1 | /* SPDX-License-Identifier: GPL-2.0-only */ |
2 | /* |
3 | * VFIO API definition |
4 | * |
5 | * Copyright (C) 2012 Red Hat, Inc. All rights reserved. |
6 | * Author: Alex Williamson <alex.williamson@redhat.com> |
7 | */ |
8 | #ifndef VFIO_H |
9 | #define VFIO_H |
10 | |
11 | |
12 | #include <linux/iommu.h> |
13 | #include <linux/mm.h> |
14 | #include <linux/workqueue.h> |
15 | #include <linux/poll.h> |
16 | #include <linux/cdev.h> |
17 | #include <uapi/linux/vfio.h> |
18 | #include <linux/iova_bitmap.h> |
19 | |
20 | struct kvm; |
21 | struct iommufd_ctx; |
22 | struct iommufd_device; |
23 | struct iommufd_access; |
24 | |
25 | /* |
26 | * VFIO devices can be placed in a set, this allows all devices to share this |
27 | * structure and the VFIO core will provide a lock that is held around |
28 | * open_device()/close_device() for all devices in the set. |
29 | */ |
30 | struct vfio_device_set { |
31 | void *set_id; |
32 | struct mutex lock; |
33 | struct list_head device_list; |
34 | unsigned int device_count; |
35 | }; |
36 | |
37 | struct vfio_device { |
38 | struct device *dev; |
39 | const struct vfio_device_ops *ops; |
40 | /* |
41 | * mig_ops/log_ops is a static property of the vfio_device which must |
42 | * be set prior to registering the vfio_device. |
43 | */ |
44 | const struct vfio_migration_ops *mig_ops; |
45 | const struct vfio_log_ops *log_ops; |
46 | #if IS_ENABLED(CONFIG_VFIO_GROUP) |
47 | struct vfio_group *group; |
48 | struct list_head group_next; |
49 | struct list_head iommu_entry; |
50 | #endif |
51 | struct vfio_device_set *dev_set; |
52 | struct list_head dev_set_list; |
53 | unsigned int migration_flags; |
54 | struct kvm *kvm; |
55 | |
56 | /* Members below here are private, not for driver use */ |
57 | unsigned int index; |
58 | struct device device; /* device.kref covers object life circle */ |
59 | #if IS_ENABLED(CONFIG_VFIO_DEVICE_CDEV) |
60 | struct cdev cdev; |
61 | #endif |
62 | refcount_t refcount; /* user count on registered device*/ |
63 | unsigned int open_count; |
64 | struct completion comp; |
65 | struct iommufd_access *iommufd_access; |
66 | void (*put_kvm)(struct kvm *kvm); |
67 | #if IS_ENABLED(CONFIG_IOMMUFD) |
68 | struct iommufd_device *iommufd_device; |
69 | u8 iommufd_attached:1; |
70 | #endif |
71 | u8 cdev_opened:1; |
72 | #ifdef CONFIG_DEBUG_FS |
73 | /* |
74 | * debug_root is a static property of the vfio_device |
75 | * which must be set prior to registering the vfio_device. |
76 | */ |
77 | struct dentry *debug_root; |
78 | #endif |
79 | }; |
80 | |
81 | /** |
82 | * struct vfio_device_ops - VFIO bus driver device callbacks |
83 | * |
84 | * @name: Name of the device driver. |
85 | * @init: initialize private fields in device structure |
86 | * @release: Reclaim private fields in device structure |
87 | * @bind_iommufd: Called when binding the device to an iommufd |
88 | * @unbind_iommufd: Opposite of bind_iommufd |
89 | * @attach_ioas: Called when attaching device to an IOAS/HWPT managed by the |
90 | * bound iommufd. Undo in unbind_iommufd if @detach_ioas is not |
91 | * called. |
92 | * @detach_ioas: Opposite of attach_ioas |
93 | * @open_device: Called when the first file descriptor is opened for this device |
94 | * @close_device: Opposite of open_device |
95 | * @read: Perform read(2) on device file descriptor |
96 | * @write: Perform write(2) on device file descriptor |
97 | * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_* |
98 | * operations documented below |
99 | * @mmap: Perform mmap(2) on a region of the device file descriptor |
100 | * @request: Request for the bus driver to release the device |
101 | * @match: Optional device name match callback (return: 0 for no-match, >0 for |
102 | * match, -errno for abort (ex. match with insufficient or incorrect |
103 | * additional args) |
104 | * @dma_unmap: Called when userspace unmaps IOVA from the container |
105 | * this device is attached to. |
106 | * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl |
107 | */ |
108 | struct vfio_device_ops { |
109 | char *name; |
110 | int (*init)(struct vfio_device *vdev); |
111 | void (*release)(struct vfio_device *vdev); |
112 | int (*bind_iommufd)(struct vfio_device *vdev, |
113 | struct iommufd_ctx *ictx, u32 *out_device_id); |
114 | void (*unbind_iommufd)(struct vfio_device *vdev); |
115 | int (*attach_ioas)(struct vfio_device *vdev, u32 *pt_id); |
116 | void (*detach_ioas)(struct vfio_device *vdev); |
117 | int (*open_device)(struct vfio_device *vdev); |
118 | void (*close_device)(struct vfio_device *vdev); |
119 | ssize_t (*read)(struct vfio_device *vdev, char __user *buf, |
120 | size_t count, loff_t *ppos); |
121 | ssize_t (*write)(struct vfio_device *vdev, const char __user *buf, |
122 | size_t count, loff_t *size); |
123 | long (*ioctl)(struct vfio_device *vdev, unsigned int cmd, |
124 | unsigned long arg); |
125 | int (*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma); |
126 | void (*request)(struct vfio_device *vdev, unsigned int count); |
127 | int (*match)(struct vfio_device *vdev, char *buf); |
128 | void (*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length); |
129 | int (*device_feature)(struct vfio_device *device, u32 flags, |
130 | void __user *arg, size_t argsz); |
131 | }; |
132 | |
133 | #if IS_ENABLED(CONFIG_IOMMUFD) |
134 | struct iommufd_ctx *vfio_iommufd_device_ictx(struct vfio_device *vdev); |
135 | int vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx); |
136 | int vfio_iommufd_physical_bind(struct vfio_device *vdev, |
137 | struct iommufd_ctx *ictx, u32 *out_device_id); |
138 | void vfio_iommufd_physical_unbind(struct vfio_device *vdev); |
139 | int vfio_iommufd_physical_attach_ioas(struct vfio_device *vdev, u32 *pt_id); |
140 | void vfio_iommufd_physical_detach_ioas(struct vfio_device *vdev); |
141 | int vfio_iommufd_emulated_bind(struct vfio_device *vdev, |
142 | struct iommufd_ctx *ictx, u32 *out_device_id); |
143 | void vfio_iommufd_emulated_unbind(struct vfio_device *vdev); |
144 | int vfio_iommufd_emulated_attach_ioas(struct vfio_device *vdev, u32 *pt_id); |
145 | void vfio_iommufd_emulated_detach_ioas(struct vfio_device *vdev); |
146 | #else |
147 | static inline struct iommufd_ctx * |
148 | vfio_iommufd_device_ictx(struct vfio_device *vdev) |
149 | { |
150 | return NULL; |
151 | } |
152 | |
153 | static inline int |
154 | vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx) |
155 | { |
156 | return VFIO_PCI_DEVID_NOT_OWNED; |
157 | } |
158 | |
159 | #define vfio_iommufd_physical_bind \ |
160 | ((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx, \ |
161 | u32 *out_device_id)) NULL) |
162 | #define vfio_iommufd_physical_unbind \ |
163 | ((void (*)(struct vfio_device *vdev)) NULL) |
164 | #define vfio_iommufd_physical_attach_ioas \ |
165 | ((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL) |
166 | #define vfio_iommufd_physical_detach_ioas \ |
167 | ((void (*)(struct vfio_device *vdev)) NULL) |
168 | #define vfio_iommufd_emulated_bind \ |
169 | ((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx, \ |
170 | u32 *out_device_id)) NULL) |
171 | #define vfio_iommufd_emulated_unbind \ |
172 | ((void (*)(struct vfio_device *vdev)) NULL) |
173 | #define vfio_iommufd_emulated_attach_ioas \ |
174 | ((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL) |
175 | #define vfio_iommufd_emulated_detach_ioas \ |
176 | ((void (*)(struct vfio_device *vdev)) NULL) |
177 | #endif |
178 | |
179 | static inline bool vfio_device_cdev_opened(struct vfio_device *device) |
180 | { |
181 | return device->cdev_opened; |
182 | } |
183 | |
184 | /** |
185 | * struct vfio_migration_ops - VFIO bus device driver migration callbacks |
186 | * |
187 | * @migration_set_state: Optional callback to change the migration state for |
188 | * devices that support migration. It's mandatory for |
189 | * VFIO_DEVICE_FEATURE_MIGRATION migration support. |
190 | * The returned FD is used for data transfer according to the FSM |
191 | * definition. The driver is responsible to ensure that FD reaches end |
192 | * of stream or error whenever the migration FSM leaves a data transfer |
193 | * state or before close_device() returns. |
194 | * @migration_get_state: Optional callback to get the migration state for |
195 | * devices that support migration. It's mandatory for |
196 | * VFIO_DEVICE_FEATURE_MIGRATION migration support. |
197 | * @migration_get_data_size: Optional callback to get the estimated data |
198 | * length that will be required to complete stop copy. It's mandatory for |
199 | * VFIO_DEVICE_FEATURE_MIGRATION migration support. |
200 | */ |
201 | struct vfio_migration_ops { |
202 | struct file *(*migration_set_state)( |
203 | struct vfio_device *device, |
204 | enum vfio_device_mig_state new_state); |
205 | int (*migration_get_state)(struct vfio_device *device, |
206 | enum vfio_device_mig_state *curr_state); |
207 | int (*migration_get_data_size)(struct vfio_device *device, |
208 | unsigned long *stop_copy_length); |
209 | }; |
210 | |
211 | /** |
212 | * struct vfio_log_ops - VFIO bus device driver logging callbacks |
213 | * |
214 | * @log_start: Optional callback to ask the device start DMA logging. |
215 | * @log_stop: Optional callback to ask the device stop DMA logging. |
216 | * @log_read_and_clear: Optional callback to ask the device read |
217 | * and clear the dirty DMAs in some given range. |
218 | * |
219 | * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set |
220 | * of features does not track logging state relative to the device, |
221 | * therefore the device implementation of vfio_log_ops must handle |
222 | * arbitrary user requests. This includes rejecting subsequent calls |
223 | * to log_start without an intervening log_stop, as well as graceful |
224 | * handling of log_stop and log_read_and_clear from invalid states. |
225 | */ |
226 | struct vfio_log_ops { |
227 | int (*log_start)(struct vfio_device *device, |
228 | struct rb_root_cached *ranges, u32 nnodes, u64 *page_size); |
229 | int (*log_stop)(struct vfio_device *device); |
230 | int (*log_read_and_clear)(struct vfio_device *device, |
231 | unsigned long iova, unsigned long length, |
232 | struct iova_bitmap *dirty); |
233 | }; |
234 | |
235 | /** |
236 | * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl |
237 | * @flags: Arg from the device_feature op |
238 | * @argsz: Arg from the device_feature op |
239 | * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver |
240 | * supports |
241 | * @minsz: Minimum data size the driver accepts |
242 | * |
243 | * For use in a driver's device_feature op. Checks that the inputs to the |
244 | * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if |
245 | * the driver should execute the get or set, otherwise the relevant |
246 | * value should be returned. |
247 | */ |
248 | static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops, |
249 | size_t minsz) |
250 | { |
251 | if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) & |
252 | ~supported_ops) |
253 | return -EINVAL; |
254 | if (flags & VFIO_DEVICE_FEATURE_PROBE) |
255 | return 0; |
256 | /* Without PROBE one of GET or SET must be requested */ |
257 | if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET))) |
258 | return -EINVAL; |
259 | if (argsz < minsz) |
260 | return -EINVAL; |
261 | return 1; |
262 | } |
263 | |
264 | struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev, |
265 | const struct vfio_device_ops *ops); |
266 | #define vfio_alloc_device(dev_struct, member, dev, ops) \ |
267 | container_of(_vfio_alloc_device(sizeof(struct dev_struct) + \ |
268 | BUILD_BUG_ON_ZERO(offsetof( \ |
269 | struct dev_struct, member)), \ |
270 | dev, ops), \ |
271 | struct dev_struct, member) |
272 | |
273 | static inline void vfio_put_device(struct vfio_device *device) |
274 | { |
275 | put_device(dev: &device->device); |
276 | } |
277 | |
278 | int vfio_register_group_dev(struct vfio_device *device); |
279 | int vfio_register_emulated_iommu_dev(struct vfio_device *device); |
280 | void vfio_unregister_group_dev(struct vfio_device *device); |
281 | |
282 | int vfio_assign_device_set(struct vfio_device *device, void *set_id); |
283 | unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set); |
284 | struct vfio_device * |
285 | vfio_find_device_in_devset(struct vfio_device_set *dev_set, |
286 | struct device *dev); |
287 | |
288 | int vfio_mig_get_next_state(struct vfio_device *device, |
289 | enum vfio_device_mig_state cur_fsm, |
290 | enum vfio_device_mig_state new_fsm, |
291 | enum vfio_device_mig_state *next_fsm); |
292 | |
293 | void vfio_combine_iova_ranges(struct rb_root_cached *root, u32 cur_nodes, |
294 | u32 req_nodes); |
295 | |
296 | /* |
297 | * External user API |
298 | */ |
299 | struct iommu_group *vfio_file_iommu_group(struct file *file); |
300 | |
301 | #if IS_ENABLED(CONFIG_VFIO_GROUP) |
302 | bool vfio_file_is_group(struct file *file); |
303 | bool vfio_file_has_dev(struct file *file, struct vfio_device *device); |
304 | #else |
305 | static inline bool vfio_file_is_group(struct file *file) |
306 | { |
307 | return false; |
308 | } |
309 | |
310 | static inline bool vfio_file_has_dev(struct file *file, struct vfio_device *device) |
311 | { |
312 | return false; |
313 | } |
314 | #endif |
315 | bool vfio_file_is_valid(struct file *file); |
316 | bool vfio_file_enforced_coherent(struct file *file); |
317 | void vfio_file_set_kvm(struct file *file, struct kvm *kvm); |
318 | |
319 | #define VFIO_PIN_PAGES_MAX_ENTRIES (PAGE_SIZE/sizeof(unsigned long)) |
320 | |
321 | int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova, |
322 | int npage, int prot, struct page **pages); |
323 | void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage); |
324 | int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova, |
325 | void *data, size_t len, bool write); |
326 | |
327 | /* |
328 | * Sub-module helpers |
329 | */ |
330 | struct vfio_info_cap { |
331 | struct vfio_info_cap_header *buf; |
332 | size_t size; |
333 | }; |
334 | struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps, |
335 | size_t size, u16 id, |
336 | u16 version); |
337 | void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset); |
338 | |
339 | int vfio_info_add_capability(struct vfio_info_cap *caps, |
340 | struct vfio_info_cap_header *cap, size_t size); |
341 | |
342 | int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr, |
343 | int num_irqs, int max_irq_type, |
344 | size_t *data_size); |
345 | |
346 | /* |
347 | * IRQfd - generic |
348 | */ |
349 | struct virqfd { |
350 | void *opaque; |
351 | struct eventfd_ctx *eventfd; |
352 | int (*handler)(void *, void *); |
353 | void (*thread)(void *, void *); |
354 | void *data; |
355 | struct work_struct inject; |
356 | wait_queue_entry_t wait; |
357 | poll_table pt; |
358 | struct work_struct shutdown; |
359 | struct work_struct flush_inject; |
360 | struct virqfd **pvirqfd; |
361 | }; |
362 | |
363 | int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *), |
364 | void (*thread)(void *, void *), void *data, |
365 | struct virqfd **pvirqfd, int fd); |
366 | void vfio_virqfd_disable(struct virqfd **pvirqfd); |
367 | void vfio_virqfd_flush_thread(struct virqfd **pvirqfd); |
368 | |
369 | #endif /* VFIO_H */ |
370 | |