xref: /linux-6.15/include/linux/vfio.h (revision 39dfd52d)
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 
18 struct kvm;
19 
20 /*
21  * VFIO devices can be placed in a set, this allows all devices to share this
22  * structure and the VFIO core will provide a lock that is held around
23  * open_device()/close_device() for all devices in the set.
24  */
25 struct vfio_device_set {
26 	void *set_id;
27 	struct mutex lock;
28 	struct list_head device_list;
29 	unsigned int device_count;
30 };
31 
32 struct vfio_device {
33 	struct device *dev;
34 	const struct vfio_device_ops *ops;
35 	/*
36 	 * mig_ops is a static property of the vfio_device which must be set
37 	 * prior to registering the vfio_device.
38 	 */
39 	const struct vfio_migration_ops *mig_ops;
40 	struct vfio_group *group;
41 	struct vfio_device_set *dev_set;
42 	struct list_head dev_set_list;
43 	unsigned int migration_flags;
44 	/* Driver must reference the kvm during open_device or never touch it */
45 	struct kvm *kvm;
46 
47 	/* Members below here are private, not for driver use */
48 	refcount_t refcount;
49 	unsigned int open_count;
50 	struct completion comp;
51 	struct list_head group_next;
52 	struct list_head iommu_entry;
53 };
54 
55 /**
56  * struct vfio_device_ops - VFIO bus driver device callbacks
57  *
58  * @open_device: Called when the first file descriptor is opened for this device
59  * @close_device: Opposite of open_device
60  * @read: Perform read(2) on device file descriptor
61  * @write: Perform write(2) on device file descriptor
62  * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
63  *         operations documented below
64  * @mmap: Perform mmap(2) on a region of the device file descriptor
65  * @request: Request for the bus driver to release the device
66  * @match: Optional device name match callback (return: 0 for no-match, >0 for
67  *         match, -errno for abort (ex. match with insufficient or incorrect
68  *         additional args)
69  * @dma_unmap: Called when userspace unmaps IOVA from the container
70  *             this device is attached to.
71  * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
72  */
73 struct vfio_device_ops {
74 	char	*name;
75 	int	(*open_device)(struct vfio_device *vdev);
76 	void	(*close_device)(struct vfio_device *vdev);
77 	ssize_t	(*read)(struct vfio_device *vdev, char __user *buf,
78 			size_t count, loff_t *ppos);
79 	ssize_t	(*write)(struct vfio_device *vdev, const char __user *buf,
80 			 size_t count, loff_t *size);
81 	long	(*ioctl)(struct vfio_device *vdev, unsigned int cmd,
82 			 unsigned long arg);
83 	int	(*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma);
84 	void	(*request)(struct vfio_device *vdev, unsigned int count);
85 	int	(*match)(struct vfio_device *vdev, char *buf);
86 	void	(*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
87 	int	(*device_feature)(struct vfio_device *device, u32 flags,
88 				  void __user *arg, size_t argsz);
89 };
90 
91 /**
92  * @migration_set_state: Optional callback to change the migration state for
93  *         devices that support migration. It's mandatory for
94  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
95  *         The returned FD is used for data transfer according to the FSM
96  *         definition. The driver is responsible to ensure that FD reaches end
97  *         of stream or error whenever the migration FSM leaves a data transfer
98  *         state or before close_device() returns.
99  * @migration_get_state: Optional callback to get the migration state for
100  *         devices that support migration. It's mandatory for
101  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
102  */
103 struct vfio_migration_ops {
104 	struct file *(*migration_set_state)(
105 		struct vfio_device *device,
106 		enum vfio_device_mig_state new_state);
107 	int (*migration_get_state)(struct vfio_device *device,
108 				   enum vfio_device_mig_state *curr_state);
109 };
110 
111 /**
112  * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
113  * @flags: Arg from the device_feature op
114  * @argsz: Arg from the device_feature op
115  * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
116  *                 supports
117  * @minsz: Minimum data size the driver accepts
118  *
119  * For use in a driver's device_feature op. Checks that the inputs to the
120  * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
121  * the driver should execute the get or set, otherwise the relevant
122  * value should be returned.
123  */
124 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
125 				    size_t minsz)
126 {
127 	if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
128 	    ~supported_ops)
129 		return -EINVAL;
130 	if (flags & VFIO_DEVICE_FEATURE_PROBE)
131 		return 0;
132 	/* Without PROBE one of GET or SET must be requested */
133 	if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
134 		return -EINVAL;
135 	if (argsz < minsz)
136 		return -EINVAL;
137 	return 1;
138 }
139 
140 void vfio_init_group_dev(struct vfio_device *device, struct device *dev,
141 			 const struct vfio_device_ops *ops);
142 void vfio_uninit_group_dev(struct vfio_device *device);
143 int vfio_register_group_dev(struct vfio_device *device);
144 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
145 void vfio_unregister_group_dev(struct vfio_device *device);
146 
147 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
148 
149 int vfio_mig_get_next_state(struct vfio_device *device,
150 			    enum vfio_device_mig_state cur_fsm,
151 			    enum vfio_device_mig_state new_fsm,
152 			    enum vfio_device_mig_state *next_fsm);
153 
154 /*
155  * External user API
156  */
157 struct iommu_group *vfio_file_iommu_group(struct file *file);
158 bool vfio_file_enforced_coherent(struct file *file);
159 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
160 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
161 
162 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
163 
164 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
165 		   int npage, int prot, struct page **pages);
166 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
167 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
168 		void *data, size_t len, bool write);
169 
170 /*
171  * Sub-module helpers
172  */
173 struct vfio_info_cap {
174 	struct vfio_info_cap_header *buf;
175 	size_t size;
176 };
177 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
178 					       size_t size, u16 id,
179 					       u16 version);
180 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
181 
182 int vfio_info_add_capability(struct vfio_info_cap *caps,
183 			     struct vfio_info_cap_header *cap, size_t size);
184 
185 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
186 				       int num_irqs, int max_irq_type,
187 				       size_t *data_size);
188 
189 struct pci_dev;
190 #if IS_ENABLED(CONFIG_VFIO_SPAPR_EEH)
191 void vfio_spapr_pci_eeh_open(struct pci_dev *pdev);
192 void vfio_spapr_pci_eeh_release(struct pci_dev *pdev);
193 long vfio_spapr_iommu_eeh_ioctl(struct iommu_group *group, unsigned int cmd,
194 				unsigned long arg);
195 #else
196 static inline void vfio_spapr_pci_eeh_open(struct pci_dev *pdev)
197 {
198 }
199 
200 static inline void vfio_spapr_pci_eeh_release(struct pci_dev *pdev)
201 {
202 }
203 
204 static inline long vfio_spapr_iommu_eeh_ioctl(struct iommu_group *group,
205 					      unsigned int cmd,
206 					      unsigned long arg)
207 {
208 	return -ENOTTY;
209 }
210 #endif /* CONFIG_VFIO_SPAPR_EEH */
211 
212 /*
213  * IRQfd - generic
214  */
215 struct virqfd {
216 	void			*opaque;
217 	struct eventfd_ctx	*eventfd;
218 	int			(*handler)(void *, void *);
219 	void			(*thread)(void *, void *);
220 	void			*data;
221 	struct work_struct	inject;
222 	wait_queue_entry_t		wait;
223 	poll_table		pt;
224 	struct work_struct	shutdown;
225 	struct virqfd		**pvirqfd;
226 };
227 
228 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
229 		       void (*thread)(void *, void *), void *data,
230 		       struct virqfd **pvirqfd, int fd);
231 void vfio_virqfd_disable(struct virqfd **pvirqfd);
232 
233 #endif /* VFIO_H */
234