1 /* SPDX-License-Identifier: GPL-2.0-only */
5 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
6 * Author: Alex Williamson <alex.williamson@redhat.com>
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>
22 * VFIO devices can be placed in a set, this allows all devices to share this
23 * structure and the VFIO core will provide a lock that is held around
24 * open_device()/close_device() for all devices in the set.
26 struct vfio_device_set {
29 struct list_head device_list;
30 unsigned int device_count;
35 const struct vfio_device_ops *ops;
37 * mig_ops/log_ops is a static property of the vfio_device which must
38 * be set prior to registering the vfio_device.
40 const struct vfio_migration_ops *mig_ops;
41 const struct vfio_log_ops *log_ops;
42 struct vfio_group *group;
43 struct vfio_device_set *dev_set;
44 struct list_head dev_set_list;
45 unsigned int migration_flags;
46 /* Driver must reference the kvm during open_device or never touch it */
49 /* Members below here are private, not for driver use */
51 struct device device; /* device.kref covers object life circle */
52 refcount_t refcount; /* user count on registered device*/
53 unsigned int open_count;
54 struct completion comp;
55 struct list_head group_next;
56 struct list_head iommu_entry;
60 * struct vfio_device_ops - VFIO bus driver device callbacks
62 * @init: initialize private fields in device structure
63 * @release: Reclaim private fields in device structure
64 * @open_device: Called when the first file descriptor is opened for this device
65 * @close_device: Opposite of open_device
66 * @read: Perform read(2) on device file descriptor
67 * @write: Perform write(2) on device file descriptor
68 * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
69 * operations documented below
70 * @mmap: Perform mmap(2) on a region of the device file descriptor
71 * @request: Request for the bus driver to release the device
72 * @match: Optional device name match callback (return: 0 for no-match, >0 for
73 * match, -errno for abort (ex. match with insufficient or incorrect
75 * @dma_unmap: Called when userspace unmaps IOVA from the container
76 * this device is attached to.
77 * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
79 struct vfio_device_ops {
81 int (*init)(struct vfio_device *vdev);
82 void (*release)(struct vfio_device *vdev);
83 int (*open_device)(struct vfio_device *vdev);
84 void (*close_device)(struct vfio_device *vdev);
85 ssize_t (*read)(struct vfio_device *vdev, char __user *buf,
86 size_t count, loff_t *ppos);
87 ssize_t (*write)(struct vfio_device *vdev, const char __user *buf,
88 size_t count, loff_t *size);
89 long (*ioctl)(struct vfio_device *vdev, unsigned int cmd,
91 int (*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma);
92 void (*request)(struct vfio_device *vdev, unsigned int count);
93 int (*match)(struct vfio_device *vdev, char *buf);
94 void (*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
95 int (*device_feature)(struct vfio_device *device, u32 flags,
96 void __user *arg, size_t argsz);
100 * @migration_set_state: Optional callback to change the migration state for
101 * devices that support migration. It's mandatory for
102 * VFIO_DEVICE_FEATURE_MIGRATION migration support.
103 * The returned FD is used for data transfer according to the FSM
104 * definition. The driver is responsible to ensure that FD reaches end
105 * of stream or error whenever the migration FSM leaves a data transfer
106 * state or before close_device() returns.
107 * @migration_get_state: Optional callback to get the migration state for
108 * devices that support migration. It's mandatory for
109 * VFIO_DEVICE_FEATURE_MIGRATION migration support.
111 struct vfio_migration_ops {
112 struct file *(*migration_set_state)(
113 struct vfio_device *device,
114 enum vfio_device_mig_state new_state);
115 int (*migration_get_state)(struct vfio_device *device,
116 enum vfio_device_mig_state *curr_state);
120 * @log_start: Optional callback to ask the device start DMA logging.
121 * @log_stop: Optional callback to ask the device stop DMA logging.
122 * @log_read_and_clear: Optional callback to ask the device read
123 * and clear the dirty DMAs in some given range.
125 * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set
126 * of features does not track logging state relative to the device,
127 * therefore the device implementation of vfio_log_ops must handle
128 * arbitrary user requests. This includes rejecting subsequent calls
129 * to log_start without an intervening log_stop, as well as graceful
130 * handling of log_stop and log_read_and_clear from invalid states.
132 struct vfio_log_ops {
133 int (*log_start)(struct vfio_device *device,
134 struct rb_root_cached *ranges, u32 nnodes, u64 *page_size);
135 int (*log_stop)(struct vfio_device *device);
136 int (*log_read_and_clear)(struct vfio_device *device,
137 unsigned long iova, unsigned long length,
138 struct iova_bitmap *dirty);
142 * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
143 * @flags: Arg from the device_feature op
144 * @argsz: Arg from the device_feature op
145 * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
147 * @minsz: Minimum data size the driver accepts
149 * For use in a driver's device_feature op. Checks that the inputs to the
150 * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
151 * the driver should execute the get or set, otherwise the relevant
152 * value should be returned.
154 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
157 if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
160 if (flags & VFIO_DEVICE_FEATURE_PROBE)
162 /* Without PROBE one of GET or SET must be requested */
163 if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
170 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
171 const struct vfio_device_ops *ops);
172 #define vfio_alloc_device(dev_struct, member, dev, ops) \
173 container_of(_vfio_alloc_device(sizeof(struct dev_struct) + \
174 BUILD_BUG_ON_ZERO(offsetof( \
175 struct dev_struct, member)), \
177 struct dev_struct, member)
179 int vfio_init_device(struct vfio_device *device, struct device *dev,
180 const struct vfio_device_ops *ops);
181 void vfio_free_device(struct vfio_device *device);
182 static inline void vfio_put_device(struct vfio_device *device)
184 put_device(&device->device);
187 int vfio_register_group_dev(struct vfio_device *device);
188 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
189 void vfio_unregister_group_dev(struct vfio_device *device);
191 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
192 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
194 int vfio_mig_get_next_state(struct vfio_device *device,
195 enum vfio_device_mig_state cur_fsm,
196 enum vfio_device_mig_state new_fsm,
197 enum vfio_device_mig_state *next_fsm);
202 struct iommu_group *vfio_file_iommu_group(struct file *file);
203 bool vfio_file_is_group(struct file *file);
204 bool vfio_file_enforced_coherent(struct file *file);
205 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
206 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
208 #define VFIO_PIN_PAGES_MAX_ENTRIES (PAGE_SIZE/sizeof(unsigned long))
210 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
211 int npage, int prot, struct page **pages);
212 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
213 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
214 void *data, size_t len, bool write);
219 struct vfio_info_cap {
220 struct vfio_info_cap_header *buf;
223 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
226 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
228 int vfio_info_add_capability(struct vfio_info_cap *caps,
229 struct vfio_info_cap_header *cap, size_t size);
231 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
232 int num_irqs, int max_irq_type,
236 #if IS_ENABLED(CONFIG_VFIO_SPAPR_EEH)
237 void vfio_spapr_pci_eeh_open(struct pci_dev *pdev);
238 void vfio_spapr_pci_eeh_release(struct pci_dev *pdev);
239 long vfio_spapr_iommu_eeh_ioctl(struct iommu_group *group, unsigned int cmd,
242 static inline void vfio_spapr_pci_eeh_open(struct pci_dev *pdev)
246 static inline void vfio_spapr_pci_eeh_release(struct pci_dev *pdev)
250 static inline long vfio_spapr_iommu_eeh_ioctl(struct iommu_group *group,
256 #endif /* CONFIG_VFIO_SPAPR_EEH */
263 struct eventfd_ctx *eventfd;
264 int (*handler)(void *, void *);
265 void (*thread)(void *, void *);
267 struct work_struct inject;
268 wait_queue_entry_t wait;
270 struct work_struct shutdown;
271 struct virqfd **pvirqfd;
274 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
275 void (*thread)(void *, void *), void *data,
276 struct virqfd **pvirqfd, int fd);
277 void vfio_virqfd_disable(struct virqfd **pvirqfd);