include/linux/vfio.h

   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
  20 struct kvm;
  21 struct iommufd_ctx;
  22 struct iommufd_device;
  23 struct 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  */
  30 struct vfio_device_set {
  31         void *set_id;
  32         struct mutex lock;
  33         struct list_head device_list;
  34         unsigned int device_count;
  35 };
  36
  37 struct 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         struct inode *inode;
  68 #if IS_ENABLED(CONFIG_IOMMUFD)
  69         struct iommufd_device *iommufd_device;
  70         u8 iommufd_attached:1;
  71 #endif
  72         u8 cdev_opened:1;
  73 #ifdef CONFIG_DEBUG_FS
  74         /*
  75          * debug_root is a static property of the vfio_device
  76          * which must be set prior to registering the vfio_device.
  77          */
  78         struct dentry *debug_root;
  79 #endif
  80 };
  81
  82 /**
  83  * struct vfio_device_ops - VFIO bus driver device callbacks
  84  *
  85  * @name: Name of the device driver.
  86  * @init: initialize private fields in device structure
  87  * @release: Reclaim private fields in device structure
  88  * @bind_iommufd: Called when binding the device to an iommufd
  89  * @unbind_iommufd: Opposite of bind_iommufd
  90  * @attach_ioas: Called when attaching device to an IOAS/HWPT managed by the
  91  *               bound iommufd. Undo in unbind_iommufd if @detach_ioas is not
  92  *               called.
  93  * @detach_ioas: Opposite of attach_ioas
  94  * @open_device: Called when the first file descriptor is opened for this device
  95  * @close_device: Opposite of open_device
  96  * @read: Perform read(2) on device file descriptor
  97  * @write: Perform write(2) on device file descriptor
  98  * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
  99  *         operations documented below
 100  * @mmap: Perform mmap(2) on a region of the device file descriptor
 101  * @request: Request for the bus driver to release the device
 102  * @match: Optional device name match callback (return: 0 for no-match, >0 for
 103  *         match, -errno for abort (ex. match with insufficient or incorrect
 104  *         additional args)
 105  * @dma_unmap: Called when userspace unmaps IOVA from the container
 106  *             this device is attached to.
 107  * @device_feature: Optional, fill in the VFIO_DEVICE_FEATURE ioctl
 108  */
 109 struct vfio_device_ops {
 110         char    *name;
 111         int     (*init)(struct vfio_device *vdev);
 112         void    (*release)(struct vfio_device *vdev);
 113         int     (*bind_iommufd)(struct vfio_device *vdev,
 114                                 struct iommufd_ctx *ictx, u32 *out_device_id);
 115         void    (*unbind_iommufd)(struct vfio_device *vdev);
 116         int     (*attach_ioas)(struct vfio_device *vdev, u32 *pt_id);
 117         void    (*detach_ioas)(struct vfio_device *vdev);
 118         int     (*open_device)(struct vfio_device *vdev);
 119         void    (*close_device)(struct vfio_device *vdev);
 120         ssize_t (*read)(struct vfio_device *vdev, char __user *buf,
 121                         size_t count, loff_t *ppos);
 122         ssize_t (*write)(struct vfio_device *vdev, const char __user *buf,
 123                          size_t count, loff_t *size);
 124         long    (*ioctl)(struct vfio_device *vdev, unsigned int cmd,
 125                          unsigned long arg);
 126         int     (*mmap)(struct vfio_device *vdev, struct vm_area_struct *vma);
 127         void    (*request)(struct vfio_device *vdev, unsigned int count);
 128         int     (*match)(struct vfio_device *vdev, char *buf);
 129         void    (*dma_unmap)(struct vfio_device *vdev, u64 iova, u64 length);
 130         int     (*device_feature)(struct vfio_device *device, u32 flags,
 131                                   void __user *arg, size_t argsz);
 132 };
 133
 134 #if IS_ENABLED(CONFIG_IOMMUFD)
 135 struct iommufd_ctx *vfio_iommufd_device_ictx(struct vfio_device *vdev);
 136 int vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx);
 137 int vfio_iommufd_physical_bind(struct vfio_device *vdev,
 138                                struct iommufd_ctx *ictx, u32 *out_device_id);
 139 void vfio_iommufd_physical_unbind(struct vfio_device *vdev);
 140 int vfio_iommufd_physical_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
 141 void vfio_iommufd_physical_detach_ioas(struct vfio_device *vdev);
 142 int vfio_iommufd_emulated_bind(struct vfio_device *vdev,
 143                                struct iommufd_ctx *ictx, u32 *out_device_id);
 144 void vfio_iommufd_emulated_unbind(struct vfio_device *vdev);
 145 int vfio_iommufd_emulated_attach_ioas(struct vfio_device *vdev, u32 *pt_id);
 146 void vfio_iommufd_emulated_detach_ioas(struct vfio_device *vdev);
 147 #else
 148 static inline struct iommufd_ctx *
 149 vfio_iommufd_device_ictx(struct vfio_device *vdev)
 150 {
 151         return NULL;
 152 }
 153
 154 static inline int
 155 vfio_iommufd_get_dev_id(struct vfio_device *vdev, struct iommufd_ctx *ictx)
 156 {
 157         return VFIO_PCI_DEVID_NOT_OWNED;
 158 }
 159
 160 #define vfio_iommufd_physical_bind                                      \
 161         ((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx,   \
 162                   u32 *out_device_id)) NULL)
 163 #define vfio_iommufd_physical_unbind \
 164         ((void (*)(struct vfio_device *vdev)) NULL)
 165 #define vfio_iommufd_physical_attach_ioas \
 166         ((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
 167 #define vfio_iommufd_physical_detach_ioas \
 168         ((void (*)(struct vfio_device *vdev)) NULL)
 169 #define vfio_iommufd_emulated_bind                                      \
 170         ((int (*)(struct vfio_device *vdev, struct iommufd_ctx *ictx,   \
 171                   u32 *out_device_id)) NULL)
 172 #define vfio_iommufd_emulated_unbind \
 173         ((void (*)(struct vfio_device *vdev)) NULL)
 174 #define vfio_iommufd_emulated_attach_ioas \
 175         ((int (*)(struct vfio_device *vdev, u32 *pt_id)) NULL)
 176 #define vfio_iommufd_emulated_detach_ioas \
 177         ((void (*)(struct vfio_device *vdev)) NULL)
 178 #endif
 179
 180 static inline bool vfio_device_cdev_opened(struct vfio_device *device)
 181 {
 182         return device->cdev_opened;
 183 }
 184
 185 /**
 186  * struct vfio_migration_ops - VFIO bus device driver migration callbacks
 187  *
 188  * @migration_set_state: Optional callback to change the migration state for
 189  *         devices that support migration. It's mandatory for
 190  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
 191  *         The returned FD is used for data transfer according to the FSM
 192  *         definition. The driver is responsible to ensure that FD reaches end
 193  *         of stream or error whenever the migration FSM leaves a data transfer
 194  *         state or before close_device() returns.
 195  * @migration_get_state: Optional callback to get the migration state for
 196  *         devices that support migration. It's mandatory for
 197  *         VFIO_DEVICE_FEATURE_MIGRATION migration support.
 198  * @migration_get_data_size: Optional callback to get the estimated data
 199  *          length that will be required to complete stop copy. It's mandatory for
 200  *          VFIO_DEVICE_FEATURE_MIGRATION migration support.
 201  */
 202 struct vfio_migration_ops {
 203         struct file *(*migration_set_state)(
 204                 struct vfio_device *device,
 205                 enum vfio_device_mig_state new_state);
 206         int (*migration_get_state)(struct vfio_device *device,
 207                                    enum vfio_device_mig_state *curr_state);
 208         int (*migration_get_data_size)(struct vfio_device *device,
 209                                        unsigned long *stop_copy_length);
 210 };
 211
 212 /**
 213  * struct vfio_log_ops - VFIO bus device driver logging callbacks
 214  *
 215  * @log_start: Optional callback to ask the device start DMA logging.
 216  * @log_stop: Optional callback to ask the device stop DMA logging.
 217  * @log_read_and_clear: Optional callback to ask the device read
 218  *         and clear the dirty DMAs in some given range.
 219  *
 220  * The vfio core implementation of the DEVICE_FEATURE_DMA_LOGGING_ set
 221  * of features does not track logging state relative to the device,
 222  * therefore the device implementation of vfio_log_ops must handle
 223  * arbitrary user requests. This includes rejecting subsequent calls
 224  * to log_start without an intervening log_stop, as well as graceful
 225  * handling of log_stop and log_read_and_clear from invalid states.
 226  */
 227 struct vfio_log_ops {
 228         int (*log_start)(struct vfio_device *device,
 229                 struct rb_root_cached *ranges, u32 nnodes, u64 *page_size);
 230         int (*log_stop)(struct vfio_device *device);
 231         int (*log_read_and_clear)(struct vfio_device *device,
 232                 unsigned long iova, unsigned long length,
 233                 struct iova_bitmap *dirty);
 234 };
 235
 236 /**
 237  * vfio_check_feature - Validate user input for the VFIO_DEVICE_FEATURE ioctl
 238  * @flags: Arg from the device_feature op
 239  * @argsz: Arg from the device_feature op
 240  * @supported_ops: Combination of VFIO_DEVICE_FEATURE_GET and SET the driver
 241  *                 supports
 242  * @minsz: Minimum data size the driver accepts
 243  *
 244  * For use in a driver's device_feature op. Checks that the inputs to the
 245  * VFIO_DEVICE_FEATURE ioctl are correct for the driver's feature. Returns 1 if
 246  * the driver should execute the get or set, otherwise the relevant
 247  * value should be returned.
 248  */
 249 static inline int vfio_check_feature(u32 flags, size_t argsz, u32 supported_ops,
 250                                     size_t minsz)
 251 {
 252         if ((flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)) &
 253             ~supported_ops)
 254                 return -EINVAL;
 255         if (flags & VFIO_DEVICE_FEATURE_PROBE)
 256                 return 0;
 257         /* Without PROBE one of GET or SET must be requested */
 258         if (!(flags & (VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_SET)))
 259                 return -EINVAL;
 260         if (argsz < minsz)
 261                 return -EINVAL;
 262         return 1;
 263 }
 264
 265 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
 266                                        const struct vfio_device_ops *ops);
 267 #define vfio_alloc_device(dev_struct, member, dev, ops)                         \
 268         container_of(_vfio_alloc_device(sizeof(struct dev_struct) +             \
 269                                         BUILD_BUG_ON_ZERO(offsetof(             \
 270                                                 struct dev_struct, member)),    \
 271                                         dev, ops),                              \
 272                      struct dev_struct, member)
 273
 274 static inline void vfio_put_device(struct vfio_device *device)
 275 {
 276         put_device(&device->device);
 277 }
 278
 279 int vfio_register_group_dev(struct vfio_device *device);
 280 int vfio_register_emulated_iommu_dev(struct vfio_device *device);
 281 void vfio_unregister_group_dev(struct vfio_device *device);
 282
 283 int vfio_assign_device_set(struct vfio_device *device, void *set_id);
 284 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set);
 285 struct vfio_device *
 286 vfio_find_device_in_devset(struct vfio_device_set *dev_set,
 287                            struct device *dev);
 288
 289 int vfio_mig_get_next_state(struct vfio_device *device,
 290                             enum vfio_device_mig_state cur_fsm,
 291                             enum vfio_device_mig_state new_fsm,
 292                             enum vfio_device_mig_state *next_fsm);
 293
 294 void vfio_combine_iova_ranges(struct rb_root_cached *root, u32 cur_nodes,
 295                               u32 req_nodes);
 296
 297 /*
 298  * External user API
 299  */
 300 struct iommu_group *vfio_file_iommu_group(struct file *file);
 301
 302 #if IS_ENABLED(CONFIG_VFIO_GROUP)
 303 bool vfio_file_is_group(struct file *file);
 304 bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 305 #else
 306 static inline bool vfio_file_is_group(struct file *file)
 307 {
 308         return false;
 309 }
 310
 311 static inline bool vfio_file_has_dev(struct file *file, struct vfio_device *device)
 312 {
 313         return false;
 314 }
 315 #endif
 316 bool vfio_file_is_valid(struct file *file);
 317 bool vfio_file_enforced_coherent(struct file *file);
 318 void vfio_file_set_kvm(struct file *file, struct kvm *kvm);
 319
 320 #define VFIO_PIN_PAGES_MAX_ENTRIES      (PAGE_SIZE/sizeof(unsigned long))
 321
 322 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 323                    int npage, int prot, struct page **pages);
 324 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
 325 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
 326                 void *data, size_t len, bool write);
 327
 328 /*
 329  * Sub-module helpers
 330  */
 331 struct vfio_info_cap {
 332         struct vfio_info_cap_header *buf;
 333         size_t size;
 334 };
 335 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
 336                                                size_t size, u16 id,
 337                                                u16 version);
 338 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset);
 339
 340 int vfio_info_add_capability(struct vfio_info_cap *caps,
 341                              struct vfio_info_cap_header *cap, size_t size);
 342
 343 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr,
 344                                        int num_irqs, int max_irq_type,
 345                                        size_t *data_size);
 346
 347 /*
 348  * IRQfd - generic
 349  */
 350 struct virqfd {
 351         void                    *opaque;
 352         struct eventfd_ctx      *eventfd;
 353         int                     (*handler)(void *, void *);
 354         void                    (*thread)(void *, void *);
 355         void                    *data;
 356         struct work_struct      inject;
 357         wait_queue_entry_t              wait;
 358         poll_table              pt;
 359         struct work_struct      shutdown;
 360         struct work_struct      flush_inject;
 361         struct virqfd           **pvirqfd;
 362 };
 363
 364 int vfio_virqfd_enable(void *opaque, int (*handler)(void *, void *),
 365                        void (*thread)(void *, void *), void *data,
 366                        struct virqfd **pvirqfd, int fd);
 367 void vfio_virqfd_disable(struct virqfd **pvirqfd);
 368 void vfio_virqfd_flush_thread(struct virqfd **pvirqfd);
 369
 370 #endif /* VFIO_H */