @data_type is supported or not by kernel.
886 */
887struct iommu_hwpt_invalidate {
888 __u32 size;
889 __u32 hwpt_id;
890 __aligned_u64 data_uptr;
891 __u32 data_type;
892 __u32 entry_len;
893 __u32 entry_num;
894 __u32 __reserved;
895};
896#define IOMMU_HWPT_INVALIDATE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_INVALIDATE)
897
898/**
899 * enum iommu_hwpt_pgfault_flags - flags for struct iommu_hwpt_pgfault
900 * @IOMMU_PGFAULT_FLAGS_PASID_VALID: The pasid field of the fault data is
901 * valid.
902 * @IOMMU_PGFAULT_FLAGS_LAST_PAGE: It's the last fault of a fault group.
903 */
904enum iommu_hwpt_pgfault_flags {
905 IOMMU_PGFAULT_FLAGS_PASID_VALID = (1 << 0),
906 IOMMU_PGFAULT_FLAGS_LAST_PAGE = (1 << 1),
907};
908
909/**
910 * enum iommu_hwpt_pgfault_perm - perm bits for struct iommu_hwpt_pgfault
911 * @IOMMU_PGFAULT_PERM_READ: request for read permission
912 * @IOMMU_PGFAULT_PERM_WRITE: request for write permission
913 * @IOMMU_PGFAULT_PERM_EXEC: (PCIE 10.4.1) request with a PASID that has the
914 * Execute Requested bit set in PASID TLP Prefix.
915 * @IOMMU_PGFAULT_PERM_PRIV: (PCIE 10.4.1) request with a PASID that has the
916 * Privileged Mode Requested bit set in PASID TLP
917 * Prefix.
918 */
919enum iommu_hwpt_pgfault_perm {
920 IOMMU_PGFAULT_PERM_READ = (1 << 0),
921 IOMMU_PGFAULT_PERM_WRITE = (1 << 1),
922 IOMMU_PGFAULT_PERM_EXEC = (1 << 2),
923 IOMMU_PGFAULT_PERM_PRIV = (1 << 3),
924};
925
926/**
927 * struct iommu_hwpt_pgfault - iommu page fault data
928 * @flags: Combination of enum iommu_hwpt_pgfault_flags
929 * @dev_id: id of the originated device
930 * @pasid: Process Address Space ID
931 * @grpid: Page Request Group Index
932 * @perm: Combination of enum iommu_hwpt_pgfault_perm
933 * @__reserved: Must be 0.
934 * @addr: Fault address
935 * @length: a hint of how much data the requestor is expecting to fetch. For
936 * example, if the PRI initiator knows it is going to do a 10MB
937 * transfer, it could fill in 10MB and the OS could pre-fault in
938 * 10MB of IOVA. It's default to 0 if there's no such hint.
939 * @cookie: kernel-managed cookie identifying a group of fault messages. The
940 * cookie number encoded in the last page fault of the group should
941 * be echoed back in the response message.
942 */
943struct iommu_hwpt_pgfault {
944 __u32 flags;
945 __u32 dev_id;
946 __u32 pasid;
947 __u32 grpid;
948 __u32 perm;
949 __u32 __reserved;
950 __aligned_u64 addr;
951 __u32 length;
952 __u32 cookie;
953};
954
955/**
956 * enum iommufd_page_response_code - Return status of fault handlers
957 * @IOMMUFD_PAGE_RESP_SUCCESS: Fault has been handled and the page tables
958 * populated, retry the access. This is the
959 * "Success" defined in PCI 10.4.2.1.
960 * @IOMMUFD_PAGE_RESP_INVALID: Could not handle this fault, don't retry the
961 * access. This is the "Invalid Request" in PCI
962 * 10.4.2.1.
963 */
964enum iommufd_page_response_code {
965 IOMMUFD_PAGE_RESP_SUCCESS = 0,
966 IOMMUFD_PAGE_RESP_INVALID = 1,
967};
968
969/**
970 * struct iommu_hwpt_page_response - IOMMU page fault response
971 * @cookie: The kernel-managed cookie reported in the fault message.
972 * @code: One of response code in enum iommufd_page_response_code.
973 */
974struct iommu_hwpt_page_response {
975 __u32 cookie;
976 __u32 code;
977};
978
979/**
980 * struct iommu_fault_alloc - ioctl(IOMMU_FAULT_QUEUE_ALLOC)
981 * @size: sizeof(struct iommu_fault_alloc)
982 * @flags: Must be 0
983 * @out_fault_id: The ID of the new FAULT
984 * @out_fault_fd: The fd of the new FAULT
985 *
986 * Explicitly allocate a fault handling object.
987 */
988struct iommu_fault_alloc {
989 __u32 size;
990 __u32 flags;
991 __u32 out_fault_id;
992 __u32 out_fault_fd;
993};
994#define IOMMU_FAULT_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_FAULT_QUEUE_ALLOC)
995
996/**
997 * enum iommu_viommu_type - Virtual IOMMU Type
998 * @IOMMU_VIOMMU_TYPE_DEFAULT: Reserved for future use
999 * @IOMMU_VIOMMU_TYPE_ARM_SMMUV3: ARM SMMUv3 driver specific type
1000 * @IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM
1001 * SMMUv3) enabled ARM SMMUv3 type
1002 */
1003enum iommu_viommu_type {
1004 IOMMU_VIOMMU_TYPE_DEFAULT = 0,
1005 IOMMU_VIOMMU_TYPE_ARM_SMMUV3 = 1,
1006 IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV = 2,
1007};
1008
1009/**
1010 * struct iommu_viommu_tegra241_cmdqv - NVIDIA Tegra241 CMDQV Virtual Interface
1011 * (IOMMU_VIOMMU_TYPE_TEGRA241_CMDQV)
1012 * @out_vintf_mmap_offset: mmap offset argument for VINTF's page0
1013 * @out_vintf_mmap_length: mmap length argument for VINTF's page0
1014 *
1015 * Both @out_vintf_mmap_offset and @out_vintf_mmap_length are reported by kernel
1016 * for user space to mmap the VINTF page0 from the host physical address space
1017 * to the guest physical address space so that a guest kernel can directly R/W
1018 * access to the VINTF page0 in order to control its virtual command queues.
1019 */
1020struct iommu_viommu_tegra241_cmdqv {
1021 __aligned_u64 out_vintf_mmap_offset;
1022 __aligned_u64 out_vintf_mmap_length;
1023};
1024
1025/**
1026 * struct iommu_viommu_alloc - ioctl(IOMMU_VIOMMU_ALLOC)
1027 * @size: sizeof(struct iommu_viommu_alloc)
1028 * @flags: Must be 0
1029 * @type: Type of the virtual IOMMU. Must be defined in enum iommu_viommu_type
1030 * @dev_id: The device's physical IOMMU will be used to back the virtual IOMMU
1031 * @hwpt_id: ID of a nesting parent HWPT to associate to
1032 * @out_viommu_id: Output virtual IOMMU ID for the allocated object
1033 * @data_len: Length of the type specific data
1034 * @__reserved: Must be 0
1035 * @data_uptr: User pointer to a driver-specific virtual IOMMU data
1036 *
1037 * Allocate a virtual IOMMU object, representing the underlying physical IOMMU's
1038 * virtualization support that is a security-isolated slice of the real IOMMU HW
1039 * that is unique to a specific VM. Operations global to the IOMMU are connected
1040 * to the vIOMMU, such as:
1041 * - Security namespace for guest owned ID, e.g. guest-controlled cache tags
1042 * - Non-device-affiliated event reporting, e.g. invalidation queue errors
1043 * - Access to a sharable nesting parent pagetable across physical IOMMUs
1044 * - Virtualization of various platforms IDs, e.g. RIDs and others
1045 * - Delivery of paravirtualized invalidation
1046 * - Direct assigned invalidation queues
1047 * - Direct assigned interrupts
1048 */
1049struct iommu_viommu_alloc {
1050 __u32 size;
1051 __u32 flags;
1052 __u32 type;
1053 __u32 dev_id;
1054 __u32 hwpt_id;
1055 __u32 out_viommu_id;
1056 __u32 data_len;
1057 __u32 __reserved;
1058 __aligned_u64 data_uptr;
1059};
1060#define IOMMU_VIOMMU_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VIOMMU_ALLOC)
1061
1062/**
1063 * struct iommu_vdevice_alloc - ioctl(IOMMU_VDEVICE_ALLOC)
1064 * @size: sizeof(struct iommu_vdevice_alloc)
1065 * @viommu_id: vIOMMU ID to associate with the virtual device
1066 * @dev_id: The physical device to allocate a virtual instance on the vIOMMU
1067 * @out_vdevice_id: Object handle for the vDevice. Pass to IOMMU_DESTORY
1068 * @virt_id: Virtual device ID per vIOMMU, e.g. vSID of ARM SMMUv3, vDeviceID
1069 * of AMD IOMMU, and vRID of Intel VT-d
1070 *
1071 * Allocate a virtual device instance (for a physical device) against a vIOMMU.
1072 * This instance holds the device's information (related to its vIOMMU) in a VM.
1073 * User should use IOMMU_DESTROY to destroy the virtual device before
1074 * destroying the physical device (by closing vfio_cdev fd). Otherwise the
1075 * virtual device would be forcibly destroyed on physical device destruction,
1076 * its vdevice_id would be permanently leaked (unremovable & unreusable) until
1077 * iommu fd closed.
1078 */
1079struct iommu_vdevice_alloc {
1080 __u32 size;
1081 __u32 viommu_id;
1082 __u32 dev_id;
1083 __u32 out_vdevice_id;
1084 __aligned_u64 virt_id;
1085};
1086#define IOMMU_VDEVICE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VDEVICE_ALLOC)
1087
1088/**
1089 * struct iommu_ioas_change_process - ioctl(VFIO_IOAS_CHANGE_PROCESS)
1090 * @size: sizeof(struct iommu_ioas_change_process)
1091 * @__reserved: Must be 0
1092 *
1093 * This transfers pinned memory counts for every memory map in every IOAS
1094 * in the context to the current process. This only supports maps created
1095 * with IOMMU_IOAS_MAP_FILE, and returns EINVAL if other maps are present.
1096 * If the ioctl returns a failure status, then nothing is changed.
1097 *
1098 * This API is useful for transferring operation of a device from one process
1099 * to another, such as during userland live update.
1100 */
1101struct iommu_ioas_change_process {
1102 __u32 size;
1103 __u32 __reserved;
1104};
1105
1106#define IOMMU_IOAS_CHANGE_PROCESS \
1107 _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_CHANGE_PROCESS)
1108
1109/**
1110 * enum iommu_veventq_flag - flag for struct iommufd_vevent_header
1111 * @IOMMU_VEVENTQ_FLAG_LOST_EVENTS: vEVENTQ has lost vEVENTs
1112 */
1113enum iommu_veventq_flag {
1114 IOMMU_VEVENTQ_FLAG_LOST_EVENTS = (1U << 0),
1115};
1116
1117/**
1118 * struct iommufd_vevent_header - Virtual Event Header for a vEVENTQ Status
1119 * @flags: Combination of enum iommu_veventq_flag
1120 * @sequence: The sequence index of a vEVENT in the vEVENTQ, with a range of
1121 * [0, INT_MAX] where the following index of INT_MAX is 0
1122 *
1123 * Each iommufd_vevent_header reports a sequence index of the following vEVENT:
1124 *
1125 * +----------------------+-------+----------------------+-------+---+-------+
1126 * | header0 {sequence=0} | data0 | header1 {sequence=1} | data1 |...| dataN |
1127 * +----------------------+-------+----------------------+-------+---+-------+
1128 *
1129 * And this sequence index is expected to be monotonic to the sequence index of
1130 * the previous vEVENT. If two adjacent sequence indexes has a delta larger than
1131 * 1, it means that delta - 1 number of vEVENTs has lost, e.g. two lost vEVENTs:
1132 *
1133 * +-----+----------------------+-------+----------------------+-------+-----+
1134 * | ... | header3 {sequence=3} | data3 | header6 {sequence=6} | data6 | ... |
1135 * +-----+----------------------+-------+----------------------+-------+-----+
1136 *
1137 * If a vEVENT lost at the tail of the vEVENTQ and there is no following vEVENT
1138 * providing the next sequence index, an IOMMU_VEVENTQ_FLAG_LOST_EVENTS header
1139 * would be added to the tail, and no data would follow this header:
1140 *
1141 * +--+----------------------+-------+-----------------------------------------+
1142 * |..| header3 {sequence=3} | data3 | header4 {flags=LOST_EVENTS, sequence=4} |
1143 * +--+----------------------+-------+-----------------------------------------+
1144 */
1145struct iommufd_vevent_header {
1146 __u32 flags;
1147 __u32 sequence;
1148};
1149
1150/**
1151 * enum iommu_veventq_type - Virtual Event Queue Type
1152 * @IOMMU_VEVENTQ_TYPE_DEFAULT: Reserved for future use
1153 * @IOMMU_VEVENTQ_TYPE_ARM_SMMUV3: ARM SMMUv3 Virtual Event Queue
1154 * @IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV Extension IRQ
1155 */
1156enum iommu_veventq_type {
1157 IOMMU_VEVENTQ_TYPE_DEFAULT = 0,
1158 IOMMU_VEVENTQ_TYPE_ARM_SMMUV3 = 1,
1159 IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV = 2,
1160};
1161
1162/**
1163 * struct iommu_vevent_arm_smmuv3 - ARM SMMUv3 Virtual Event
1164 * (IOMMU_VEVENTQ_TYPE_ARM_SMMUV3)
1165 * @evt: 256-bit ARM SMMUv3 Event record, little-endian.
1166 * Reported event records: (Refer to "7.3 Event records" in SMMUv3 HW Spec)
1167 * - 0x04 C_BAD_STE
1168 * - 0x06 F_STREAM_DISABLED
1169 * - 0x08 C_BAD_SUBSTREAMID
1170 * - 0x0a C_BAD_CD
1171 * - 0x10 F_TRANSLATION
1172 * - 0x11 F_ADDR_SIZE
1173 * - 0x12 F_ACCESS
1174 * - 0x13 F_PERMISSION
1175 *
1176 * StreamID field reports a virtual device ID. To receive a virtual event for a
1177 * device, a vDEVICE must be allocated via IOMMU_VDEVICE_ALLOC.
1178 */
1179struct iommu_vevent_arm_smmuv3 {
1180 __aligned_le64 evt[4];
1181};
1182
1183/**
1184 * struct iommu_vevent_tegra241_cmdqv - Tegra241 CMDQV IRQ
1185 * (IOMMU_VEVENTQ_TYPE_TEGRA241_CMDQV)
1186 * @lvcmdq_err_map: 128-bit logical vcmdq error map, little-endian.
1187 * (Refer to register LVCMDQ_ERR_MAPs per VINTF )
1188 *
1189 * The 128-bit register value from HW exclusively reflect the error bits for a
1190 * Virtual Interface represented by a vIOMMU object. Read and report directly.
1191 */
1192struct iommu_vevent_tegra241_cmdqv {
1193 __aligned_le64 lvcmdq_err_map[2];
1194};
1195
1196/**
1197 * struct iommu_veventq_alloc - ioctl(IOMMU_VEVENTQ_ALLOC)
1198 * @size: sizeof(struct iommu_veventq_alloc)
1199 * @flags: Must be 0
1200 * @viommu_id: virtual IOMMU ID to associate the vEVENTQ with
1201 * @type: Type of the vEVENTQ. Must be defined in enum iommu_veventq_type
1202 * @veventq_depth: Maximum number of events in the vEVENTQ
1203 * @out_veventq_id: The ID of the new vEVENTQ
1204 * @out_veventq_fd: The fd of the new vEVENTQ. User space must close the
1205 * successfully returned fd after using it
1206 * @__reserved: Must be 0
1207 *
1208 * Explicitly allocate a virtual event queue interface for a vIOMMU. A vIOMMU
1209 * can have multiple FDs for different types, but is confined to one per @type.
1210 * User space should open the @out_veventq_fd to read vEVENTs out of a vEVENTQ,
1211 * if there are vEVENTs available. A vEVENTQ will lose events due to overflow,
1212 * if the number of the vEVENTs hits @veventq_depth.
1213 *
1214 * Each vEVENT in a vEVENTQ encloses a struct iommufd_vevent_header followed by
1215 * a type-specific data structure, in a normal case:
1216 *
1217 * +-+---------+-------+---------+-------+-----+---------+-------+-+
1218 * | | header0 | data0 | header1 | data1 | ... | headerN | dataN | |
1219 * +-+---------+-------+---------+-------+-----+---------+-------+-+
1220 *
1221 * unless a tailing IOMMU_VEVENTQ_FLAG_LOST_EVENTS header is logged (refer to
1222 * struct iommufd_vevent_header).
1223 */
1224struct iommu_veventq_alloc {
1225 __u32 size;
1226 __u32 flags;
1227 __u32 viommu_id;
1228 __u32 type;
1229 __u32 veventq_depth;
1230 __u32 out_veventq_id;
1231 __u32 out_veventq_fd;
1232 __u32 __reserved;
1233};
1234#define IOMMU_VEVENTQ_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VEVENTQ_ALLOC)
1235
1236/**
1237 * enum iommu_hw_queue_type - HW Queue Type
1238 * @IOMMU_HW_QUEUE_TYPE_DEFAULT: Reserved for future use
1239 * @IOMMU_HW_QUEUE_TYPE_TEGRA241_CMDQV: NVIDIA Tegra241 CMDQV (extension for ARM
1240 * SMMUv3) Virtual Command Queue (VCMDQ)
1241 */
1242enum iommu_hw_queue_type {
1243 IOMMU_HW_QUEUE_TYPE_DEFAULT = 0,
1244 /*
1245 * TEGRA241_CMDQV requirements (otherwise, allocation will fail)
1246 * - alloc starts from the lowest @index=0 in ascending order
1247 * - destroy starts from the last allocated @index in descending order
1248 * - @base_addr must be aligned to @length in bytes and mapped in IOAS
1249 * - @length must be a power of 2, with a minimum 32 bytes and a maximum
1250 * 2 ^ idr[1].CMDQS * 16 bytes (use GET_HW_INFO call to read idr[1]
1251 * from struct iommu_hw_info_arm_smmuv3)
1252 * - suggest to back the queue memory with contiguous physical pages or
1253 * a single huge page with alignment of the queue size, and limit the
1254 * emulated vSMMU's IDR1.CMDQS to log2(huge page size / 16 bytes)
1255 */
1256 IOMMU_HW_QUEUE_TYPE_TEGRA241_CMDQV = 1,
1257};
1258
1259/**
1260 * struct iommu_hw_queue_alloc - ioctl(IOMMU_HW_QUEUE_ALLOC)
1261 * @size: sizeof(struct iommu_hw_queue_alloc)
1262 * @flags: Must be 0
1263 * @viommu_id: Virtual IOMMU ID to associate the HW queue with
1264 * @type: One of enum iommu_hw_queue_type
1265 * @index: The logical index to the HW queue per virtual IOMMU for a multi-queue
1266 * model
1267 * @out_hw_queue_id: The ID of the new HW queue
1268 * @nesting_parent_iova: Base address of the queue memory in the guest physical
1269 * address space
1270 * @length: Length of the queue memory
1271 *
1272 * Allocate a HW queue object for a vIOMMU-specific HW-accelerated queue, which
1273 * allows HW to access a guest queue memory described using @nesting_parent_iova
1274 * and @length.
1275 *
1276 * A vIOMMU can allocate multiple queues, but it must use a different @index per
1277 * type to separate each allocation, e.g::
1278 *
1279 * Type1 HW queue0, Type1 HW queue1, Type2 HW queue0, ...
1280 */
1281struct iommu_hw_queue_alloc {
1282 __u32 size;
1283 __u32 flags;
1284 __u32 viommu_id;
1285 __u32 type;
1286 __u32 index;
1287 __u32 out_hw_queue_id;
1288 __aligned_u64 nesting_parent_iova;
1289 __aligned_u64 length;
1290};
1291#define IOMMU_HW_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HW_QUEUE_ALLOC)
1292#endif
1293