include/linux/vfio.h

   1 /*
   2  * VFIO API definition
   3  *
   4  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
   5  *     Author: Alex Williamson <alex.williamson@redhat.com>
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #ifndef VFIO_H
  12 #define VFIO_H
  13
  14 #include <linux/types.h>
  15 #include <linux/ioctl.h>
  16
  17 #define VFIO_API_VERSION        0
  18
  19 #ifdef __KERNEL__       /* Internal VFIO-core/bus driver API */
  20
  21 #include <linux/iommu.h>
  22 #include <linux/mm.h>
  23
  24 /**
  25  * struct vfio_device_ops - VFIO bus driver device callbacks
  26  *
  27  * @open: Called when userspace creates new file descriptor for device
  28  * @release: Called when userspace releases file descriptor for device
  29  * @read: Perform read(2) on device file descriptor
  30  * @write: Perform write(2) on device file descriptor
  31  * @ioctl: Perform ioctl(2) on device file descriptor, supporting VFIO_DEVICE_*
  32  *         operations documented below
  33  * @mmap: Perform mmap(2) on a region of the device file descriptor
  34  */
  35 struct vfio_device_ops {
  36         char    *name;
  37         int     (*open)(void *device_data);
  38         void    (*release)(void *device_data);
  39         ssize_t (*read)(void *device_data, char __user *buf,
  40                         size_t count, loff_t *ppos);
  41         ssize_t (*write)(void *device_data, const char __user *buf,
  42                          size_t count, loff_t *size);
  43         long    (*ioctl)(void *device_data, unsigned int cmd,
  44                          unsigned long arg);
  45         int     (*mmap)(void *device_data, struct vm_area_struct *vma);
  46 };
  47
  48 extern int vfio_add_group_dev(struct device *dev,
  49                               const struct vfio_device_ops *ops,
  50                               void *device_data);
  51
  52 extern void *vfio_del_group_dev(struct device *dev);
  53
  54 /**
  55  * struct vfio_iommu_driver_ops - VFIO IOMMU driver callbacks
  56  */
  57 struct vfio_iommu_driver_ops {
  58         char            *name;
  59         struct module   *owner;
  60         void            *(*open)(unsigned long arg);
  61         void            (*release)(void *iommu_data);
  62         ssize_t         (*read)(void *iommu_data, char __user *buf,
  63                                 size_t count, loff_t *ppos);
  64         ssize_t         (*write)(void *iommu_data, const char __user *buf,
  65                                  size_t count, loff_t *size);
  66         long            (*ioctl)(void *iommu_data, unsigned int cmd,
  67                                  unsigned long arg);
  68         int             (*mmap)(void *iommu_data, struct vm_area_struct *vma);
  69         int             (*attach_group)(void *iommu_data,
  70                                         struct iommu_group *group);
  71         void            (*detach_group)(void *iommu_data,
  72                                         struct iommu_group *group);
  73
  74 };
  75
  76 extern int vfio_register_iommu_driver(const struct vfio_iommu_driver_ops *ops);
  77
  78 extern void vfio_unregister_iommu_driver(
  79                                 const struct vfio_iommu_driver_ops *ops);
  80
  81 /**
  82  * offsetofend(TYPE, MEMBER)
  83  *
  84  * @TYPE: The type of the structure
  85  * @MEMBER: The member within the structure to get the end offset of
  86  *
  87  * Simple helper macro for dealing with variable sized structures passed
  88  * from user space.  This allows us to easily determine if the provided
  89  * structure is sized to include various fields.
  90  */
  91 #define offsetofend(TYPE, MEMBER) ({                            \
  92         TYPE tmp;                                               \
  93         offsetof(TYPE, MEMBER) + sizeof(tmp.MEMBER); })         \
  94
  95 #endif /* __KERNEL__ */
  96
  97 /* Kernel & User level defines for VFIO IOCTLs. */
  98
  99 /* Extensions */
 100
 101 #define VFIO_TYPE1_IOMMU                1
 102
 103 /*
 104  * The IOCTL interface is designed for extensibility by embedding the
 105  * structure length (argsz) and flags into structures passed between
 106  * kernel and userspace.  We therefore use the _IO() macro for these
 107  * defines to avoid implicitly embedding a size into the ioctl request.
 108  * As structure fields are added, argsz will increase to match and flag
 109  * bits will be defined to indicate additional fields with valid data.
 110  * It's *always* the caller's responsibility to indicate the size of
 111  * the structure passed by setting argsz appropriately.
 112  */
 113
 114 #define VFIO_TYPE       (';')
 115 #define VFIO_BASE       100
 116
 117 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
 118
 119 /**
 120  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
 121  *
 122  * Report the version of the VFIO API.  This allows us to bump the entire
 123  * API version should we later need to add or change features in incompatible
 124  * ways.
 125  * Return: VFIO_API_VERSION
 126  * Availability: Always
 127  */
 128 #define VFIO_GET_API_VERSION            _IO(VFIO_TYPE, VFIO_BASE + 0)
 129
 130 /**
 131  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
 132  *
 133  * Check whether an extension is supported.
 134  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
 135  * Availability: Always
 136  */
 137 #define VFIO_CHECK_EXTENSION            _IO(VFIO_TYPE, VFIO_BASE + 1)
 138
 139 /**
 140  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
 141  *
 142  * Set the iommu to the given type.  The type must be supported by an
 143  * iommu driver as verified by calling CHECK_EXTENSION using the same
 144  * type.  A group must be set to this file descriptor before this
 145  * ioctl is available.  The IOMMU interfaces enabled by this call are
 146  * specific to the value set.
 147  * Return: 0 on success, -errno on failure
 148  * Availability: When VFIO group attached
 149  */
 150 #define VFIO_SET_IOMMU                  _IO(VFIO_TYPE, VFIO_BASE + 2)
 151
 152 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
 153
 154 /**
 155  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
 156  *                                              struct vfio_group_status)
 157  *
 158  * Retrieve information about the group.  Fills in provided
 159  * struct vfio_group_info.  Caller sets argsz.
 160  * Return: 0 on succes, -errno on failure.
 161  * Availability: Always
 162  */
 163 struct vfio_group_status {
 164         __u32   argsz;
 165         __u32   flags;
 166 #define VFIO_GROUP_FLAGS_VIABLE         (1 << 0)
 167 #define VFIO_GROUP_FLAGS_CONTAINER_SET  (1 << 1)
 168 };
 169 #define VFIO_GROUP_GET_STATUS           _IO(VFIO_TYPE, VFIO_BASE + 3)
 170
 171 /**
 172  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
 173  *
 174  * Set the container for the VFIO group to the open VFIO file
 175  * descriptor provided.  Groups may only belong to a single
 176  * container.  Containers may, at their discretion, support multiple
 177  * groups.  Only when a container is set are all of the interfaces
 178  * of the VFIO file descriptor and the VFIO group file descriptor
 179  * available to the user.
 180  * Return: 0 on success, -errno on failure.
 181  * Availability: Always
 182  */
 183 #define VFIO_GROUP_SET_CONTAINER        _IO(VFIO_TYPE, VFIO_BASE + 4)
 184
 185 /**
 186  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
 187  *
 188  * Remove the group from the attached container.  This is the
 189  * opposite of the SET_CONTAINER call and returns the group to
 190  * an initial state.  All device file descriptors must be released
 191  * prior to calling this interface.  When removing the last group
 192  * from a container, the IOMMU will be disabled and all state lost,
 193  * effectively also returning the VFIO file descriptor to an initial
 194  * state.
 195  * Return: 0 on success, -errno on failure.
 196  * Availability: When attached to container
 197  */
 198 #define VFIO_GROUP_UNSET_CONTAINER      _IO(VFIO_TYPE, VFIO_BASE + 5)
 199
 200 /**
 201  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
 202  *
 203  * Return a new file descriptor for the device object described by
 204  * the provided string.  The string should match a device listed in
 205  * the devices subdirectory of the IOMMU group sysfs entry.  The
 206  * group containing the device must already be added to this context.
 207  * Return: new file descriptor on success, -errno on failure.
 208  * Availability: When attached to container
 209  */
 210 #define VFIO_GROUP_GET_DEVICE_FD        _IO(VFIO_TYPE, VFIO_BASE + 6)
 211
 212 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
 213
 214 /**
 215  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
 216  *                                              struct vfio_device_info)
 217  *
 218  * Retrieve information about the device.  Fills in provided
 219  * struct vfio_device_info.  Caller sets argsz.
 220  * Return: 0 on success, -errno on failure.
 221  */
 222 struct vfio_device_info {
 223         __u32   argsz;
 224         __u32   flags;
 225 #define VFIO_DEVICE_FLAGS_RESET (1 << 0)        /* Device supports reset */
 226 #define VFIO_DEVICE_FLAGS_PCI   (1 << 1)        /* vfio-pci device */
 227         __u32   num_regions;    /* Max region index + 1 */
 228         __u32   num_irqs;       /* Max IRQ index + 1 */
 229 };
 230 #define VFIO_DEVICE_GET_INFO            _IO(VFIO_TYPE, VFIO_BASE + 7)
 231
 232 /**
 233  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
 234  *                                     struct vfio_region_info)
 235  *
 236  * Retrieve information about a device region.  Caller provides
 237  * struct vfio_region_info with index value set.  Caller sets argsz.
 238  * Implementation of region mapping is bus driver specific.  This is
 239  * intended to describe MMIO, I/O port, as well as bus specific
 240  * regions (ex. PCI config space).  Zero sized regions may be used
 241  * to describe unimplemented regions (ex. unimplemented PCI BARs).
 242  * Return: 0 on success, -errno on failure.
 243  */
 244 struct vfio_region_info {
 245         __u32   argsz;
 246         __u32   flags;
 247 #define VFIO_REGION_INFO_FLAG_READ      (1 << 0) /* Region supports read */
 248 #define VFIO_REGION_INFO_FLAG_WRITE     (1 << 1) /* Region supports write */
 249 #define VFIO_REGION_INFO_FLAG_MMAP      (1 << 2) /* Region supports mmap */
 250         __u32   index;          /* Region index */
 251         __u32   resv;           /* Reserved for alignment */
 252         __u64   size;           /* Region size (bytes) */
 253         __u64   offset;         /* Region offset from start of device fd */
 254 };
 255 #define VFIO_DEVICE_GET_REGION_INFO     _IO(VFIO_TYPE, VFIO_BASE + 8)
 256
 257 /**
 258  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
 259  *                                  struct vfio_irq_info)
 260  *
 261  * Retrieve information about a device IRQ.  Caller provides
 262  * struct vfio_irq_info with index value set.  Caller sets argsz.
 263  * Implementation of IRQ mapping is bus driver specific.  Indexes
 264  * using multiple IRQs are primarily intended to support MSI-like
 265  * interrupt blocks.  Zero count irq blocks may be used to describe
 266  * unimplemented interrupt types.
 267  *
 268  * The EVENTFD flag indicates the interrupt index supports eventfd based
 269  * signaling.
 270  *
 271  * The MASKABLE flags indicates the index supports MASK and UNMASK
 272  * actions described below.
 273  *
 274  * AUTOMASKED indicates that after signaling, the interrupt line is
 275  * automatically masked by VFIO and the user needs to unmask the line
 276  * to receive new interrupts.  This is primarily intended to distinguish
 277  * level triggered interrupts.
 278  *
 279  * The NORESIZE flag indicates that the interrupt lines within the index
 280  * are setup as a set and new subindexes cannot be enabled without first
 281  * disabling the entire index.  This is used for interrupts like PCI MSI
 282  * and MSI-X where the driver may only use a subset of the available
 283  * indexes, but VFIO needs to enable a specific number of vectors
 284  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
 285  * then add and unmask vectors, it's up to userspace to make the decision
 286  * whether to allocate the maximum supported number of vectors or tear
 287  * down setup and incrementally increase the vectors as each is enabled.
 288  */
 289 struct vfio_irq_info {
 290         __u32   argsz;
 291         __u32   flags;
 292 #define VFIO_IRQ_INFO_EVENTFD           (1 << 0)
 293 #define VFIO_IRQ_INFO_MASKABLE          (1 << 1)
 294 #define VFIO_IRQ_INFO_AUTOMASKED        (1 << 2)
 295 #define VFIO_IRQ_INFO_NORESIZE          (1 << 3)
 296         __u32   index;          /* IRQ index */
 297         __u32   count;          /* Number of IRQs within this index */
 298 };
 299 #define VFIO_DEVICE_GET_IRQ_INFO        _IO(VFIO_TYPE, VFIO_BASE + 9)
 300
 301 /**
 302  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
 303  *
 304  * Set signaling, masking, and unmasking of interrupts.  Caller provides
 305  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
 306  * the range of subindexes being specified.
 307  *
 308  * The DATA flags specify the type of data provided.  If DATA_NONE, the
 309  * operation performs the specified action immediately on the specified
 310  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
 311  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
 312  *
 313  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
 314  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
 315  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
 316  * data = {1,0,1}
 317  *
 318  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
 319  * A value of -1 can be used to either de-assign interrupts if already
 320  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
 321  * to be trigger for interrupts [0,0] and [0,2]:
 322  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
 323  * data = {fd1, -1, fd2}
 324  * If index [0,1] is previously set, two count = 1 ioctls calls would be
 325  * required to set [0,0] and [0,2] without changing [0,1].
 326  *
 327  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
 328  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
 329  * from userspace (ie. simulate hardware triggering).
 330  *
 331  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
 332  * enables the interrupt index for the device.  Individual subindex interrupts
 333  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
 334  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
 335  *
 336  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
 337  * ACTION_TRIGGER specifies kernel->user signaling.
 338  */
 339 struct vfio_irq_set {
 340         __u32   argsz;
 341         __u32   flags;
 342 #define VFIO_IRQ_SET_DATA_NONE          (1 << 0) /* Data not present */
 343 #define VFIO_IRQ_SET_DATA_BOOL          (1 << 1) /* Data is bool (u8) */
 344 #define VFIO_IRQ_SET_DATA_EVENTFD       (1 << 2) /* Data is eventfd (s32) */
 345 #define VFIO_IRQ_SET_ACTION_MASK        (1 << 3) /* Mask interrupt */
 346 #define VFIO_IRQ_SET_ACTION_UNMASK      (1 << 4) /* Unmask interrupt */
 347 #define VFIO_IRQ_SET_ACTION_TRIGGER     (1 << 5) /* Trigger interrupt */
 348         __u32   index;
 349         __u32   start;
 350         __u32   count;
 351         __u8    data[];
 352 };
 353 #define VFIO_DEVICE_SET_IRQS            _IO(VFIO_TYPE, VFIO_BASE + 10)
 354
 355 #define VFIO_IRQ_SET_DATA_TYPE_MASK     (VFIO_IRQ_SET_DATA_NONE | \
 356                                          VFIO_IRQ_SET_DATA_BOOL | \
 357                                          VFIO_IRQ_SET_DATA_EVENTFD)
 358 #define VFIO_IRQ_SET_ACTION_TYPE_MASK   (VFIO_IRQ_SET_ACTION_MASK | \
 359                                          VFIO_IRQ_SET_ACTION_UNMASK | \
 360                                          VFIO_IRQ_SET_ACTION_TRIGGER)
 361 /**
 362  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
 363  *
 364  * Reset a device.
 365  */
 366 #define VFIO_DEVICE_RESET               _IO(VFIO_TYPE, VFIO_BASE + 11)
 367
 368 /*
 369  * The VFIO-PCI bus driver makes use of the following fixed region and
 370  * IRQ index mapping.  Unimplemented regions return a size of zero.
 371  * Unimplemented IRQ types return a count of zero.
 372  */
 373
 374 enum {
 375         VFIO_PCI_BAR0_REGION_INDEX,
 376         VFIO_PCI_BAR1_REGION_INDEX,
 377         VFIO_PCI_BAR2_REGION_INDEX,
 378         VFIO_PCI_BAR3_REGION_INDEX,
 379         VFIO_PCI_BAR4_REGION_INDEX,
 380         VFIO_PCI_BAR5_REGION_INDEX,
 381         VFIO_PCI_ROM_REGION_INDEX,
 382         VFIO_PCI_CONFIG_REGION_INDEX,
 383         VFIO_PCI_NUM_REGIONS
 384 };
 385
 386 enum {
 387         VFIO_PCI_INTX_IRQ_INDEX,
 388         VFIO_PCI_MSI_IRQ_INDEX,
 389         VFIO_PCI_MSIX_IRQ_INDEX,
 390         VFIO_PCI_NUM_IRQS
 391 };
 392
 393 /* -------- API for Type1 VFIO IOMMU -------- */
 394
 395 /**
 396  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
 397  *
 398  * Retrieve information about the IOMMU object. Fills in provided
 399  * struct vfio_iommu_info. Caller sets argsz.
 400  *
 401  * XXX Should we do these by CHECK_EXTENSION too?
 402  */
 403 struct vfio_iommu_type1_info {
 404         __u32   argsz;
 405         __u32   flags;
 406 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)        /* supported page sizes info */
 407         __u64   iova_pgsizes;           /* Bitmap of supported page sizes */
 408 };
 409
 410 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
 411
 412 /**
 413  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
 414  *
 415  * Map process virtual addresses to IO virtual addresses using the
 416  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
 417  */
 418 struct vfio_iommu_type1_dma_map {
 419         __u32   argsz;
 420         __u32   flags;
 421 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)         /* readable from device */
 422 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)        /* writable from device */
 423         __u64   vaddr;                          /* Process virtual address */
 424         __u64   iova;                           /* IO virtual address */
 425         __u64   size;                           /* Size of mapping (bytes) */
 426 };
 427
 428 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
 429
 430 /**
 431  * VFIO_IOMMU_UNMAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 14, struct vfio_dma_unmap)
 432  *
 433  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
 434  * Caller sets argsz.
 435  */
 436 struct vfio_iommu_type1_dma_unmap {
 437         __u32   argsz;
 438         __u32   flags;
 439         __u64   iova;                           /* IO virtual address */
 440         __u64   size;                           /* Size of mapping (bytes) */
 441 };
 442
 443 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 444
 445 #endif /* VFIO_H */