1 /* SPDX-License-Identifier: GPL-2.0-only */
5 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
12 #include <linux/iommu.h>
14 #include <linux/workqueue.h>
15 #include <linux/poll.h>
16 #include <uapi/linux/vfio.h>
17 #include <linux/iova_bitmap.h>
21 struct iommufd_device;
22 struct iommufd_access;
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.
29 struct vfio_device_set {
32 struct list_head device_list;
33 unsigned int device_count;
38 const struct vfio_device_ops *ops;
40 * mig_ops/log_ops is a static property of the vfio_device which must
41 * be set prior to registering the vfio_device.
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 */
52 /* Members below here are private, not for driver use */
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;
69 * struct vfio_device_ops - VFIO bus driver device callbacks
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
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
88 struct vfio_device_ops {
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,
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);
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);
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)
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 * @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.
153 struct vfio_migration_ops {
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);
159 int (*migration_get_data_size)(struct vfio_device *device,
160 unsigned long *stop_copy_length);
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.
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.
176 struct 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);
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
191 * @minsz: Minimum data size the driver accepts
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.
198 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
201 if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
204 if (flags & VFIO_DEVICE_FEATURE_PROBE)
206 /* Without PROBE one of GET or SET must be requested */
207 if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
214 struct 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)), \
221 struct dev_struct, member)
223 static inline void vfio_put_device(struct vfio_device *device)
225 put_device(&device->device);
228 int vfio_register_group_dev(struct vfio_device *device);
229 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
230 void vfio_unregister_group_dev(struct vfio_device *device);
232 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
233 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
235 int 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);
243 struct iommu_group *vfio_file_iommu_group(struct file *file);
244 bool vfio_file_is_group(struct file *file);
245 bool vfio_file_enforced_coherent(struct file *file);
246 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
247 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
249 #define VFIO_PIN_PAGES_MAX_ENTRIES (PAGE_SIZE/sizeof(unsigned long))
251 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
252 int npage, int prot, struct page **pages);
253 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
254 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
255 void *data, size_t len, bool write);
260 struct vfio_info_cap {
261 struct vfio_info_cap_header *buf;
264 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
267 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
269 int vfio_info_add_capability(struct vfio_info_cap *caps,
270 struct vfio_info_cap_header *cap, size_t size);
272 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
273 int num_irqs, int max_irq_type,
281 struct eventfd_ctx *eventfd;
282 int (*handler)(void *, void *);
283 void (*thread)(void *, void *);
285 struct work_struct inject;
286 wait_queue_entry_t wait;
288 struct work_struct shutdown;
289 struct virqfd **pvirqfd;
292 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
293 void (*thread)(void *, void *), void *data,
294 struct virqfd **pvirqfd, int fd);
295 void vfio_virqfd_disable(struct virqfd **pvirqfd);