xref: /qemu/subprojects/libvduse/linux-headers/linux/iommufd.h (revision 7be29f2f1a3f5b037d27eedbd5df9f441e8c8c16)
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/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  */
37 enum {
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 };
60 
61 /**
62  * struct iommu_destroy - ioctl(IOMMU_DESTROY)
63  * @size: sizeof(struct iommu_destroy)
64  * @id: iommufd object ID to destroy. Can be any destroyable object type.
65  *
66  * Destroy any object held within iommufd.
67  */
68 struct iommu_destroy {
69 	__u32 size;
70 	__u32 id;
71 };
72 #define IOMMU_DESTROY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DESTROY)
73 
74 /**
75  * struct iommu_ioas_alloc - ioctl(IOMMU_IOAS_ALLOC)
76  * @size: sizeof(struct iommu_ioas_alloc)
77  * @flags: Must be 0
78  * @out_ioas_id: Output IOAS ID for the allocated object
79  *
80  * Allocate an IO Address Space (IOAS) which holds an IO Virtual Address (IOVA)
81  * to memory mapping.
82  */
83 struct iommu_ioas_alloc {
84 	__u32 size;
85 	__u32 flags;
86 	__u32 out_ioas_id;
87 };
88 #define IOMMU_IOAS_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOC)
89 
90 /**
91  * struct iommu_iova_range - ioctl(IOMMU_IOVA_RANGE)
92  * @start: First IOVA
93  * @last: Inclusive last IOVA
94  *
95  * An interval in IOVA space.
96  */
97 struct iommu_iova_range {
98 	__aligned_u64 start;
99 	__aligned_u64 last;
100 };
101 
102 /**
103  * struct iommu_ioas_iova_ranges - ioctl(IOMMU_IOAS_IOVA_RANGES)
104  * @size: sizeof(struct iommu_ioas_iova_ranges)
105  * @ioas_id: IOAS ID to read ranges from
106  * @num_iovas: Input/Output total number of ranges in the IOAS
107  * @__reserved: Must be 0
108  * @allowed_iovas: Pointer to the output array of struct iommu_iova_range
109  * @out_iova_alignment: Minimum alignment required for mapping IOVA
110  *
111  * Query an IOAS for ranges of allowed IOVAs. Mapping IOVA outside these ranges
112  * is not allowed. num_iovas will be set to the total number of iovas and
113  * the allowed_iovas[] will be filled in as space permits.
114  *
115  * The allowed ranges are dependent on the HW path the DMA operation takes, and
116  * can change during the lifetime of the IOAS. A fresh empty IOAS will have a
117  * full range, and each attached device will narrow the ranges based on that
118  * device's HW restrictions. Detaching a device can widen the ranges. Userspace
119  * should query ranges after every attach/detach to know what IOVAs are valid
120  * for mapping.
121  *
122  * On input num_iovas is the length of the allowed_iovas array. On output it is
123  * the total number of iovas filled in. The ioctl will return -EMSGSIZE and set
124  * num_iovas to the required value if num_iovas is too small. In this case the
125  * caller should allocate a larger output array and re-issue the ioctl.
126  *
127  * out_iova_alignment returns the minimum IOVA alignment that can be given
128  * to IOMMU_IOAS_MAP/COPY. IOVA's must satisfy::
129  *
130  *   starting_iova % out_iova_alignment == 0
131  *   (starting_iova + length) % out_iova_alignment == 0
132  *
133  * out_iova_alignment can be 1 indicating any IOVA is allowed. It cannot
134  * be higher than the system PAGE_SIZE.
135  */
136 struct iommu_ioas_iova_ranges {
137 	__u32 size;
138 	__u32 ioas_id;
139 	__u32 num_iovas;
140 	__u32 __reserved;
141 	__aligned_u64 allowed_iovas;
142 	__aligned_u64 out_iova_alignment;
143 };
144 #define IOMMU_IOAS_IOVA_RANGES _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_IOVA_RANGES)
145 
146 /**
147  * struct iommu_ioas_allow_iovas - ioctl(IOMMU_IOAS_ALLOW_IOVAS)
148  * @size: sizeof(struct iommu_ioas_allow_iovas)
149  * @ioas_id: IOAS ID to allow IOVAs from
150  * @num_iovas: Input/Output total number of ranges in the IOAS
151  * @__reserved: Must be 0
152  * @allowed_iovas: Pointer to array of struct iommu_iova_range
153  *
154  * Ensure a range of IOVAs are always available for allocation. If this call
155  * succeeds then IOMMU_IOAS_IOVA_RANGES will never return a list of IOVA ranges
156  * that are narrower than the ranges provided here. This call will fail if
157  * IOMMU_IOAS_IOVA_RANGES is currently narrower than the given ranges.
158  *
159  * When an IOAS is first created the IOVA_RANGES will be maximally sized, and as
160  * devices are attached the IOVA will narrow based on the device restrictions.
161  * When an allowed range is specified any narrowing will be refused, ie device
162  * attachment can fail if the device requires limiting within the allowed range.
163  *
164  * Automatic IOVA allocation is also impacted by this call. MAP will only
165  * allocate within the allowed IOVAs if they are present.
166  *
167  * This call replaces the entire allowed list with the given list.
168  */
169 struct iommu_ioas_allow_iovas {
170 	__u32 size;
171 	__u32 ioas_id;
172 	__u32 num_iovas;
173 	__u32 __reserved;
174 	__aligned_u64 allowed_iovas;
175 };
176 #define IOMMU_IOAS_ALLOW_IOVAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_ALLOW_IOVAS)
177 
178 /**
179  * enum iommufd_ioas_map_flags - Flags for map and copy
180  * @IOMMU_IOAS_MAP_FIXED_IOVA: If clear the kernel will compute an appropriate
181  *                             IOVA to place the mapping at
182  * @IOMMU_IOAS_MAP_WRITEABLE: DMA is allowed to write to this mapping
183  * @IOMMU_IOAS_MAP_READABLE: DMA is allowed to read from this mapping
184  */
185 enum iommufd_ioas_map_flags {
186 	IOMMU_IOAS_MAP_FIXED_IOVA = 1 << 0,
187 	IOMMU_IOAS_MAP_WRITEABLE = 1 << 1,
188 	IOMMU_IOAS_MAP_READABLE = 1 << 2,
189 };
190 
191 /**
192  * struct iommu_ioas_map - ioctl(IOMMU_IOAS_MAP)
193  * @size: sizeof(struct iommu_ioas_map)
194  * @flags: Combination of enum iommufd_ioas_map_flags
195  * @ioas_id: IOAS ID to change the mapping of
196  * @__reserved: Must be 0
197  * @user_va: Userspace pointer to start mapping from
198  * @length: Number of bytes to map
199  * @iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is set
200  *        then this must be provided as input.
201  *
202  * Set an IOVA mapping from a user pointer. If FIXED_IOVA is specified then the
203  * mapping will be established at iova, otherwise a suitable location based on
204  * the reserved and allowed lists will be automatically selected and returned in
205  * iova.
206  *
207  * If IOMMU_IOAS_MAP_FIXED_IOVA is specified then the iova range must currently
208  * be unused, existing IOVA cannot be replaced.
209  */
210 struct iommu_ioas_map {
211 	__u32 size;
212 	__u32 flags;
213 	__u32 ioas_id;
214 	__u32 __reserved;
215 	__aligned_u64 user_va;
216 	__aligned_u64 length;
217 	__aligned_u64 iova;
218 };
219 #define IOMMU_IOAS_MAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP)
220 
221 /**
222  * struct iommu_ioas_map_file - ioctl(IOMMU_IOAS_MAP_FILE)
223  * @size: sizeof(struct iommu_ioas_map_file)
224  * @flags: same as for iommu_ioas_map
225  * @ioas_id: same as for iommu_ioas_map
226  * @fd: the memfd to map
227  * @start: byte offset from start of file to map from
228  * @length: same as for iommu_ioas_map
229  * @iova: same as for iommu_ioas_map
230  *
231  * Set an IOVA mapping from a memfd file.  All other arguments and semantics
232  * match those of IOMMU_IOAS_MAP.
233  */
234 struct iommu_ioas_map_file {
235 	__u32 size;
236 	__u32 flags;
237 	__u32 ioas_id;
238 	__s32 fd;
239 	__aligned_u64 start;
240 	__aligned_u64 length;
241 	__aligned_u64 iova;
242 };
243 #define IOMMU_IOAS_MAP_FILE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_MAP_FILE)
244 
245 /**
246  * struct iommu_ioas_copy - ioctl(IOMMU_IOAS_COPY)
247  * @size: sizeof(struct iommu_ioas_copy)
248  * @flags: Combination of enum iommufd_ioas_map_flags
249  * @dst_ioas_id: IOAS ID to change the mapping of
250  * @src_ioas_id: IOAS ID to copy from
251  * @length: Number of bytes to copy and map
252  * @dst_iova: IOVA the mapping was placed at. If IOMMU_IOAS_MAP_FIXED_IOVA is
253  *            set then this must be provided as input.
254  * @src_iova: IOVA to start the copy
255  *
256  * Copy an already existing mapping from src_ioas_id and establish it in
257  * dst_ioas_id. The src iova/length must exactly match a range used with
258  * IOMMU_IOAS_MAP.
259  *
260  * This may be used to efficiently clone a subset of an IOAS to another, or as a
261  * kind of 'cache' to speed up mapping. Copy has an efficiency advantage over
262  * establishing equivalent new mappings, as internal resources are shared, and
263  * the kernel will pin the user memory only once.
264  */
265 struct iommu_ioas_copy {
266 	__u32 size;
267 	__u32 flags;
268 	__u32 dst_ioas_id;
269 	__u32 src_ioas_id;
270 	__aligned_u64 length;
271 	__aligned_u64 dst_iova;
272 	__aligned_u64 src_iova;
273 };
274 #define IOMMU_IOAS_COPY _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_COPY)
275 
276 /**
277  * struct iommu_ioas_unmap - ioctl(IOMMU_IOAS_UNMAP)
278  * @size: sizeof(struct iommu_ioas_unmap)
279  * @ioas_id: IOAS ID to change the mapping of
280  * @iova: IOVA to start the unmapping at
281  * @length: Number of bytes to unmap, and return back the bytes unmapped
282  *
283  * Unmap an IOVA range. The iova/length must be a superset of a previously
284  * mapped range used with IOMMU_IOAS_MAP or IOMMU_IOAS_COPY. Splitting or
285  * truncating ranges is not allowed. The values 0 to U64_MAX will unmap
286  * everything.
287  */
288 struct iommu_ioas_unmap {
289 	__u32 size;
290 	__u32 ioas_id;
291 	__aligned_u64 iova;
292 	__aligned_u64 length;
293 };
294 #define IOMMU_IOAS_UNMAP _IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_UNMAP)
295 
296 /**
297  * enum iommufd_option - ioctl(IOMMU_OPTION_RLIMIT_MODE) and
298  *                       ioctl(IOMMU_OPTION_HUGE_PAGES)
299  * @IOMMU_OPTION_RLIMIT_MODE:
300  *    Change how RLIMIT_MEMLOCK accounting works. The caller must have privilege
301  *    to invoke this. Value 0 (default) is user based accounting, 1 uses process
302  *    based accounting. Global option, object_id must be 0
303  * @IOMMU_OPTION_HUGE_PAGES:
304  *    Value 1 (default) allows contiguous pages to be combined when generating
305  *    iommu mappings. Value 0 disables combining, everything is mapped to
306  *    PAGE_SIZE. This can be useful for benchmarking.  This is a per-IOAS
307  *    option, the object_id must be the IOAS ID.
308  */
309 enum iommufd_option {
310 	IOMMU_OPTION_RLIMIT_MODE = 0,
311 	IOMMU_OPTION_HUGE_PAGES = 1,
312 };
313 
314 /**
315  * enum iommufd_option_ops - ioctl(IOMMU_OPTION_OP_SET) and
316  *                           ioctl(IOMMU_OPTION_OP_GET)
317  * @IOMMU_OPTION_OP_SET: Set the option's value
318  * @IOMMU_OPTION_OP_GET: Get the option's value
319  */
320 enum iommufd_option_ops {
321 	IOMMU_OPTION_OP_SET = 0,
322 	IOMMU_OPTION_OP_GET = 1,
323 };
324 
325 /**
326  * struct iommu_option - iommu option multiplexer
327  * @size: sizeof(struct iommu_option)
328  * @option_id: One of enum iommufd_option
329  * @op: One of enum iommufd_option_ops
330  * @__reserved: Must be 0
331  * @object_id: ID of the object if required
332  * @val64: Option value to set or value returned on get
333  *
334  * Change a simple option value. This multiplexor allows controlling options
335  * on objects. IOMMU_OPTION_OP_SET will load an option and IOMMU_OPTION_OP_GET
336  * will return the current value.
337  */
338 struct iommu_option {
339 	__u32 size;
340 	__u32 option_id;
341 	__u16 op;
342 	__u16 __reserved;
343 	__u32 object_id;
344 	__aligned_u64 val64;
345 };
346 #define IOMMU_OPTION _IO(IOMMUFD_TYPE, IOMMUFD_CMD_OPTION)
347 
348 /**
349  * enum iommufd_vfio_ioas_op - IOMMU_VFIO_IOAS_* ioctls
350  * @IOMMU_VFIO_IOAS_GET: Get the current compatibility IOAS
351  * @IOMMU_VFIO_IOAS_SET: Change the current compatibility IOAS
352  * @IOMMU_VFIO_IOAS_CLEAR: Disable VFIO compatibility
353  */
354 enum iommufd_vfio_ioas_op {
355 	IOMMU_VFIO_IOAS_GET = 0,
356 	IOMMU_VFIO_IOAS_SET = 1,
357 	IOMMU_VFIO_IOAS_CLEAR = 2,
358 };
359 
360 /**
361  * struct iommu_vfio_ioas - ioctl(IOMMU_VFIO_IOAS)
362  * @size: sizeof(struct iommu_vfio_ioas)
363  * @ioas_id: For IOMMU_VFIO_IOAS_SET the input IOAS ID to set
364  *           For IOMMU_VFIO_IOAS_GET will output the IOAS ID
365  * @op: One of enum iommufd_vfio_ioas_op
366  * @__reserved: Must be 0
367  *
368  * The VFIO compatibility support uses a single ioas because VFIO APIs do not
369  * support the ID field. Set or Get the IOAS that VFIO compatibility will use.
370  * When VFIO_GROUP_SET_CONTAINER is used on an iommufd it will get the
371  * compatibility ioas, either by taking what is already set, or auto creating
372  * one. From then on VFIO will continue to use that ioas and is not effected by
373  * this ioctl. SET or CLEAR does not destroy any auto-created IOAS.
374  */
375 struct iommu_vfio_ioas {
376 	__u32 size;
377 	__u32 ioas_id;
378 	__u16 op;
379 	__u16 __reserved;
380 };
381 #define IOMMU_VFIO_IOAS _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VFIO_IOAS)
382 
383 /**
384  * enum iommufd_hwpt_alloc_flags - Flags for HWPT allocation
385  * @IOMMU_HWPT_ALLOC_NEST_PARENT: If set, allocate a HWPT that can serve as
386  *                                the parent HWPT in a nesting configuration.
387  * @IOMMU_HWPT_ALLOC_DIRTY_TRACKING: Dirty tracking support for device IOMMU is
388  *                                   enforced on device attachment
389  * @IOMMU_HWPT_FAULT_ID_VALID: The fault_id field of hwpt allocation data is
390  *                             valid.
391  * @IOMMU_HWPT_ALLOC_PASID: Requests a domain that can be used with PASID. The
392  *                          domain can be attached to any PASID on the device.
393  *                          Any domain attached to the non-PASID part of the
394  *                          device must also be flagged, otherwise attaching a
395  *                          PASID will blocked.
396  *                          For the user that wants to attach PASID, ioas is
397  *                          not recommended for both the non-PASID part
398  *                          and PASID part of the device.
399  *                          If IOMMU does not support PASID it will return
400  *                          error (-EOPNOTSUPP).
401  */
402 enum iommufd_hwpt_alloc_flags {
403 	IOMMU_HWPT_ALLOC_NEST_PARENT = 1 << 0,
404 	IOMMU_HWPT_ALLOC_DIRTY_TRACKING = 1 << 1,
405 	IOMMU_HWPT_FAULT_ID_VALID = 1 << 2,
406 	IOMMU_HWPT_ALLOC_PASID = 1 << 3,
407 };
408 
409 /**
410  * enum iommu_hwpt_vtd_s1_flags - Intel VT-d stage-1 page table
411  *                                entry attributes
412  * @IOMMU_VTD_S1_SRE: Supervisor request
413  * @IOMMU_VTD_S1_EAFE: Extended access enable
414  * @IOMMU_VTD_S1_WPE: Write protect enable
415  */
416 enum iommu_hwpt_vtd_s1_flags {
417 	IOMMU_VTD_S1_SRE = 1 << 0,
418 	IOMMU_VTD_S1_EAFE = 1 << 1,
419 	IOMMU_VTD_S1_WPE = 1 << 2,
420 };
421 
422 /**
423  * struct iommu_hwpt_vtd_s1 - Intel VT-d stage-1 page table
424  *                            info (IOMMU_HWPT_DATA_VTD_S1)
425  * @flags: Combination of enum iommu_hwpt_vtd_s1_flags
426  * @pgtbl_addr: The base address of the stage-1 page table.
427  * @addr_width: The address width of the stage-1 page table
428  * @__reserved: Must be 0
429  */
430 struct iommu_hwpt_vtd_s1 {
431 	__aligned_u64 flags;
432 	__aligned_u64 pgtbl_addr;
433 	__u32 addr_width;
434 	__u32 __reserved;
435 };
436 
437 /**
438  * struct iommu_hwpt_arm_smmuv3 - ARM SMMUv3 nested STE
439  *                                (IOMMU_HWPT_DATA_ARM_SMMUV3)
440  *
441  * @ste: The first two double words of the user space Stream Table Entry for
442  *       the translation. Must be little-endian.
443  *       Allowed fields: (Refer to "5.2 Stream Table Entry" in SMMUv3 HW Spec)
444  *       - word-0: V, Cfg, S1Fmt, S1ContextPtr, S1CDMax
445  *       - word-1: EATS, S1DSS, S1CIR, S1COR, S1CSH, S1STALLD
446  *
447  * -EIO will be returned if @ste is not legal or contains any non-allowed field.
448  * Cfg can be used to select a S1, Bypass or Abort configuration. A Bypass
449  * nested domain will translate the same as the nesting parent. The S1 will
450  * install a Context Descriptor Table pointing at userspace memory translated
451  * by the nesting parent.
452  */
453 struct iommu_hwpt_arm_smmuv3 {
454 	__aligned_le64 ste[2];
455 };
456 
457 /**
458  * enum iommu_hwpt_data_type - IOMMU HWPT Data Type
459  * @IOMMU_HWPT_DATA_NONE: no data
460  * @IOMMU_HWPT_DATA_VTD_S1: Intel VT-d stage-1 page table
461  * @IOMMU_HWPT_DATA_ARM_SMMUV3: ARM SMMUv3 Context Descriptor Table
462  */
463 enum iommu_hwpt_data_type {
464 	IOMMU_HWPT_DATA_NONE = 0,
465 	IOMMU_HWPT_DATA_VTD_S1 = 1,
466 	IOMMU_HWPT_DATA_ARM_SMMUV3 = 2,
467 };
468 
469 /**
470  * struct iommu_hwpt_alloc - ioctl(IOMMU_HWPT_ALLOC)
471  * @size: sizeof(struct iommu_hwpt_alloc)
472  * @flags: Combination of enum iommufd_hwpt_alloc_flags
473  * @dev_id: The device to allocate this HWPT for
474  * @pt_id: The IOAS or HWPT or vIOMMU to connect this HWPT to
475  * @out_hwpt_id: The ID of the new HWPT
476  * @__reserved: Must be 0
477  * @data_type: One of enum iommu_hwpt_data_type
478  * @data_len: Length of the type specific data
479  * @data_uptr: User pointer to the type specific data
480  * @fault_id: The ID of IOMMUFD_FAULT object. Valid only if flags field of
481  *            IOMMU_HWPT_FAULT_ID_VALID is set.
482  * @__reserved2: Padding to 64-bit alignment. Must be 0.
483  *
484  * Explicitly allocate a hardware page table object. This is the same object
485  * type that is returned by iommufd_device_attach() and represents the
486  * underlying iommu driver's iommu_domain kernel object.
487  *
488  * A kernel-managed HWPT will be created with the mappings from the given
489  * IOAS via the @pt_id. The @data_type for this allocation must be set to
490  * IOMMU_HWPT_DATA_NONE. The HWPT can be allocated as a parent HWPT for a
491  * nesting configuration by passing IOMMU_HWPT_ALLOC_NEST_PARENT via @flags.
492  *
493  * A user-managed nested HWPT will be created from a given vIOMMU (wrapping a
494  * parent HWPT) or a parent HWPT via @pt_id, in which the parent HWPT must be
495  * allocated previously via the same ioctl from a given IOAS (@pt_id). In this
496  * case, the @data_type must be set to a pre-defined type corresponding to an
497  * I/O page table type supported by the underlying IOMMU hardware. The device
498  * via @dev_id and the vIOMMU via @pt_id must be associated to the same IOMMU
499  * instance.
500  *
501  * If the @data_type is set to IOMMU_HWPT_DATA_NONE, @data_len and
502  * @data_uptr should be zero. Otherwise, both @data_len and @data_uptr
503  * must be given.
504  */
505 struct iommu_hwpt_alloc {
506 	__u32 size;
507 	__u32 flags;
508 	__u32 dev_id;
509 	__u32 pt_id;
510 	__u32 out_hwpt_id;
511 	__u32 __reserved;
512 	__u32 data_type;
513 	__u32 data_len;
514 	__aligned_u64 data_uptr;
515 	__u32 fault_id;
516 	__u32 __reserved2;
517 };
518 #define IOMMU_HWPT_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_ALLOC)
519 
520 /**
521  * enum iommu_hw_info_vtd_flags - Flags for VT-d hw_info
522  * @IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17: If set, disallow read-only mappings
523  *                                         on a nested_parent domain.
524  *                                         https://www.intel.com/content/www/us/en/content-details/772415/content-details.html
525  */
526 enum iommu_hw_info_vtd_flags {
527 	IOMMU_HW_INFO_VTD_ERRATA_772415_SPR17 = 1 << 0,
528 };
529 
530 /**
531  * struct iommu_hw_info_vtd - Intel VT-d hardware information
532  *
533  * @flags: Combination of enum iommu_hw_info_vtd_flags
534  * @__reserved: Must be 0
535  *
536  * @cap_reg: Value of Intel VT-d capability register defined in VT-d spec
537  *           section 11.4.2 Capability Register.
538  * @ecap_reg: Value of Intel VT-d capability register defined in VT-d spec
539  *            section 11.4.3 Extended Capability Register.
540  *
541  * User needs to understand the Intel VT-d specification to decode the
542  * register value.
543  */
544 struct iommu_hw_info_vtd {
545 	__u32 flags;
546 	__u32 __reserved;
547 	__aligned_u64 cap_reg;
548 	__aligned_u64 ecap_reg;
549 };
550 
551 /**
552  * struct iommu_hw_info_arm_smmuv3 - ARM SMMUv3 hardware information
553  *                                   (IOMMU_HW_INFO_TYPE_ARM_SMMUV3)
554  *
555  * @flags: Must be set to 0
556  * @__reserved: Must be 0
557  * @idr: Implemented features for ARM SMMU Non-secure programming interface
558  * @iidr: Information about the implementation and implementer of ARM SMMU,
559  *        and architecture version supported
560  * @aidr: ARM SMMU architecture version
561  *
562  * For the details of @idr, @iidr and @aidr, please refer to the chapters
563  * from 6.3.1 to 6.3.6 in the SMMUv3 Spec.
564  *
565  * This reports the raw HW capability, and not all bits are meaningful to be
566  * read by userspace. Only the following fields should be used:
567  *
568  * idr[0]: ST_LEVEL, TERM_MODEL, STALL_MODEL, TTENDIAN , CD2L, ASID16, TTF
569  * idr[1]: SIDSIZE, SSIDSIZE
570  * idr[3]: BBML, RIL
571  * idr[5]: VAX, GRAN64K, GRAN16K, GRAN4K
572  *
573  * - S1P should be assumed to be true if a NESTED HWPT can be created
574  * - VFIO/iommufd only support platforms with COHACC, it should be assumed to be
575  *   true.
576  * - ATS is a per-device property. If the VMM describes any devices as ATS
577  *   capable in ACPI/DT it should set the corresponding idr.
578  *
579  * This list may expand in future (eg E0PD, AIE, PBHA, D128, DS etc). It is
580  * important that VMMs do not read bits outside the list to allow for
581  * compatibility with future kernels. Several features in the SMMUv3
582  * architecture are not currently supported by the kernel for nesting: HTTU,
583  * BTM, MPAM and others.
584  */
585 struct iommu_hw_info_arm_smmuv3 {
586 	__u32 flags;
587 	__u32 __reserved;
588 	__u32 idr[6];
589 	__u32 iidr;
590 	__u32 aidr;
591 };
592 
593 /**
594  * enum iommu_hw_info_type - IOMMU Hardware Info Types
595  * @IOMMU_HW_INFO_TYPE_NONE: Used by the drivers that do not report hardware
596  *                           info
597  * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type
598  * @IOMMU_HW_INFO_TYPE_ARM_SMMUV3: ARM SMMUv3 iommu info type
599  */
600 enum iommu_hw_info_type {
601 	IOMMU_HW_INFO_TYPE_NONE = 0,
602 	IOMMU_HW_INFO_TYPE_INTEL_VTD = 1,
603 	IOMMU_HW_INFO_TYPE_ARM_SMMUV3 = 2,
604 };
605 
606 /**
607  * enum iommufd_hw_capabilities
608  * @IOMMU_HW_CAP_DIRTY_TRACKING: IOMMU hardware support for dirty tracking
609  *                               If available, it means the following APIs
610  *                               are supported:
611  *
612  *                                   IOMMU_HWPT_GET_DIRTY_BITMAP
613  *                                   IOMMU_HWPT_SET_DIRTY_TRACKING
614  *
615  * @IOMMU_HW_CAP_PCI_PASID_EXEC: Execute Permission Supported, user ignores it
616  *                               when the struct
617  *                               iommu_hw_info::out_max_pasid_log2 is zero.
618  * @IOMMU_HW_CAP_PCI_PASID_PRIV: Privileged Mode Supported, user ignores it
619  *                               when the struct
620  *                               iommu_hw_info::out_max_pasid_log2 is zero.
621  */
622 enum iommufd_hw_capabilities {
623 	IOMMU_HW_CAP_DIRTY_TRACKING = 1 << 0,
624 	IOMMU_HW_CAP_PCI_PASID_EXEC = 1 << 1,
625 	IOMMU_HW_CAP_PCI_PASID_PRIV = 1 << 2,
626 };
627 
628 /**
629  * struct iommu_hw_info - ioctl(IOMMU_GET_HW_INFO)
630  * @size: sizeof(struct iommu_hw_info)
631  * @flags: Must be 0
632  * @dev_id: The device bound to the iommufd
633  * @data_len: Input the length of a user buffer in bytes. Output the length of
634  *            data that kernel supports
635  * @data_uptr: User pointer to a user-space buffer used by the kernel to fill
636  *             the iommu type specific hardware information data
637  * @out_data_type: Output the iommu hardware info type as defined in the enum
638  *                 iommu_hw_info_type.
639  * @out_capabilities: Output the generic iommu capability info type as defined
640  *                    in the enum iommu_hw_capabilities.
641  * @out_max_pasid_log2: Output the width of PASIDs. 0 means no PASID support.
642  *                      PCI devices turn to out_capabilities to check if the
643  *                      specific capabilities is supported or not.
644  * @__reserved: Must be 0
645  *
646  * Query an iommu type specific hardware information data from an iommu behind
647  * a given device that has been bound to iommufd. This hardware info data will
648  * be used to sync capabilities between the virtual iommu and the physical
649  * iommu, e.g. a nested translation setup needs to check the hardware info, so
650  * a guest stage-1 page table can be compatible with the physical iommu.
651  *
652  * To capture an iommu type specific hardware information data, @data_uptr and
653  * its length @data_len must be provided. Trailing bytes will be zeroed if the
654  * user buffer is larger than the data that kernel has. Otherwise, kernel only
655  * fills the buffer using the given length in @data_len. If the ioctl succeeds,
656  * @data_len will be updated to the length that kernel actually supports,
657  * @out_data_type will be filled to decode the data filled in the buffer
658  * pointed by @data_uptr. Input @data_len == zero is allowed.
659  */
660 struct iommu_hw_info {
661 	__u32 size;
662 	__u32 flags;
663 	__u32 dev_id;
664 	__u32 data_len;
665 	__aligned_u64 data_uptr;
666 	__u32 out_data_type;
667 	__u8 out_max_pasid_log2;
668 	__u8 __reserved[3];
669 	__aligned_u64 out_capabilities;
670 };
671 #define IOMMU_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_GET_HW_INFO)
672 
673 /*
674  * enum iommufd_hwpt_set_dirty_tracking_flags - Flags for steering dirty
675  *                                              tracking
676  * @IOMMU_HWPT_DIRTY_TRACKING_ENABLE: Enable dirty tracking
677  */
678 enum iommufd_hwpt_set_dirty_tracking_flags {
679 	IOMMU_HWPT_DIRTY_TRACKING_ENABLE = 1,
680 };
681 
682 /**
683  * struct iommu_hwpt_set_dirty_tracking - ioctl(IOMMU_HWPT_SET_DIRTY_TRACKING)
684  * @size: sizeof(struct iommu_hwpt_set_dirty_tracking)
685  * @flags: Combination of enum iommufd_hwpt_set_dirty_tracking_flags
686  * @hwpt_id: HW pagetable ID that represents the IOMMU domain
687  * @__reserved: Must be 0
688  *
689  * Toggle dirty tracking on an HW pagetable.
690  */
691 struct iommu_hwpt_set_dirty_tracking {
692 	__u32 size;
693 	__u32 flags;
694 	__u32 hwpt_id;
695 	__u32 __reserved;
696 };
697 #define IOMMU_HWPT_SET_DIRTY_TRACKING _IO(IOMMUFD_TYPE, \
698 					  IOMMUFD_CMD_HWPT_SET_DIRTY_TRACKING)
699 
700 /**
701  * enum iommufd_hwpt_get_dirty_bitmap_flags - Flags for getting dirty bits
702  * @IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR: Just read the PTEs without clearing
703  *                                        any dirty bits metadata. This flag
704  *                                        can be passed in the expectation
705  *                                        where the next operation is an unmap
706  *                                        of the same IOVA range.
707  *
708  */
709 enum iommufd_hwpt_get_dirty_bitmap_flags {
710 	IOMMU_HWPT_GET_DIRTY_BITMAP_NO_CLEAR = 1,
711 };
712 
713 /**
714  * struct iommu_hwpt_get_dirty_bitmap - ioctl(IOMMU_HWPT_GET_DIRTY_BITMAP)
715  * @size: sizeof(struct iommu_hwpt_get_dirty_bitmap)
716  * @hwpt_id: HW pagetable ID that represents the IOMMU domain
717  * @flags: Combination of enum iommufd_hwpt_get_dirty_bitmap_flags
718  * @__reserved: Must be 0
719  * @iova: base IOVA of the bitmap first bit
720  * @length: IOVA range size
721  * @page_size: page size granularity of each bit in the bitmap
722  * @data: bitmap where to set the dirty bits. The bitmap bits each
723  *        represent a page_size which you deviate from an arbitrary iova.
724  *
725  * Checking a given IOVA is dirty:
726  *
727  *  data[(iova / page_size) / 64] & (1ULL << ((iova / page_size) % 64))
728  *
729  * Walk the IOMMU pagetables for a given IOVA range to return a bitmap
730  * with the dirty IOVAs. In doing so it will also by default clear any
731  * dirty bit metadata set in the IOPTE.
732  */
733 struct iommu_hwpt_get_dirty_bitmap {
734 	__u32 size;
735 	__u32 hwpt_id;
736 	__u32 flags;
737 	__u32 __reserved;
738 	__aligned_u64 iova;
739 	__aligned_u64 length;
740 	__aligned_u64 page_size;
741 	__aligned_u64 data;
742 };
743 #define IOMMU_HWPT_GET_DIRTY_BITMAP _IO(IOMMUFD_TYPE, \
744 					IOMMUFD_CMD_HWPT_GET_DIRTY_BITMAP)
745 
746 /**
747  * enum iommu_hwpt_invalidate_data_type - IOMMU HWPT Cache Invalidation
748  *                                        Data Type
749  * @IOMMU_HWPT_INVALIDATE_DATA_VTD_S1: Invalidation data for VTD_S1
750  * @IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3: Invalidation data for ARM SMMUv3
751  */
752 enum iommu_hwpt_invalidate_data_type {
753 	IOMMU_HWPT_INVALIDATE_DATA_VTD_S1 = 0,
754 	IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3 = 1,
755 };
756 
757 /**
758  * enum iommu_hwpt_vtd_s1_invalidate_flags - Flags for Intel VT-d
759  *                                           stage-1 cache invalidation
760  * @IOMMU_VTD_INV_FLAGS_LEAF: Indicates whether the invalidation applies
761  *                            to all-levels page structure cache or just
762  *                            the leaf PTE cache.
763  */
764 enum iommu_hwpt_vtd_s1_invalidate_flags {
765 	IOMMU_VTD_INV_FLAGS_LEAF = 1 << 0,
766 };
767 
768 /**
769  * struct iommu_hwpt_vtd_s1_invalidate - Intel VT-d cache invalidation
770  *                                       (IOMMU_HWPT_INVALIDATE_DATA_VTD_S1)
771  * @addr: The start address of the range to be invalidated. It needs to
772  *        be 4KB aligned.
773  * @npages: Number of contiguous 4K pages to be invalidated.
774  * @flags: Combination of enum iommu_hwpt_vtd_s1_invalidate_flags
775  * @__reserved: Must be 0
776  *
777  * The Intel VT-d specific invalidation data for user-managed stage-1 cache
778  * invalidation in nested translation. Userspace uses this structure to
779  * tell the impacted cache scope after modifying the stage-1 page table.
780  *
781  * Invalidating all the caches related to the page table by setting @addr
782  * to be 0 and @npages to be U64_MAX.
783  *
784  * The device TLB will be invalidated automatically if ATS is enabled.
785  */
786 struct iommu_hwpt_vtd_s1_invalidate {
787 	__aligned_u64 addr;
788 	__aligned_u64 npages;
789 	__u32 flags;
790 	__u32 __reserved;
791 };
792 
793 /**
794  * struct iommu_viommu_arm_smmuv3_invalidate - ARM SMMUv3 cache invalidation
795  *         (IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3)
796  * @cmd: 128-bit cache invalidation command that runs in SMMU CMDQ.
797  *       Must be little-endian.
798  *
799  * Supported command list only when passing in a vIOMMU via @hwpt_id:
800  *     CMDQ_OP_TLBI_NSNH_ALL
801  *     CMDQ_OP_TLBI_NH_VA
802  *     CMDQ_OP_TLBI_NH_VAA
803  *     CMDQ_OP_TLBI_NH_ALL
804  *     CMDQ_OP_TLBI_NH_ASID
805  *     CMDQ_OP_ATC_INV
806  *     CMDQ_OP_CFGI_CD
807  *     CMDQ_OP_CFGI_CD_ALL
808  *
809  * -EIO will be returned if the command is not supported.
810  */
811 struct iommu_viommu_arm_smmuv3_invalidate {
812 	__aligned_le64 cmd[2];
813 };
814 
815 /**
816  * struct iommu_hwpt_invalidate - ioctl(IOMMU_HWPT_INVALIDATE)
817  * @size: sizeof(struct iommu_hwpt_invalidate)
818  * @hwpt_id: ID of a nested HWPT or a vIOMMU, for cache invalidation
819  * @data_uptr: User pointer to an array of driver-specific cache invalidation
820  *             data.
821  * @data_type: One of enum iommu_hwpt_invalidate_data_type, defining the data
822  *             type of all the entries in the invalidation request array. It
823  *             should be a type supported by the hwpt pointed by @hwpt_id.
824  * @entry_len: Length (in bytes) of a request entry in the request array
825  * @entry_num: Input the number of cache invalidation requests in the array.
826  *             Output the number of requests successfully handled by kernel.
827  * @__reserved: Must be 0.
828  *
829  * Invalidate iommu cache for user-managed page table or vIOMMU. Modifications
830  * on a user-managed page table should be followed by this operation, if a HWPT
831  * is passed in via @hwpt_id. Other caches, such as device cache or descriptor
832  * cache can be flushed if a vIOMMU is passed in via the @hwpt_id field.
833  *
834  * Each ioctl can support one or more cache invalidation requests in the array
835  * that has a total size of @entry_len * @entry_num.
836  *
837  * An empty invalidation request array by setting @entry_num==0 is allowed, and
838  * @entry_len and @data_uptr would be ignored in this case. This can be used to
839  * check if the given @data_type is supported or not by kernel.
840  */
841 struct iommu_hwpt_invalidate {
842 	__u32 size;
843 	__u32 hwpt_id;
844 	__aligned_u64 data_uptr;
845 	__u32 data_type;
846 	__u32 entry_len;
847 	__u32 entry_num;
848 	__u32 __reserved;
849 };
850 #define IOMMU_HWPT_INVALIDATE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_HWPT_INVALIDATE)
851 
852 /**
853  * enum iommu_hwpt_pgfault_flags - flags for struct iommu_hwpt_pgfault
854  * @IOMMU_PGFAULT_FLAGS_PASID_VALID: The pasid field of the fault data is
855  *                                   valid.
856  * @IOMMU_PGFAULT_FLAGS_LAST_PAGE: It's the last fault of a fault group.
857  */
858 enum iommu_hwpt_pgfault_flags {
859 	IOMMU_PGFAULT_FLAGS_PASID_VALID		= (1 << 0),
860 	IOMMU_PGFAULT_FLAGS_LAST_PAGE		= (1 << 1),
861 };
862 
863 /**
864  * enum iommu_hwpt_pgfault_perm - perm bits for struct iommu_hwpt_pgfault
865  * @IOMMU_PGFAULT_PERM_READ: request for read permission
866  * @IOMMU_PGFAULT_PERM_WRITE: request for write permission
867  * @IOMMU_PGFAULT_PERM_EXEC: (PCIE 10.4.1) request with a PASID that has the
868  *                           Execute Requested bit set in PASID TLP Prefix.
869  * @IOMMU_PGFAULT_PERM_PRIV: (PCIE 10.4.1) request with a PASID that has the
870  *                           Privileged Mode Requested bit set in PASID TLP
871  *                           Prefix.
872  */
873 enum iommu_hwpt_pgfault_perm {
874 	IOMMU_PGFAULT_PERM_READ			= (1 << 0),
875 	IOMMU_PGFAULT_PERM_WRITE		= (1 << 1),
876 	IOMMU_PGFAULT_PERM_EXEC			= (1 << 2),
877 	IOMMU_PGFAULT_PERM_PRIV			= (1 << 3),
878 };
879 
880 /**
881  * struct iommu_hwpt_pgfault - iommu page fault data
882  * @flags: Combination of enum iommu_hwpt_pgfault_flags
883  * @dev_id: id of the originated device
884  * @pasid: Process Address Space ID
885  * @grpid: Page Request Group Index
886  * @perm: Combination of enum iommu_hwpt_pgfault_perm
887  * @__reserved: Must be 0.
888  * @addr: Fault address
889  * @length: a hint of how much data the requestor is expecting to fetch. For
890  *          example, if the PRI initiator knows it is going to do a 10MB
891  *          transfer, it could fill in 10MB and the OS could pre-fault in
892  *          10MB of IOVA. It's default to 0 if there's no such hint.
893  * @cookie: kernel-managed cookie identifying a group of fault messages. The
894  *          cookie number encoded in the last page fault of the group should
895  *          be echoed back in the response message.
896  */
897 struct iommu_hwpt_pgfault {
898 	__u32 flags;
899 	__u32 dev_id;
900 	__u32 pasid;
901 	__u32 grpid;
902 	__u32 perm;
903 	__u32 __reserved;
904 	__aligned_u64 addr;
905 	__u32 length;
906 	__u32 cookie;
907 };
908 
909 /**
910  * enum iommufd_page_response_code - Return status of fault handlers
911  * @IOMMUFD_PAGE_RESP_SUCCESS: Fault has been handled and the page tables
912  *                             populated, retry the access. This is the
913  *                             "Success" defined in PCI 10.4.2.1.
914  * @IOMMUFD_PAGE_RESP_INVALID: Could not handle this fault, don't retry the
915  *                             access. This is the "Invalid Request" in PCI
916  *                             10.4.2.1.
917  */
918 enum iommufd_page_response_code {
919 	IOMMUFD_PAGE_RESP_SUCCESS = 0,
920 	IOMMUFD_PAGE_RESP_INVALID = 1,
921 };
922 
923 /**
924  * struct iommu_hwpt_page_response - IOMMU page fault response
925  * @cookie: The kernel-managed cookie reported in the fault message.
926  * @code: One of response code in enum iommufd_page_response_code.
927  */
928 struct iommu_hwpt_page_response {
929 	__u32 cookie;
930 	__u32 code;
931 };
932 
933 /**
934  * struct iommu_fault_alloc - ioctl(IOMMU_FAULT_QUEUE_ALLOC)
935  * @size: sizeof(struct iommu_fault_alloc)
936  * @flags: Must be 0
937  * @out_fault_id: The ID of the new FAULT
938  * @out_fault_fd: The fd of the new FAULT
939  *
940  * Explicitly allocate a fault handling object.
941  */
942 struct iommu_fault_alloc {
943 	__u32 size;
944 	__u32 flags;
945 	__u32 out_fault_id;
946 	__u32 out_fault_fd;
947 };
948 #define IOMMU_FAULT_QUEUE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_FAULT_QUEUE_ALLOC)
949 
950 /**
951  * enum iommu_viommu_type - Virtual IOMMU Type
952  * @IOMMU_VIOMMU_TYPE_DEFAULT: Reserved for future use
953  * @IOMMU_VIOMMU_TYPE_ARM_SMMUV3: ARM SMMUv3 driver specific type
954  */
955 enum iommu_viommu_type {
956 	IOMMU_VIOMMU_TYPE_DEFAULT = 0,
957 	IOMMU_VIOMMU_TYPE_ARM_SMMUV3 = 1,
958 };
959 
960 /**
961  * struct iommu_viommu_alloc - ioctl(IOMMU_VIOMMU_ALLOC)
962  * @size: sizeof(struct iommu_viommu_alloc)
963  * @flags: Must be 0
964  * @type: Type of the virtual IOMMU. Must be defined in enum iommu_viommu_type
965  * @dev_id: The device's physical IOMMU will be used to back the virtual IOMMU
966  * @hwpt_id: ID of a nesting parent HWPT to associate to
967  * @out_viommu_id: Output virtual IOMMU ID for the allocated object
968  *
969  * Allocate a virtual IOMMU object, representing the underlying physical IOMMU's
970  * virtualization support that is a security-isolated slice of the real IOMMU HW
971  * that is unique to a specific VM. Operations global to the IOMMU are connected
972  * to the vIOMMU, such as:
973  * - Security namespace for guest owned ID, e.g. guest-controlled cache tags
974  * - Non-device-affiliated event reporting, e.g. invalidation queue errors
975  * - Access to a sharable nesting parent pagetable across physical IOMMUs
976  * - Virtualization of various platforms IDs, e.g. RIDs and others
977  * - Delivery of paravirtualized invalidation
978  * - Direct assigned invalidation queues
979  * - Direct assigned interrupts
980  */
981 struct iommu_viommu_alloc {
982 	__u32 size;
983 	__u32 flags;
984 	__u32 type;
985 	__u32 dev_id;
986 	__u32 hwpt_id;
987 	__u32 out_viommu_id;
988 };
989 #define IOMMU_VIOMMU_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VIOMMU_ALLOC)
990 
991 /**
992  * struct iommu_vdevice_alloc - ioctl(IOMMU_VDEVICE_ALLOC)
993  * @size: sizeof(struct iommu_vdevice_alloc)
994  * @viommu_id: vIOMMU ID to associate with the virtual device
995  * @dev_id: The physical device to allocate a virtual instance on the vIOMMU
996  * @out_vdevice_id: Object handle for the vDevice. Pass to IOMMU_DESTORY
997  * @virt_id: Virtual device ID per vIOMMU, e.g. vSID of ARM SMMUv3, vDeviceID
998  *           of AMD IOMMU, and vRID of a nested Intel VT-d to a Context Table
999  *
1000  * Allocate a virtual device instance (for a physical device) against a vIOMMU.
1001  * This instance holds the device's information (related to its vIOMMU) in a VM.
1002  */
1003 struct iommu_vdevice_alloc {
1004 	__u32 size;
1005 	__u32 viommu_id;
1006 	__u32 dev_id;
1007 	__u32 out_vdevice_id;
1008 	__aligned_u64 virt_id;
1009 };
1010 #define IOMMU_VDEVICE_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VDEVICE_ALLOC)
1011 
1012 /**
1013  * struct iommu_ioas_change_process - ioctl(VFIO_IOAS_CHANGE_PROCESS)
1014  * @size: sizeof(struct iommu_ioas_change_process)
1015  * @__reserved: Must be 0
1016  *
1017  * This transfers pinned memory counts for every memory map in every IOAS
1018  * in the context to the current process.  This only supports maps created
1019  * with IOMMU_IOAS_MAP_FILE, and returns EINVAL if other maps are present.
1020  * If the ioctl returns a failure status, then nothing is changed.
1021  *
1022  * This API is useful for transferring operation of a device from one process
1023  * to another, such as during userland live update.
1024  */
1025 struct iommu_ioas_change_process {
1026 	__u32 size;
1027 	__u32 __reserved;
1028 };
1029 
1030 #define IOMMU_IOAS_CHANGE_PROCESS \
1031 	_IO(IOMMUFD_TYPE, IOMMUFD_CMD_IOAS_CHANGE_PROCESS)
1032 
1033 /**
1034  * enum iommu_veventq_flag - flag for struct iommufd_vevent_header
1035  * @IOMMU_VEVENTQ_FLAG_LOST_EVENTS: vEVENTQ has lost vEVENTs
1036  */
1037 enum iommu_veventq_flag {
1038 	IOMMU_VEVENTQ_FLAG_LOST_EVENTS = (1U << 0),
1039 };
1040 
1041 /**
1042  * struct iommufd_vevent_header - Virtual Event Header for a vEVENTQ Status
1043  * @flags: Combination of enum iommu_veventq_flag
1044  * @sequence: The sequence index of a vEVENT in the vEVENTQ, with a range of
1045  *            [0, INT_MAX] where the following index of INT_MAX is 0
1046  *
1047  * Each iommufd_vevent_header reports a sequence index of the following vEVENT:
1048  *
1049  * +----------------------+-------+----------------------+-------+---+-------+
1050  * | header0 {sequence=0} | data0 | header1 {sequence=1} | data1 |...| dataN |
1051  * +----------------------+-------+----------------------+-------+---+-------+
1052  *
1053  * And this sequence index is expected to be monotonic to the sequence index of
1054  * the previous vEVENT. If two adjacent sequence indexes has a delta larger than
1055  * 1, it means that delta - 1 number of vEVENTs has lost, e.g. two lost vEVENTs:
1056  *
1057  * +-----+----------------------+-------+----------------------+-------+-----+
1058  * | ... | header3 {sequence=3} | data3 | header6 {sequence=6} | data6 | ... |
1059  * +-----+----------------------+-------+----------------------+-------+-----+
1060  *
1061  * If a vEVENT lost at the tail of the vEVENTQ and there is no following vEVENT
1062  * providing the next sequence index, an IOMMU_VEVENTQ_FLAG_LOST_EVENTS header
1063  * would be added to the tail, and no data would follow this header:
1064  *
1065  * +--+----------------------+-------+-----------------------------------------+
1066  * |..| header3 {sequence=3} | data3 | header4 {flags=LOST_EVENTS, sequence=4} |
1067  * +--+----------------------+-------+-----------------------------------------+
1068  */
1069 struct iommufd_vevent_header {
1070 	__u32 flags;
1071 	__u32 sequence;
1072 };
1073 
1074 /**
1075  * enum iommu_veventq_type - Virtual Event Queue Type
1076  * @IOMMU_VEVENTQ_TYPE_DEFAULT: Reserved for future use
1077  * @IOMMU_VEVENTQ_TYPE_ARM_SMMUV3: ARM SMMUv3 Virtual Event Queue
1078  */
1079 enum iommu_veventq_type {
1080 	IOMMU_VEVENTQ_TYPE_DEFAULT = 0,
1081 	IOMMU_VEVENTQ_TYPE_ARM_SMMUV3 = 1,
1082 };
1083 
1084 /**
1085  * struct iommu_vevent_arm_smmuv3 - ARM SMMUv3 Virtual Event
1086  *                                  (IOMMU_VEVENTQ_TYPE_ARM_SMMUV3)
1087  * @evt: 256-bit ARM SMMUv3 Event record, little-endian.
1088  *       Reported event records: (Refer to "7.3 Event records" in SMMUv3 HW Spec)
1089  *       - 0x04 C_BAD_STE
1090  *       - 0x06 F_STREAM_DISABLED
1091  *       - 0x08 C_BAD_SUBSTREAMID
1092  *       - 0x0a C_BAD_CD
1093  *       - 0x10 F_TRANSLATION
1094  *       - 0x11 F_ADDR_SIZE
1095  *       - 0x12 F_ACCESS
1096  *       - 0x13 F_PERMISSION
1097  *
1098  * StreamID field reports a virtual device ID. To receive a virtual event for a
1099  * device, a vDEVICE must be allocated via IOMMU_VDEVICE_ALLOC.
1100  */
1101 struct iommu_vevent_arm_smmuv3 {
1102 	__aligned_le64 evt[4];
1103 };
1104 
1105 /**
1106  * struct iommu_veventq_alloc - ioctl(IOMMU_VEVENTQ_ALLOC)
1107  * @size: sizeof(struct iommu_veventq_alloc)
1108  * @flags: Must be 0
1109  * @viommu_id: virtual IOMMU ID to associate the vEVENTQ with
1110  * @type: Type of the vEVENTQ. Must be defined in enum iommu_veventq_type
1111  * @veventq_depth: Maximum number of events in the vEVENTQ
1112  * @out_veventq_id: The ID of the new vEVENTQ
1113  * @out_veventq_fd: The fd of the new vEVENTQ. User space must close the
1114  *                  successfully returned fd after using it
1115  * @__reserved: Must be 0
1116  *
1117  * Explicitly allocate a virtual event queue interface for a vIOMMU. A vIOMMU
1118  * can have multiple FDs for different types, but is confined to one per @type.
1119  * User space should open the @out_veventq_fd to read vEVENTs out of a vEVENTQ,
1120  * if there are vEVENTs available. A vEVENTQ will lose events due to overflow,
1121  * if the number of the vEVENTs hits @veventq_depth.
1122  *
1123  * Each vEVENT in a vEVENTQ encloses a struct iommufd_vevent_header followed by
1124  * a type-specific data structure, in a normal case:
1125  *
1126  * +-+---------+-------+---------+-------+-----+---------+-------+-+
1127  * | | header0 | data0 | header1 | data1 | ... | headerN | dataN | |
1128  * +-+---------+-------+---------+-------+-----+---------+-------+-+
1129  *
1130  * unless a tailing IOMMU_VEVENTQ_FLAG_LOST_EVENTS header is logged (refer to
1131  * struct iommufd_vevent_header).
1132  */
1133 struct iommu_veventq_alloc {
1134 	__u32 size;
1135 	__u32 flags;
1136 	__u32 viommu_id;
1137 	__u32 type;
1138 	__u32 veventq_depth;
1139 	__u32 out_veventq_id;
1140 	__u32 out_veventq_fd;
1141 	__u32 __reserved;
1142 };
1143 #define IOMMU_VEVENTQ_ALLOC _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VEVENTQ_ALLOC)
1144 #endif
1145