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 _UAPIVFIO_H 13 #define _UAPIVFIO_H 14 15 #include <linux/types.h> 16 #include <linux/ioctl.h> 17 #include <linux/stddef.h> 18 19 #define VFIO_API_VERSION 0 20 21 22 /* Kernel & User level defines for VFIO IOCTLs. */ 23 24 /* Extensions */ 25 26 #define VFIO_TYPE1_IOMMU 1 27 #define VFIO_SPAPR_TCE_IOMMU 2 28 #define VFIO_TYPE1v2_IOMMU 3 29 /* 30 * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping). This 31 * capability is subject to change as groups are added or removed. 32 */ 33 #define VFIO_DMA_CC_IOMMU 4 34 35 /* Check if EEH is supported */ 36 #define VFIO_EEH 5 37 38 /* Two-stage IOMMU */ 39 #define __VFIO_RESERVED_TYPE1_NESTING_IOMMU 6 /* Implies v2 */ 40 41 #define VFIO_SPAPR_TCE_v2_IOMMU 7 42 43 /* 44 * The No-IOMMU IOMMU offers no translation or isolation for devices and 45 * supports no ioctls outside of VFIO_CHECK_EXTENSION. Use of VFIO's No-IOMMU 46 * code will taint the host kernel and should be used with extreme caution. 47 */ 48 #define VFIO_NOIOMMU_IOMMU 8 49 50 /* Supports VFIO_DMA_UNMAP_FLAG_ALL */ 51 #define VFIO_UNMAP_ALL 9 52 53 /* 54 * Supports the vaddr flag for DMA map and unmap. Not supported for mediated 55 * devices, so this capability is subject to change as groups are added or 56 * removed. 57 */ 58 #define VFIO_UPDATE_VADDR 10 59 60 /* 61 * The IOCTL interface is designed for extensibility by embedding the 62 * structure length (argsz) and flags into structures passed between 63 * kernel and userspace. We therefore use the _IO() macro for these 64 * defines to avoid implicitly embedding a size into the ioctl request. 65 * As structure fields are added, argsz will increase to match and flag 66 * bits will be defined to indicate additional fields with valid data. 67 * It's *always* the caller's responsibility to indicate the size of 68 * the structure passed by setting argsz appropriately. 69 */ 70 71 #define VFIO_TYPE (';') 72 #define VFIO_BASE 100 73 74 /* 75 * For extension of INFO ioctls, VFIO makes use of a capability chain 76 * designed after PCI/e capabilities. A flag bit indicates whether 77 * this capability chain is supported and a field defined in the fixed 78 * structure defines the offset of the first capability in the chain. 79 * This field is only valid when the corresponding bit in the flags 80 * bitmap is set. This offset field is relative to the start of the 81 * INFO buffer, as is the next field within each capability header. 82 * The id within the header is a shared address space per INFO ioctl, 83 * while the version field is specific to the capability id. The 84 * contents following the header are specific to the capability id. 85 */ 86 struct vfio_info_cap_header { 87 __u16 id; /* Identifies capability */ 88 __u16 version; /* Version specific to the capability ID */ 89 __u32 next; /* Offset of next capability */ 90 }; 91 92 /* 93 * Callers of INFO ioctls passing insufficiently sized buffers will see 94 * the capability chain flag bit set, a zero value for the first capability 95 * offset (if available within the provided argsz), and argsz will be 96 * updated to report the necessary buffer size. For compatibility, the 97 * INFO ioctl will not report error in this case, but the capability chain 98 * will not be available. 99 */ 100 101 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */ 102 103 /** 104 * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0) 105 * 106 * Report the version of the VFIO API. This allows us to bump the entire 107 * API version should we later need to add or change features in incompatible 108 * ways. 109 * Return: VFIO_API_VERSION 110 * Availability: Always 111 */ 112 #define VFIO_GET_API_VERSION _IO(VFIO_TYPE, VFIO_BASE + 0) 113 114 /** 115 * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32) 116 * 117 * Check whether an extension is supported. 118 * Return: 0 if not supported, 1 (or some other positive integer) if supported. 119 * Availability: Always 120 */ 121 #define VFIO_CHECK_EXTENSION _IO(VFIO_TYPE, VFIO_BASE + 1) 122 123 /** 124 * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32) 125 * 126 * Set the iommu to the given type. The type must be supported by an 127 * iommu driver as verified by calling CHECK_EXTENSION using the same 128 * type. A group must be set to this file descriptor before this 129 * ioctl is available. The IOMMU interfaces enabled by this call are 130 * specific to the value set. 131 * Return: 0 on success, -errno on failure 132 * Availability: When VFIO group attached 133 */ 134 #define VFIO_SET_IOMMU _IO(VFIO_TYPE, VFIO_BASE + 2) 135 136 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */ 137 138 /** 139 * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3, 140 * struct vfio_group_status) 141 * 142 * Retrieve information about the group. Fills in provided 143 * struct vfio_group_info. Caller sets argsz. 144 * Return: 0 on succes, -errno on failure. 145 * Availability: Always 146 */ 147 struct vfio_group_status { 148 __u32 argsz; 149 __u32 flags; 150 #define VFIO_GROUP_FLAGS_VIABLE (1 << 0) 151 #define VFIO_GROUP_FLAGS_CONTAINER_SET (1 << 1) 152 }; 153 #define VFIO_GROUP_GET_STATUS _IO(VFIO_TYPE, VFIO_BASE + 3) 154 155 /** 156 * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32) 157 * 158 * Set the container for the VFIO group to the open VFIO file 159 * descriptor provided. Groups may only belong to a single 160 * container. Containers may, at their discretion, support multiple 161 * groups. Only when a container is set are all of the interfaces 162 * of the VFIO file descriptor and the VFIO group file descriptor 163 * available to the user. 164 * Return: 0 on success, -errno on failure. 165 * Availability: Always 166 */ 167 #define VFIO_GROUP_SET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 4) 168 169 /** 170 * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5) 171 * 172 * Remove the group from the attached container. This is the 173 * opposite of the SET_CONTAINER call and returns the group to 174 * an initial state. All device file descriptors must be released 175 * prior to calling this interface. When removing the last group 176 * from a container, the IOMMU will be disabled and all state lost, 177 * effectively also returning the VFIO file descriptor to an initial 178 * state. 179 * Return: 0 on success, -errno on failure. 180 * Availability: When attached to container 181 */ 182 #define VFIO_GROUP_UNSET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 5) 183 184 /** 185 * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char) 186 * 187 * Return a new file descriptor for the device object described by 188 * the provided string. The string should match a device listed in 189 * the devices subdirectory of the IOMMU group sysfs entry. The 190 * group containing the device must already be added to this context. 191 * Return: new file descriptor on success, -errno on failure. 192 * Availability: When attached to container 193 */ 194 #define VFIO_GROUP_GET_DEVICE_FD _IO(VFIO_TYPE, VFIO_BASE + 6) 195 196 /* --------------- IOCTLs for DEVICE file descriptors --------------- */ 197 198 /** 199 * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7, 200 * struct vfio_device_info) 201 * 202 * Retrieve information about the device. Fills in provided 203 * struct vfio_device_info. Caller sets argsz. 204 * Return: 0 on success, -errno on failure. 205 */ 206 struct vfio_device_info { 207 __u32 argsz; 208 __u32 flags; 209 #define VFIO_DEVICE_FLAGS_RESET (1 << 0) /* Device supports reset */ 210 #define VFIO_DEVICE_FLAGS_PCI (1 << 1) /* vfio-pci device */ 211 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2) /* vfio-platform device */ 212 #define VFIO_DEVICE_FLAGS_AMBA (1 << 3) /* vfio-amba device */ 213 #define VFIO_DEVICE_FLAGS_CCW (1 << 4) /* vfio-ccw device */ 214 #define VFIO_DEVICE_FLAGS_AP (1 << 5) /* vfio-ap device */ 215 #define VFIO_DEVICE_FLAGS_FSL_MC (1 << 6) /* vfio-fsl-mc device */ 216 #define VFIO_DEVICE_FLAGS_CAPS (1 << 7) /* Info supports caps */ 217 #define VFIO_DEVICE_FLAGS_CDX (1 << 8) /* vfio-cdx device */ 218 __u32 num_regions; /* Max region index + 1 */ 219 __u32 num_irqs; /* Max IRQ index + 1 */ 220 __u32 cap_offset; /* Offset within info struct of first cap */ 221 __u32 pad; 222 }; 223 #define VFIO_DEVICE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 7) 224 225 /* 226 * Vendor driver using Mediated device framework should provide device_api 227 * attribute in supported type attribute groups. Device API string should be one 228 * of the following corresponding to device flags in vfio_device_info structure. 229 */ 230 231 #define VFIO_DEVICE_API_PCI_STRING "vfio-pci" 232 #define VFIO_DEVICE_API_PLATFORM_STRING "vfio-platform" 233 #define VFIO_DEVICE_API_AMBA_STRING "vfio-amba" 234 #define VFIO_DEVICE_API_CCW_STRING "vfio-ccw" 235 #define VFIO_DEVICE_API_AP_STRING "vfio-ap" 236 237 /* 238 * The following capabilities are unique to s390 zPCI devices. Their contents 239 * are further-defined in vfio_zdev.h 240 */ 241 #define VFIO_DEVICE_INFO_CAP_ZPCI_BASE 1 242 #define VFIO_DEVICE_INFO_CAP_ZPCI_GROUP 2 243 #define VFIO_DEVICE_INFO_CAP_ZPCI_UTIL 3 244 #define VFIO_DEVICE_INFO_CAP_ZPCI_PFIP 4 245 246 /* 247 * The following VFIO_DEVICE_INFO capability reports support for PCIe AtomicOp 248 * completion to the root bus with supported widths provided via flags. 249 */ 250 #define VFIO_DEVICE_INFO_CAP_PCI_ATOMIC_COMP 5 251 struct vfio_device_info_cap_pci_atomic_comp { 252 struct vfio_info_cap_header header; 253 __u32 flags; 254 #define VFIO_PCI_ATOMIC_COMP32 (1 << 0) 255 #define VFIO_PCI_ATOMIC_COMP64 (1 << 1) 256 #define VFIO_PCI_ATOMIC_COMP128 (1 << 2) 257 __u32 reserved; 258 }; 259 260 /** 261 * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8, 262 * struct vfio_region_info) 263 * 264 * Retrieve information about a device region. Caller provides 265 * struct vfio_region_info with index value set. Caller sets argsz. 266 * Implementation of region mapping is bus driver specific. This is 267 * intended to describe MMIO, I/O port, as well as bus specific 268 * regions (ex. PCI config space). Zero sized regions may be used 269 * to describe unimplemented regions (ex. unimplemented PCI BARs). 270 * Return: 0 on success, -errno on failure. 271 */ 272 struct vfio_region_info { 273 __u32 argsz; 274 __u32 flags; 275 #define VFIO_REGION_INFO_FLAG_READ (1 << 0) /* Region supports read */ 276 #define VFIO_REGION_INFO_FLAG_WRITE (1 << 1) /* Region supports write */ 277 #define VFIO_REGION_INFO_FLAG_MMAP (1 << 2) /* Region supports mmap */ 278 #define VFIO_REGION_INFO_FLAG_CAPS (1 << 3) /* Info supports caps */ 279 __u32 index; /* Region index */ 280 __u32 cap_offset; /* Offset within info struct of first cap */ 281 __aligned_u64 size; /* Region size (bytes) */ 282 __aligned_u64 offset; /* Region offset from start of device fd */ 283 }; 284 #define VFIO_DEVICE_GET_REGION_INFO _IO(VFIO_TYPE, VFIO_BASE + 8) 285 286 /* 287 * The sparse mmap capability allows finer granularity of specifying areas 288 * within a region with mmap support. When specified, the user should only 289 * mmap the offset ranges specified by the areas array. mmaps outside of the 290 * areas specified may fail (such as the range covering a PCI MSI-X table) or 291 * may result in improper device behavior. 292 * 293 * The structures below define version 1 of this capability. 294 */ 295 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP 1 296 297 struct vfio_region_sparse_mmap_area { 298 __aligned_u64 offset; /* Offset of mmap'able area within region */ 299 __aligned_u64 size; /* Size of mmap'able area */ 300 }; 301 302 struct vfio_region_info_cap_sparse_mmap { 303 struct vfio_info_cap_header header; 304 __u32 nr_areas; 305 __u32 reserved; 306 struct vfio_region_sparse_mmap_area areas[]; 307 }; 308 309 /* 310 * The device specific type capability allows regions unique to a specific 311 * device or class of devices to be exposed. This helps solve the problem for 312 * vfio bus drivers of defining which region indexes correspond to which region 313 * on the device, without needing to resort to static indexes, as done by 314 * vfio-pci. For instance, if we were to go back in time, we might remove 315 * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes 316 * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd 317 * make a "VGA" device specific type to describe the VGA access space. This 318 * means that non-VGA devices wouldn't need to waste this index, and thus the 319 * address space associated with it due to implementation of device file 320 * descriptor offsets in vfio-pci. 321 * 322 * The current implementation is now part of the user ABI, so we can't use this 323 * for VGA, but there are other upcoming use cases, such as opregions for Intel 324 * IGD devices and framebuffers for vGPU devices. We missed VGA, but we'll 325 * use this for future additions. 326 * 327 * The structure below defines version 1 of this capability. 328 */ 329 #define VFIO_REGION_INFO_CAP_TYPE 2 330 331 struct vfio_region_info_cap_type { 332 struct vfio_info_cap_header header; 333 __u32 type; /* global per bus driver */ 334 __u32 subtype; /* type specific */ 335 }; 336 337 /* 338 * List of region types, global per bus driver. 339 * If you introduce a new type, please add it here. 340 */ 341 342 /* PCI region type containing a PCI vendor part */ 343 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE (1 << 31) 344 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK (0xffff) 345 #define VFIO_REGION_TYPE_GFX (1) 346 #define VFIO_REGION_TYPE_CCW (2) 347 #define VFIO_REGION_TYPE_MIGRATION_DEPRECATED (3) 348 349 /* sub-types for VFIO_REGION_TYPE_PCI_* */ 350 351 /* 8086 vendor PCI sub-types */ 352 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION (1) 353 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG (2) 354 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG (3) 355 356 /* 10de vendor PCI sub-types */ 357 /* 358 * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space. 359 * 360 * Deprecated, region no longer provided 361 */ 362 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM (1) 363 364 /* 1014 vendor PCI sub-types */ 365 /* 366 * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU 367 * to do TLB invalidation on a GPU. 368 * 369 * Deprecated, region no longer provided 370 */ 371 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD (1) 372 373 /* sub-types for VFIO_REGION_TYPE_GFX */ 374 #define VFIO_REGION_SUBTYPE_GFX_EDID (1) 375 376 /** 377 * struct vfio_region_gfx_edid - EDID region layout. 378 * 379 * Set display link state and EDID blob. 380 * 381 * The EDID blob has monitor information such as brand, name, serial 382 * number, physical size, supported video modes and more. 383 * 384 * This special region allows userspace (typically qemu) set a virtual 385 * EDID for the virtual monitor, which allows a flexible display 386 * configuration. 387 * 388 * For the edid blob spec look here: 389 * https://en.wikipedia.org/wiki/Extended_Display_Identification_Data 390 * 391 * On linux systems you can find the EDID blob in sysfs: 392 * /sys/class/drm/${card}/${connector}/edid 393 * 394 * You can use the edid-decode ulility (comes with xorg-x11-utils) to 395 * decode the EDID blob. 396 * 397 * @edid_offset: location of the edid blob, relative to the 398 * start of the region (readonly). 399 * @edid_max_size: max size of the edid blob (readonly). 400 * @edid_size: actual edid size (read/write). 401 * @link_state: display link state (read/write). 402 * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on. 403 * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off. 404 * @max_xres: max display width (0 == no limitation, readonly). 405 * @max_yres: max display height (0 == no limitation, readonly). 406 * 407 * EDID update protocol: 408 * (1) set link-state to down. 409 * (2) update edid blob and size. 410 * (3) set link-state to up. 411 */ 412 struct vfio_region_gfx_edid { 413 __u32 edid_offset; 414 __u32 edid_max_size; 415 __u32 edid_size; 416 __u32 max_xres; 417 __u32 max_yres; 418 __u32 link_state; 419 #define VFIO_DEVICE_GFX_LINK_STATE_UP 1 420 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN 2 421 }; 422 423 /* sub-types for VFIO_REGION_TYPE_CCW */ 424 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD (1) 425 #define VFIO_REGION_SUBTYPE_CCW_SCHIB (2) 426 #define VFIO_REGION_SUBTYPE_CCW_CRW (3) 427 428 /* sub-types for VFIO_REGION_TYPE_MIGRATION */ 429 #define VFIO_REGION_SUBTYPE_MIGRATION_DEPRECATED (1) 430 431 struct vfio_device_migration_info { 432 __u32 device_state; /* VFIO device state */ 433 #define VFIO_DEVICE_STATE_V1_STOP (0) 434 #define VFIO_DEVICE_STATE_V1_RUNNING (1 << 0) 435 #define VFIO_DEVICE_STATE_V1_SAVING (1 << 1) 436 #define VFIO_DEVICE_STATE_V1_RESUMING (1 << 2) 437 #define VFIO_DEVICE_STATE_MASK (VFIO_DEVICE_STATE_V1_RUNNING | \ 438 VFIO_DEVICE_STATE_V1_SAVING | \ 439 VFIO_DEVICE_STATE_V1_RESUMING) 440 441 #define VFIO_DEVICE_STATE_VALID(state) \ 442 (state & VFIO_DEVICE_STATE_V1_RESUMING ? \ 443 (state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_V1_RESUMING : 1) 444 445 #define VFIO_DEVICE_STATE_IS_ERROR(state) \ 446 ((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_V1_SAVING | \ 447 VFIO_DEVICE_STATE_V1_RESUMING)) 448 449 #define VFIO_DEVICE_STATE_SET_ERROR(state) \ 450 ((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_STATE_V1_SAVING | \ 451 VFIO_DEVICE_STATE_V1_RESUMING) 452 453 __u32 reserved; 454 __aligned_u64 pending_bytes; 455 __aligned_u64 data_offset; 456 __aligned_u64 data_size; 457 }; 458 459 /* 460 * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped 461 * which allows direct access to non-MSIX registers which happened to be within 462 * the same system page. 463 * 464 * Even though the userspace gets direct access to the MSIX data, the existing 465 * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration. 466 */ 467 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE 3 468 469 /* 470 * Capability with compressed real address (aka SSA - small system address) 471 * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing 472 * and by the userspace to associate a NVLink bridge with a GPU. 473 * 474 * Deprecated, capability no longer provided 475 */ 476 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT 4 477 478 struct vfio_region_info_cap_nvlink2_ssatgt { 479 struct vfio_info_cap_header header; 480 __aligned_u64 tgt; 481 }; 482 483 /* 484 * Capability with an NVLink link speed. The value is read by 485 * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed" 486 * property in the device tree. The value is fixed in the hardware 487 * and failing to provide the correct value results in the link 488 * not working with no indication from the driver why. 489 * 490 * Deprecated, capability no longer provided 491 */ 492 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD 5 493 494 struct vfio_region_info_cap_nvlink2_lnkspd { 495 struct vfio_info_cap_header header; 496 __u32 link_speed; 497 __u32 __pad; 498 }; 499 500 /** 501 * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9, 502 * struct vfio_irq_info) 503 * 504 * Retrieve information about a device IRQ. Caller provides 505 * struct vfio_irq_info with index value set. Caller sets argsz. 506 * Implementation of IRQ mapping is bus driver specific. Indexes 507 * using multiple IRQs are primarily intended to support MSI-like 508 * interrupt blocks. Zero count irq blocks may be used to describe 509 * unimplemented interrupt types. 510 * 511 * The EVENTFD flag indicates the interrupt index supports eventfd based 512 * signaling. 513 * 514 * The MASKABLE flags indicates the index supports MASK and UNMASK 515 * actions described below. 516 * 517 * AUTOMASKED indicates that after signaling, the interrupt line is 518 * automatically masked by VFIO and the user needs to unmask the line 519 * to receive new interrupts. This is primarily intended to distinguish 520 * level triggered interrupts. 521 * 522 * The NORESIZE flag indicates that the interrupt lines within the index 523 * are setup as a set and new subindexes cannot be enabled without first 524 * disabling the entire index. This is used for interrupts like PCI MSI 525 * and MSI-X where the driver may only use a subset of the available 526 * indexes, but VFIO needs to enable a specific number of vectors 527 * upfront. In the case of MSI-X, where the user can enable MSI-X and 528 * then add and unmask vectors, it's up to userspace to make the decision 529 * whether to allocate the maximum supported number of vectors or tear 530 * down setup and incrementally increase the vectors as each is enabled. 531 * Absence of the NORESIZE flag indicates that vectors can be enabled 532 * and disabled dynamically without impacting other vectors within the 533 * index. 534 */ 535 struct vfio_irq_info { 536 __u32 argsz; 537 __u32 flags; 538 #define VFIO_IRQ_INFO_EVENTFD (1 << 0) 539 #define VFIO_IRQ_INFO_MASKABLE (1 << 1) 540 #define VFIO_IRQ_INFO_AUTOMASKED (1 << 2) 541 #define VFIO_IRQ_INFO_NORESIZE (1 << 3) 542 __u32 index; /* IRQ index */ 543 __u32 count; /* Number of IRQs within this index */ 544 }; 545 #define VFIO_DEVICE_GET_IRQ_INFO _IO(VFIO_TYPE, VFIO_BASE + 9) 546 547 /** 548 * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set) 549 * 550 * Set signaling, masking, and unmasking of interrupts. Caller provides 551 * struct vfio_irq_set with all fields set. 'start' and 'count' indicate 552 * the range of subindexes being specified. 553 * 554 * The DATA flags specify the type of data provided. If DATA_NONE, the 555 * operation performs the specified action immediately on the specified 556 * interrupt(s). For example, to unmask AUTOMASKED interrupt [0,0]: 557 * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1. 558 * 559 * DATA_BOOL allows sparse support for the same on arrays of interrupts. 560 * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]): 561 * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3, 562 * data = {1,0,1} 563 * 564 * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd. 565 * A value of -1 can be used to either de-assign interrupts if already 566 * assigned or skip un-assigned interrupts. For example, to set an eventfd 567 * to be trigger for interrupts [0,0] and [0,2]: 568 * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3, 569 * data = {fd1, -1, fd2} 570 * If index [0,1] is previously set, two count = 1 ioctls calls would be 571 * required to set [0,0] and [0,2] without changing [0,1]. 572 * 573 * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used 574 * with ACTION_TRIGGER to perform kernel level interrupt loopback testing 575 * from userspace (ie. simulate hardware triggering). 576 * 577 * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER 578 * enables the interrupt index for the device. Individual subindex interrupts 579 * can be disabled using the -1 value for DATA_EVENTFD or the index can be 580 * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0. 581 * 582 * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while 583 * ACTION_TRIGGER specifies kernel->user signaling. 584 */ 585 struct vfio_irq_set { 586 __u32 argsz; 587 __u32 flags; 588 #define VFIO_IRQ_SET_DATA_NONE (1 << 0) /* Data not present */ 589 #define VFIO_IRQ_SET_DATA_BOOL (1 << 1) /* Data is bool (u8) */ 590 #define VFIO_IRQ_SET_DATA_EVENTFD (1 << 2) /* Data is eventfd (s32) */ 591 #define VFIO_IRQ_SET_ACTION_MASK (1 << 3) /* Mask interrupt */ 592 #define VFIO_IRQ_SET_ACTION_UNMASK (1 << 4) /* Unmask interrupt */ 593 #define VFIO_IRQ_SET_ACTION_TRIGGER (1 << 5) /* Trigger interrupt */ 594 __u32 index; 595 __u32 start; 596 __u32 count; 597 __u8 data[]; 598 }; 599 #define VFIO_DEVICE_SET_IRQS _IO(VFIO_TYPE, VFIO_BASE + 10) 600 601 #define VFIO_IRQ_SET_DATA_TYPE_MASK (VFIO_IRQ_SET_DATA_NONE | \ 602 VFIO_IRQ_SET_DATA_BOOL | \ 603 VFIO_IRQ_SET_DATA_EVENTFD) 604 #define VFIO_IRQ_SET_ACTION_TYPE_MASK (VFIO_IRQ_SET_ACTION_MASK | \ 605 VFIO_IRQ_SET_ACTION_UNMASK | \ 606 VFIO_IRQ_SET_ACTION_TRIGGER) 607 /** 608 * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11) 609 * 610 * Reset a device. 611 */ 612 #define VFIO_DEVICE_RESET _IO(VFIO_TYPE, VFIO_BASE + 11) 613 614 /* 615 * The VFIO-PCI bus driver makes use of the following fixed region and 616 * IRQ index mapping. Unimplemented regions return a size of zero. 617 * Unimplemented IRQ types return a count of zero. 618 */ 619 620 enum { 621 VFIO_PCI_BAR0_REGION_INDEX, 622 VFIO_PCI_BAR1_REGION_INDEX, 623 VFIO_PCI_BAR2_REGION_INDEX, 624 VFIO_PCI_BAR3_REGION_INDEX, 625 VFIO_PCI_BAR4_REGION_INDEX, 626 VFIO_PCI_BAR5_REGION_INDEX, 627 VFIO_PCI_ROM_REGION_INDEX, 628 VFIO_PCI_CONFIG_REGION_INDEX, 629 /* 630 * Expose VGA regions defined for PCI base class 03, subclass 00. 631 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df 632 * as well as the MMIO range 0xa0000 to 0xbffff. Each implemented 633 * range is found at it's identity mapped offset from the region 634 * offset, for example 0x3b0 is region_info.offset + 0x3b0. Areas 635 * between described ranges are unimplemented. 636 */ 637 VFIO_PCI_VGA_REGION_INDEX, 638 VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */ 639 /* device specific cap to define content. */ 640 }; 641 642 enum { 643 VFIO_PCI_INTX_IRQ_INDEX, 644 VFIO_PCI_MSI_IRQ_INDEX, 645 VFIO_PCI_MSIX_IRQ_INDEX, 646 VFIO_PCI_ERR_IRQ_INDEX, 647 VFIO_PCI_REQ_IRQ_INDEX, 648 VFIO_PCI_NUM_IRQS 649 }; 650 651 /* 652 * The vfio-ccw bus driver makes use of the following fixed region and 653 * IRQ index mapping. Unimplemented regions return a size of zero. 654 * Unimplemented IRQ types return a count of zero. 655 */ 656 657 enum { 658 VFIO_CCW_CONFIG_REGION_INDEX, 659 VFIO_CCW_NUM_REGIONS 660 }; 661 662 enum { 663 VFIO_CCW_IO_IRQ_INDEX, 664 VFIO_CCW_CRW_IRQ_INDEX, 665 VFIO_CCW_REQ_IRQ_INDEX, 666 VFIO_CCW_NUM_IRQS 667 }; 668 669 /* 670 * The vfio-ap bus driver makes use of the following IRQ index mapping. 671 * Unimplemented IRQ types return a count of zero. 672 */ 673 enum { 674 VFIO_AP_REQ_IRQ_INDEX, 675 VFIO_AP_CFG_CHG_IRQ_INDEX, 676 VFIO_AP_NUM_IRQS 677 }; 678 679 /** 680 * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12, 681 * struct vfio_pci_hot_reset_info) 682 * 683 * This command is used to query the affected devices in the hot reset for 684 * a given device. 685 * 686 * This command always reports the segment, bus, and devfn information for 687 * each affected device, and selectively reports the group_id or devid per 688 * the way how the calling device is opened. 689 * 690 * - If the calling device is opened via the traditional group/container 691 * API, group_id is reported. User should check if it has owned all 692 * the affected devices and provides a set of group fds to prove the 693 * ownership in VFIO_DEVICE_PCI_HOT_RESET ioctl. 694 * 695 * - If the calling device is opened as a cdev, devid is reported. 696 * Flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set to indicate this 697 * data type. All the affected devices should be represented in 698 * the dev_set, ex. bound to a vfio driver, and also be owned by 699 * this interface which is determined by the following conditions: 700 * 1) Has a valid devid within the iommufd_ctx of the calling device. 701 * Ownership cannot be determined across separate iommufd_ctx and 702 * the cdev calling conventions do not support a proof-of-ownership 703 * model as provided in the legacy group interface. In this case 704 * valid devid with value greater than zero is provided in the return 705 * structure. 706 * 2) Does not have a valid devid within the iommufd_ctx of the calling 707 * device, but belongs to the same IOMMU group as the calling device 708 * or another opened device that has a valid devid within the 709 * iommufd_ctx of the calling device. This provides implicit ownership 710 * for devices within the same DMA isolation context. In this case 711 * the devid value of VFIO_PCI_DEVID_OWNED is provided in the return 712 * structure. 713 * 714 * A devid value of VFIO_PCI_DEVID_NOT_OWNED is provided in the return 715 * structure for affected devices where device is NOT represented in the 716 * dev_set or ownership is not available. Such devices prevent the use 717 * of VFIO_DEVICE_PCI_HOT_RESET ioctl outside of the proof-of-ownership 718 * calling conventions (ie. via legacy group accessed devices). Flag 719 * VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED would be set when all the 720 * affected devices are represented in the dev_set and also owned by 721 * the user. This flag is available only when 722 * flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set, otherwise reserved. 723 * When set, user could invoke VFIO_DEVICE_PCI_HOT_RESET with a zero 724 * length fd array on the calling device as the ownership is validated 725 * by iommufd_ctx. 726 * 727 * Return: 0 on success, -errno on failure: 728 * -enospc = insufficient buffer, -enodev = unsupported for device. 729 */ 730 struct vfio_pci_dependent_device { 731 union { 732 __u32 group_id; 733 __u32 devid; 734 #define VFIO_PCI_DEVID_OWNED 0 735 #define VFIO_PCI_DEVID_NOT_OWNED -1 736 }; 737 __u16 segment; 738 __u8 bus; 739 __u8 devfn; /* Use PCI_SLOT/PCI_FUNC */ 740 }; 741 742 struct vfio_pci_hot_reset_info { 743 __u32 argsz; 744 __u32 flags; 745 #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID (1 << 0) 746 #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED (1 << 1) 747 __u32 count; 748 struct vfio_pci_dependent_device devices[]; 749 }; 750 751 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 752 753 /** 754 * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13, 755 * struct vfio_pci_hot_reset) 756 * 757 * A PCI hot reset results in either a bus or slot reset which may affect 758 * other devices sharing the bus/slot. The calling user must have 759 * ownership of the full set of affected devices as determined by the 760 * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO ioctl. 761 * 762 * When called on a device file descriptor acquired through the vfio 763 * group interface, the user is required to provide proof of ownership 764 * of those affected devices via the group_fds array in struct 765 * vfio_pci_hot_reset. 766 * 767 * When called on a direct cdev opened vfio device, the flags field of 768 * struct vfio_pci_hot_reset_info reports the ownership status of the 769 * affected devices and this ioctl must be called with an empty group_fds 770 * array. See above INFO ioctl definition for ownership requirements. 771 * 772 * Mixed usage of legacy groups and cdevs across the set of affected 773 * devices is not supported. 774 * 775 * Return: 0 on success, -errno on failure. 776 */ 777 struct vfio_pci_hot_reset { 778 __u32 argsz; 779 __u32 flags; 780 __u32 count; 781 __s32 group_fds[]; 782 }; 783 784 #define VFIO_DEVICE_PCI_HOT_RESET _IO(VFIO_TYPE, VFIO_BASE + 13) 785 786 /** 787 * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14, 788 * struct vfio_device_query_gfx_plane) 789 * 790 * Set the drm_plane_type and flags, then retrieve the gfx plane info. 791 * 792 * flags supported: 793 * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set 794 * to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no 795 * support for dma-buf. 796 * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set 797 * to ask if the mdev supports region. 0 on support, -EINVAL on no 798 * support for region. 799 * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set 800 * with each call to query the plane info. 801 * - Others are invalid and return -EINVAL. 802 * 803 * Note: 804 * 1. Plane could be disabled by guest. In that case, success will be 805 * returned with zero-initialized drm_format, size, width and height 806 * fields. 807 * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available 808 * 809 * Return: 0 on success, -errno on other failure. 810 */ 811 struct vfio_device_gfx_plane_info { 812 __u32 argsz; 813 __u32 flags; 814 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0) 815 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1) 816 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2) 817 /* in */ 818 __u32 drm_plane_type; /* type of plane: DRM_PLANE_TYPE_* */ 819 /* out */ 820 __u32 drm_format; /* drm format of plane */ 821 __aligned_u64 drm_format_mod; /* tiled mode */ 822 __u32 width; /* width of plane */ 823 __u32 height; /* height of plane */ 824 __u32 stride; /* stride of plane */ 825 __u32 size; /* size of plane in bytes, align on page*/ 826 __u32 x_pos; /* horizontal position of cursor plane */ 827 __u32 y_pos; /* vertical position of cursor plane*/ 828 __u32 x_hot; /* horizontal position of cursor hotspot */ 829 __u32 y_hot; /* vertical position of cursor hotspot */ 830 union { 831 __u32 region_index; /* region index */ 832 __u32 dmabuf_id; /* dma-buf id */ 833 }; 834 __u32 reserved; 835 }; 836 837 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14) 838 839 /** 840 * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32) 841 * 842 * Return a new dma-buf file descriptor for an exposed guest framebuffer 843 * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_ 844 * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer. 845 */ 846 847 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15) 848 849 /** 850 * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16, 851 * struct vfio_device_ioeventfd) 852 * 853 * Perform a write to the device at the specified device fd offset, with 854 * the specified data and width when the provided eventfd is triggered. 855 * vfio bus drivers may not support this for all regions, for all widths, 856 * or at all. vfio-pci currently only enables support for BAR regions, 857 * excluding the MSI-X vector table. 858 * 859 * Return: 0 on success, -errno on failure. 860 */ 861 struct vfio_device_ioeventfd { 862 __u32 argsz; 863 __u32 flags; 864 #define VFIO_DEVICE_IOEVENTFD_8 (1 << 0) /* 1-byte write */ 865 #define VFIO_DEVICE_IOEVENTFD_16 (1 << 1) /* 2-byte write */ 866 #define VFIO_DEVICE_IOEVENTFD_32 (1 << 2) /* 4-byte write */ 867 #define VFIO_DEVICE_IOEVENTFD_64 (1 << 3) /* 8-byte write */ 868 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK (0xf) 869 __aligned_u64 offset; /* device fd offset of write */ 870 __aligned_u64 data; /* data to be written */ 871 __s32 fd; /* -1 for de-assignment */ 872 __u32 reserved; 873 }; 874 875 #define VFIO_DEVICE_IOEVENTFD _IO(VFIO_TYPE, VFIO_BASE + 16) 876 877 /** 878 * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17, 879 * struct vfio_device_feature) 880 * 881 * Get, set, or probe feature data of the device. The feature is selected 882 * using the FEATURE_MASK portion of the flags field. Support for a feature 883 * can be probed by setting both the FEATURE_MASK and PROBE bits. A probe 884 * may optionally include the GET and/or SET bits to determine read vs write 885 * access of the feature respectively. Probing a feature will return success 886 * if the feature is supported and all of the optionally indicated GET/SET 887 * methods are supported. The format of the data portion of the structure is 888 * specific to the given feature. The data portion is not required for 889 * probing. GET and SET are mutually exclusive, except for use with PROBE. 890 * 891 * Return 0 on success, -errno on failure. 892 */ 893 struct vfio_device_feature { 894 __u32 argsz; 895 __u32 flags; 896 #define VFIO_DEVICE_FEATURE_MASK (0xffff) /* 16-bit feature index */ 897 #define VFIO_DEVICE_FEATURE_GET (1 << 16) /* Get feature into data[] */ 898 #define VFIO_DEVICE_FEATURE_SET (1 << 17) /* Set feature from data[] */ 899 #define VFIO_DEVICE_FEATURE_PROBE (1 << 18) /* Probe feature support */ 900 __u8 data[]; 901 }; 902 903 #define VFIO_DEVICE_FEATURE _IO(VFIO_TYPE, VFIO_BASE + 17) 904 905 /* 906 * VFIO_DEVICE_BIND_IOMMUFD - _IOR(VFIO_TYPE, VFIO_BASE + 18, 907 * struct vfio_device_bind_iommufd) 908 * @argsz: User filled size of this data. 909 * @flags: Must be 0 or a bit flags of VFIO_DEVICE_BIND_* 910 * @iommufd: iommufd to bind. 911 * @out_devid: The device id generated by this bind. devid is a handle for 912 * this device/iommufd bond and can be used in IOMMUFD commands. 913 * @token_uuid_ptr: Valid if VFIO_DEVICE_BIND_FLAG_TOKEN. Points to a 16 byte 914 * UUID in the same format as VFIO_DEVICE_FEATURE_PCI_VF_TOKEN. 915 * 916 * Bind a vfio_device to the specified iommufd. 917 * 918 * User is restricted from accessing the device before the binding operation 919 * is completed. Only allowed on cdev fds. 920 * 921 * Unbind is automatically conducted when device fd is closed. 922 * 923 * A token is sometimes required to open the device, unless this is known to be 924 * needed VFIO_DEVICE_BIND_FLAG_TOKEN should not be set and token_uuid_ptr is 925 * ignored. The only case today is a PF/VF relationship where the VF bind must 926 * be provided the same token as VFIO_DEVICE_FEATURE_PCI_VF_TOKEN provided to 927 * the PF. 928 * 929 * Return: 0 on success, -errno on failure. 930 */ 931 struct vfio_device_bind_iommufd { 932 __u32 argsz; 933 __u32 flags; 934 #define VFIO_DEVICE_BIND_FLAG_TOKEN (1 << 0) 935 __s32 iommufd; 936 __u32 out_devid; 937 __aligned_u64 token_uuid_ptr; 938 }; 939 940 #define VFIO_DEVICE_BIND_IOMMUFD _IO(VFIO_TYPE, VFIO_BASE + 18) 941 942 /* 943 * VFIO_DEVICE_ATTACH_IOMMUFD_PT - _IOW(VFIO_TYPE, VFIO_BASE + 19, 944 * struct vfio_device_attach_iommufd_pt) 945 * @argsz: User filled size of this data. 946 * @flags: Flags for attach. 947 * @pt_id: Input the target id which can represent an ioas or a hwpt 948 * allocated via iommufd subsystem. 949 * Output the input ioas id or the attached hwpt id which could 950 * be the specified hwpt itself or a hwpt automatically created 951 * for the specified ioas by kernel during the attachment. 952 * @pasid: The pasid to be attached, only meaningful when 953 * VFIO_DEVICE_ATTACH_PASID is set in @flags 954 * 955 * Associate the device with an address space within the bound iommufd. 956 * Undo by VFIO_DEVICE_DETACH_IOMMUFD_PT or device fd close. This is only 957 * allowed on cdev fds. 958 * 959 * If a vfio device or a pasid of this device is currently attached to a valid 960 * hw_pagetable (hwpt), without doing a VFIO_DEVICE_DETACH_IOMMUFD_PT, a second 961 * VFIO_DEVICE_ATTACH_IOMMUFD_PT ioctl passing in another hwpt id is allowed. 962 * This action, also known as a hw_pagetable replacement, will replace the 963 * currently attached hwpt of the device or the pasid of this device with a new 964 * hwpt corresponding to the given pt_id. 965 * 966 * Return: 0 on success, -errno on failure. 967 * 968 * When a device is resetting, -EBUSY will be returned to reject any concurrent 969 * attachment to the resetting device itself or any sibling device in the IOMMU 970 * group having the resetting device. 971 */ 972 struct vfio_device_attach_iommufd_pt { 973 __u32 argsz; 974 __u32 flags; 975 #define VFIO_DEVICE_ATTACH_PASID (1 << 0) 976 __u32 pt_id; 977 __u32 pasid; 978 }; 979 980 #define VFIO_DEVICE_ATTACH_IOMMUFD_PT _IO(VFIO_TYPE, VFIO_BASE + 19) 981 982 /* 983 * VFIO_DEVICE_DETACH_IOMMUFD_PT - _IOW(VFIO_TYPE, VFIO_BASE + 20, 984 * struct vfio_device_detach_iommufd_pt) 985 * @argsz: User filled size of this data. 986 * @flags: Flags for detach. 987 * @pasid: The pasid to be detached, only meaningful when 988 * VFIO_DEVICE_DETACH_PASID is set in @flags 989 * 990 * Remove the association of the device or a pasid of the device and its current 991 * associated address space. After it, the device or the pasid should be in a 992 * blocking DMA state. This is only allowed on cdev fds. 993 * 994 * Return: 0 on success, -errno on failure. 995 */ 996 struct vfio_device_detach_iommufd_pt { 997 __u32 argsz; 998 __u32 flags; 999 #define VFIO_DEVICE_DETACH_PASID (1 << 0) 1000 __u32 pasid; 1001 }; 1002 1003 #define VFIO_DEVICE_DETACH_IOMMUFD_PT _IO(VFIO_TYPE, VFIO_BASE + 20) 1004 1005 /* 1006 * Provide support for setting a PCI VF Token, which is used as a shared 1007 * secret between PF and VF drivers. This feature may only be set on a 1008 * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing 1009 * open VFs. Data provided when setting this feature is a 16-byte array 1010 * (__u8 b[16]), representing a UUID. 1011 */ 1012 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN (0) 1013 1014 /* 1015 * Indicates the device can support the migration API through 1016 * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and 1017 * ERROR states are always supported. Support for additional states is 1018 * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be 1019 * set. 1020 * 1021 * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and 1022 * RESUMING are supported. 1023 * 1024 * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P 1025 * is supported in addition to the STOP_COPY states. 1026 * 1027 * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that 1028 * PRE_COPY is supported in addition to the STOP_COPY states. 1029 * 1030 * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY 1031 * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported 1032 * in addition to the STOP_COPY states. 1033 * 1034 * Other combinations of flags have behavior to be defined in the future. 1035 */ 1036 struct vfio_device_feature_migration { 1037 __aligned_u64 flags; 1038 #define VFIO_MIGRATION_STOP_COPY (1 << 0) 1039 #define VFIO_MIGRATION_P2P (1 << 1) 1040 #define VFIO_MIGRATION_PRE_COPY (1 << 2) 1041 }; 1042 #define VFIO_DEVICE_FEATURE_MIGRATION 1 1043 1044 /* 1045 * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO 1046 * device. The new state is supplied in device_state, see enum 1047 * vfio_device_mig_state for details 1048 * 1049 * The kernel migration driver must fully transition the device to the new state 1050 * value before the operation returns to the user. 1051 * 1052 * The kernel migration driver must not generate asynchronous device state 1053 * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET 1054 * ioctl as described above. 1055 * 1056 * If this function fails then current device_state may be the original 1057 * operating state or some other state along the combination transition path. 1058 * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt 1059 * to return to the original state, or attempt to return to some other state 1060 * such as RUNNING or STOP. 1061 * 1062 * If the new_state starts a new data transfer session then the FD associated 1063 * with that session is returned in data_fd. The user is responsible to close 1064 * this FD when it is finished. The user must consider the migration data stream 1065 * carried over the FD to be opaque and must preserve the byte order of the 1066 * stream. The user is not required to preserve buffer segmentation when writing 1067 * the data stream during the RESUMING operation. 1068 * 1069 * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO 1070 * device, data_fd will be -1. 1071 */ 1072 struct vfio_device_feature_mig_state { 1073 __u32 device_state; /* From enum vfio_device_mig_state */ 1074 __s32 data_fd; 1075 }; 1076 #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2 1077 1078 /* 1079 * The device migration Finite State Machine is described by the enum 1080 * vfio_device_mig_state. Some of the FSM arcs will create a migration data 1081 * transfer session by returning a FD, in this case the migration data will 1082 * flow over the FD using read() and write() as discussed below. 1083 * 1084 * There are 5 states to support VFIO_MIGRATION_STOP_COPY: 1085 * RUNNING - The device is running normally 1086 * STOP - The device does not change the internal or external state 1087 * STOP_COPY - The device internal state can be read out 1088 * RESUMING - The device is stopped and is loading a new internal state 1089 * ERROR - The device has failed and must be reset 1090 * 1091 * And optional states to support VFIO_MIGRATION_P2P: 1092 * RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA 1093 * And VFIO_MIGRATION_PRE_COPY: 1094 * PRE_COPY - The device is running normally but tracking internal state 1095 * changes 1096 * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY: 1097 * PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA 1098 * 1099 * The FSM takes actions on the arcs between FSM states. The driver implements 1100 * the following behavior for the FSM arcs: 1101 * 1102 * RUNNING_P2P -> STOP 1103 * STOP_COPY -> STOP 1104 * While in STOP the device must stop the operation of the device. The device 1105 * must not generate interrupts, DMA, or any other change to external state. 1106 * It must not change its internal state. When stopped the device and kernel 1107 * migration driver must accept and respond to interaction to support external 1108 * subsystems in the STOP state, for example PCI MSI-X and PCI config space. 1109 * Failure by the user to restrict device access while in STOP must not result 1110 * in error conditions outside the user context (ex. host system faults). 1111 * 1112 * The STOP_COPY arc will terminate a data transfer session. 1113 * 1114 * RESUMING -> STOP 1115 * Leaving RESUMING terminates a data transfer session and indicates the 1116 * device should complete processing of the data delivered by write(). The 1117 * kernel migration driver should complete the incorporation of data written 1118 * to the data transfer FD into the device internal state and perform 1119 * final validity and consistency checking of the new device state. If the 1120 * user provided data is found to be incomplete, inconsistent, or otherwise 1121 * invalid, the migration driver must fail the SET_STATE ioctl and 1122 * optionally go to the ERROR state as described below. 1123 * 1124 * While in STOP the device has the same behavior as other STOP states 1125 * described above. 1126 * 1127 * To abort a RESUMING session the device must be reset. 1128 * 1129 * PRE_COPY -> RUNNING 1130 * RUNNING_P2P -> RUNNING 1131 * While in RUNNING the device is fully operational, the device may generate 1132 * interrupts, DMA, respond to MMIO, all vfio device regions are functional, 1133 * and the device may advance its internal state. 1134 * 1135 * The PRE_COPY arc will terminate a data transfer session. 1136 * 1137 * PRE_COPY_P2P -> RUNNING_P2P 1138 * RUNNING -> RUNNING_P2P 1139 * STOP -> RUNNING_P2P 1140 * While in RUNNING_P2P the device is partially running in the P2P quiescent 1141 * state defined below. 1142 * 1143 * The PRE_COPY_P2P arc will terminate a data transfer session. 1144 * 1145 * RUNNING -> PRE_COPY 1146 * RUNNING_P2P -> PRE_COPY_P2P 1147 * STOP -> STOP_COPY 1148 * PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states 1149 * which share a data transfer session. Moving between these states alters 1150 * what is streamed in session, but does not terminate or otherwise affect 1151 * the associated fd. 1152 * 1153 * These arcs begin the process of saving the device state and will return a 1154 * new data_fd. The migration driver may perform actions such as enabling 1155 * dirty logging of device state when entering PRE_COPY or PER_COPY_P2P. 1156 * 1157 * Each arc does not change the device operation, the device remains 1158 * RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below 1159 * in PRE_COPY_P2P -> STOP_COPY. 1160 * 1161 * PRE_COPY -> PRE_COPY_P2P 1162 * Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above. 1163 * However, while in the PRE_COPY_P2P state, the device is partially running 1164 * in the P2P quiescent state defined below, like RUNNING_P2P. 1165 * 1166 * PRE_COPY_P2P -> PRE_COPY 1167 * This arc allows returning the device to a full RUNNING behavior while 1168 * continuing all the behaviors of PRE_COPY. 1169 * 1170 * PRE_COPY_P2P -> STOP_COPY 1171 * While in the STOP_COPY state the device has the same behavior as STOP 1172 * with the addition that the data transfers session continues to stream the 1173 * migration state. End of stream on the FD indicates the entire device 1174 * state has been transferred. 1175 * 1176 * The user should take steps to restrict access to vfio device regions while 1177 * the device is in STOP_COPY or risk corruption of the device migration data 1178 * stream. 1179 * 1180 * STOP -> RESUMING 1181 * Entering the RESUMING state starts a process of restoring the device state 1182 * and will return a new data_fd. The data stream fed into the data_fd should 1183 * be taken from the data transfer output of a single FD during saving from 1184 * a compatible device. The migration driver may alter/reset the internal 1185 * device state for this arc if required to prepare the device to receive the 1186 * migration data. 1187 * 1188 * STOP_COPY -> PRE_COPY 1189 * STOP_COPY -> PRE_COPY_P2P 1190 * These arcs are not permitted and return error if requested. Future 1191 * revisions of this API may define behaviors for these arcs, in this case 1192 * support will be discoverable by a new flag in 1193 * VFIO_DEVICE_FEATURE_MIGRATION. 1194 * 1195 * any -> ERROR 1196 * ERROR cannot be specified as a device state, however any transition request 1197 * can be failed with an errno return and may then move the device_state into 1198 * ERROR. In this case the device was unable to execute the requested arc and 1199 * was also unable to restore the device to any valid device_state. 1200 * To recover from ERROR VFIO_DEVICE_RESET must be used to return the 1201 * device_state back to RUNNING. 1202 * 1203 * The optional peer to peer (P2P) quiescent state is intended to be a quiescent 1204 * state for the device for the purposes of managing multiple devices within a 1205 * user context where peer-to-peer DMA between devices may be active. The 1206 * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating 1207 * any new P2P DMA transactions. If the device can identify P2P transactions 1208 * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration 1209 * driver must complete any such outstanding operations prior to completing the 1210 * FSM arc into a P2P state. For the purpose of specification the states 1211 * behave as though the device was fully running if not supported. Like while in 1212 * STOP or STOP_COPY the user must not touch the device, otherwise the state 1213 * can be exited. 1214 * 1215 * The remaining possible transitions are interpreted as combinations of the 1216 * above FSM arcs. As there are multiple paths through the FSM arcs the path 1217 * should be selected based on the following rules: 1218 * - Select the shortest path. 1219 * - The path cannot have saving group states as interior arcs, only 1220 * starting/end states. 1221 * Refer to vfio_mig_get_next_state() for the result of the algorithm. 1222 * 1223 * The automatic transit through the FSM arcs that make up the combination 1224 * transition is invisible to the user. When working with combination arcs the 1225 * user may see any step along the path in the device_state if SET_STATE 1226 * fails. When handling these types of errors users should anticipate future 1227 * revisions of this protocol using new states and those states becoming 1228 * visible in this case. 1229 * 1230 * The optional states cannot be used with SET_STATE if the device does not 1231 * support them. The user can discover if these states are supported by using 1232 * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can 1233 * avoid knowing about these optional states if the kernel driver supports them. 1234 * 1235 * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY 1236 * is not present. 1237 */ 1238 enum vfio_device_mig_state { 1239 VFIO_DEVICE_STATE_ERROR = 0, 1240 VFIO_DEVICE_STATE_STOP = 1, 1241 VFIO_DEVICE_STATE_RUNNING = 2, 1242 VFIO_DEVICE_STATE_STOP_COPY = 3, 1243 VFIO_DEVICE_STATE_RESUMING = 4, 1244 VFIO_DEVICE_STATE_RUNNING_P2P = 5, 1245 VFIO_DEVICE_STATE_PRE_COPY = 6, 1246 VFIO_DEVICE_STATE_PRE_COPY_P2P = 7, 1247 VFIO_DEVICE_STATE_NR, 1248 }; 1249 1250 /** 1251 * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21) 1252 * 1253 * This ioctl is used on the migration data FD in the precopy phase of the 1254 * migration data transfer. It returns an estimate of the current data sizes 1255 * remaining to be transferred. It allows the user to judge when it is 1256 * appropriate to leave PRE_COPY for STOP_COPY. 1257 * 1258 * This ioctl is valid only in PRE_COPY states and kernel driver should 1259 * return -EINVAL from any other migration state. 1260 * 1261 * The vfio_precopy_info data structure returned by this ioctl provides 1262 * estimates of data available from the device during the PRE_COPY states. 1263 * This estimate is split into two categories, initial_bytes and 1264 * dirty_bytes. 1265 * 1266 * The initial_bytes field indicates the amount of initial precopy 1267 * data available from the device. This field should have a non-zero initial 1268 * value and decrease as migration data is read from the device. 1269 * It is recommended to leave PRE_COPY for STOP_COPY only after this field 1270 * reaches zero. Leaving PRE_COPY earlier might make things slower. 1271 * 1272 * The dirty_bytes field tracks device state changes relative to data 1273 * previously retrieved. This field starts at zero and may increase as 1274 * the internal device state is modified or decrease as that modified 1275 * state is read from the device. 1276 * 1277 * Userspace may use the combination of these fields to estimate the 1278 * potential data size available during the PRE_COPY phases, as well as 1279 * trends relative to the rate the device is dirtying its internal 1280 * state, but these fields are not required to have any bearing relative 1281 * to the data size available during the STOP_COPY phase. 1282 * 1283 * Drivers have a lot of flexibility in when and what they transfer during the 1284 * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO. 1285 * 1286 * During pre-copy the migration data FD has a temporary "end of stream" that is 1287 * reached when both initial_bytes and dirty_byte are zero. For instance, this 1288 * may indicate that the device is idle and not currently dirtying any internal 1289 * state. When read() is done on this temporary end of stream the kernel driver 1290 * should return ENOMSG from read(). Userspace can wait for more data (which may 1291 * never come) by using poll. 1292 * 1293 * Once in STOP_COPY the migration data FD has a permanent end of stream 1294 * signaled in the usual way by read() always returning 0 and poll always 1295 * returning readable. ENOMSG may not be returned in STOP_COPY. 1296 * Support for this ioctl is mandatory if a driver claims to support 1297 * VFIO_MIGRATION_PRE_COPY. 1298 * 1299 * Return: 0 on success, -1 and errno set on failure. 1300 */ 1301 struct vfio_precopy_info { 1302 __u32 argsz; 1303 __u32 flags; 1304 __aligned_u64 initial_bytes; 1305 __aligned_u64 dirty_bytes; 1306 }; 1307 1308 #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21) 1309 1310 /* 1311 * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power 1312 * state with the platform-based power management. Device use of lower power 1313 * states depends on factors managed by the runtime power management core, 1314 * including system level support and coordinating support among dependent 1315 * devices. Enabling device low power entry does not guarantee lower power 1316 * usage by the device, nor is a mechanism provided through this feature to 1317 * know the current power state of the device. If any device access happens 1318 * (either from the host or through the vfio uAPI) when the device is in the 1319 * low power state, then the host will move the device out of the low power 1320 * state as necessary prior to the access. Once the access is completed, the 1321 * device may re-enter the low power state. For single shot low power support 1322 * with wake-up notification, see 1323 * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below. Access to mmap'd 1324 * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after 1325 * calling LOW_POWER_EXIT. 1326 */ 1327 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3 1328 1329 /* 1330 * This device feature has the same behavior as 1331 * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user 1332 * provides an eventfd for wake-up notification. When the device moves out of 1333 * the low power state for the wake-up, the host will not allow the device to 1334 * re-enter a low power state without a subsequent user call to one of the low 1335 * power entry device feature IOCTLs. Access to mmap'd device regions is 1336 * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the 1337 * low power exit. The low power exit can happen either through LOW_POWER_EXIT 1338 * or through any other access (where the wake-up notification has been 1339 * generated). The access to mmap'd device regions will not trigger low power 1340 * exit. 1341 * 1342 * The notification through the provided eventfd will be generated only when 1343 * the device has entered and is resumed from a low power state after 1344 * calling this device feature IOCTL. A device that has not entered low power 1345 * state, as managed through the runtime power management core, will not 1346 * generate a notification through the provided eventfd on access. Calling the 1347 * LOW_POWER_EXIT feature is optional in the case where notification has been 1348 * signaled on the provided eventfd that a resume from low power has occurred. 1349 */ 1350 struct vfio_device_low_power_entry_with_wakeup { 1351 __s32 wakeup_eventfd; 1352 __u32 reserved; 1353 }; 1354 1355 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4 1356 1357 /* 1358 * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as 1359 * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or 1360 * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features. 1361 * This device feature IOCTL may itself generate a wakeup eventfd notification 1362 * in the latter case if the device had previously entered a low power state. 1363 */ 1364 #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5 1365 1366 /* 1367 * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging. 1368 * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports 1369 * DMA logging. 1370 * 1371 * DMA logging allows a device to internally record what DMAs the device is 1372 * initiating and report them back to userspace. It is part of the VFIO 1373 * migration infrastructure that allows implementing dirty page tracking 1374 * during the pre copy phase of live migration. Only DMA WRITEs are logged, 1375 * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. 1376 * 1377 * When DMA logging is started a range of IOVAs to monitor is provided and the 1378 * device can optimize its logging to cover only the IOVA range given. Each 1379 * DMA that the device initiates inside the range will be logged by the device 1380 * for later retrieval. 1381 * 1382 * page_size is an input that hints what tracking granularity the device 1383 * should try to achieve. If the device cannot do the hinted page size then 1384 * it's the driver choice which page size to pick based on its support. 1385 * On output the device will return the page size it selected. 1386 * 1387 * ranges is a pointer to an array of 1388 * struct vfio_device_feature_dma_logging_range. 1389 * 1390 * The core kernel code guarantees to support by minimum num_ranges that fit 1391 * into a single kernel page. User space can try higher values but should give 1392 * up if the above can't be achieved as of some driver limitations. 1393 * 1394 * A single call to start device DMA logging can be issued and a matching stop 1395 * should follow at the end. Another start is not allowed in the meantime. 1396 */ 1397 struct vfio_device_feature_dma_logging_control { 1398 __aligned_u64 page_size; 1399 __u32 num_ranges; 1400 __u32 __reserved; 1401 __aligned_u64 ranges; 1402 }; 1403 1404 struct vfio_device_feature_dma_logging_range { 1405 __aligned_u64 iova; 1406 __aligned_u64 length; 1407 }; 1408 1409 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6 1410 1411 /* 1412 * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started 1413 * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START 1414 */ 1415 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7 1416 1417 /* 1418 * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log 1419 * 1420 * Query the device's DMA log for written pages within the given IOVA range. 1421 * During querying the log is cleared for the IOVA range. 1422 * 1423 * bitmap is a pointer to an array of u64s that will hold the output bitmap 1424 * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits 1425 * is given by: 1426 * bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64)) 1427 * 1428 * The input page_size can be any power of two value and does not have to 1429 * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver 1430 * will format its internal logging to match the reporting page size, possibly 1431 * by replicating bits if the internal page size is lower than requested. 1432 * 1433 * The LOGGING_REPORT will only set bits in the bitmap and never clear or 1434 * perform any initialization of the user provided bitmap. 1435 * 1436 * If any error is returned userspace should assume that the dirty log is 1437 * corrupted. Error recovery is to consider all memory dirty and try to 1438 * restart the dirty tracking, or to abort/restart the whole migration. 1439 * 1440 * If DMA logging is not enabled, an error will be returned. 1441 * 1442 */ 1443 struct vfio_device_feature_dma_logging_report { 1444 __aligned_u64 iova; 1445 __aligned_u64 length; 1446 __aligned_u64 page_size; 1447 __aligned_u64 bitmap; 1448 }; 1449 1450 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8 1451 1452 /* 1453 * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will 1454 * be required to complete stop copy. 1455 * 1456 * Note: Can be called on each device state. 1457 */ 1458 1459 struct vfio_device_feature_mig_data_size { 1460 __aligned_u64 stop_copy_length; 1461 }; 1462 1463 #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9 1464 1465 /** 1466 * Upon VFIO_DEVICE_FEATURE_SET, set or clear the BUS mastering for the device 1467 * based on the operation specified in op flag. 1468 * 1469 * The functionality is incorporated for devices that needs bus master control, 1470 * but the in-band device interface lacks the support. Consequently, it is not 1471 * applicable to PCI devices, as bus master control for PCI devices is managed 1472 * in-band through the configuration space. At present, this feature is supported 1473 * only for CDX devices. 1474 * When the device's BUS MASTER setting is configured as CLEAR, it will result in 1475 * blocking all incoming DMA requests from the device. On the other hand, configuring 1476 * the device's BUS MASTER setting as SET (enable) will grant the device the 1477 * capability to perform DMA to the host memory. 1478 */ 1479 struct vfio_device_feature_bus_master { 1480 __u32 op; 1481 #define VFIO_DEVICE_FEATURE_CLEAR_MASTER 0 /* Clear Bus Master */ 1482 #define VFIO_DEVICE_FEATURE_SET_MASTER 1 /* Set Bus Master */ 1483 }; 1484 #define VFIO_DEVICE_FEATURE_BUS_MASTER 10 1485 1486 /** 1487 * Upon VFIO_DEVICE_FEATURE_GET create a dma_buf fd for the 1488 * regions selected. 1489 * 1490 * open_flags are the typical flags passed to open(2), eg O_RDWR, O_CLOEXEC, 1491 * etc. offset/length specify a slice of the region to create the dmabuf from. 1492 * nr_ranges is the total number of (P2P DMA) ranges that comprise the dmabuf. 1493 * 1494 * flags should be 0. 1495 * 1496 * Return: The fd number on success, -1 and errno is set on failure. 1497 */ 1498 #define VFIO_DEVICE_FEATURE_DMA_BUF 11 1499 1500 struct vfio_region_dma_range { 1501 __u64 offset; 1502 __u64 length; 1503 }; 1504 1505 struct vfio_device_feature_dma_buf { 1506 __u32 region_index; 1507 __u32 open_flags; 1508 __u32 flags; 1509 __u32 nr_ranges; 1510 struct vfio_region_dma_range dma_ranges[] __counted_by(nr_ranges); 1511 }; 1512 1513 /* -------- API for Type1 VFIO IOMMU -------- */ 1514 1515 /** 1516 * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info) 1517 * 1518 * Retrieve information about the IOMMU object. Fills in provided 1519 * struct vfio_iommu_info. Caller sets argsz. 1520 * 1521 * XXX Should we do these by CHECK_EXTENSION too? 1522 */ 1523 struct vfio_iommu_type1_info { 1524 __u32 argsz; 1525 __u32 flags; 1526 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0) /* supported page sizes info */ 1527 #define VFIO_IOMMU_INFO_CAPS (1 << 1) /* Info supports caps */ 1528 __aligned_u64 iova_pgsizes; /* Bitmap of supported page sizes */ 1529 __u32 cap_offset; /* Offset within info struct of first cap */ 1530 __u32 pad; 1531 }; 1532 1533 /* 1534 * The IOVA capability allows to report the valid IOVA range(s) 1535 * excluding any non-relaxable reserved regions exposed by 1536 * devices attached to the container. Any DMA map attempt 1537 * outside the valid iova range will return error. 1538 * 1539 * The structures below define version 1 of this capability. 1540 */ 1541 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE 1 1542 1543 struct vfio_iova_range { 1544 __u64 start; 1545 __u64 end; 1546 }; 1547 1548 struct vfio_iommu_type1_info_cap_iova_range { 1549 struct vfio_info_cap_header header; 1550 __u32 nr_iovas; 1551 __u32 reserved; 1552 struct vfio_iova_range iova_ranges[]; 1553 }; 1554 1555 /* 1556 * The migration capability allows to report supported features for migration. 1557 * 1558 * The structures below define version 1 of this capability. 1559 * 1560 * The existence of this capability indicates that IOMMU kernel driver supports 1561 * dirty page logging. 1562 * 1563 * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty 1564 * page logging. 1565 * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap 1566 * size in bytes that can be used by user applications when getting the dirty 1567 * bitmap. 1568 */ 1569 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION 2 1570 1571 struct vfio_iommu_type1_info_cap_migration { 1572 struct vfio_info_cap_header header; 1573 __u32 flags; 1574 __u64 pgsize_bitmap; 1575 __u64 max_dirty_bitmap_size; /* in bytes */ 1576 }; 1577 1578 /* 1579 * The DMA available capability allows to report the current number of 1580 * simultaneously outstanding DMA mappings that are allowed. 1581 * 1582 * The structure below defines version 1 of this capability. 1583 * 1584 * avail: specifies the current number of outstanding DMA mappings allowed. 1585 */ 1586 #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3 1587 1588 struct vfio_iommu_type1_info_dma_avail { 1589 struct vfio_info_cap_header header; 1590 __u32 avail; 1591 }; 1592 1593 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 1594 1595 /** 1596 * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map) 1597 * 1598 * Map process virtual addresses to IO virtual addresses using the 1599 * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required. 1600 * 1601 * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr 1602 * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR. To 1603 * maintain memory consistency within the user application, the updated vaddr 1604 * must address the same memory object as originally mapped. Failure to do so 1605 * will result in user memory corruption and/or device misbehavior. iova and 1606 * size must match those in the original MAP_DMA call. Protection is not 1607 * changed, and the READ & WRITE flags must be 0. 1608 */ 1609 struct vfio_iommu_type1_dma_map { 1610 __u32 argsz; 1611 __u32 flags; 1612 #define VFIO_DMA_MAP_FLAG_READ (1 << 0) /* readable from device */ 1613 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1) /* writable from device */ 1614 #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2) 1615 __u64 vaddr; /* Process virtual address */ 1616 __u64 iova; /* IO virtual address */ 1617 __u64 size; /* Size of mapping (bytes) */ 1618 }; 1619 1620 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13) 1621 1622 struct vfio_bitmap { 1623 __u64 pgsize; /* page size for bitmap in bytes */ 1624 __u64 size; /* in bytes */ 1625 __u64 __user *data; /* one bit per page */ 1626 }; 1627 1628 /** 1629 * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14, 1630 * struct vfio_dma_unmap) 1631 * 1632 * Unmap IO virtual addresses using the provided struct vfio_dma_unmap. 1633 * Caller sets argsz. The actual unmapped size is returned in the size 1634 * field. No guarantee is made to the user that arbitrary unmaps of iova 1635 * or size different from those used in the original mapping call will 1636 * succeed. 1637 * 1638 * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap 1639 * before unmapping IO virtual addresses. When this flag is set, the user must 1640 * provide a struct vfio_bitmap in data[]. User must provide zero-allocated 1641 * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field. 1642 * A bit in the bitmap represents one page, of user provided page size in 1643 * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set 1644 * indicates that the page at that offset from iova is dirty. A Bitmap of the 1645 * pages in the range of unmapped size is returned in the user-provided 1646 * vfio_bitmap.data. 1647 * 1648 * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses. iova and size 1649 * must be 0. This cannot be combined with the get-dirty-bitmap flag. 1650 * 1651 * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host 1652 * virtual addresses in the iova range. DMA to already-mapped pages continues. 1653 * Groups may not be added to the container while any addresses are invalid. 1654 * This cannot be combined with the get-dirty-bitmap flag. 1655 */ 1656 struct vfio_iommu_type1_dma_unmap { 1657 __u32 argsz; 1658 __u32 flags; 1659 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0) 1660 #define VFIO_DMA_UNMAP_FLAG_ALL (1 << 1) 1661 #define VFIO_DMA_UNMAP_FLAG_VADDR (1 << 2) 1662 __u64 iova; /* IO virtual address */ 1663 __u64 size; /* Size of mapping (bytes) */ 1664 __u8 data[]; 1665 }; 1666 1667 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14) 1668 1669 /* 1670 * IOCTLs to enable/disable IOMMU container usage. 1671 * No parameters are supported. 1672 */ 1673 #define VFIO_IOMMU_ENABLE _IO(VFIO_TYPE, VFIO_BASE + 15) 1674 #define VFIO_IOMMU_DISABLE _IO(VFIO_TYPE, VFIO_BASE + 16) 1675 1676 /** 1677 * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17, 1678 * struct vfio_iommu_type1_dirty_bitmap) 1679 * IOCTL is used for dirty pages logging. 1680 * Caller should set flag depending on which operation to perform, details as 1681 * below: 1682 * 1683 * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs 1684 * the IOMMU driver to log pages that are dirtied or potentially dirtied by 1685 * the device; designed to be used when a migration is in progress. Dirty pages 1686 * are logged until logging is disabled by user application by calling the IOCTL 1687 * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag. 1688 * 1689 * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs 1690 * the IOMMU driver to stop logging dirtied pages. 1691 * 1692 * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set 1693 * returns the dirty pages bitmap for IOMMU container for a given IOVA range. 1694 * The user must specify the IOVA range and the pgsize through the structure 1695 * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface 1696 * supports getting a bitmap of the smallest supported pgsize only and can be 1697 * modified in future to get a bitmap of any specified supported pgsize. The 1698 * user must provide a zeroed memory area for the bitmap memory and specify its 1699 * size in bitmap.size. One bit is used to represent one page consecutively 1700 * starting from iova offset. The user should provide page size in bitmap.pgsize 1701 * field. A bit set in the bitmap indicates that the page at that offset from 1702 * iova is dirty. The caller must set argsz to a value including the size of 1703 * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the 1704 * actual bitmap. If dirty pages logging is not enabled, an error will be 1705 * returned. 1706 * 1707 * Only one of the flags _START, _STOP and _GET may be specified at a time. 1708 * 1709 */ 1710 struct vfio_iommu_type1_dirty_bitmap { 1711 __u32 argsz; 1712 __u32 flags; 1713 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START (1 << 0) 1714 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP (1 << 1) 1715 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP (1 << 2) 1716 __u8 data[]; 1717 }; 1718 1719 struct vfio_iommu_type1_dirty_bitmap_get { 1720 __u64 iova; /* IO virtual address */ 1721 __u64 size; /* Size of iova range */ 1722 struct vfio_bitmap bitmap; 1723 }; 1724 1725 #define VFIO_IOMMU_DIRTY_PAGES _IO(VFIO_TYPE, VFIO_BASE + 17) 1726 1727 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */ 1728 1729 /* 1730 * The SPAPR TCE DDW info struct provides the information about 1731 * the details of Dynamic DMA window capability. 1732 * 1733 * @pgsizes contains a page size bitmask, 4K/64K/16M are supported. 1734 * @max_dynamic_windows_supported tells the maximum number of windows 1735 * which the platform can create. 1736 * @levels tells the maximum number of levels in multi-level IOMMU tables; 1737 * this allows splitting a table into smaller chunks which reduces 1738 * the amount of physically contiguous memory required for the table. 1739 */ 1740 struct vfio_iommu_spapr_tce_ddw_info { 1741 __u64 pgsizes; /* Bitmap of supported page sizes */ 1742 __u32 max_dynamic_windows_supported; 1743 __u32 levels; 1744 }; 1745 1746 /* 1747 * The SPAPR TCE info struct provides the information about the PCI bus 1748 * address ranges available for DMA, these values are programmed into 1749 * the hardware so the guest has to know that information. 1750 * 1751 * The DMA 32 bit window start is an absolute PCI bus address. 1752 * The IOVA address passed via map/unmap ioctls are absolute PCI bus 1753 * addresses too so the window works as a filter rather than an offset 1754 * for IOVA addresses. 1755 * 1756 * Flags supported: 1757 * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows 1758 * (DDW) support is present. @ddw is only supported when DDW is present. 1759 */ 1760 struct vfio_iommu_spapr_tce_info { 1761 __u32 argsz; 1762 __u32 flags; 1763 #define VFIO_IOMMU_SPAPR_INFO_DDW (1 << 0) /* DDW supported */ 1764 __u32 dma32_window_start; /* 32 bit window start (bytes) */ 1765 __u32 dma32_window_size; /* 32 bit window size (bytes) */ 1766 struct vfio_iommu_spapr_tce_ddw_info ddw; 1767 }; 1768 1769 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 1770 1771 /* 1772 * EEH PE operation struct provides ways to: 1773 * - enable/disable EEH functionality; 1774 * - unfreeze IO/DMA for frozen PE; 1775 * - read PE state; 1776 * - reset PE; 1777 * - configure PE; 1778 * - inject EEH error. 1779 */ 1780 struct vfio_eeh_pe_err { 1781 __u32 type; 1782 __u32 func; 1783 __u64 addr; 1784 __u64 mask; 1785 }; 1786 1787 struct vfio_eeh_pe_op { 1788 __u32 argsz; 1789 __u32 flags; 1790 __u32 op; 1791 union { 1792 struct vfio_eeh_pe_err err; 1793 }; 1794 }; 1795 1796 #define VFIO_EEH_PE_DISABLE 0 /* Disable EEH functionality */ 1797 #define VFIO_EEH_PE_ENABLE 1 /* Enable EEH functionality */ 1798 #define VFIO_EEH_PE_UNFREEZE_IO 2 /* Enable IO for frozen PE */ 1799 #define VFIO_EEH_PE_UNFREEZE_DMA 3 /* Enable DMA for frozen PE */ 1800 #define VFIO_EEH_PE_GET_STATE 4 /* PE state retrieval */ 1801 #define VFIO_EEH_PE_STATE_NORMAL 0 /* PE in functional state */ 1802 #define VFIO_EEH_PE_STATE_RESET 1 /* PE reset in progress */ 1803 #define VFIO_EEH_PE_STATE_STOPPED 2 /* Stopped DMA and IO */ 1804 #define VFIO_EEH_PE_STATE_STOPPED_DMA 4 /* Stopped DMA only */ 1805 #define VFIO_EEH_PE_STATE_UNAVAIL 5 /* State unavailable */ 1806 #define VFIO_EEH_PE_RESET_DEACTIVATE 5 /* Deassert PE reset */ 1807 #define VFIO_EEH_PE_RESET_HOT 6 /* Assert hot reset */ 1808 #define VFIO_EEH_PE_RESET_FUNDAMENTAL 7 /* Assert fundamental reset */ 1809 #define VFIO_EEH_PE_CONFIGURE 8 /* PE configuration */ 1810 #define VFIO_EEH_PE_INJECT_ERR 9 /* Inject EEH error */ 1811 1812 #define VFIO_EEH_PE_OP _IO(VFIO_TYPE, VFIO_BASE + 21) 1813 1814 /** 1815 * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory) 1816 * 1817 * Registers user space memory where DMA is allowed. It pins 1818 * user pages and does the locked memory accounting so 1819 * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls 1820 * get faster. 1821 */ 1822 struct vfio_iommu_spapr_register_memory { 1823 __u32 argsz; 1824 __u32 flags; 1825 __u64 vaddr; /* Process virtual address */ 1826 __u64 size; /* Size of mapping (bytes) */ 1827 }; 1828 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 17) 1829 1830 /** 1831 * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory) 1832 * 1833 * Unregisters user space memory registered with 1834 * VFIO_IOMMU_SPAPR_REGISTER_MEMORY. 1835 * Uses vfio_iommu_spapr_register_memory for parameters. 1836 */ 1837 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 18) 1838 1839 /** 1840 * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create) 1841 * 1842 * Creates an additional TCE table and programs it (sets a new DMA window) 1843 * to every IOMMU group in the container. It receives page shift, window 1844 * size and number of levels in the TCE table being created. 1845 * 1846 * It allocates and returns an offset on a PCI bus of the new DMA window. 1847 */ 1848 struct vfio_iommu_spapr_tce_create { 1849 __u32 argsz; 1850 __u32 flags; 1851 /* in */ 1852 __u32 page_shift; 1853 __u32 __resv1; 1854 __u64 window_size; 1855 __u32 levels; 1856 __u32 __resv2; 1857 /* out */ 1858 __u64 start_addr; 1859 }; 1860 #define VFIO_IOMMU_SPAPR_TCE_CREATE _IO(VFIO_TYPE, VFIO_BASE + 19) 1861 1862 /** 1863 * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove) 1864 * 1865 * Unprograms a TCE table from all groups in the container and destroys it. 1866 * It receives a PCI bus offset as a window id. 1867 */ 1868 struct vfio_iommu_spapr_tce_remove { 1869 __u32 argsz; 1870 __u32 flags; 1871 /* in */ 1872 __u64 start_addr; 1873 }; 1874 #define VFIO_IOMMU_SPAPR_TCE_REMOVE _IO(VFIO_TYPE, VFIO_BASE + 20) 1875 1876 /* ***************************************************************** */ 1877 1878 #endif /* _UAPIVFIO_H */ 1879