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 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
  63
  64 /**
  65  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
  66  *
  67  * Report the version of the VFIO API.  This allows us to bump the entire
  68  * API version should we later need to add or change features in incompatible
  69  * ways.
  70  * Return: VFIO_API_VERSION
  71  * Availability: Always
  72  */
  73 #define VFIO_GET_API_VERSION            _IO(VFIO_TYPE, VFIO_BASE + 0)
  74
  75 /**
  76  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
  77  *
  78  * Check whether an extension is supported.
  79  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
  80  * Availability: Always
  81  */
  82 #define VFIO_CHECK_EXTENSION            _IO(VFIO_TYPE, VFIO_BASE + 1)
  83
  84 /**
  85  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
  86  *
  87  * Set the iommu to the given type.  The type must be supported by an
  88  * iommu driver as verified by calling CHECK_EXTENSION using the same
  89  * type.  A group must be set to this file descriptor before this
  90  * ioctl is available.  The IOMMU interfaces enabled by this call are
  91  * specific to the value set.
  92  * Return: 0 on success, -errno on failure
  93  * Availability: When VFIO group attached
  94  */
  95 #define VFIO_SET_IOMMU                  _IO(VFIO_TYPE, VFIO_BASE + 2)
  96
  97 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
  98
  99 /**
 100  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
 101  *                                              struct vfio_group_status)
 102  *
 103  * Retrieve information about the group.  Fills in provided
 104  * struct vfio_group_info.  Caller sets argsz.
 105  * Return: 0 on succes, -errno on failure.
 106  * Availability: Always
 107  */
 108 struct vfio_group_status {
 109         __u32   argsz;
 110         __u32   flags;
 111 #define VFIO_GROUP_FLAGS_VIABLE         (1 << 0)
 112 #define VFIO_GROUP_FLAGS_CONTAINER_SET  (1 << 1)
 113 };
 114 #define VFIO_GROUP_GET_STATUS           _IO(VFIO_TYPE, VFIO_BASE + 3)
 115
 116 /**
 117  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
 118  *
 119  * Set the container for the VFIO group to the open VFIO file
 120  * descriptor provided.  Groups may only belong to a single
 121  * container.  Containers may, at their discretion, support multiple
 122  * groups.  Only when a container is set are all of the interfaces
 123  * of the VFIO file descriptor and the VFIO group file descriptor
 124  * available to the user.
 125  * Return: 0 on success, -errno on failure.
 126  * Availability: Always
 127  */
 128 #define VFIO_GROUP_SET_CONTAINER        _IO(VFIO_TYPE, VFIO_BASE + 4)
 129
 130 /**
 131  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
 132  *
 133  * Remove the group from the attached container.  This is the
 134  * opposite of the SET_CONTAINER call and returns the group to
 135  * an initial state.  All device file descriptors must be released
 136  * prior to calling this interface.  When removing the last group
 137  * from a container, the IOMMU will be disabled and all state lost,
 138  * effectively also returning the VFIO file descriptor to an initial
 139  * state.
 140  * Return: 0 on success, -errno on failure.
 141  * Availability: When attached to container
 142  */
 143 #define VFIO_GROUP_UNSET_CONTAINER      _IO(VFIO_TYPE, VFIO_BASE + 5)
 144
 145 /**
 146  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
 147  *
 148  * Return a new file descriptor for the device object described by
 149  * the provided string.  The string should match a device listed in
 150  * the devices subdirectory of the IOMMU group sysfs entry.  The
 151  * group containing the device must already be added to this context.
 152  * Return: new file descriptor on success, -errno on failure.
 153  * Availability: When attached to container
 154  */
 155 #define VFIO_GROUP_GET_DEVICE_FD        _IO(VFIO_TYPE, VFIO_BASE + 6)
 156
 157 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
 158
 159 /**
 160  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
 161  *                                              struct vfio_device_info)
 162  *
 163  * Retrieve information about the device.  Fills in provided
 164  * struct vfio_device_info.  Caller sets argsz.
 165  * Return: 0 on success, -errno on failure.
 166  */
 167 struct vfio_device_info {
 168         __u32   argsz;
 169         __u32   flags;
 170 #define VFIO_DEVICE_FLAGS_RESET (1 << 0)        /* Device supports reset */
 171 #define VFIO_DEVICE_FLAGS_PCI   (1 << 1)        /* vfio-pci device */
 172 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2)     /* vfio-platform device */
 173 #define VFIO_DEVICE_FLAGS_AMBA  (1 << 3)        /* vfio-amba device */
 174         __u32   num_regions;    /* Max region index + 1 */
 175         __u32   num_irqs;       /* Max IRQ index + 1 */
 176 };
 177 #define VFIO_DEVICE_GET_INFO            _IO(VFIO_TYPE, VFIO_BASE + 7)
 178
 179 /**
 180  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
 181  *                                     struct vfio_region_info)
 182  *
 183  * Retrieve information about a device region.  Caller provides
 184  * struct vfio_region_info with index value set.  Caller sets argsz.
 185  * Implementation of region mapping is bus driver specific.  This is
 186  * intended to describe MMIO, I/O port, as well as bus specific
 187  * regions (ex. PCI config space).  Zero sized regions may be used
 188  * to describe unimplemented regions (ex. unimplemented PCI BARs).
 189  * Return: 0 on success, -errno on failure.
 190  */
 191 struct vfio_region_info {
 192         __u32   argsz;
 193         __u32   flags;
 194 #define VFIO_REGION_INFO_FLAG_READ      (1 << 0) /* Region supports read */
 195 #define VFIO_REGION_INFO_FLAG_WRITE     (1 << 1) /* Region supports write */
 196 #define VFIO_REGION_INFO_FLAG_MMAP      (1 << 2) /* Region supports mmap */
 197         __u32   index;          /* Region index */
 198         __u32   resv;           /* Reserved for alignment */
 199         __u64   size;           /* Region size (bytes) */
 200         __u64   offset;         /* Region offset from start of device fd */
 201 };
 202 #define VFIO_DEVICE_GET_REGION_INFO     _IO(VFIO_TYPE, VFIO_BASE + 8)
 203
 204 /**
 205  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
 206  *                                  struct vfio_irq_info)
 207  *
 208  * Retrieve information about a device IRQ.  Caller provides
 209  * struct vfio_irq_info with index value set.  Caller sets argsz.
 210  * Implementation of IRQ mapping is bus driver specific.  Indexes
 211  * using multiple IRQs are primarily intended to support MSI-like
 212  * interrupt blocks.  Zero count irq blocks may be used to describe
 213  * unimplemented interrupt types.
 214  *
 215  * The EVENTFD flag indicates the interrupt index supports eventfd based
 216  * signaling.
 217  *
 218  * The MASKABLE flags indicates the index supports MASK and UNMASK
 219  * actions described below.
 220  *
 221  * AUTOMASKED indicates that after signaling, the interrupt line is
 222  * automatically masked by VFIO and the user needs to unmask the line
 223  * to receive new interrupts.  This is primarily intended to distinguish
 224  * level triggered interrupts.
 225  *
 226  * The NORESIZE flag indicates that the interrupt lines within the index
 227  * are setup as a set and new subindexes cannot be enabled without first
 228  * disabling the entire index.  This is used for interrupts like PCI MSI
 229  * and MSI-X where the driver may only use a subset of the available
 230  * indexes, but VFIO needs to enable a specific number of vectors
 231  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
 232  * then add and unmask vectors, it's up to userspace to make the decision
 233  * whether to allocate the maximum supported number of vectors or tear
 234  * down setup and incrementally increase the vectors as each is enabled.
 235  */
 236 struct vfio_irq_info {
 237         __u32   argsz;
 238         __u32   flags;
 239 #define VFIO_IRQ_INFO_EVENTFD           (1 << 0)
 240 #define VFIO_IRQ_INFO_MASKABLE          (1 << 1)
 241 #define VFIO_IRQ_INFO_AUTOMASKED        (1 << 2)
 242 #define VFIO_IRQ_INFO_NORESIZE          (1 << 3)
 243         __u32   index;          /* IRQ index */
 244         __u32   count;          /* Number of IRQs within this index */
 245 };
 246 #define VFIO_DEVICE_GET_IRQ_INFO        _IO(VFIO_TYPE, VFIO_BASE + 9)
 247
 248 /**
 249  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
 250  *
 251  * Set signaling, masking, and unmasking of interrupts.  Caller provides
 252  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
 253  * the range of subindexes being specified.
 254  *
 255  * The DATA flags specify the type of data provided.  If DATA_NONE, the
 256  * operation performs the specified action immediately on the specified
 257  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
 258  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
 259  *
 260  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
 261  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
 262  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
 263  * data = {1,0,1}
 264  *
 265  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
 266  * A value of -1 can be used to either de-assign interrupts if already
 267  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
 268  * to be trigger for interrupts [0,0] and [0,2]:
 269  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
 270  * data = {fd1, -1, fd2}
 271  * If index [0,1] is previously set, two count = 1 ioctls calls would be
 272  * required to set [0,0] and [0,2] without changing [0,1].
 273  *
 274  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
 275  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
 276  * from userspace (ie. simulate hardware triggering).
 277  *
 278  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
 279  * enables the interrupt index for the device.  Individual subindex interrupts
 280  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
 281  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
 282  *
 283  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
 284  * ACTION_TRIGGER specifies kernel->user signaling.
 285  */
 286 struct vfio_irq_set {
 287         __u32   argsz;
 288         __u32   flags;
 289 #define VFIO_IRQ_SET_DATA_NONE          (1 << 0) /* Data not present */
 290 #define VFIO_IRQ_SET_DATA_BOOL          (1 << 1) /* Data is bool (u8) */
 291 #define VFIO_IRQ_SET_DATA_EVENTFD       (1 << 2) /* Data is eventfd (s32) */
 292 #define VFIO_IRQ_SET_ACTION_MASK        (1 << 3) /* Mask interrupt */
 293 #define VFIO_IRQ_SET_ACTION_UNMASK      (1 << 4) /* Unmask interrupt */
 294 #define VFIO_IRQ_SET_ACTION_TRIGGER     (1 << 5) /* Trigger interrupt */
 295         __u32   index;
 296         __u32   start;
 297         __u32   count;
 298         __u8    data[];
 299 };
 300 #define VFIO_DEVICE_SET_IRQS            _IO(VFIO_TYPE, VFIO_BASE + 10)
 301
 302 #define VFIO_IRQ_SET_DATA_TYPE_MASK     (VFIO_IRQ_SET_DATA_NONE | \
 303                                          VFIO_IRQ_SET_DATA_BOOL | \
 304                                          VFIO_IRQ_SET_DATA_EVENTFD)
 305 #define VFIO_IRQ_SET_ACTION_TYPE_MASK   (VFIO_IRQ_SET_ACTION_MASK | \
 306                                          VFIO_IRQ_SET_ACTION_UNMASK | \
 307                                          VFIO_IRQ_SET_ACTION_TRIGGER)
 308 /**
 309  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
 310  *
 311  * Reset a device.
 312  */
 313 #define VFIO_DEVICE_RESET               _IO(VFIO_TYPE, VFIO_BASE + 11)
 314
 315 /*
 316  * The VFIO-PCI bus driver makes use of the following fixed region and
 317  * IRQ index mapping.  Unimplemented regions return a size of zero.
 318  * Unimplemented IRQ types return a count of zero.
 319  */
 320
 321 enum {
 322         VFIO_PCI_BAR0_REGION_INDEX,
 323         VFIO_PCI_BAR1_REGION_INDEX,
 324         VFIO_PCI_BAR2_REGION_INDEX,
 325         VFIO_PCI_BAR3_REGION_INDEX,
 326         VFIO_PCI_BAR4_REGION_INDEX,
 327         VFIO_PCI_BAR5_REGION_INDEX,
 328         VFIO_PCI_ROM_REGION_INDEX,
 329         VFIO_PCI_CONFIG_REGION_INDEX,
 330         /*
 331          * Expose VGA regions defined for PCI base class 03, subclass 00.
 332          * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
 333          * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
 334          * range is found at it's identity mapped offset from the region
 335          * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
 336          * between described ranges are unimplemented.
 337          */
 338         VFIO_PCI_VGA_REGION_INDEX,
 339         VFIO_PCI_NUM_REGIONS
 340 };
 341
 342 enum {
 343         VFIO_PCI_INTX_IRQ_INDEX,
 344         VFIO_PCI_MSI_IRQ_INDEX,
 345         VFIO_PCI_MSIX_IRQ_INDEX,
 346         VFIO_PCI_ERR_IRQ_INDEX,
 347         VFIO_PCI_REQ_IRQ_INDEX,
 348         VFIO_PCI_NUM_IRQS
 349 };
 350
 351 /**
 352  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
 353  *                                            struct vfio_pci_hot_reset_info)
 354  *
 355  * Return: 0 on success, -errno on failure:
 356  *      -enospc = insufficient buffer, -enodev = unsupported for device.
 357  */
 358 struct vfio_pci_dependent_device {
 359         __u32   group_id;
 360         __u16   segment;
 361         __u8    bus;
 362         __u8    devfn; /* Use PCI_SLOT/PCI_FUNC */
 363 };
 364
 365 struct vfio_pci_hot_reset_info {
 366         __u32   argsz;
 367         __u32   flags;
 368         __u32   count;
 369         struct vfio_pci_dependent_device        devices[];
 370 };
 371
 372 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO      _IO(VFIO_TYPE, VFIO_BASE + 12)
 373
 374 /**
 375  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
 376  *                                  struct vfio_pci_hot_reset)
 377  *
 378  * Return: 0 on success, -errno on failure.
 379  */
 380 struct vfio_pci_hot_reset {
 381         __u32   argsz;
 382         __u32   flags;
 383         __u32   count;
 384         __s32   group_fds[];
 385 };
 386
 387 #define VFIO_DEVICE_PCI_HOT_RESET       _IO(VFIO_TYPE, VFIO_BASE + 13)
 388
 389 /* -------- API for Type1 VFIO IOMMU -------- */
 390
 391 /**
 392  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
 393  *
 394  * Retrieve information about the IOMMU object. Fills in provided
 395  * struct vfio_iommu_info. Caller sets argsz.
 396  *
 397  * XXX Should we do these by CHECK_EXTENSION too?
 398  */
 399 struct vfio_iommu_type1_info {
 400         __u32   argsz;
 401         __u32   flags;
 402 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)        /* supported page sizes info */
 403         __u64   iova_pgsizes;           /* Bitmap of supported page sizes */
 404 };
 405
 406 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
 407
 408 /**
 409  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
 410  *
 411  * Map process virtual addresses to IO virtual addresses using the
 412  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
 413  */
 414 struct vfio_iommu_type1_dma_map {
 415         __u32   argsz;
 416         __u32   flags;
 417 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)         /* readable from device */
 418 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)        /* writable from device */
 419         __u64   vaddr;                          /* Process virtual address */
 420         __u64   iova;                           /* IO virtual address */
 421         __u64   size;                           /* Size of mapping (bytes) */
 422 };
 423
 424 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
 425
 426 /**
 427  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
 428  *                                                      struct vfio_dma_unmap)
 429  *
 430  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
 431  * Caller sets argsz.  The actual unmapped size is returned in the size
 432  * field.  No guarantee is made to the user that arbitrary unmaps of iova
 433  * or size different from those used in the original mapping call will
 434  * succeed.
 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 /*
 446  * IOCTLs to enable/disable IOMMU container usage.
 447  * No parameters are supported.
 448  */
 449 #define VFIO_IOMMU_ENABLE       _IO(VFIO_TYPE, VFIO_BASE + 15)
 450 #define VFIO_IOMMU_DISABLE      _IO(VFIO_TYPE, VFIO_BASE + 16)
 451
 452 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
 453
 454 /*
 455  * The SPAPR TCE DDW info struct provides the information about
 456  * the details of Dynamic DMA window capability.
 457  *
 458  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
 459  * @max_dynamic_windows_supported tells the maximum number of windows
 460  * which the platform can create.
 461  * @levels tells the maximum number of levels in multi-level IOMMU tables;
 462  * this allows splitting a table into smaller chunks which reduces
 463  * the amount of physically contiguous memory required for the table.
 464  */
 465 struct vfio_iommu_spapr_tce_ddw_info {
 466         __u64 pgsizes;                  /* Bitmap of supported page sizes */
 467         __u32 max_dynamic_windows_supported;
 468         __u32 levels;
 469 };
 470
 471 /*
 472  * The SPAPR TCE info struct provides the information about the PCI bus
 473  * address ranges available for DMA, these values are programmed into
 474  * the hardware so the guest has to know that information.
 475  *
 476  * The DMA 32 bit window start is an absolute PCI bus address.
 477  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
 478  * addresses too so the window works as a filter rather than an offset
 479  * for IOVA addresses.
 480  *
 481  * Flags supported:
 482  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
 483  *   (DDW) support is present. @ddw is only supported when DDW is present.
 484  */
 485 struct vfio_iommu_spapr_tce_info {
 486         __u32 argsz;
 487         __u32 flags;
 488 #define VFIO_IOMMU_SPAPR_INFO_DDW       (1 << 0)        /* DDW supported */
 489         __u32 dma32_window_start;       /* 32 bit window start (bytes) */
 490         __u32 dma32_window_size;        /* 32 bit window size (bytes) */
 491         struct vfio_iommu_spapr_tce_ddw_info ddw;
 492 };
 493
 494 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO   _IO(VFIO_TYPE, VFIO_BASE + 12)
 495
 496 /*
 497  * EEH PE operation struct provides ways to:
 498  * - enable/disable EEH functionality;
 499  * - unfreeze IO/DMA for frozen PE;
 500  * - read PE state;
 501  * - reset PE;
 502  * - configure PE;
 503  * - inject EEH error.
 504  */
 505 struct vfio_eeh_pe_err {
 506         __u32 type;
 507         __u32 func;
 508         __u64 addr;
 509         __u64 mask;
 510 };
 511
 512 struct vfio_eeh_pe_op {
 513         __u32 argsz;
 514         __u32 flags;
 515         __u32 op;
 516         union {
 517                 struct vfio_eeh_pe_err err;
 518         };
 519 };
 520
 521 #define VFIO_EEH_PE_DISABLE             0       /* Disable EEH functionality */
 522 #define VFIO_EEH_PE_ENABLE              1       /* Enable EEH functionality  */
 523 #define VFIO_EEH_PE_UNFREEZE_IO         2       /* Enable IO for frozen PE   */
 524 #define VFIO_EEH_PE_UNFREEZE_DMA        3       /* Enable DMA for frozen PE  */
 525 #define VFIO_EEH_PE_GET_STATE           4       /* PE state retrieval        */
 526 #define  VFIO_EEH_PE_STATE_NORMAL       0       /* PE in functional state    */
 527 #define  VFIO_EEH_PE_STATE_RESET        1       /* PE reset in progress      */
 528 #define  VFIO_EEH_PE_STATE_STOPPED      2       /* Stopped DMA and IO        */
 529 #define  VFIO_EEH_PE_STATE_STOPPED_DMA  4       /* Stopped DMA only          */
 530 #define  VFIO_EEH_PE_STATE_UNAVAIL      5       /* State unavailable         */
 531 #define VFIO_EEH_PE_RESET_DEACTIVATE    5       /* Deassert PE reset         */
 532 #define VFIO_EEH_PE_RESET_HOT           6       /* Assert hot reset          */
 533 #define VFIO_EEH_PE_RESET_FUNDAMENTAL   7       /* Assert fundamental reset  */
 534 #define VFIO_EEH_PE_CONFIGURE           8       /* PE configuration          */
 535 #define VFIO_EEH_PE_INJECT_ERR          9       /* Inject EEH error          */
 536
 537 #define VFIO_EEH_PE_OP                  _IO(VFIO_TYPE, VFIO_BASE + 21)
 538
 539 /**
 540  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
 541  *
 542  * Registers user space memory where DMA is allowed. It pins
 543  * user pages and does the locked memory accounting so
 544  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
 545  * get faster.
 546  */
 547 struct vfio_iommu_spapr_register_memory {
 548         __u32   argsz;
 549         __u32   flags;
 550         __u64   vaddr;                          /* Process virtual address */
 551         __u64   size;                           /* Size of mapping (bytes) */
 552 };
 553 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY        _IO(VFIO_TYPE, VFIO_BASE + 17)
 554
 555 /**
 556  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
 557  *
 558  * Unregisters user space memory registered with
 559  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
 560  * Uses vfio_iommu_spapr_register_memory for parameters.
 561  */
 562 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY      _IO(VFIO_TYPE, VFIO_BASE + 18)
 563
 564 /**
 565  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
 566  *
 567  * Creates an additional TCE table and programs it (sets a new DMA window)
 568  * to every IOMMU group in the container. It receives page shift, window
 569  * size and number of levels in the TCE table being created.
 570  *
 571  * It allocates and returns an offset on a PCI bus of the new DMA window.
 572  */
 573 struct vfio_iommu_spapr_tce_create {
 574         __u32 argsz;
 575         __u32 flags;
 576         /* in */
 577         __u32 page_shift;
 578         __u32 __resv1;
 579         __u64 window_size;
 580         __u32 levels;
 581         __u32 __resv2;
 582         /* out */
 583         __u64 start_addr;
 584 };
 585 #define VFIO_IOMMU_SPAPR_TCE_CREATE     _IO(VFIO_TYPE, VFIO_BASE + 19)
 586
 587 /**
 588  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
 589  *
 590  * Unprograms a TCE table from all groups in the container and destroys it.
 591  * It receives a PCI bus offset as a window id.
 592  */
 593 struct vfio_iommu_spapr_tce_remove {
 594         __u32 argsz;
 595         __u32 flags;
 596         /* in */
 597         __u64 start_addr;
 598 };
 599 #define VFIO_IOMMU_SPAPR_TCE_REMOVE     _IO(VFIO_TYPE, VFIO_BASE + 20)
 600
 601 /* ***************************************************************** */
 602
 603 #endif /* VFIO_H */