linux-headers/linux/vfio.h

   1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
   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  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12 #ifndef VFIO_H
  13 #define VFIO_H
  14
  15 #include <linux/types.h>
  16 #include <linux/ioctl.h>
  17
  18 #define VFIO_API_VERSION        0
  19
  20
  21 /* Kernel & User level defines for VFIO IOCTLs. */
  22
  23 /* Extensions */
  24
  25 #define VFIO_TYPE1_IOMMU                1
  26 #define VFIO_SPAPR_TCE_IOMMU            2
  27 #define VFIO_TYPE1v2_IOMMU              3
  28 /*
  29  * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping).  This
  30  * capability is subject to change as groups are added or removed.
  31  */
  32 #define VFIO_DMA_CC_IOMMU               4
  33
  34 /* Check if EEH is supported */
  35 #define VFIO_EEH                        5
  36
  37 /* Two-stage IOMMU */
  38 #define VFIO_TYPE1_NESTING_IOMMU        6       /* Implies v2 */
  39
  40 #define VFIO_SPAPR_TCE_v2_IOMMU         7
  41
  42 /*
  43  * The No-IOMMU IOMMU offers no translation or isolation for devices and
  44  * supports no ioctls outside of VFIO_CHECK_EXTENSION.  Use of VFIO's No-IOMMU
  45  * code will taint the host kernel and should be used with extreme caution.
  46  */
  47 #define VFIO_NOIOMMU_IOMMU              8
  48
  49 /*
  50  * The IOCTL interface is designed for extensibility by embedding the
  51  * structure length (argsz) and flags into structures passed between
  52  * kernel and userspace.  We therefore use the _IO() macro for these
  53  * defines to avoid implicitly embedding a size into the ioctl request.
  54  * As structure fields are added, argsz will increase to match and flag
  55  * bits will be defined to indicate additional fields with valid data.
  56  * It's *always* the caller's responsibility to indicate the size of
  57  * the structure passed by setting argsz appropriately.
  58  */
  59
  60 #define VFIO_TYPE       (';')
  61 #define VFIO_BASE       100
  62
  63 /*
  64  * For extension of INFO ioctls, VFIO makes use of a capability chain
  65  * designed after PCI/e capabilities.  A flag bit indicates whether
  66  * this capability chain is supported and a field defined in the fixed
  67  * structure defines the offset of the first capability in the chain.
  68  * This field is only valid when the corresponding bit in the flags
  69  * bitmap is set.  This offset field is relative to the start of the
  70  * INFO buffer, as is the next field within each capability header.
  71  * The id within the header is a shared address space per INFO ioctl,
  72  * while the version field is specific to the capability id.  The
  73  * contents following the header are specific to the capability id.
  74  */
  75 struct vfio_info_cap_header {
  76         __u16   id;             /* Identifies capability */
  77         __u16   version;        /* Version specific to the capability ID */
  78         __u32   next;           /* Offset of next capability */
  79 };
  80
  81 /*
  82  * Callers of INFO ioctls passing insufficiently sized buffers will see
  83  * the capability chain flag bit set, a zero value for the first capability
  84  * offset (if available within the provided argsz), and argsz will be
  85  * updated to report the necessary buffer size.  For compatibility, the
  86  * INFO ioctl will not report error in this case, but the capability chain
  87  * will not be available.
  88  */
  89
  90 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
  91
  92 /**
  93  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
  94  *
  95  * Report the version of the VFIO API.  This allows us to bump the entire
  96  * API version should we later need to add or change features in incompatible
  97  * ways.
  98  * Return: VFIO_API_VERSION
  99  * Availability: Always
 100  */
 101 #define VFIO_GET_API_VERSION            _IO(VFIO_TYPE, VFIO_BASE + 0)
 102
 103 /**
 104  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
 105  *
 106  * Check whether an extension is supported.
 107  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
 108  * Availability: Always
 109  */
 110 #define VFIO_CHECK_EXTENSION            _IO(VFIO_TYPE, VFIO_BASE + 1)
 111
 112 /**
 113  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
 114  *
 115  * Set the iommu to the given type.  The type must be supported by an
 116  * iommu driver as verified by calling CHECK_EXTENSION using the same
 117  * type.  A group must be set to this file descriptor before this
 118  * ioctl is available.  The IOMMU interfaces enabled by this call are
 119  * specific to the value set.
 120  * Return: 0 on success, -errno on failure
 121  * Availability: When VFIO group attached
 122  */
 123 #define VFIO_SET_IOMMU                  _IO(VFIO_TYPE, VFIO_BASE + 2)
 124
 125 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
 126
 127 /**
 128  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
 129  *                                              struct vfio_group_status)
 130  *
 131  * Retrieve information about the group.  Fills in provided
 132  * struct vfio_group_info.  Caller sets argsz.
 133  * Return: 0 on succes, -errno on failure.
 134  * Availability: Always
 135  */
 136 struct vfio_group_status {
 137         __u32   argsz;
 138         __u32   flags;
 139 #define VFIO_GROUP_FLAGS_VIABLE         (1 << 0)
 140 #define VFIO_GROUP_FLAGS_CONTAINER_SET  (1 << 1)
 141 };
 142 #define VFIO_GROUP_GET_STATUS           _IO(VFIO_TYPE, VFIO_BASE + 3)
 143
 144 /**
 145  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
 146  *
 147  * Set the container for the VFIO group to the open VFIO file
 148  * descriptor provided.  Groups may only belong to a single
 149  * container.  Containers may, at their discretion, support multiple
 150  * groups.  Only when a container is set are all of the interfaces
 151  * of the VFIO file descriptor and the VFIO group file descriptor
 152  * available to the user.
 153  * Return: 0 on success, -errno on failure.
 154  * Availability: Always
 155  */
 156 #define VFIO_GROUP_SET_CONTAINER        _IO(VFIO_TYPE, VFIO_BASE + 4)
 157
 158 /**
 159  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
 160  *
 161  * Remove the group from the attached container.  This is the
 162  * opposite of the SET_CONTAINER call and returns the group to
 163  * an initial state.  All device file descriptors must be released
 164  * prior to calling this interface.  When removing the last group
 165  * from a container, the IOMMU will be disabled and all state lost,
 166  * effectively also returning the VFIO file descriptor to an initial
 167  * state.
 168  * Return: 0 on success, -errno on failure.
 169  * Availability: When attached to container
 170  */
 171 #define VFIO_GROUP_UNSET_CONTAINER      _IO(VFIO_TYPE, VFIO_BASE + 5)
 172
 173 /**
 174  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
 175  *
 176  * Return a new file descriptor for the device object described by
 177  * the provided string.  The string should match a device listed in
 178  * the devices subdirectory of the IOMMU group sysfs entry.  The
 179  * group containing the device must already be added to this context.
 180  * Return: new file descriptor on success, -errno on failure.
 181  * Availability: When attached to container
 182  */
 183 #define VFIO_GROUP_GET_DEVICE_FD        _IO(VFIO_TYPE, VFIO_BASE + 6)
 184
 185 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
 186
 187 /**
 188  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
 189  *                                              struct vfio_device_info)
 190  *
 191  * Retrieve information about the device.  Fills in provided
 192  * struct vfio_device_info.  Caller sets argsz.
 193  * Return: 0 on success, -errno on failure.
 194  */
 195 struct vfio_device_info {
 196         __u32   argsz;
 197         __u32   flags;
 198 #define VFIO_DEVICE_FLAGS_RESET (1 << 0)        /* Device supports reset */
 199 #define VFIO_DEVICE_FLAGS_PCI   (1 << 1)        /* vfio-pci device */
 200 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2)     /* vfio-platform device */
 201 #define VFIO_DEVICE_FLAGS_AMBA  (1 << 3)        /* vfio-amba device */
 202 #define VFIO_DEVICE_FLAGS_CCW   (1 << 4)        /* vfio-ccw device */
 203         __u32   num_regions;    /* Max region index + 1 */
 204         __u32   num_irqs;       /* Max IRQ index + 1 */
 205 };
 206 #define VFIO_DEVICE_GET_INFO            _IO(VFIO_TYPE, VFIO_BASE + 7)
 207
 208 /*
 209  * Vendor driver using Mediated device framework should provide device_api
 210  * attribute in supported type attribute groups. Device API string should be one
 211  * of the following corresponding to device flags in vfio_device_info structure.
 212  */
 213
 214 #define VFIO_DEVICE_API_PCI_STRING              "vfio-pci"
 215 #define VFIO_DEVICE_API_PLATFORM_STRING         "vfio-platform"
 216 #define VFIO_DEVICE_API_AMBA_STRING             "vfio-amba"
 217 #define VFIO_DEVICE_API_CCW_STRING              "vfio-ccw"
 218
 219 /**
 220  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
 221  *                                     struct vfio_region_info)
 222  *
 223  * Retrieve information about a device region.  Caller provides
 224  * struct vfio_region_info with index value set.  Caller sets argsz.
 225  * Implementation of region mapping is bus driver specific.  This is
 226  * intended to describe MMIO, I/O port, as well as bus specific
 227  * regions (ex. PCI config space).  Zero sized regions may be used
 228  * to describe unimplemented regions (ex. unimplemented PCI BARs).
 229  * Return: 0 on success, -errno on failure.
 230  */
 231 struct vfio_region_info {
 232         __u32   argsz;
 233         __u32   flags;
 234 #define VFIO_REGION_INFO_FLAG_READ      (1 << 0) /* Region supports read */
 235 #define VFIO_REGION_INFO_FLAG_WRITE     (1 << 1) /* Region supports write */
 236 #define VFIO_REGION_INFO_FLAG_MMAP      (1 << 2) /* Region supports mmap */
 237 #define VFIO_REGION_INFO_FLAG_CAPS      (1 << 3) /* Info supports caps */
 238         __u32   index;          /* Region index */
 239         __u32   cap_offset;     /* Offset within info struct of first cap */
 240         __u64   size;           /* Region size (bytes) */
 241         __u64   offset;         /* Region offset from start of device fd */
 242 };
 243 #define VFIO_DEVICE_GET_REGION_INFO     _IO(VFIO_TYPE, VFIO_BASE + 8)
 244
 245 /*
 246  * The sparse mmap capability allows finer granularity of specifying areas
 247  * within a region with mmap support.  When specified, the user should only
 248  * mmap the offset ranges specified by the areas array.  mmaps outside of the
 249  * areas specified may fail (such as the range covering a PCI MSI-X table) or
 250  * may result in improper device behavior.
 251  *
 252  * The structures below define version 1 of this capability.
 253  */
 254 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP        1
 255
 256 struct vfio_region_sparse_mmap_area {
 257         __u64   offset; /* Offset of mmap'able area within region */
 258         __u64   size;   /* Size of mmap'able area */
 259 };
 260
 261 struct vfio_region_info_cap_sparse_mmap {
 262         struct vfio_info_cap_header header;
 263         __u32   nr_areas;
 264         __u32   reserved;
 265         struct vfio_region_sparse_mmap_area areas[];
 266 };
 267
 268 /*
 269  * The device specific type capability allows regions unique to a specific
 270  * device or class of devices to be exposed.  This helps solve the problem for
 271  * vfio bus drivers of defining which region indexes correspond to which region
 272  * on the device, without needing to resort to static indexes, as done by
 273  * vfio-pci.  For instance, if we were to go back in time, we might remove
 274  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
 275  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
 276  * make a "VGA" device specific type to describe the VGA access space.  This
 277  * means that non-VGA devices wouldn't need to waste this index, and thus the
 278  * address space associated with it due to implementation of device file
 279  * descriptor offsets in vfio-pci.
 280  *
 281  * The current implementation is now part of the user ABI, so we can't use this
 282  * for VGA, but there are other upcoming use cases, such as opregions for Intel
 283  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
 284  * use this for future additions.
 285  *
 286  * The structure below defines version 1 of this capability.
 287  */
 288 #define VFIO_REGION_INFO_CAP_TYPE       2
 289
 290 struct vfio_region_info_cap_type {
 291         struct vfio_info_cap_header header;
 292         __u32 type;     /* global per bus driver */
 293         __u32 subtype;  /* type specific */
 294 };
 295
 296 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE        (1 << 31)
 297 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK        (0xffff)
 298
 299 /* 8086 Vendor sub-types */
 300 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION  (1)
 301 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG  (2)
 302 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG   (3)
 303
 304 /*
 305  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
 306  * which allows direct access to non-MSIX registers which happened to be within
 307  * the same system page.
 308  *
 309  * Even though the userspace gets direct access to the MSIX data, the existing
 310  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
 311  */
 312 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE      3
 313
 314 /**
 315  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
 316  *                                  struct vfio_irq_info)
 317  *
 318  * Retrieve information about a device IRQ.  Caller provides
 319  * struct vfio_irq_info with index value set.  Caller sets argsz.
 320  * Implementation of IRQ mapping is bus driver specific.  Indexes
 321  * using multiple IRQs are primarily intended to support MSI-like
 322  * interrupt blocks.  Zero count irq blocks may be used to describe
 323  * unimplemented interrupt types.
 324  *
 325  * The EVENTFD flag indicates the interrupt index supports eventfd based
 326  * signaling.
 327  *
 328  * The MASKABLE flags indicates the index supports MASK and UNMASK
 329  * actions described below.
 330  *
 331  * AUTOMASKED indicates that after signaling, the interrupt line is
 332  * automatically masked by VFIO and the user needs to unmask the line
 333  * to receive new interrupts.  This is primarily intended to distinguish
 334  * level triggered interrupts.
 335  *
 336  * The NORESIZE flag indicates that the interrupt lines within the index
 337  * are setup as a set and new subindexes cannot be enabled without first
 338  * disabling the entire index.  This is used for interrupts like PCI MSI
 339  * and MSI-X where the driver may only use a subset of the available
 340  * indexes, but VFIO needs to enable a specific number of vectors
 341  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
 342  * then add and unmask vectors, it's up to userspace to make the decision
 343  * whether to allocate the maximum supported number of vectors or tear
 344  * down setup and incrementally increase the vectors as each is enabled.
 345  */
 346 struct vfio_irq_info {
 347         __u32   argsz;
 348         __u32   flags;
 349 #define VFIO_IRQ_INFO_EVENTFD           (1 << 0)
 350 #define VFIO_IRQ_INFO_MASKABLE          (1 << 1)
 351 #define VFIO_IRQ_INFO_AUTOMASKED        (1 << 2)
 352 #define VFIO_IRQ_INFO_NORESIZE          (1 << 3)
 353         __u32   index;          /* IRQ index */
 354         __u32   count;          /* Number of IRQs within this index */
 355 };
 356 #define VFIO_DEVICE_GET_IRQ_INFO        _IO(VFIO_TYPE, VFIO_BASE + 9)
 357
 358 /**
 359  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
 360  *
 361  * Set signaling, masking, and unmasking of interrupts.  Caller provides
 362  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
 363  * the range of subindexes being specified.
 364  *
 365  * The DATA flags specify the type of data provided.  If DATA_NONE, the
 366  * operation performs the specified action immediately on the specified
 367  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
 368  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
 369  *
 370  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
 371  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
 372  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
 373  * data = {1,0,1}
 374  *
 375  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
 376  * A value of -1 can be used to either de-assign interrupts if already
 377  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
 378  * to be trigger for interrupts [0,0] and [0,2]:
 379  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
 380  * data = {fd1, -1, fd2}
 381  * If index [0,1] is previously set, two count = 1 ioctls calls would be
 382  * required to set [0,0] and [0,2] without changing [0,1].
 383  *
 384  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
 385  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
 386  * from userspace (ie. simulate hardware triggering).
 387  *
 388  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
 389  * enables the interrupt index for the device.  Individual subindex interrupts
 390  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
 391  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
 392  *
 393  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
 394  * ACTION_TRIGGER specifies kernel->user signaling.
 395  */
 396 struct vfio_irq_set {
 397         __u32   argsz;
 398         __u32   flags;
 399 #define VFIO_IRQ_SET_DATA_NONE          (1 << 0) /* Data not present */
 400 #define VFIO_IRQ_SET_DATA_BOOL          (1 << 1) /* Data is bool (u8) */
 401 #define VFIO_IRQ_SET_DATA_EVENTFD       (1 << 2) /* Data is eventfd (s32) */
 402 #define VFIO_IRQ_SET_ACTION_MASK        (1 << 3) /* Mask interrupt */
 403 #define VFIO_IRQ_SET_ACTION_UNMASK      (1 << 4) /* Unmask interrupt */
 404 #define VFIO_IRQ_SET_ACTION_TRIGGER     (1 << 5) /* Trigger interrupt */
 405         __u32   index;
 406         __u32   start;
 407         __u32   count;
 408         __u8    data[];
 409 };
 410 #define VFIO_DEVICE_SET_IRQS            _IO(VFIO_TYPE, VFIO_BASE + 10)
 411
 412 #define VFIO_IRQ_SET_DATA_TYPE_MASK     (VFIO_IRQ_SET_DATA_NONE | \
 413                                          VFIO_IRQ_SET_DATA_BOOL | \
 414                                          VFIO_IRQ_SET_DATA_EVENTFD)
 415 #define VFIO_IRQ_SET_ACTION_TYPE_MASK   (VFIO_IRQ_SET_ACTION_MASK | \
 416                                          VFIO_IRQ_SET_ACTION_UNMASK | \
 417                                          VFIO_IRQ_SET_ACTION_TRIGGER)
 418 /**
 419  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
 420  *
 421  * Reset a device.
 422  */
 423 #define VFIO_DEVICE_RESET               _IO(VFIO_TYPE, VFIO_BASE + 11)
 424
 425 /*
 426  * The VFIO-PCI bus driver makes use of the following fixed region and
 427  * IRQ index mapping.  Unimplemented regions return a size of zero.
 428  * Unimplemented IRQ types return a count of zero.
 429  */
 430
 431 enum {
 432         VFIO_PCI_BAR0_REGION_INDEX,
 433         VFIO_PCI_BAR1_REGION_INDEX,
 434         VFIO_PCI_BAR2_REGION_INDEX,
 435         VFIO_PCI_BAR3_REGION_INDEX,
 436         VFIO_PCI_BAR4_REGION_INDEX,
 437         VFIO_PCI_BAR5_REGION_INDEX,
 438         VFIO_PCI_ROM_REGION_INDEX,
 439         VFIO_PCI_CONFIG_REGION_INDEX,
 440         /*
 441          * Expose VGA regions defined for PCI base class 03, subclass 00.
 442          * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
 443          * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
 444          * range is found at it's identity mapped offset from the region
 445          * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
 446          * between described ranges are unimplemented.
 447          */
 448         VFIO_PCI_VGA_REGION_INDEX,
 449         VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
 450                                  /* device specific cap to define content. */
 451 };
 452
 453 enum {
 454         VFIO_PCI_INTX_IRQ_INDEX,
 455         VFIO_PCI_MSI_IRQ_INDEX,
 456         VFIO_PCI_MSIX_IRQ_INDEX,
 457         VFIO_PCI_ERR_IRQ_INDEX,
 458         VFIO_PCI_REQ_IRQ_INDEX,
 459         VFIO_PCI_NUM_IRQS
 460 };
 461
 462 /*
 463  * The vfio-ccw bus driver makes use of the following fixed region and
 464  * IRQ index mapping. Unimplemented regions return a size of zero.
 465  * Unimplemented IRQ types return a count of zero.
 466  */
 467
 468 enum {
 469         VFIO_CCW_CONFIG_REGION_INDEX,
 470         VFIO_CCW_NUM_REGIONS
 471 };
 472
 473 enum {
 474         VFIO_CCW_IO_IRQ_INDEX,
 475         VFIO_CCW_NUM_IRQS
 476 };
 477
 478 /**
 479  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
 480  *                                            struct vfio_pci_hot_reset_info)
 481  *
 482  * Return: 0 on success, -errno on failure:
 483  *      -enospc = insufficient buffer, -enodev = unsupported for device.
 484  */
 485 struct vfio_pci_dependent_device {
 486         __u32   group_id;
 487         __u16   segment;
 488         __u8    bus;
 489         __u8    devfn; /* Use PCI_SLOT/PCI_FUNC */
 490 };
 491
 492 struct vfio_pci_hot_reset_info {
 493         __u32   argsz;
 494         __u32   flags;
 495         __u32   count;
 496         struct vfio_pci_dependent_device        devices[];
 497 };
 498
 499 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO      _IO(VFIO_TYPE, VFIO_BASE + 12)
 500
 501 /**
 502  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
 503  *                                  struct vfio_pci_hot_reset)
 504  *
 505  * Return: 0 on success, -errno on failure.
 506  */
 507 struct vfio_pci_hot_reset {
 508         __u32   argsz;
 509         __u32   flags;
 510         __u32   count;
 511         __s32   group_fds[];
 512 };
 513
 514 #define VFIO_DEVICE_PCI_HOT_RESET       _IO(VFIO_TYPE, VFIO_BASE + 13)
 515
 516 /**
 517  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
 518  *                                    struct vfio_device_query_gfx_plane)
 519  *
 520  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
 521  *
 522  * flags supported:
 523  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
 524  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
 525  *   support for dma-buf.
 526  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
 527  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
 528  *   support for region.
 529  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
 530  *   with each call to query the plane info.
 531  * - Others are invalid and return -EINVAL.
 532  *
 533  * Note:
 534  * 1. Plane could be disabled by guest. In that case, success will be
 535  *    returned with zero-initialized drm_format, size, width and height
 536  *    fields.
 537  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
 538  *
 539  * Return: 0 on success, -errno on other failure.
 540  */
 541 struct vfio_device_gfx_plane_info {
 542         __u32 argsz;
 543         __u32 flags;
 544 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
 545 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
 546 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
 547         /* in */
 548         __u32 drm_plane_type;   /* type of plane: DRM_PLANE_TYPE_* */
 549         /* out */
 550         __u32 drm_format;       /* drm format of plane */
 551         __u64 drm_format_mod;   /* tiled mode */
 552         __u32 width;    /* width of plane */
 553         __u32 height;   /* height of plane */
 554         __u32 stride;   /* stride of plane */
 555         __u32 size;     /* size of plane in bytes, align on page*/
 556         __u32 x_pos;    /* horizontal position of cursor plane */
 557         __u32 y_pos;    /* vertical position of cursor plane*/
 558         __u32 x_hot;    /* horizontal position of cursor hotspot */
 559         __u32 y_hot;    /* vertical position of cursor hotspot */
 560         union {
 561                 __u32 region_index;     /* region index */
 562                 __u32 dmabuf_id;        /* dma-buf id */
 563         };
 564 };
 565
 566 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
 567
 568 /**
 569  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
 570  *
 571  * Return a new dma-buf file descriptor for an exposed guest framebuffer
 572  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
 573  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
 574  */
 575
 576 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
 577
 578 /**
 579  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
 580  *                              struct vfio_device_ioeventfd)
 581  *
 582  * Perform a write to the device at the specified device fd offset, with
 583  * the specified data and width when the provided eventfd is triggered.
 584  * vfio bus drivers may not support this for all regions, for all widths,
 585  * or at all.  vfio-pci currently only enables support for BAR regions,
 586  * excluding the MSI-X vector table.
 587  *
 588  * Return: 0 on success, -errno on failure.
 589  */
 590 struct vfio_device_ioeventfd {
 591         __u32   argsz;
 592         __u32   flags;
 593 #define VFIO_DEVICE_IOEVENTFD_8         (1 << 0) /* 1-byte write */
 594 #define VFIO_DEVICE_IOEVENTFD_16        (1 << 1) /* 2-byte write */
 595 #define VFIO_DEVICE_IOEVENTFD_32        (1 << 2) /* 4-byte write */
 596 #define VFIO_DEVICE_IOEVENTFD_64        (1 << 3) /* 8-byte write */
 597 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK (0xf)
 598         __u64   offset;                 /* device fd offset of write */
 599         __u64   data;                   /* data to be written */
 600         __s32   fd;                     /* -1 for de-assignment */
 601 };
 602
 603 #define VFIO_DEVICE_IOEVENTFD           _IO(VFIO_TYPE, VFIO_BASE + 16)
 604
 605 /* -------- API for Type1 VFIO IOMMU -------- */
 606
 607 /**
 608  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
 609  *
 610  * Retrieve information about the IOMMU object. Fills in provided
 611  * struct vfio_iommu_info. Caller sets argsz.
 612  *
 613  * XXX Should we do these by CHECK_EXTENSION too?
 614  */
 615 struct vfio_iommu_type1_info {
 616         __u32   argsz;
 617         __u32   flags;
 618 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)        /* supported page sizes info */
 619         __u64   iova_pgsizes;           /* Bitmap of supported page sizes */
 620 };
 621
 622 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
 623
 624 /**
 625  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
 626  *
 627  * Map process virtual addresses to IO virtual addresses using the
 628  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
 629  */
 630 struct vfio_iommu_type1_dma_map {
 631         __u32   argsz;
 632         __u32   flags;
 633 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)         /* readable from device */
 634 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)        /* writable from device */
 635         __u64   vaddr;                          /* Process virtual address */
 636         __u64   iova;                           /* IO virtual address */
 637         __u64   size;                           /* Size of mapping (bytes) */
 638 };
 639
 640 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
 641
 642 /**
 643  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
 644  *                                                      struct vfio_dma_unmap)
 645  *
 646  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
 647  * Caller sets argsz.  The actual unmapped size is returned in the size
 648  * field.  No guarantee is made to the user that arbitrary unmaps of iova
 649  * or size different from those used in the original mapping call will
 650  * succeed.
 651  */
 652 struct vfio_iommu_type1_dma_unmap {
 653         __u32   argsz;
 654         __u32   flags;
 655         __u64   iova;                           /* IO virtual address */
 656         __u64   size;                           /* Size of mapping (bytes) */
 657 };
 658
 659 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 660
 661 /*
 662  * IOCTLs to enable/disable IOMMU container usage.
 663  * No parameters are supported.
 664  */
 665 #define VFIO_IOMMU_ENABLE       _IO(VFIO_TYPE, VFIO_BASE + 15)
 666 #define VFIO_IOMMU_DISABLE      _IO(VFIO_TYPE, VFIO_BASE + 16)
 667
 668 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
 669
 670 /*
 671  * The SPAPR TCE DDW info struct provides the information about
 672  * the details of Dynamic DMA window capability.
 673  *
 674  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
 675  * @max_dynamic_windows_supported tells the maximum number of windows
 676  * which the platform can create.
 677  * @levels tells the maximum number of levels in multi-level IOMMU tables;
 678  * this allows splitting a table into smaller chunks which reduces
 679  * the amount of physically contiguous memory required for the table.
 680  */
 681 struct vfio_iommu_spapr_tce_ddw_info {
 682         __u64 pgsizes;                  /* Bitmap of supported page sizes */
 683         __u32 max_dynamic_windows_supported;
 684         __u32 levels;
 685 };
 686
 687 /*
 688  * The SPAPR TCE info struct provides the information about the PCI bus
 689  * address ranges available for DMA, these values are programmed into
 690  * the hardware so the guest has to know that information.
 691  *
 692  * The DMA 32 bit window start is an absolute PCI bus address.
 693  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
 694  * addresses too so the window works as a filter rather than an offset
 695  * for IOVA addresses.
 696  *
 697  * Flags supported:
 698  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
 699  *   (DDW) support is present. @ddw is only supported when DDW is present.
 700  */
 701 struct vfio_iommu_spapr_tce_info {
 702         __u32 argsz;
 703         __u32 flags;
 704 #define VFIO_IOMMU_SPAPR_INFO_DDW       (1 << 0)        /* DDW supported */
 705         __u32 dma32_window_start;       /* 32 bit window start (bytes) */
 706         __u32 dma32_window_size;        /* 32 bit window size (bytes) */
 707         struct vfio_iommu_spapr_tce_ddw_info ddw;
 708 };
 709
 710 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO   _IO(VFIO_TYPE, VFIO_BASE + 12)
 711
 712 /*
 713  * EEH PE operation struct provides ways to:
 714  * - enable/disable EEH functionality;
 715  * - unfreeze IO/DMA for frozen PE;
 716  * - read PE state;
 717  * - reset PE;
 718  * - configure PE;
 719  * - inject EEH error.
 720  */
 721 struct vfio_eeh_pe_err {
 722         __u32 type;
 723         __u32 func;
 724         __u64 addr;
 725         __u64 mask;
 726 };
 727
 728 struct vfio_eeh_pe_op {
 729         __u32 argsz;
 730         __u32 flags;
 731         __u32 op;
 732         union {
 733                 struct vfio_eeh_pe_err err;
 734         };
 735 };
 736
 737 #define VFIO_EEH_PE_DISABLE             0       /* Disable EEH functionality */
 738 #define VFIO_EEH_PE_ENABLE              1       /* Enable EEH functionality  */
 739 #define VFIO_EEH_PE_UNFREEZE_IO         2       /* Enable IO for frozen PE   */
 740 #define VFIO_EEH_PE_UNFREEZE_DMA        3       /* Enable DMA for frozen PE  */
 741 #define VFIO_EEH_PE_GET_STATE           4       /* PE state retrieval        */
 742 #define  VFIO_EEH_PE_STATE_NORMAL       0       /* PE in functional state    */
 743 #define  VFIO_EEH_PE_STATE_RESET        1       /* PE reset in progress      */
 744 #define  VFIO_EEH_PE_STATE_STOPPED      2       /* Stopped DMA and IO        */
 745 #define  VFIO_EEH_PE_STATE_STOPPED_DMA  4       /* Stopped DMA only          */
 746 #define  VFIO_EEH_PE_STATE_UNAVAIL      5       /* State unavailable         */
 747 #define VFIO_EEH_PE_RESET_DEACTIVATE    5       /* Deassert PE reset         */
 748 #define VFIO_EEH_PE_RESET_HOT           6       /* Assert hot reset          */
 749 #define VFIO_EEH_PE_RESET_FUNDAMENTAL   7       /* Assert fundamental reset  */
 750 #define VFIO_EEH_PE_CONFIGURE           8       /* PE configuration          */
 751 #define VFIO_EEH_PE_INJECT_ERR          9       /* Inject EEH error          */
 752
 753 #define VFIO_EEH_PE_OP                  _IO(VFIO_TYPE, VFIO_BASE + 21)
 754
 755 /**
 756  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
 757  *
 758  * Registers user space memory where DMA is allowed. It pins
 759  * user pages and does the locked memory accounting so
 760  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
 761  * get faster.
 762  */
 763 struct vfio_iommu_spapr_register_memory {
 764         __u32   argsz;
 765         __u32   flags;
 766         __u64   vaddr;                          /* Process virtual address */
 767         __u64   size;                           /* Size of mapping (bytes) */
 768 };
 769 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY        _IO(VFIO_TYPE, VFIO_BASE + 17)
 770
 771 /**
 772  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
 773  *
 774  * Unregisters user space memory registered with
 775  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
 776  * Uses vfio_iommu_spapr_register_memory for parameters.
 777  */
 778 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY      _IO(VFIO_TYPE, VFIO_BASE + 18)
 779
 780 /**
 781  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
 782  *
 783  * Creates an additional TCE table and programs it (sets a new DMA window)
 784  * to every IOMMU group in the container. It receives page shift, window
 785  * size and number of levels in the TCE table being created.
 786  *
 787  * It allocates and returns an offset on a PCI bus of the new DMA window.
 788  */
 789 struct vfio_iommu_spapr_tce_create {
 790         __u32 argsz;
 791         __u32 flags;
 792         /* in */
 793         __u32 page_shift;
 794         __u32 __resv1;
 795         __u64 window_size;
 796         __u32 levels;
 797         __u32 __resv2;
 798         /* out */
 799         __u64 start_addr;
 800 };
 801 #define VFIO_IOMMU_SPAPR_TCE_CREATE     _IO(VFIO_TYPE, VFIO_BASE + 19)
 802
 803 /**
 804  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
 805  *
 806  * Unprograms a TCE table from all groups in the container and destroys it.
 807  * It receives a PCI bus offset as a window id.
 808  */
 809 struct vfio_iommu_spapr_tce_remove {
 810         __u32 argsz;
 811         __u32 flags;
 812         /* in */
 813         __u64 start_addr;
 814 };
 815 #define VFIO_IOMMU_SPAPR_TCE_REMOVE     _IO(VFIO_TYPE, VFIO_BASE + 20)
 816
 817 /* ***************************************************************** */
 818
 819 #endif /* VFIO_H */