Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[linux-block.git] / include / linux / vfio.h
CommitLineData
d2912cb1 1/* SPDX-License-Identifier: GPL-2.0-only */
cba3345c
AW
2/*
3 * VFIO API definition
4 *
5 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
6 * Author: Alex Williamson <alex.williamson@redhat.com>
cba3345c
AW
7 */
8#ifndef VFIO_H
9#define VFIO_H
10
cba3345c
AW
11
12#include <linux/iommu.h>
13#include <linux/mm.h>
7e992d69
AM
14#include <linux/workqueue.h>
15#include <linux/poll.h>
607ca46e 16#include <uapi/linux/vfio.h>
80c4b92a 17#include <linux/iova_bitmap.h>
cba3345c 18
ba70a89f 19struct kvm;
a4d1f91d
JG
20struct iommufd_ctx;
21struct iommufd_device;
4741f2e9 22struct iommufd_access;
ba70a89f 23
2fd585f4
JG
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 */
29struct vfio_device_set {
30 void *set_id;
31 struct mutex lock;
32 struct list_head device_list;
33 unsigned int device_count;
34};
35
0bfc6a4e
JG
36struct vfio_device {
37 struct device *dev;
38 const struct vfio_device_ops *ops;
6e97eba8 39 /*
80c4b92a
YH
40 * mig_ops/log_ops is a static property of the vfio_device which must
41 * be set prior to registering the vfio_device.
6e97eba8
YH
42 */
43 const struct vfio_migration_ops *mig_ops;
80c4b92a 44 const struct vfio_log_ops *log_ops;
0bfc6a4e 45 struct vfio_group *group;
2fd585f4
JG
46 struct vfio_device_set *dev_set;
47 struct list_head dev_set_list;
8cb3d83b 48 unsigned int migration_flags;
421cfe65
MR
49 /* Driver must reference the kvm during open_device or never touch it */
50 struct kvm *kvm;
0bfc6a4e
JG
51
52 /* Members below here are private, not for driver use */
3c28a761
YL
53 unsigned int index;
54 struct device device; /* device.kref covers object life circle */
cb9ff3f3 55 refcount_t refcount; /* user count on registered device*/
2fd585f4 56 unsigned int open_count;
0bfc6a4e
JG
57 struct completion comp;
58 struct list_head group_next;
8cfc5b60 59 struct list_head iommu_entry;
4741f2e9 60 struct iommufd_access *iommufd_access;
a4d1f91d
JG
61#if IS_ENABLED(CONFIG_IOMMUFD)
62 struct iommufd_device *iommufd_device;
4741f2e9 63 struct iommufd_ctx *iommufd_ictx;
a4d1f91d
JG
64 bool iommufd_attached;
65#endif
0bfc6a4e
JG
66};
67
cba3345c
AW
68/**
69 * struct vfio_device_ops - VFIO bus driver device callbacks
70 *
cb9ff3f3
KT
71 * @init: initialize private fields in device structure
72 * @release: Reclaim private fields in device structure
2fd585f4
JG
73 * @open_device: Called when the first file descriptor is opened for this device
74 * @close_device: Opposite of open_device
cba3345c
AW
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
13060b64 80 * @request: Request for the bus driver to release the device
5f3874c2
AW
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)
ce4b4657
JG
84 * @dma_unmap: Called when userspace unmaps IOVA from the container
85 * this device is attached to.
445ad495 86 * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
cba3345c
AW
87 */
88struct vfio_device_ops {
89 char *name;
cb9ff3f3
KT
90 int (*init)(struct vfio_device *vdev);
91 void (*release)(struct vfio_device *vdev);
a4d1f91d
JG
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);
2fd585f4
JG
96 int (*open_device)(struct vfio_device *vdev);
97 void (*close_device)(struct vfio_device *vdev);
6df62c5b 98 ssize_t (*read)(struct vfio_device *vdev, char __user *buf,
cba3345c 99 size_t count, loff_t *ppos);
6df62c5b 100 ssize_t (*write)(struct vfio_device *vdev, const char __user *buf,
cba3345c 101 size_t count, loff_t *size);
6df62c5b 102 long (*ioctl)(struct vfio_device *vdev, unsigned int cmd,
cba3345c 103 unsigned long arg);
6df62c5b
JG
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);
ce4b4657 107 void (*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
445ad495
JG
108 int (*device_feature)(struct vfio_device *device, u32 flags,
109 void __user *arg, size_t argsz);
6e97eba8
YH
110};
111
a4d1f91d
JG
112#if IS_ENABLED(CONFIG_IOMMUFD)
113int vfio_iommufd_physical_bind(struct vfio_device *vdev,
114 struct iommufd_ctx *ictx, u32 *out_device_id);
115void vfio_iommufd_physical_unbind(struct vfio_device *vdev);
116int vfio_iommufd_physical_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
4741f2e9
JG
117int vfio_iommufd_emulated_bind(struct vfio_device *vdev,
118 struct iommufd_ctx *ictx, u32 *out_device_id);
119void vfio_iommufd_emulated_unbind(struct vfio_device *vdev);
120int vfio_iommufd_emulated_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
a4d1f91d
JG
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)
4741f2e9
JG
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)
a4d1f91d
JG
136#endif
137
6e97eba8
YH
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.
4e016f96
YH
149 * @migration_get_data_size: Optional callback to get the estimated data
150 * length that will be required to complete stop copy. It's mandatory for
151 * VFIO_DEVICE_FEATURE_MIGRATION migration support.
6e97eba8
YH
152 */
153struct vfio_migration_ops {
115dcec6
JG
154 struct file *(*migration_set_state)(
155 struct vfio_device *device,
156 enum vfio_device_mig_state new_state);
157 int (*migration_get_state)(struct vfio_device *device,
158 enum vfio_device_mig_state *curr_state);
4e016f96
YH
159 int (*migration_get_data_size)(struct vfio_device *device,
160 unsigned long *stop_copy_length);
cba3345c
AW
161};
162
80c4b92a
YH
163/**
164 * @log_start: Optional callback to ask the device start DMA logging.
165 * @log_stop: Optional callback to ask the device stop DMA logging.
166 * @log_read_and_clear: Optional callback to ask the device read
167 * and clear the dirty DMAs in some given range.
168 *
169 * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set
170 * of features does not track logging state relative to the device,
171 * therefore the device implementation of vfio_log_ops must handle
172 * arbitrary user requests. This includes rejecting subsequent calls
173 * to log_start without an intervening log_stop, as well as graceful
174 * handling of log_stop and log_read_and_clear from invalid states.
175 */
176struct vfio_log_ops {
177 int (*log_start)(struct vfio_device *device,
178 struct rb_root_cached *ranges, u32 nnodes, u64 *page_size);
179 int (*log_stop)(struct vfio_device *device);
180 int (*log_read_and_clear)(struct vfio_device *device,
181 unsigned long iova, unsigned long length,
182 struct iova_bitmap *dirty);
183};
184
445ad495
JG
185/**
186 * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
187 * @flags: Arg from the device_feature op
188 * @argsz: Arg from the device_feature op
189 * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
190 * supports
191 * @minsz: Minimum data size the driver accepts
192 *
193 * For use in a driver's device_feature op. Checks that the inputs to the
194 * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
195 * the driver should execute the get or set, otherwise the relevant
196 * value should be returned.
197 */
198static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
199 size_t minsz)
200{
201 if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
202 ~supported_ops)
203 return -EINVAL;
204 if (flags & VFIO_DEVICE_FEATURE_PROBE)
205 return 0;
206 /* Without PROBE one of GET or SET must be requested */
207 if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
208 return -EINVAL;
209 if (argsz < minsz)
210 return -EINVAL;
211 return 1;
212}
213
cb9ff3f3
KT
214struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
215 const struct vfio_device_ops *ops);
216#define vfio_alloc_device(dev_struct, member, dev, ops) \
217 container_of(_vfio_alloc_device(sizeof(struct dev_struct) + \
218 BUILD_BUG_ON_ZERO(offsetof( \
219 struct dev_struct, member)), \
220 dev, ops), \
221 struct dev_struct, member)
222
cb9ff3f3
KT
223static inline void vfio_put_device(struct vfio_device *device)
224{
3c28a761 225 put_device(&device->device);
cb9ff3f3
KT
226}
227
0bfc6a4e 228int vfio_register_group_dev(struct vfio_device *device);
c68ea0d0 229int vfio_register_emulated_iommu_dev(struct vfio_device *device);
0bfc6a4e 230void vfio_unregister_group_dev(struct vfio_device *device);
cba3345c 231
2fd585f4 232int vfio_assign_device_set(struct vfio_device *device, void *set_id);
5cd189e4 233unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
2fd585f4 234
115dcec6
JG
235int vfio_mig_get_next_state(struct vfio_device *device,
236 enum vfio_device_mig_state cur_fsm,
237 enum vfio_device_mig_state new_fsm,
238 enum vfio_device_mig_state *next_fsm);
239
6cdd9782
AK
240/*
241 * External user API
242 */
d1877e63 243struct iommu_group *vfio_file_iommu_group(struct file *file);
4b22ef04 244bool vfio_file_is_group(struct file *file);
d1877e63
AW
245bool vfio_file_enforced_coherent(struct file *file);
246void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
247bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
6cdd9782 248
2169037d
KW
249#define VFIO_PIN_PAGES_MAX_ENTRIES (PAGE_SIZE/sizeof(unsigned long))
250
44abdd16 251int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
34a255e6 252 int npage, int prot, struct page **pages);
44abdd16 253void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
8561aa4f 254int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
d1877e63 255 void *data, size_t len, bool write);
8d46c0cc 256
d7a8d5ed
AW
257/*
258 * Sub-module helpers
259 */
260struct vfio_info_cap {
261 struct vfio_info_cap_header *buf;
262 size_t size;
263};
d1877e63
AW
264struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
265 size_t size, u16 id,
266 u16 version);
267void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
d7a8d5ed 268
d1877e63
AW
269int vfio_info_add_capability(struct vfio_info_cap *caps,
270 struct vfio_info_cap_header *cap, size_t size);
b3c0a866 271
d1877e63
AW
272int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
273 int num_irqs, int max_irq_type,
274 size_t *data_size);
c747f08a 275
7e992d69
AM
276/*
277 * IRQfd - generic
278 */
279struct virqfd {
280 void *opaque;
281 struct eventfd_ctx *eventfd;
282 int (*handler)(void *, void *);
283 void (*thread)(void *, void *);
284 void *data;
285 struct work_struct inject;
ac6424b9 286 wait_queue_entry_t wait;
7e992d69
AM
287 poll_table pt;
288 struct work_struct shutdown;
289 struct virqfd **pvirqfd;
290};
291
d1877e63
AW
292int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
293 void (*thread)(void *, void *), void *data,
294 struct virqfd **pvirqfd, int fd);
295void vfio_virqfd_disable(struct virqfd **pvirqfd);
7e992d69 296
cba3345c 297#endif /* VFIO_H */