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

source code of linux/include/linux/vfio.h