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