xref: /linux/include/linux/vfio.h (revision 0074281bb6316108e0cff094bd4db78ab3eee236)
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 	struct ida pasids;
71 	u8 iommufd_attached:1;
72 #endif
73 	u8 cdev_opened:1;
74 #ifdef CONFIG_DEBUG_FS
75 	/*
76 	 * debug_root is a static property of the vfio_device
77 	 * which must be set prior to registering the vfio_device.
78 	 */
79 	struct dentry *debug_root;
80 #endif
81 };
82 
83 /**
84  * struct vfio_device_ops - VFIO bus driver device callbacks
85  *
86  * @name: Name of the device driver.
87  * @init: initialize private fields in device structure
88  * @release: Reclaim private fields in device structure
89  * @bind_iommufd: Called when binding the device to an iommufd
90  * @unbind_iommufd: Opposite of bind_iommufd
91  * @attach_ioas: Called when attaching device to an IOAS/HWPT managed by the
92  *		 bound iommufd. Undo in unbind_iommufd if @detach_ioas is not
93  *		 called.
94  * @detach_ioas: Opposite of attach_ioas
95  * @pasid_attach_ioas: The pasid variation of attach_ioas
96  * @pasid_detach_ioas: Opposite of pasid_attach_ioas
97  * @open_device: Called when the first file descriptor is opened for this device
98  * @close_device: Opposite of open_device
99  * @read: Perform read(2) on device file descriptor
100  * @write: Perform write(2) on device file descriptor
101  * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
102  *         operations documented below
103  * @mmap: Perform mmap(2) on a region of the device file descriptor
104  * @request: Request for the bus driver to release the device
105  * @match: Optional device name match callback (return: 0 for no-match, >0 for
106  *         match, -errno for abort (ex. match with insufficient or incorrect
107  *         additional args)
108  * @match_token_uuid: Optional device token match/validation. Return 0
109  *         if the uuid is valid for the device, -errno otherwise. uuid is NULL
110  *         if none was provided.
111  * @dma_unmap: Called when userspace unmaps IOVA from the container
112  *             this device is attached to.
113  * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
114  */
115 struct vfio_device_ops {
116 	char	*name;
117 	int	(*init)(struct vfio_device *vdev);
118 	void	(*release)(struct vfio_device *vdev);
119 	int	(*bind_iommufd)(struct vfio_device *vdev,
120 				struct iommufd_ctx *ictx, u32 *out_device_id);
121 	void	(*unbind_iommufd)(struct vfio_device *vdev);
122 	int	(*attach_ioas)(struct vfio_device *vdev, u32 *pt_id);
123 	void	(*detach_ioas)(struct vfio_device *vdev);
124 	int	(*pasid_attach_ioas)(struct vfio_device *vdev, u32 pasid,
125 				     u32 *pt_id);
126 	void	(*pasid_detach_ioas)(struct vfio_device *vdev, u32 pasid);
127 	int	(*open_device)(struct vfio_device *vdev);
128 	void	(*close_device)(struct vfio_device *vdev);
129 	ssize_t	(*read)(struct vfio_device *vdev, char __user *buf,
130 			size_t count, loff_t *ppos);
131 	ssize_t	(*write)(struct vfio_device *vdev, const char __user *buf,
132 			 size_t count, loff_t *size);
133 	long	(*ioctl)(struct vfio_device *vdev, unsigned int cmd,
134 			 unsigned long arg);
135 	int	(*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma);
136 	void	(*request)(struct vfio_device *vdev, unsigned int count);
137 	int	(*match)(struct vfio_device *vdev, char *buf);
138 	int	(*match_token_uuid)(struct vfio_device *vdev, const uuid_t *uuid);
139 	void	(*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
140 	int	(*device_feature)(struct vfio_device *device, u32 flags,
141 				  void __user *arg, size_t argsz);
142 };
143 
144 #if IS_ENABLED(CONFIG_IOMMUFD)
145 struct iommufd_ctx *vfio_iommufd_device_ictx(struct vfio_device *vdev);
146 int vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx);
147 int vfio_iommufd_physical_bind(struct vfio_device *vdev,
148 			       struct iommufd_ctx *ictx, u32 *out_device_id);
149 void vfio_iommufd_physical_unbind(struct vfio_device *vdev);
150 int vfio_iommufd_physical_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
151 void vfio_iommufd_physical_detach_ioas(struct vfio_device *vdev);
152 int vfio_iommufd_physical_pasid_attach_ioas(struct vfio_device *vdev,
153 					    u32 pasid, u32 *pt_id);
154 void vfio_iommufd_physical_pasid_detach_ioas(struct vfio_device *vdev,
155 					     u32 pasid);
156 int vfio_iommufd_emulated_bind(struct vfio_device *vdev,
157 			       struct iommufd_ctx *ictx, u32 *out_device_id);
158 void vfio_iommufd_emulated_unbind(struct vfio_device *vdev);
159 int vfio_iommufd_emulated_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
160 void vfio_iommufd_emulated_detach_ioas(struct vfio_device *vdev);
161 #else
162 static inline struct iommufd_ctx *
vfio_iommufd_device_ictx(struct vfio_device * vdev)163 vfio_iommufd_device_ictx(struct vfio_device *vdev)
164 {
165 	return NULL;
166 }
167 
168 static inline int
vfio_iommufd_get_dev_id(struct vfio_device * vdev,struct iommufd_ctx * ictx)169 vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx)
170 {
171 	return VFIO_PCI_DEVID_NOT_OWNED;
172 }
173 
174 #define vfio_iommufd_physical_bind                                      \
175 	((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx,   \
176 		  u32 *out_device_id)) NULL)
177 #define vfio_iommufd_physical_unbind \
178 	((void (*)(struct vfio_device *vdev)) NULL)
179 #define vfio_iommufd_physical_attach_ioas \
180 	((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
181 #define vfio_iommufd_physical_detach_ioas \
182 	((void (*)(struct vfio_device *vdev)) NULL)
183 #define vfio_iommufd_physical_pasid_attach_ioas \
184 	((int (*)(struct vfio_device *vdev, u32 pasid, u32 *pt_id)) NULL)
185 #define vfio_iommufd_physical_pasid_detach_ioas \
186 	((void (*)(struct vfio_device *vdev, u32 pasid)) NULL)
187 #define vfio_iommufd_emulated_bind                                      \
188 	((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx,   \
189 		  u32 *out_device_id)) NULL)
190 #define vfio_iommufd_emulated_unbind \
191 	((void (*)(struct vfio_device *vdev)) NULL)
192 #define vfio_iommufd_emulated_attach_ioas \
193 	((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
194 #define vfio_iommufd_emulated_detach_ioas \
195 	((void (*)(struct vfio_device *vdev)) NULL)
196 #endif
197 
vfio_device_cdev_opened(struct vfio_device * device)198 static inline bool vfio_device_cdev_opened(struct vfio_device *device)
199 {
200 	return device->cdev_opened;
201 }
202 
203 /**
204  * struct vfio_migration_ops - VFIO bus device driver migration callbacks
205  *
206  * @migration_set_state: Optional callback to change the migration state for
207  *         devices that support migration. It's mandatory for
208  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
209  *         The returned FD is used for data transfer according to the FSM
210  *         definition. The driver is responsible to ensure that FD reaches end
211  *         of stream or error whenever the migration FSM leaves a data transfer
212  *         state or before close_device() returns.
213  * @migration_get_state: Optional callback to get the migration state for
214  *         devices that support migration. It's mandatory for
215  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
216  * @migration_get_data_size: Optional callback to get the estimated data
217  *          length that will be required to complete stop copy. It's mandatory for
218  *          VFIO_DEVICE_FEATURE_MIGRATION migration support.
219  */
220 struct vfio_migration_ops {
221 	struct file *(*migration_set_state)(
222 		struct vfio_device *device,
223 		enum vfio_device_mig_state new_state);
224 	int (*migration_get_state)(struct vfio_device *device,
225 				   enum vfio_device_mig_state *curr_state);
226 	int (*migration_get_data_size)(struct vfio_device *device,
227 				       unsigned long *stop_copy_length);
228 };
229 
230 /**
231  * struct vfio_log_ops - VFIO bus device driver logging callbacks
232  *
233  * @log_start: Optional callback to ask the device start DMA logging.
234  * @log_stop: Optional callback to ask the device stop DMA logging.
235  * @log_read_and_clear: Optional callback to ask the device read
236  *         and clear the dirty DMAs in some given range.
237  *
238  * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set
239  * of features does not track logging state relative to the device,
240  * therefore the device implementation of vfio_log_ops must handle
241  * arbitrary user requests. This includes rejecting subsequent calls
242  * to log_start without an intervening log_stop, as well as graceful
243  * handling of log_stop and log_read_and_clear from invalid states.
244  */
245 struct vfio_log_ops {
246 	int (*log_start)(struct vfio_device *device,
247 		struct rb_root_cached *ranges, u32 nnodes, u64 *page_size);
248 	int (*log_stop)(struct vfio_device *device);
249 	int (*log_read_and_clear)(struct vfio_device *device,
250 		unsigned long iova, unsigned long length,
251 		struct iova_bitmap *dirty);
252 };
253 
254 /**
255  * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
256  * @flags: Arg from the device_feature op
257  * @argsz: Arg from the device_feature op
258  * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
259  *                 supports
260  * @minsz: Minimum data size the driver accepts
261  *
262  * For use in a driver's device_feature op. Checks that the inputs to the
263  * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
264  * the driver should execute the get or set, otherwise the relevant
265  * value should be returned.
266  */
vfio_check_feature(u32 flags,size_t argsz,u32 supported_ops,size_t minsz)267 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
268 				    size_t minsz)
269 {
270 	if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
271 	    ~supported_ops)
272 		return -EINVAL;
273 	if (flags & VFIO_DEVICE_FEATURE_PROBE)
274 		return 0;
275 	/* Without PROBE one of GET or SET must be requested */
276 	if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
277 		return -EINVAL;
278 	if (argsz < minsz)
279 		return -EINVAL;
280 	return 1;
281 }
282 
283 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
284 				       const struct vfio_device_ops *ops);
285 #define vfio_alloc_device(dev_struct, member, dev, ops)				\
286 	container_of(_vfio_alloc_device(sizeof(struct dev_struct) +		\
287 					BUILD_BUG_ON_ZERO(offsetof(		\
288 						struct dev_struct, member)),	\
289 					dev, ops),				\
290 		     struct dev_struct, member)
291 
vfio_put_device(struct vfio_device * device)292 static inline void vfio_put_device(struct vfio_device *device)
293 {
294 	put_device(&device->device);
295 }
296 
297 int vfio_register_group_dev(struct vfio_device *device);
298 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
299 void vfio_unregister_group_dev(struct vfio_device *device);
300 
301 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
302 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
303 struct vfio_device *
304 vfio_find_device_in_devset(struct vfio_device_set *dev_set,
305 			   struct device *dev);
306 
307 int vfio_mig_get_next_state(struct vfio_device *device,
308 			    enum vfio_device_mig_state cur_fsm,
309 			    enum vfio_device_mig_state new_fsm,
310 			    enum vfio_device_mig_state *next_fsm);
311 
312 void vfio_combine_iova_ranges(struct rb_root_cached *root, u32 cur_nodes,
313 			      u32 req_nodes);
314 
315 /*
316  * External user API
317  */
318 struct iommu_group *vfio_file_iommu_group(struct file *file);
319 
320 #if IS_ENABLED(CONFIG_VFIO_GROUP)
321 bool vfio_file_is_group(struct file *file);
322 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
323 #else
vfio_file_is_group(struct file * file)324 static inline bool vfio_file_is_group(struct file *file)
325 {
326 	return false;
327 }
328 
vfio_file_has_dev(struct file * file,struct vfio_device * device)329 static inline bool vfio_file_has_dev(struct file *file, struct vfio_device *device)
330 {
331 	return false;
332 }
333 #endif
334 bool vfio_file_is_valid(struct file *file);
335 bool vfio_file_enforced_coherent(struct file *file);
336 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
337 
338 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
339 
340 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
341 		   int npage, int prot, struct page **pages);
342 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
343 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
344 		void *data, size_t len, bool write);
345 
346 /*
347  * Sub-module helpers
348  */
349 struct vfio_info_cap {
350 	struct vfio_info_cap_header *buf;
351 	size_t size;
352 };
353 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
354 					       size_t size, u16 id,
355 					       u16 version);
356 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
357 
358 int vfio_info_add_capability(struct vfio_info_cap *caps,
359 			     struct vfio_info_cap_header *cap, size_t size);
360 
361 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
362 				       int num_irqs, int max_irq_type,
363 				       size_t *data_size);
364 
365 /*
366  * IRQfd - generic
367  */
368 struct virqfd {
369 	void			*opaque;
370 	struct eventfd_ctx	*eventfd;
371 	int			(*handler)(void *, void *);
372 	void			(*thread)(void *, void *);
373 	void			*data;
374 	struct work_struct	inject;
375 	wait_queue_entry_t		wait;
376 	poll_table		pt;
377 	struct work_struct	shutdown;
378 	struct work_struct	flush_inject;
379 	struct virqfd		**pvirqfd;
380 };
381 
382 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
383 		       void (*thread)(void *, void *), void *data,
384 		       struct virqfd **pvirqfd, int fd);
385 void vfio_virqfd_disable(struct virqfd **pvirqfd);
386 void vfio_virqfd_flush_thread(struct virqfd **pvirqfd);
387 
388 #endif /* VFIO_H */
389