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 struct inode *inode;
68 #if IS_ENABLED(CONFIG_IOMMUFD)
69 struct iommufd_device *iommufd_device;
70 u8 iommufd_attached:1;
71 #endif
72 u8 cdev_opened:1;
73 #ifdef CONFIG_DEBUG_FS
74 /*
75 * debug_root is a static property of the vfio_device
76 * which must be set prior to registering the vfio_device.
77 */
78 struct dentry *debug_root;
79 #endif
80 /* protected by more privileged entity(hypervisor). */
81 bool protected;
82 };
83
84 /**
85 * struct vfio_device_ops - VFIO bus driver device callbacks
86 *
87 * @name: Name of the device driver.
88 * @init: initialize private fields in device structure
89 * @release: Reclaim private fields in device structure
90 * @bind_iommufd: Called when binding the device to an iommufd
91 * @unbind_iommufd: Opposite of bind_iommufd
92 * @attach_ioas: Called when attaching device to an IOAS/HWPT managed by the
93 * bound iommufd. Undo in unbind_iommufd if @detach_ioas is not
94 * called.
95 * @detach_ioas: Opposite of attach_ioas
96 * @open_device: Called when the first file descriptor is opened for this device
97 * @close_device: Opposite of open_device
98 * @read: Perform read(2) on device file descriptor
99 * @write: Perform write(2) on device file descriptor
100 * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
101 * operations documented below
102 * @mmap: Perform mmap(2) on a region of the device file descriptor
103 * @request: Request for the bus driver to release the device
104 * @match: Optional device name match callback (return: 0 for no-match, >0 for
105 * match, -errno for abort (ex. match with insufficient or incorrect
106 * additional args)
107 * @dma_unmap: Called when userspace unmaps IOVA from the container
108 * this device is attached to.
109 * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
110 */
111 struct vfio_device_ops {
112 char *name;
113 int (*init)(struct vfio_device *vdev);
114 void (*release)(struct vfio_device *vdev);
115 int (*bind_iommufd)(struct vfio_device *vdev,
116 struct iommufd_ctx *ictx, u32 *out_device_id);
117 void (*unbind_iommufd)(struct vfio_device *vdev);
118 int (*attach_ioas)(struct vfio_device *vdev, u32 *pt_id);
119 void (*detach_ioas)(struct vfio_device *vdev);
120 int (*open_device)(struct vfio_device *vdev);
121 void (*close_device)(struct vfio_device *vdev);
122 ssize_t (*read)(struct vfio_device *vdev, char __user *buf,
123 size_t count, loff_t *ppos);
124 ssize_t (*write)(struct vfio_device *vdev, const char __user *buf,
125 size_t count, loff_t *size);
126 long (*ioctl)(struct vfio_device *vdev, unsigned int cmd,
127 unsigned long arg);
128 int (*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma);
129 void (*request)(struct vfio_device *vdev, unsigned int count);
130 int (*match)(struct vfio_device *vdev, char *buf);
131 void (*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
132 int (*device_feature)(struct vfio_device *device, u32 flags,
133 void __user *arg, size_t argsz);
134 };
135
136 #if IS_ENABLED(CONFIG_IOMMUFD)
137 struct iommufd_ctx *vfio_iommufd_device_ictx(struct vfio_device *vdev);
138 int vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx);
139 int vfio_iommufd_physical_bind(struct vfio_device *vdev,
140 struct iommufd_ctx *ictx, u32 *out_device_id);
141 void vfio_iommufd_physical_unbind(struct vfio_device *vdev);
142 int vfio_iommufd_physical_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
143 void vfio_iommufd_physical_detach_ioas(struct vfio_device *vdev);
144 int vfio_iommufd_emulated_bind(struct vfio_device *vdev,
145 struct iommufd_ctx *ictx, u32 *out_device_id);
146 void vfio_iommufd_emulated_unbind(struct vfio_device *vdev);
147 int vfio_iommufd_emulated_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
148 void vfio_iommufd_emulated_detach_ioas(struct vfio_device *vdev);
149 #else
150 static inline struct iommufd_ctx *
vfio_iommufd_device_ictx(struct vfio_device * vdev)151 vfio_iommufd_device_ictx(struct vfio_device *vdev)
152 {
153 return NULL;
154 }
155
156 static inline int
vfio_iommufd_get_dev_id(struct vfio_device * vdev,struct iommufd_ctx * ictx)157 vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx)
158 {
159 return VFIO_PCI_DEVID_NOT_OWNED;
160 }
161
162 #define vfio_iommufd_physical_bind \
163 ((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx, \
164 u32 *out_device_id)) NULL)
165 #define vfio_iommufd_physical_unbind \
166 ((void (*)(struct vfio_device *vdev)) NULL)
167 #define vfio_iommufd_physical_attach_ioas \
168 ((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
169 #define vfio_iommufd_physical_detach_ioas \
170 ((void (*)(struct vfio_device *vdev)) NULL)
171 #define vfio_iommufd_emulated_bind \
172 ((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx, \
173 u32 *out_device_id)) NULL)
174 #define vfio_iommufd_emulated_unbind \
175 ((void (*)(struct vfio_device *vdev)) NULL)
176 #define vfio_iommufd_emulated_attach_ioas \
177 ((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
178 #define vfio_iommufd_emulated_detach_ioas \
179 ((void (*)(struct vfio_device *vdev)) NULL)
180 #endif
181
vfio_device_cdev_opened(struct vfio_device * device)182 static inline bool vfio_device_cdev_opened(struct vfio_device *device)
183 {
184 return device->cdev_opened;
185 }
186
187 /**
188 * struct vfio_migration_ops - VFIO bus device driver migration callbacks
189 *
190 * @migration_set_state: Optional callback to change the migration state for
191 * devices that support migration. It's mandatory for
192 * VFIO_DEVICE_FEATURE_MIGRATION migration support.
193 * The returned FD is used for data transfer according to the FSM
194 * definition. The driver is responsible to ensure that FD reaches end
195 * of stream or error whenever the migration FSM leaves a data transfer
196 * state or before close_device() returns.
197 * @migration_get_state: Optional callback to get the migration state for
198 * devices that support migration. It's mandatory for
199 * VFIO_DEVICE_FEATURE_MIGRATION migration support.
200 * @migration_get_data_size: Optional callback to get the estimated data
201 * length that will be required to complete stop copy. It's mandatory for
202 * VFIO_DEVICE_FEATURE_MIGRATION migration support.
203 */
204 struct vfio_migration_ops {
205 struct file *(*migration_set_state)(
206 struct vfio_device *device,
207 enum vfio_device_mig_state new_state);
208 int (*migration_get_state)(struct vfio_device *device,
209 enum vfio_device_mig_state *curr_state);
210 int (*migration_get_data_size)(struct vfio_device *device,
211 unsigned long *stop_copy_length);
212 };
213
214 /**
215 * struct vfio_log_ops - VFIO bus device driver logging callbacks
216 *
217 * @log_start: Optional callback to ask the device start DMA logging.
218 * @log_stop: Optional callback to ask the device stop DMA logging.
219 * @log_read_and_clear: Optional callback to ask the device read
220 * and clear the dirty DMAs in some given range.
221 *
222 * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set
223 * of features does not track logging state relative to the device,
224 * therefore the device implementation of vfio_log_ops must handle
225 * arbitrary user requests. This includes rejecting subsequent calls
226 * to log_start without an intervening log_stop, as well as graceful
227 * handling of log_stop and log_read_and_clear from invalid states.
228 */
229 struct vfio_log_ops {
230 int (*log_start)(struct vfio_device *device,
231 struct rb_root_cached *ranges, u32 nnodes, u64 *page_size);
232 int (*log_stop)(struct vfio_device *device);
233 int (*log_read_and_clear)(struct vfio_device *device,
234 unsigned long iova, unsigned long length,
235 struct iova_bitmap *dirty);
236 };
237
238 /**
239 * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
240 * @flags: Arg from the device_feature op
241 * @argsz: Arg from the device_feature op
242 * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
243 * supports
244 * @minsz: Minimum data size the driver accepts
245 *
246 * For use in a driver's device_feature op. Checks that the inputs to the
247 * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
248 * the driver should execute the get or set, otherwise the relevant
249 * value should be returned.
250 */
vfio_check_feature(u32 flags,size_t argsz,u32 supported_ops,size_t minsz)251 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
252 size_t minsz)
253 {
254 if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
255 ~supported_ops)
256 return -EINVAL;
257 if (flags & VFIO_DEVICE_FEATURE_PROBE)
258 return 0;
259 /* Without PROBE one of GET or SET must be requested */
260 if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
261 return -EINVAL;
262 if (argsz < minsz)
263 return -EINVAL;
264 return 1;
265 }
266
267 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
268 const struct vfio_device_ops *ops);
269 #define vfio_alloc_device(dev_struct, member, dev, ops) \
270 container_of(_vfio_alloc_device(sizeof(struct dev_struct) + \
271 BUILD_BUG_ON_ZERO(offsetof( \
272 struct dev_struct, member)), \
273 dev, ops), \
274 struct dev_struct, member)
275
vfio_put_device(struct vfio_device * device)276 static inline void vfio_put_device(struct vfio_device *device)
277 {
278 put_device(&device->device);
279 }
280
281 int vfio_register_group_dev(struct vfio_device *device);
282 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
283 void vfio_unregister_group_dev(struct vfio_device *device);
284
285 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
286 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
287 struct vfio_device *
288 vfio_find_device_in_devset(struct vfio_device_set *dev_set,
289 struct device *dev);
290
291 int vfio_mig_get_next_state(struct vfio_device *device,
292 enum vfio_device_mig_state cur_fsm,
293 enum vfio_device_mig_state new_fsm,
294 enum vfio_device_mig_state *next_fsm);
295
296 void vfio_combine_iova_ranges(struct rb_root_cached *root, u32 cur_nodes,
297 u32 req_nodes);
298
299 /*
300 * External user API
301 */
302 struct iommu_group *vfio_file_iommu_group(struct file *file);
303
304 #if IS_ENABLED(CONFIG_VFIO_GROUP)
305 bool vfio_file_is_group(struct file *file);
306 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
307 #else
vfio_file_is_group(struct file * file)308 static inline bool vfio_file_is_group(struct file *file)
309 {
310 return false;
311 }
312
vfio_file_has_dev(struct file * file,struct vfio_device * device)313 static inline bool vfio_file_has_dev(struct file *file, struct vfio_device *device)
314 {
315 return false;
316 }
317 #endif
318 bool vfio_file_is_valid(struct file *file);
319 bool vfio_file_enforced_coherent(struct file *file);
320 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
321 struct device *vfio_file_get_device(struct file *file);
322
323 #define VFIO_PIN_PAGES_MAX_ENTRIES (PAGE_SIZE/sizeof(unsigned long))
324
325 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
326 int npage, int prot, struct page **pages);
327 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
328 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
329 void *data, size_t len, bool write);
330
331 /*
332 * Sub-module helpers
333 */
334 struct vfio_info_cap {
335 struct vfio_info_cap_header *buf;
336 size_t size;
337 };
338 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
339 size_t size, u16 id,
340 u16 version);
341 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
342
343 int vfio_info_add_capability(struct vfio_info_cap *caps,
344 struct vfio_info_cap_header *cap, size_t size);
345
346 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
347 int num_irqs, int max_irq_type,
348 size_t *data_size);
349
350 /*
351 * IRQfd - generic
352 */
353 struct virqfd {
354 void *opaque;
355 struct eventfd_ctx *eventfd;
356 int (*handler)(void *, void *);
357 void (*thread)(void *, void *);
358 void *data;
359 struct work_struct inject;
360 wait_queue_entry_t wait;
361 poll_table pt;
362 struct work_struct shutdown;
363 struct work_struct flush_inject;
364 struct virqfd **pvirqfd;
365 };
366
367 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
368 void (*thread)(void *, void *), void *data,
369 struct virqfd **pvirqfd, int fd);
370 void vfio_virqfd_disable(struct virqfd **pvirqfd);
371 void vfio_virqfd_flush_thread(struct virqfd **pvirqfd);
372
373 #endif /* VFIO_H */
374