xref: /linux-6.15/include/linux/vfio.h (revision 71c53e19)
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 <[email protected]>
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 <uapi/linux/vfio.h>
17 #include <linux/iova_bitmap.h>
18 
19 struct kvm;
20 struct iommufd_ctx;
21 struct iommufd_device;
22 struct iommufd_access;
23 
24 /*
25  * VFIO devices can be placed in a set, this allows all devices to share this
26  * structure and the VFIO core will provide a lock that is held around
27  * open_device()/close_device() for all devices in the set.
28  */
29 struct vfio_device_set {
30 	void *set_id;
31 	struct mutex lock;
32 	struct list_head device_list;
33 	unsigned int device_count;
34 };
35 
36 struct vfio_device {
37 	struct device *dev;
38 	const struct vfio_device_ops *ops;
39 	/*
40 	 * mig_ops/log_ops is a static property of the vfio_device which must
41 	 * be set prior to registering the vfio_device.
42 	 */
43 	const struct vfio_migration_ops *mig_ops;
44 	const struct vfio_log_ops *log_ops;
45 	struct vfio_group *group;
46 	struct vfio_device_set *dev_set;
47 	struct list_head dev_set_list;
48 	unsigned int migration_flags;
49 	/* Driver must reference the kvm during open_device or never touch it */
50 	struct kvm *kvm;
51 
52 	/* Members below here are private, not for driver use */
53 	unsigned int index;
54 	struct device device;	/* device.kref covers object life circle */
55 	refcount_t refcount;	/* user count on registered device*/
56 	unsigned int open_count;
57 	struct completion comp;
58 	struct list_head group_next;
59 	struct list_head iommu_entry;
60 	struct iommufd_access *iommufd_access;
61 #if IS_ENABLED(CONFIG_IOMMUFD)
62 	struct iommufd_device *iommufd_device;
63 	struct iommufd_ctx *iommufd_ictx;
64 	bool iommufd_attached;
65 #endif
66 };
67 
68 /**
69  * struct vfio_device_ops - VFIO bus driver device callbacks
70  *
71  * @init: initialize private fields in device structure
72  * @release: Reclaim private fields in device structure
73  * @open_device: Called when the first file descriptor is opened for this device
74  * @close_device: Opposite of open_device
75  * @read: Perform read(2) on device file descriptor
76  * @write: Perform write(2) on device file descriptor
77  * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
78  *         operations documented below
79  * @mmap: Perform mmap(2) on a region of the device file descriptor
80  * @request: Request for the bus driver to release the device
81  * @match: Optional device name match callback (return: 0 for no-match, >0 for
82  *         match, -errno for abort (ex. match with insufficient or incorrect
83  *         additional args)
84  * @dma_unmap: Called when userspace unmaps IOVA from the container
85  *             this device is attached to.
86  * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
87  */
88 struct vfio_device_ops {
89 	char	*name;
90 	int	(*init)(struct vfio_device *vdev);
91 	void	(*release)(struct vfio_device *vdev);
92 	int	(*bind_iommufd)(struct vfio_device *vdev,
93 				struct iommufd_ctx *ictx, u32 *out_device_id);
94 	void	(*unbind_iommufd)(struct vfio_device *vdev);
95 	int	(*attach_ioas)(struct vfio_device *vdev, u32 *pt_id);
96 	int	(*open_device)(struct vfio_device *vdev);
97 	void	(*close_device)(struct vfio_device *vdev);
98 	ssize_t	(*read)(struct vfio_device *vdev, char __user *buf,
99 			size_t count, loff_t *ppos);
100 	ssize_t	(*write)(struct vfio_device *vdev, const char __user *buf,
101 			 size_t count, loff_t *size);
102 	long	(*ioctl)(struct vfio_device *vdev, unsigned int cmd,
103 			 unsigned long arg);
104 	int	(*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma);
105 	void	(*request)(struct vfio_device *vdev, unsigned int count);
106 	int	(*match)(struct vfio_device *vdev, char *buf);
107 	void	(*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
108 	int	(*device_feature)(struct vfio_device *device, u32 flags,
109 				  void __user *arg, size_t argsz);
110 };
111 
112 #if IS_ENABLED(CONFIG_IOMMUFD)
113 int vfio_iommufd_physical_bind(struct vfio_device *vdev,
114 			       struct iommufd_ctx *ictx, u32 *out_device_id);
115 void vfio_iommufd_physical_unbind(struct vfio_device *vdev);
116 int vfio_iommufd_physical_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
117 int vfio_iommufd_emulated_bind(struct vfio_device *vdev,
118 			       struct iommufd_ctx *ictx, u32 *out_device_id);
119 void vfio_iommufd_emulated_unbind(struct vfio_device *vdev);
120 int vfio_iommufd_emulated_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
121 #else
122 #define vfio_iommufd_physical_bind                                      \
123 	((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx,   \
124 		  u32 *out_device_id)) NULL)
125 #define vfio_iommufd_physical_unbind \
126 	((void (*)(struct vfio_device *vdev)) NULL)
127 #define vfio_iommufd_physical_attach_ioas \
128 	((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
129 #define vfio_iommufd_emulated_bind                                      \
130 	((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx,   \
131 		  u32 *out_device_id)) NULL)
132 #define vfio_iommufd_emulated_unbind \
133 	((void (*)(struct vfio_device *vdev)) NULL)
134 #define vfio_iommufd_emulated_attach_ioas \
135 	((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
136 #endif
137 
138 /**
139  * @migration_set_state: Optional callback to change the migration state for
140  *         devices that support migration. It's mandatory for
141  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
142  *         The returned FD is used for data transfer according to the FSM
143  *         definition. The driver is responsible to ensure that FD reaches end
144  *         of stream or error whenever the migration FSM leaves a data transfer
145  *         state or before close_device() returns.
146  * @migration_get_state: Optional callback to get the migration state for
147  *         devices that support migration. It's mandatory for
148  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
149  */
150 struct vfio_migration_ops {
151 	struct file *(*migration_set_state)(
152 		struct vfio_device *device,
153 		enum vfio_device_mig_state new_state);
154 	int (*migration_get_state)(struct vfio_device *device,
155 				   enum vfio_device_mig_state *curr_state);
156 };
157 
158 /**
159  * @log_start: Optional callback to ask the device start DMA logging.
160  * @log_stop: Optional callback to ask the device stop DMA logging.
161  * @log_read_and_clear: Optional callback to ask the device read
162  *         and clear the dirty DMAs in some given range.
163  *
164  * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set
165  * of features does not track logging state relative to the device,
166  * therefore the device implementation of vfio_log_ops must handle
167  * arbitrary user requests. This includes rejecting subsequent calls
168  * to log_start without an intervening log_stop, as well as graceful
169  * handling of log_stop and log_read_and_clear from invalid states.
170  */
171 struct vfio_log_ops {
172 	int (*log_start)(struct vfio_device *device,
173 		struct rb_root_cached *ranges, u32 nnodes, u64 *page_size);
174 	int (*log_stop)(struct vfio_device *device);
175 	int (*log_read_and_clear)(struct vfio_device *device,
176 		unsigned long iova, unsigned long length,
177 		struct iova_bitmap *dirty);
178 };
179 
180 /**
181  * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
182  * @flags: Arg from the device_feature op
183  * @argsz: Arg from the device_feature op
184  * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
185  *                 supports
186  * @minsz: Minimum data size the driver accepts
187  *
188  * For use in a driver's device_feature op. Checks that the inputs to the
189  * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
190  * the driver should execute the get or set, otherwise the relevant
191  * value should be returned.
192  */
193 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
194 				    size_t minsz)
195 {
196 	if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
197 	    ~supported_ops)
198 		return -EINVAL;
199 	if (flags & VFIO_DEVICE_FEATURE_PROBE)
200 		return 0;
201 	/* Without PROBE one of GET or SET must be requested */
202 	if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
203 		return -EINVAL;
204 	if (argsz < minsz)
205 		return -EINVAL;
206 	return 1;
207 }
208 
209 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
210 				       const struct vfio_device_ops *ops);
211 #define vfio_alloc_device(dev_struct, member, dev, ops)				\
212 	container_of(_vfio_alloc_device(sizeof(struct dev_struct) +		\
213 					BUILD_BUG_ON_ZERO(offsetof(		\
214 						struct dev_struct, member)),	\
215 					dev, ops),				\
216 		     struct dev_struct, member)
217 
218 int vfio_init_device(struct vfio_device *device, struct device *dev,
219 		     const struct vfio_device_ops *ops);
220 void vfio_free_device(struct vfio_device *device);
221 static inline void vfio_put_device(struct vfio_device *device)
222 {
223 	put_device(&device->device);
224 }
225 
226 int vfio_register_group_dev(struct vfio_device *device);
227 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
228 void vfio_unregister_group_dev(struct vfio_device *device);
229 
230 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
231 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
232 
233 int vfio_mig_get_next_state(struct vfio_device *device,
234 			    enum vfio_device_mig_state cur_fsm,
235 			    enum vfio_device_mig_state new_fsm,
236 			    enum vfio_device_mig_state *next_fsm);
237 
238 /*
239  * External user API
240  */
241 struct iommu_group *vfio_file_iommu_group(struct file *file);
242 bool vfio_file_is_group(struct file *file);
243 bool vfio_file_enforced_coherent(struct file *file);
244 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
245 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
246 
247 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
248 
249 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
250 		   int npage, int prot, struct page **pages);
251 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
252 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
253 		void *data, size_t len, bool write);
254 
255 /*
256  * Sub-module helpers
257  */
258 struct vfio_info_cap {
259 	struct vfio_info_cap_header *buf;
260 	size_t size;
261 };
262 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
263 					       size_t size, u16 id,
264 					       u16 version);
265 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
266 
267 int vfio_info_add_capability(struct vfio_info_cap *caps,
268 			     struct vfio_info_cap_header *cap, size_t size);
269 
270 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
271 				       int num_irqs, int max_irq_type,
272 				       size_t *data_size);
273 
274 struct pci_dev;
275 #if IS_ENABLED(CONFIG_VFIO_SPAPR_EEH)
276 void vfio_spapr_pci_eeh_open(struct pci_dev *pdev);
277 void vfio_spapr_pci_eeh_release(struct pci_dev *pdev);
278 long vfio_spapr_iommu_eeh_ioctl(struct iommu_group *group, unsigned int cmd,
279 				unsigned long arg);
280 #else
281 static inline void vfio_spapr_pci_eeh_open(struct pci_dev *pdev)
282 {
283 }
284 
285 static inline void vfio_spapr_pci_eeh_release(struct pci_dev *pdev)
286 {
287 }
288 
289 static inline long vfio_spapr_iommu_eeh_ioctl(struct iommu_group *group,
290 					      unsigned int cmd,
291 					      unsigned long arg)
292 {
293 	return -ENOTTY;
294 }
295 #endif /* CONFIG_VFIO_SPAPR_EEH */
296 
297 /*
298  * IRQfd - generic
299  */
300 struct virqfd {
301 	void			*opaque;
302 	struct eventfd_ctx	*eventfd;
303 	int			(*handler)(void *, void *);
304 	void			(*thread)(void *, void *);
305 	void			*data;
306 	struct work_struct	inject;
307 	wait_queue_entry_t		wait;
308 	poll_table		pt;
309 	struct work_struct	shutdown;
310 	struct virqfd		**pvirqfd;
311 };
312 
313 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
314 		       void (*thread)(void *, void *), void *data,
315 		       struct virqfd **pvirqfd, int fd);
316 void vfio_virqfd_disable(struct virqfd **pvirqfd);
317 
318 #endif /* VFIO_H */
319