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 */ 75struct 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 */ 136struct 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 */ 195struct 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#define VFIO_DEVICE_FLAGS_AP (1 << 5) /* vfio-ap device */ 204 __u32 num_regions; /* Max region index + 1 */ 205 __u32 num_irqs; /* Max IRQ index + 1 */ 206}; 207#define VFIO_DEVICE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 7) 208 209/* 210 * Vendor driver using Mediated device framework should provide device_api 211 * attribute in supported type attribute groups. Device API string should be one 212 * of the following corresponding to device flags in vfio_device_info structure. 213 */ 214 215#define VFIO_DEVICE_API_PCI_STRING "vfio-pci" 216#define VFIO_DEVICE_API_PLATFORM_STRING "vfio-platform" 217#define VFIO_DEVICE_API_AMBA_STRING "vfio-amba" 218#define VFIO_DEVICE_API_CCW_STRING "vfio-ccw" 219#define VFIO_DEVICE_API_AP_STRING "vfio-ap" 220 221/** 222 * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8, 223 * struct vfio_region_info) 224 * 225 * Retrieve information about a device region. Caller provides 226 * struct vfio_region_info with index value set. Caller sets argsz. 227 * Implementation of region mapping is bus driver specific. This is 228 * intended to describe MMIO, I/O port, as well as bus specific 229 * regions (ex. PCI config space). Zero sized regions may be used 230 * to describe unimplemented regions (ex. unimplemented PCI BARs). 231 * Return: 0 on success, -errno on failure. 232 */ 233struct vfio_region_info { 234 __u32 argsz; 235 __u32 flags; 236#define VFIO_REGION_INFO_FLAG_READ (1 << 0) /* Region supports read */ 237#define VFIO_REGION_INFO_FLAG_WRITE (1 << 1) /* Region supports write */ 238#define VFIO_REGION_INFO_FLAG_MMAP (1 << 2) /* Region supports mmap */ 239#define VFIO_REGION_INFO_FLAG_CAPS (1 << 3) /* Info supports caps */ 240 __u32 index; /* Region index */ 241 __u32 cap_offset; /* Offset within info struct of first cap */ 242 __u64 size; /* Region size (bytes) */ 243 __u64 offset; /* Region offset from start of device fd */ 244}; 245#define VFIO_DEVICE_GET_REGION_INFO _IO(VFIO_TYPE, VFIO_BASE + 8) 246 247/* 248 * The sparse mmap capability allows finer granularity of specifying areas 249 * within a region with mmap support. When specified, the user should only 250 * mmap the offset ranges specified by the areas array. mmaps outside of the 251 * areas specified may fail (such as the range covering a PCI MSI-X table) or 252 * may result in improper device behavior. 253 * 254 * The structures below define version 1 of this capability. 255 */ 256#define VFIO_REGION_INFO_CAP_SPARSE_MMAP 1 257 258struct vfio_region_sparse_mmap_area { 259 __u64 offset; /* Offset of mmap'able area within region */ 260 __u64 size; /* Size of mmap'able area */ 261}; 262 263struct vfio_region_info_cap_sparse_mmap { 264 struct vfio_info_cap_header header; 265 __u32 nr_areas; 266 __u32 reserved; 267 struct vfio_region_sparse_mmap_area areas[]; 268}; 269 270/* 271 * The device specific type capability allows regions unique to a specific 272 * device or class of devices to be exposed. This helps solve the problem for 273 * vfio bus drivers of defining which region indexes correspond to which region 274 * on the device, without needing to resort to static indexes, as done by 275 * vfio-pci. For instance, if we were to go back in time, we might remove 276 * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes 277 * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd 278 * make a "VGA" device specific type to describe the VGA access space. This 279 * means that non-VGA devices wouldn't need to waste this index, and thus the 280 * address space associated with it due to implementation of device file 281 * descriptor offsets in vfio-pci. 282 * 283 * The current implementation is now part of the user ABI, so we can't use this 284 * for VGA, but there are other upcoming use cases, such as opregions for Intel 285 * IGD devices and framebuffers for vGPU devices. We missed VGA, but we'll 286 * use this for future additions. 287 * 288 * The structure below defines version 1 of this capability. 289 */ 290#define VFIO_REGION_INFO_CAP_TYPE 2 291 292struct vfio_region_info_cap_type { 293 struct vfio_info_cap_header header; 294 __u32 type; /* global per bus driver */ 295 __u32 subtype; /* type specific */ 296}; 297 298#define VFIO_REGION_TYPE_PCI_VENDOR_TYPE (1 << 31) 299#define VFIO_REGION_TYPE_PCI_VENDOR_MASK (0xffff) 300 301/* 8086 Vendor sub-types */ 302#define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION (1) 303#define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG (2) 304#define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG (3) 305 306/* 307 * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped 308 * which allows direct access to non-MSIX registers which happened to be within 309 * the same system page. 310 * 311 * Even though the userspace gets direct access to the MSIX data, the existing 312 * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration. 313 */ 314#define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE 3 315 316/** 317 * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9, 318 * struct vfio_irq_info) 319 * 320 * Retrieve information about a device IRQ. Caller provides 321 * struct vfio_irq_info with index value set. Caller sets argsz. 322 * Implementation of IRQ mapping is bus driver specific. Indexes 323 * using multiple IRQs are primarily intended to support MSI-like 324 * interrupt blocks. Zero count irq blocks may be used to describe 325 * unimplemented interrupt types. 326 * 327 * The EVENTFD flag indicates the interrupt index supports eventfd based 328 * signaling. 329 * 330 * The MASKABLE flags indicates the index supports MASK and UNMASK 331 * actions described below. 332 * 333 * AUTOMASKED indicates that after signaling, the interrupt line is 334 * automatically masked by VFIO and the user needs to unmask the line 335 * to receive new interrupts. This is primarily intended to distinguish 336 * level triggered interrupts. 337 * 338 * The NORESIZE flag indicates that the interrupt lines within the index 339 * are setup as a set and new subindexes cannot be enabled without first 340 * disabling the entire index. This is used for interrupts like PCI MSI 341 * and MSI-X where the driver may only use a subset of the available 342 * indexes, but VFIO needs to enable a specific number of vectors 343 * upfront. In the case of MSI-X, where the user can enable MSI-X and 344 * then add and unmask vectors, it's up to userspace to make the decision 345 * whether to allocate the maximum supported number of vectors or tear 346 * down setup and incrementally increase the vectors as each is enabled. 347 */ 348struct vfio_irq_info { 349 __u32 argsz; 350 __u32 flags; 351#define VFIO_IRQ_INFO_EVENTFD (1 << 0) 352#define VFIO_IRQ_INFO_MASKABLE (1 << 1) 353#define VFIO_IRQ_INFO_AUTOMASKED (1 << 2) 354#define VFIO_IRQ_INFO_NORESIZE (1 << 3) 355 __u32 index; /* IRQ index */ 356 __u32 count; /* Number of IRQs within this index */ 357}; 358#define VFIO_DEVICE_GET_IRQ_INFO _IO(VFIO_TYPE, VFIO_BASE + 9) 359 360/** 361 * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set) 362 * 363 * Set signaling, masking, and unmasking of interrupts. Caller provides 364 * struct vfio_irq_set with all fields set. 'start' and 'count' indicate 365 * the range of subindexes being specified. 366 * 367 * The DATA flags specify the type of data provided. If DATA_NONE, the 368 * operation performs the specified action immediately on the specified 369 * interrupt(s). For example, to unmask AUTOMASKED interrupt [0,0]: 370 * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1. 371 * 372 * DATA_BOOL allows sparse support for the same on arrays of interrupts. 373 * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]): 374 * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3, 375 * data = {1,0,1} 376 * 377 * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd. 378 * A value of -1 can be used to either de-assign interrupts if already 379 * assigned or skip un-assigned interrupts. For example, to set an eventfd 380 * to be trigger for interrupts [0,0] and [0,2]: 381 * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3, 382 * data = {fd1, -1, fd2} 383 * If index [0,1] is previously set, two count = 1 ioctls calls would be 384 * required to set [0,0] and [0,2] without changing [0,1]. 385 * 386 * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used 387 * with ACTION_TRIGGER to perform kernel level interrupt loopback testing 388 * from userspace (ie. simulate hardware triggering). 389 * 390 * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER 391 * enables the interrupt index for the device. Individual subindex interrupts 392 * can be disabled using the -1 value for DATA_EVENTFD or the index can be 393 * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0. 394 * 395 * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while 396 * ACTION_TRIGGER specifies kernel->user signaling. 397 */ 398struct vfio_irq_set { 399 __u32 argsz; 400 __u32 flags; 401#define VFIO_IRQ_SET_DATA_NONE (1 << 0) /* Data not present */ 402#define VFIO_IRQ_SET_DATA_BOOL (1 << 1) /* Data is bool (u8) */ 403#define VFIO_IRQ_SET_DATA_EVENTFD (1 << 2) /* Data is eventfd (s32) */ 404#define VFIO_IRQ_SET_ACTION_MASK (1 << 3) /* Mask interrupt */ 405#define VFIO_IRQ_SET_ACTION_UNMASK (1 << 4) /* Unmask interrupt */ 406#define VFIO_IRQ_SET_ACTION_TRIGGER (1 << 5) /* Trigger interrupt */ 407 __u32 index; 408 __u32 start; 409 __u32 count; 410 __u8 data[]; 411}; 412#define VFIO_DEVICE_SET_IRQS _IO(VFIO_TYPE, VFIO_BASE + 10) 413 414#define VFIO_IRQ_SET_DATA_TYPE_MASK (VFIO_IRQ_SET_DATA_NONE | \ 415 VFIO_IRQ_SET_DATA_BOOL | \ 416 VFIO_IRQ_SET_DATA_EVENTFD) 417#define VFIO_IRQ_SET_ACTION_TYPE_MASK (VFIO_IRQ_SET_ACTION_MASK | \ 418 VFIO_IRQ_SET_ACTION_UNMASK | \ 419 VFIO_IRQ_SET_ACTION_TRIGGER) 420/** 421 * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11) 422 * 423 * Reset a device. 424 */ 425#define VFIO_DEVICE_RESET _IO(VFIO_TYPE, VFIO_BASE + 11) 426 427/* 428 * The VFIO-PCI bus driver makes use of the following fixed region and 429 * IRQ index mapping. Unimplemented regions return a size of zero. 430 * Unimplemented IRQ types return a count of zero. 431 */ 432 433enum { 434 VFIO_PCI_BAR0_REGION_INDEX, 435 VFIO_PCI_BAR1_REGION_INDEX, 436 VFIO_PCI_BAR2_REGION_INDEX, 437 VFIO_PCI_BAR3_REGION_INDEX, 438 VFIO_PCI_BAR4_REGION_INDEX, 439 VFIO_PCI_BAR5_REGION_INDEX, 440 VFIO_PCI_ROM_REGION_INDEX, 441 VFIO_PCI_CONFIG_REGION_INDEX, 442 /* 443 * Expose VGA regions defined for PCI base class 03, subclass 00. 444 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df 445 * as well as the MMIO range 0xa0000 to 0xbffff. Each implemented 446 * range is found at it's identity mapped offset from the region 447 * offset, for example 0x3b0 is region_info.offset + 0x3b0. Areas 448 * between described ranges are unimplemented. 449 */ 450 VFIO_PCI_VGA_REGION_INDEX, 451 VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */ 452 /* device specific cap to define content. */ 453}; 454 455enum { 456 VFIO_PCI_INTX_IRQ_INDEX, 457 VFIO_PCI_MSI_IRQ_INDEX, 458 VFIO_PCI_MSIX_IRQ_INDEX, 459 VFIO_PCI_ERR_IRQ_INDEX, 460 VFIO_PCI_REQ_IRQ_INDEX, 461 VFIO_PCI_NUM_IRQS 462}; 463 464/* 465 * The vfio-ccw bus driver makes use of the following fixed region and 466 * IRQ index mapping. Unimplemented regions return a size of zero. 467 * Unimplemented IRQ types return a count of zero. 468 */ 469 470enum { 471 VFIO_CCW_CONFIG_REGION_INDEX, 472 VFIO_CCW_NUM_REGIONS 473}; 474 475enum { 476 VFIO_CCW_IO_IRQ_INDEX, 477 VFIO_CCW_NUM_IRQS 478}; 479 480/** 481 * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12, 482 * struct vfio_pci_hot_reset_info) 483 * 484 * Return: 0 on success, -errno on failure: 485 * -enospc = insufficient buffer, -enodev = unsupported for device. 486 */ 487struct vfio_pci_dependent_device { 488 __u32 group_id; 489 __u16 segment; 490 __u8 bus; 491 __u8 devfn; /* Use PCI_SLOT/PCI_FUNC */ 492}; 493 494struct vfio_pci_hot_reset_info { 495 __u32 argsz; 496 __u32 flags; 497 __u32 count; 498 struct vfio_pci_dependent_device devices[]; 499}; 500 501#define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 502 503/** 504 * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13, 505 * struct vfio_pci_hot_reset) 506 * 507 * Return: 0 on success, -errno on failure. 508 */ 509struct vfio_pci_hot_reset { 510 __u32 argsz; 511 __u32 flags; 512 __u32 count; 513 __s32 group_fds[]; 514}; 515 516#define VFIO_DEVICE_PCI_HOT_RESET _IO(VFIO_TYPE, VFIO_BASE + 13) 517 518/** 519 * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14, 520 * struct vfio_device_query_gfx_plane) 521 * 522 * Set the drm_plane_type and flags, then retrieve the gfx plane info. 523 * 524 * flags supported: 525 * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set 526 * to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no 527 * support for dma-buf. 528 * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set 529 * to ask if the mdev supports region. 0 on support, -EINVAL on no 530 * support for region. 531 * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set 532 * with each call to query the plane info. 533 * - Others are invalid and return -EINVAL. 534 * 535 * Note: 536 * 1. Plane could be disabled by guest. In that case, success will be 537 * returned with zero-initialized drm_format, size, width and height 538 * fields. 539 * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available 540 * 541 * Return: 0 on success, -errno on other failure. 542 */ 543struct vfio_device_gfx_plane_info { 544 __u32 argsz; 545 __u32 flags; 546#define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0) 547#define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1) 548#define VFIO_GFX_PLANE_TYPE_REGION (1 << 2) 549 /* in */ 550 __u32 drm_plane_type; /* type of plane: DRM_PLANE_TYPE_* */ 551 /* out */ 552 __u32 drm_format; /* drm format of plane */ 553 __u64 drm_format_mod; /* tiled mode */ 554 __u32 width; /* width of plane */ 555 __u32 height; /* height of plane */ 556 __u32 stride; /* stride of plane */ 557 __u32 size; /* size of plane in bytes, align on page*/ 558 __u32 x_pos; /* horizontal position of cursor plane */ 559 __u32 y_pos; /* vertical position of cursor plane*/ 560 __u32 x_hot; /* horizontal position of cursor hotspot */ 561 __u32 y_hot; /* vertical position of cursor hotspot */ 562 union { 563 __u32 region_index; /* region index */ 564 __u32 dmabuf_id; /* dma-buf id */ 565 }; 566}; 567 568#define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14) 569 570/** 571 * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32) 572 * 573 * Return a new dma-buf file descriptor for an exposed guest framebuffer 574 * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_ 575 * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer. 576 */ 577 578#define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15) 579 580/** 581 * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16, 582 * struct vfio_device_ioeventfd) 583 * 584 * Perform a write to the device at the specified device fd offset, with 585 * the specified data and width when the provided eventfd is triggered. 586 * vfio bus drivers may not support this for all regions, for all widths, 587 * or at all. vfio-pci currently only enables support for BAR regions, 588 * excluding the MSI-X vector table. 589 * 590 * Return: 0 on success, -errno on failure. 591 */ 592struct vfio_device_ioeventfd { 593 __u32 argsz; 594 __u32 flags; 595#define VFIO_DEVICE_IOEVENTFD_8 (1 << 0) /* 1-byte write */ 596#define VFIO_DEVICE_IOEVENTFD_16 (1 << 1) /* 2-byte write */ 597#define VFIO_DEVICE_IOEVENTFD_32 (1 << 2) /* 4-byte write */ 598#define VFIO_DEVICE_IOEVENTFD_64 (1 << 3) /* 8-byte write */ 599#define VFIO_DEVICE_IOEVENTFD_SIZE_MASK (0xf) 600 __u64 offset; /* device fd offset of write */ 601 __u64 data; /* data to be written */ 602 __s32 fd; /* -1 for de-assignment */ 603}; 604 605#define VFIO_DEVICE_IOEVENTFD _IO(VFIO_TYPE, VFIO_BASE + 16) 606 607/* -------- API for Type1 VFIO IOMMU -------- */ 608 609/** 610 * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info) 611 * 612 * Retrieve information about the IOMMU object. Fills in provided 613 * struct vfio_iommu_info. Caller sets argsz. 614 * 615 * XXX Should we do these by CHECK_EXTENSION too? 616 */ 617struct vfio_iommu_type1_info { 618 __u32 argsz; 619 __u32 flags; 620#define VFIO_IOMMU_INFO_PGSIZES (1 << 0) /* supported page sizes info */ 621 __u64 iova_pgsizes; /* Bitmap of supported page sizes */ 622}; 623 624#define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 625 626/** 627 * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map) 628 * 629 * Map process virtual addresses to IO virtual addresses using the 630 * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required. 631 */ 632struct vfio_iommu_type1_dma_map { 633 __u32 argsz; 634 __u32 flags; 635#define VFIO_DMA_MAP_FLAG_READ (1 << 0) /* readable from device */ 636#define VFIO_DMA_MAP_FLAG_WRITE (1 << 1) /* writable from device */ 637 __u64 vaddr; /* Process virtual address */ 638 __u64 iova; /* IO virtual address */ 639 __u64 size; /* Size of mapping (bytes) */ 640}; 641 642#define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13) 643 644/** 645 * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14, 646 * struct vfio_dma_unmap) 647 * 648 * Unmap IO virtual addresses using the provided struct vfio_dma_unmap. 649 * Caller sets argsz. The actual unmapped size is returned in the size 650 * field. No guarantee is made to the user that arbitrary unmaps of iova 651 * or size different from those used in the original mapping call will 652 * succeed. 653 */ 654struct vfio_iommu_type1_dma_unmap { 655 __u32 argsz; 656 __u32 flags; 657 __u64 iova; /* IO virtual address */ 658 __u64 size; /* Size of mapping (bytes) */ 659}; 660 661#define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14) 662 663/* 664 * IOCTLs to enable/disable IOMMU container usage. 665 * No parameters are supported. 666 */ 667#define VFIO_IOMMU_ENABLE _IO(VFIO_TYPE, VFIO_BASE + 15) 668#define VFIO_IOMMU_DISABLE _IO(VFIO_TYPE, VFIO_BASE + 16) 669 670/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */ 671 672/* 673 * The SPAPR TCE DDW info struct provides the information about 674 * the details of Dynamic DMA window capability. 675 * 676 * @pgsizes contains a page size bitmask, 4K/64K/16M are supported. 677 * @max_dynamic_windows_supported tells the maximum number of windows 678 * which the platform can create. 679 * @levels tells the maximum number of levels in multi-level IOMMU tables; 680 * this allows splitting a table into smaller chunks which reduces 681 * the amount of physically contiguous memory required for the table. 682 */ 683struct vfio_iommu_spapr_tce_ddw_info { 684 __u64 pgsizes; /* Bitmap of supported page sizes */ 685 __u32 max_dynamic_windows_supported; 686 __u32 levels; 687}; 688 689/* 690 * The SPAPR TCE info struct provides the information about the PCI bus 691 * address ranges available for DMA, these values are programmed into 692 * the hardware so the guest has to know that information. 693 * 694 * The DMA 32 bit window start is an absolute PCI bus address. 695 * The IOVA address passed via map/unmap ioctls are absolute PCI bus 696 * addresses too so the window works as a filter rather than an offset 697 * for IOVA addresses. 698 * 699 * Flags supported: 700 * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows 701 * (DDW) support is present. @ddw is only supported when DDW is present. 702 */ 703struct vfio_iommu_spapr_tce_info { 704 __u32 argsz; 705 __u32 flags; 706#define VFIO_IOMMU_SPAPR_INFO_DDW (1 << 0) /* DDW supported */ 707 __u32 dma32_window_start; /* 32 bit window start (bytes) */ 708 __u32 dma32_window_size; /* 32 bit window size (bytes) */ 709 struct vfio_iommu_spapr_tce_ddw_info ddw; 710}; 711 712#define VFIO_IOMMU_SPAPR_TCE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 713 714/* 715 * EEH PE operation struct provides ways to: 716 * - enable/disable EEH functionality; 717 * - unfreeze IO/DMA for frozen PE; 718 * - read PE state; 719 * - reset PE; 720 * - configure PE; 721 * - inject EEH error. 722 */ 723struct vfio_eeh_pe_err { 724 __u32 type; 725 __u32 func; 726 __u64 addr; 727 __u64 mask; 728}; 729 730struct vfio_eeh_pe_op { 731 __u32 argsz; 732 __u32 flags; 733 __u32 op; 734 union { 735 struct vfio_eeh_pe_err err; 736 }; 737}; 738 739#define VFIO_EEH_PE_DISABLE 0 /* Disable EEH functionality */ 740#define VFIO_EEH_PE_ENABLE 1 /* Enable EEH functionality */ 741#define VFIO_EEH_PE_UNFREEZE_IO 2 /* Enable IO for frozen PE */ 742#define VFIO_EEH_PE_UNFREEZE_DMA 3 /* Enable DMA for frozen PE */ 743#define VFIO_EEH_PE_GET_STATE 4 /* PE state retrieval */ 744#define VFIO_EEH_PE_STATE_NORMAL 0 /* PE in functional state */ 745#define VFIO_EEH_PE_STATE_RESET 1 /* PE reset in progress */ 746#define VFIO_EEH_PE_STATE_STOPPED 2 /* Stopped DMA and IO */ 747#define VFIO_EEH_PE_STATE_STOPPED_DMA 4 /* Stopped DMA only */ 748#define VFIO_EEH_PE_STATE_UNAVAIL 5 /* State unavailable */ 749#define VFIO_EEH_PE_RESET_DEACTIVATE 5 /* Deassert PE reset */ 750#define VFIO_EEH_PE_RESET_HOT 6 /* Assert hot reset */ 751#define VFIO_EEH_PE_RESET_FUNDAMENTAL 7 /* Assert fundamental reset */ 752#define VFIO_EEH_PE_CONFIGURE 8 /* PE configuration */ 753#define VFIO_EEH_PE_INJECT_ERR 9 /* Inject EEH error */ 754 755#define VFIO_EEH_PE_OP _IO(VFIO_TYPE, VFIO_BASE + 21) 756 757/** 758 * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory) 759 * 760 * Registers user space memory where DMA is allowed. It pins 761 * user pages and does the locked memory accounting so 762 * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls 763 * get faster. 764 */ 765struct vfio_iommu_spapr_register_memory { 766 __u32 argsz; 767 __u32 flags; 768 __u64 vaddr; /* Process virtual address */ 769 __u64 size; /* Size of mapping (bytes) */ 770}; 771#define VFIO_IOMMU_SPAPR_REGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 17) 772 773/** 774 * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory) 775 * 776 * Unregisters user space memory registered with 777 * VFIO_IOMMU_SPAPR_REGISTER_MEMORY. 778 * Uses vfio_iommu_spapr_register_memory for parameters. 779 */ 780#define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 18) 781 782/** 783 * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create) 784 * 785 * Creates an additional TCE table and programs it (sets a new DMA window) 786 * to every IOMMU group in the container. It receives page shift, window 787 * size and number of levels in the TCE table being created. 788 * 789 * It allocates and returns an offset on a PCI bus of the new DMA window. 790 */ 791struct vfio_iommu_spapr_tce_create { 792 __u32 argsz; 793 __u32 flags; 794 /* in */ 795 __u32 page_shift; 796 __u32 __resv1; 797 __u64 window_size; 798 __u32 levels; 799 __u32 __resv2; 800 /* out */ 801 __u64 start_addr; 802}; 803#define VFIO_IOMMU_SPAPR_TCE_CREATE _IO(VFIO_TYPE, VFIO_BASE + 19) 804 805/** 806 * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove) 807 * 808 * Unprograms a TCE table from all groups in the container and destroys it. 809 * It receives a PCI bus offset as a window id. 810 */ 811struct vfio_iommu_spapr_tce_remove { 812 __u32 argsz; 813 __u32 flags; 814 /* in */ 815 __u64 start_addr; 816}; 817#define VFIO_IOMMU_SPAPR_TCE_REMOVE _IO(VFIO_TYPE, VFIO_BASE + 20) 818 819/* ***************************************************************** */ 820 821#endif /* VFIO_H */ 822