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