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