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 /* Supports the vaddr flag for DMA map and unmap */ 53 #define VFIO_UPDATE_VADDR 10 54 55 /* 56 * The IOCTL interface is designed for extensibility by embedding the 57 * structure length (argsz) and flags into structures passed between 58 * kernel and userspace. We therefore use the _IO() macro for these 59 * defines to avoid implicitly embedding a size into the ioctl request. 60 * As structure fields are added, argsz will increase to match and flag 61 * bits will be defined to indicate additional fields with valid data. 62 * It's *always* the caller's responsibility to indicate the size of 63 * the structure passed by setting argsz appropriately. 64 */ 65 66 #define VFIO_TYPE (';') 67 #define VFIO_BASE 100 68 69 /* 70 * For extension of INFO ioctls, VFIO makes use of a capability chain 71 * designed after PCI/e capabilities. A flag bit indicates whether 72 * this capability chain is supported and a field defined in the fixed 73 * structure defines the offset of the first capability in the chain. 74 * This field is only valid when the corresponding bit in the flags 75 * bitmap is set. This offset field is relative to the start of the 76 * INFO buffer, as is the next field within each capability header. 77 * The id within the header is a shared address space per INFO ioctl, 78 * while the version field is specific to the capability id. The 79 * contents following the header are specific to the capability id. 80 */ 81 struct vfio_info_cap_header { 82 __u16 id; /* Identifies capability */ 83 __u16 version; /* Version specific to the capability ID */ 84 __u32 next; /* Offset of next capability */ 85 }; 86 87 /* 88 * Callers of INFO ioctls passing insufficiently sized buffers will see 89 * the capability chain flag bit set, a zero value for the first capability 90 * offset (if available within the provided argsz), and argsz will be 91 * updated to report the necessary buffer size. For compatibility, the 92 * INFO ioctl will not report error in this case, but the capability chain 93 * will not be available. 94 */ 95 96 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */ 97 98 /** 99 * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0) 100 * 101 * Report the version of the VFIO API. This allows us to bump the entire 102 * API version should we later need to add or change features in incompatible 103 * ways. 104 * Return: VFIO_API_VERSION 105 * Availability: Always 106 */ 107 #define VFIO_GET_API_VERSION _IO(VFIO_TYPE, VFIO_BASE + 0) 108 109 /** 110 * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32) 111 * 112 * Check whether an extension is supported. 113 * Return: 0 if not supported, 1 (or some other positive integer) if supported. 114 * Availability: Always 115 */ 116 #define VFIO_CHECK_EXTENSION _IO(VFIO_TYPE, VFIO_BASE + 1) 117 118 /** 119 * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32) 120 * 121 * Set the iommu to the given type. The type must be supported by an 122 * iommu driver as verified by calling CHECK_EXTENSION using the same 123 * type. A group must be set to this file descriptor before this 124 * ioctl is available. The IOMMU interfaces enabled by this call are 125 * specific to the value set. 126 * Return: 0 on success, -errno on failure 127 * Availability: When VFIO group attached 128 */ 129 #define VFIO_SET_IOMMU _IO(VFIO_TYPE, VFIO_BASE + 2) 130 131 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */ 132 133 /** 134 * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3, 135 * struct vfio_group_status) 136 * 137 * Retrieve information about the group. Fills in provided 138 * struct vfio_group_info. Caller sets argsz. 139 * Return: 0 on succes, -errno on failure. 140 * Availability: Always 141 */ 142 struct vfio_group_status { 143 __u32 argsz; 144 __u32 flags; 145 #define VFIO_GROUP_FLAGS_VIABLE (1 << 0) 146 #define VFIO_GROUP_FLAGS_CONTAINER_SET (1 << 1) 147 }; 148 #define VFIO_GROUP_GET_STATUS _IO(VFIO_TYPE, VFIO_BASE + 3) 149 150 /** 151 * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32) 152 * 153 * Set the container for the VFIO group to the open VFIO file 154 * descriptor provided. Groups may only belong to a single 155 * container. Containers may, at their discretion, support multiple 156 * groups. Only when a container is set are all of the interfaces 157 * of the VFIO file descriptor and the VFIO group file descriptor 158 * available to the user. 159 * Return: 0 on success, -errno on failure. 160 * Availability: Always 161 */ 162 #define VFIO_GROUP_SET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 4) 163 164 /** 165 * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5) 166 * 167 * Remove the group from the attached container. This is the 168 * opposite of the SET_CONTAINER call and returns the group to 169 * an initial state. All device file descriptors must be released 170 * prior to calling this interface. When removing the last group 171 * from a container, the IOMMU will be disabled and all state lost, 172 * effectively also returning the VFIO file descriptor to an initial 173 * state. 174 * Return: 0 on success, -errno on failure. 175 * Availability: When attached to container 176 */ 177 #define VFIO_GROUP_UNSET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 5) 178 179 /** 180 * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char) 181 * 182 * Return a new file descriptor for the device object described by 183 * the provided string. The string should match a device listed in 184 * the devices subdirectory of the IOMMU group sysfs entry. The 185 * group containing the device must already be added to this context. 186 * Return: new file descriptor on success, -errno on failure. 187 * Availability: When attached to container 188 */ 189 #define VFIO_GROUP_GET_DEVICE_FD _IO(VFIO_TYPE, VFIO_BASE + 6) 190 191 /* --------------- IOCTLs for DEVICE file descriptors --------------- */ 192 193 /** 194 * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7, 195 * struct vfio_device_info) 196 * 197 * Retrieve information about the device. Fills in provided 198 * struct vfio_device_info. Caller sets argsz. 199 * Return: 0 on success, -errno on failure. 200 */ 201 struct vfio_device_info { 202 __u32 argsz; 203 __u32 flags; 204 #define VFIO_DEVICE_FLAGS_RESET (1 << 0) /* Device supports reset */ 205 #define VFIO_DEVICE_FLAGS_PCI (1 << 1) /* vfio-pci device */ 206 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2) /* vfio-platform device */ 207 #define VFIO_DEVICE_FLAGS_AMBA (1 << 3) /* vfio-amba device */ 208 #define VFIO_DEVICE_FLAGS_CCW (1 << 4) /* vfio-ccw device */ 209 #define VFIO_DEVICE_FLAGS_AP (1 << 5) /* vfio-ap device */ 210 #define VFIO_DEVICE_FLAGS_FSL_MC (1 << 6) /* vfio-fsl-mc device */ 211 #define VFIO_DEVICE_FLAGS_CAPS (1 << 7) /* Info supports caps */ 212 __u32 num_regions; /* Max region index + 1 */ 213 __u32 num_irqs; /* Max IRQ index + 1 */ 214 __u32 cap_offset; /* Offset within info struct of first cap */ 215 }; 216 #define VFIO_DEVICE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 7) 217 218 /* 219 * Vendor driver using Mediated device framework should provide device_api 220 * attribute in supported type attribute groups. Device API string should be one 221 * of the following corresponding to device flags in vfio_device_info structure. 222 */ 223 224 #define VFIO_DEVICE_API_PCI_STRING "vfio-pci" 225 #define VFIO_DEVICE_API_PLATFORM_STRING "vfio-platform" 226 #define VFIO_DEVICE_API_AMBA_STRING "vfio-amba" 227 #define VFIO_DEVICE_API_CCW_STRING "vfio-ccw" 228 #define VFIO_DEVICE_API_AP_STRING "vfio-ap" 229 230 /* 231 * The following capabilities are unique to s390 zPCI devices. Their contents 232 * are further-defined in vfio_zdev.h 233 */ 234 #define VFIO_DEVICE_INFO_CAP_ZPCI_BASE 1 235 #define VFIO_DEVICE_INFO_CAP_ZPCI_GROUP 2 236 #define VFIO_DEVICE_INFO_CAP_ZPCI_UTIL 3 237 #define VFIO_DEVICE_INFO_CAP_ZPCI_PFIP 4 238 239 /** 240 * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8, 241 * struct vfio_region_info) 242 * 243 * Retrieve information about a device region. Caller provides 244 * struct vfio_region_info with index value set. Caller sets argsz. 245 * Implementation of region mapping is bus driver specific. This is 246 * intended to describe MMIO, I/O port, as well as bus specific 247 * regions (ex. PCI config space). Zero sized regions may be used 248 * to describe unimplemented regions (ex. unimplemented PCI BARs). 249 * Return: 0 on success, -errno on failure. 250 */ 251 struct vfio_region_info { 252 __u32 argsz; 253 __u32 flags; 254 #define VFIO_REGION_INFO_FLAG_READ (1 << 0) /* Region supports read */ 255 #define VFIO_REGION_INFO_FLAG_WRITE (1 << 1) /* Region supports write */ 256 #define VFIO_REGION_INFO_FLAG_MMAP (1 << 2) /* Region supports mmap */ 257 #define VFIO_REGION_INFO_FLAG_CAPS (1 << 3) /* Info supports caps */ 258 __u32 index; /* Region index */ 259 __u32 cap_offset; /* Offset within info struct of first cap */ 260 __u64 size; /* Region size (bytes) */ 261 __u64 offset; /* Region offset from start of device fd */ 262 }; 263 #define VFIO_DEVICE_GET_REGION_INFO _IO(VFIO_TYPE, VFIO_BASE + 8) 264 265 /* 266 * The sparse mmap capability allows finer granularity of specifying areas 267 * within a region with mmap support. When specified, the user should only 268 * mmap the offset ranges specified by the areas array. mmaps outside of the 269 * areas specified may fail (such as the range covering a PCI MSI-X table) or 270 * may result in improper device behavior. 271 * 272 * The structures below define version 1 of this capability. 273 */ 274 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP 1 275 276 struct vfio_region_sparse_mmap_area { 277 __u64 offset; /* Offset of mmap'able area within region */ 278 __u64 size; /* Size of mmap'able area */ 279 }; 280 281 struct vfio_region_info_cap_sparse_mmap { 282 struct vfio_info_cap_header header; 283 __u32 nr_areas; 284 __u32 reserved; 285 struct vfio_region_sparse_mmap_area areas[]; 286 }; 287 288 /* 289 * The device specific type capability allows regions unique to a specific 290 * device or class of devices to be exposed. This helps solve the problem for 291 * vfio bus drivers of defining which region indexes correspond to which region 292 * on the device, without needing to resort to static indexes, as done by 293 * vfio-pci. For instance, if we were to go back in time, we might remove 294 * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes 295 * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd 296 * make a "VGA" device specific type to describe the VGA access space. This 297 * means that non-VGA devices wouldn't need to waste this index, and thus the 298 * address space associated with it due to implementation of device file 299 * descriptor offsets in vfio-pci. 300 * 301 * The current implementation is now part of the user ABI, so we can't use this 302 * for VGA, but there are other upcoming use cases, such as opregions for Intel 303 * IGD devices and framebuffers for vGPU devices. We missed VGA, but we'll 304 * use this for future additions. 305 * 306 * The structure below defines version 1 of this capability. 307 */ 308 #define VFIO_REGION_INFO_CAP_TYPE 2 309 310 struct vfio_region_info_cap_type { 311 struct vfio_info_cap_header header; 312 __u32 type; /* global per bus driver */ 313 __u32 subtype; /* type specific */ 314 }; 315 316 /* 317 * List of region types, global per bus driver. 318 * If you introduce a new type, please add it here. 319 */ 320 321 /* PCI region type containing a PCI vendor part */ 322 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE (1 << 31) 323 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK (0xffff) 324 #define VFIO_REGION_TYPE_GFX (1) 325 #define VFIO_REGION_TYPE_CCW (2) 326 #define VFIO_REGION_TYPE_MIGRATION (3) 327 328 /* sub-types for VFIO_REGION_TYPE_PCI_* */ 329 330 /* 8086 vendor PCI sub-types */ 331 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION (1) 332 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG (2) 333 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG (3) 334 335 /* 10de vendor PCI sub-types */ 336 /* 337 * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space. 338 */ 339 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM (1) 340 341 /* 1014 vendor PCI sub-types */ 342 /* 343 * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU 344 * to do TLB invalidation on a GPU. 345 */ 346 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD (1) 347 348 /* sub-types for VFIO_REGION_TYPE_GFX */ 349 #define VFIO_REGION_SUBTYPE_GFX_EDID (1) 350 351 /** 352 * struct vfio_region_gfx_edid - EDID region layout. 353 * 354 * Set display link state and EDID blob. 355 * 356 * The EDID blob has monitor information such as brand, name, serial 357 * number, physical size, supported video modes and more. 358 * 359 * This special region allows userspace (typically qemu) set a virtual 360 * EDID for the virtual monitor, which allows a flexible display 361 * configuration. 362 * 363 * For the edid blob spec look here: 364 * https://en.wikipedia.org/wiki/Extended_Display_Identification_Data 365 * 366 * On linux systems you can find the EDID blob in sysfs: 367 * /sys/class/drm/${card}/${connector}/edid 368 * 369 * You can use the edid-decode ulility (comes with xorg-x11-utils) to 370 * decode the EDID blob. 371 * 372 * @edid_offset: location of the edid blob, relative to the 373 * start of the region (readonly). 374 * @edid_max_size: max size of the edid blob (readonly). 375 * @edid_size: actual edid size (read/write). 376 * @link_state: display link state (read/write). 377 * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on. 378 * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off. 379 * @max_xres: max display width (0 == no limitation, readonly). 380 * @max_yres: max display height (0 == no limitation, readonly). 381 * 382 * EDID update protocol: 383 * (1) set link-state to down. 384 * (2) update edid blob and size. 385 * (3) set link-state to up. 386 */ 387 struct vfio_region_gfx_edid { 388 __u32 edid_offset; 389 __u32 edid_max_size; 390 __u32 edid_size; 391 __u32 max_xres; 392 __u32 max_yres; 393 __u32 link_state; 394 #define VFIO_DEVICE_GFX_LINK_STATE_UP 1 395 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN 2 396 }; 397 398 /* sub-types for VFIO_REGION_TYPE_CCW */ 399 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD (1) 400 #define VFIO_REGION_SUBTYPE_CCW_SCHIB (2) 401 #define VFIO_REGION_SUBTYPE_CCW_CRW (3) 402 403 /* sub-types for VFIO_REGION_TYPE_MIGRATION */ 404 #define VFIO_REGION_SUBTYPE_MIGRATION (1) 405 406 /* 407 * The structure vfio_device_migration_info is placed at the 0th offset of 408 * the VFIO_REGION_SUBTYPE_MIGRATION region to get and set VFIO device related 409 * migration information. Field accesses from this structure are only supported 410 * at their native width and alignment. Otherwise, the result is undefined and 411 * vendor drivers should return an error. 412 * 413 * device_state: (read/write) 414 * - The user application writes to this field to inform the vendor driver 415 * about the device state to be transitioned to. 416 * - The vendor driver should take the necessary actions to change the 417 * device state. After successful transition to a given state, the 418 * vendor driver should return success on write(device_state, state) 419 * system call. If the device state transition fails, the vendor driver 420 * should return an appropriate -errno for the fault condition. 421 * - On the user application side, if the device state transition fails, 422 * that is, if write(device_state, state) returns an error, read 423 * device_state again to determine the current state of the device from 424 * the vendor driver. 425 * - The vendor driver should return previous state of the device unless 426 * the vendor driver has encountered an internal error, in which case 427 * the vendor driver may report the device_state VFIO_DEVICE_STATE_ERROR. 428 * - The user application must use the device reset ioctl to recover the 429 * device from VFIO_DEVICE_STATE_ERROR state. If the device is 430 * indicated to be in a valid device state by reading device_state, the 431 * user application may attempt to transition the device to any valid 432 * state reachable from the current state or terminate itself. 433 * 434 * device_state consists of 3 bits: 435 * - If bit 0 is set, it indicates the _RUNNING state. If bit 0 is clear, 436 * it indicates the _STOP state. When the device state is changed to 437 * _STOP, driver should stop the device before write() returns. 438 * - If bit 1 is set, it indicates the _SAVING state, which means that the 439 * driver should start gathering device state information that will be 440 * provided to the VFIO user application to save the device's state. 441 * - If bit 2 is set, it indicates the _RESUMING state, which means that 442 * the driver should prepare to resume the device. Data provided through 443 * the migration region should be used to resume the device. 444 * Bits 3 - 31 are reserved for future use. To preserve them, the user 445 * application should perform a read-modify-write operation on this 446 * field when modifying the specified bits. 447 * 448 * +------- _RESUMING 449 * |+------ _SAVING 450 * ||+----- _RUNNING 451 * ||| 452 * 000b => Device Stopped, not saving or resuming 453 * 001b => Device running, which is the default state 454 * 010b => Stop the device & save the device state, stop-and-copy state 455 * 011b => Device running and save the device state, pre-copy state 456 * 100b => Device stopped and the device state is resuming 457 * 101b => Invalid state 458 * 110b => Error state 459 * 111b => Invalid state 460 * 461 * State transitions: 462 * 463 * _RESUMING _RUNNING Pre-copy Stop-and-copy _STOP 464 * (100b) (001b) (011b) (010b) (000b) 465 * 0. Running or default state 466 * | 467 * 468 * 1. Normal Shutdown (optional) 469 * |------------------------------------->| 470 * 471 * 2. Save the state or suspend 472 * |------------------------->|---------->| 473 * 474 * 3. Save the state during live migration 475 * |----------->|------------>|---------->| 476 * 477 * 4. Resuming 478 * |<---------| 479 * 480 * 5. Resumed 481 * |--------->| 482 * 483 * 0. Default state of VFIO device is _RUNNING when the user application starts. 484 * 1. During normal shutdown of the user application, the user application may 485 * optionally change the VFIO device state from _RUNNING to _STOP. This 486 * transition is optional. The vendor driver must support this transition but 487 * must not require it. 488 * 2. When the user application saves state or suspends the application, the 489 * device state transitions from _RUNNING to stop-and-copy and then to _STOP. 490 * On state transition from _RUNNING to stop-and-copy, driver must stop the 491 * device, save the device state and send it to the application through the 492 * migration region. The sequence to be followed for such transition is given 493 * below. 494 * 3. In live migration of user application, the state transitions from _RUNNING 495 * to pre-copy, to stop-and-copy, and to _STOP. 496 * On state transition from _RUNNING to pre-copy, the driver should start 497 * gathering the device state while the application is still running and send 498 * the device state data to application through the migration region. 499 * On state transition from pre-copy to stop-and-copy, the driver must stop 500 * the device, save the device state and send it to the user application 501 * through the migration region. 502 * Vendor drivers must support the pre-copy state even for implementations 503 * where no data is provided to the user before the stop-and-copy state. The 504 * user must not be required to consume all migration data before the device 505 * transitions to a new state, including the stop-and-copy state. 506 * The sequence to be followed for above two transitions is given below. 507 * 4. To start the resuming phase, the device state should be transitioned from 508 * the _RUNNING to the _RESUMING state. 509 * In the _RESUMING state, the driver should use the device state data 510 * received through the migration region to resume the device. 511 * 5. After providing saved device data to the driver, the application should 512 * change the state from _RESUMING to _RUNNING. 513 * 514 * reserved: 515 * Reads on this field return zero and writes are ignored. 516 * 517 * pending_bytes: (read only) 518 * The number of pending bytes still to be migrated from the vendor driver. 519 * 520 * data_offset: (read only) 521 * The user application should read data_offset field from the migration 522 * region. The user application should read the device data from this 523 * offset within the migration region during the _SAVING state or write 524 * the device data during the _RESUMING state. See below for details of 525 * sequence to be followed. 526 * 527 * data_size: (read/write) 528 * The user application should read data_size to get the size in bytes of 529 * the data copied in the migration region during the _SAVING state and 530 * write the size in bytes of the data copied in the migration region 531 * during the _RESUMING state. 532 * 533 * The format of the migration region is as follows: 534 * ------------------------------------------------------------------ 535 * |vfio_device_migration_info| data section | 536 * | | /////////////////////////////// | 537 * ------------------------------------------------------------------ 538 * ^ ^ 539 * offset 0-trapped part data_offset 540 * 541 * The structure vfio_device_migration_info is always followed by the data 542 * section in the region, so data_offset will always be nonzero. The offset 543 * from where the data is copied is decided by the kernel driver. The data 544 * section can be trapped, mmapped, or partitioned, depending on how the kernel 545 * driver defines the data section. The data section partition can be defined 546 * as mapped by the sparse mmap capability. If mmapped, data_offset must be 547 * page aligned, whereas initial section which contains the 548 * vfio_device_migration_info structure, might not end at the offset, which is 549 * page aligned. The user is not required to access through mmap regardless 550 * of the capabilities of the region mmap. 551 * The vendor driver should determine whether and how to partition the data 552 * section. The vendor driver should return data_offset accordingly. 553 * 554 * The sequence to be followed while in pre-copy state and stop-and-copy state 555 * is as follows: 556 * a. Read pending_bytes, indicating the start of a new iteration to get device 557 * data. Repeated read on pending_bytes at this stage should have no side 558 * effects. 559 * If pending_bytes == 0, the user application should not iterate to get data 560 * for that device. 561 * If pending_bytes > 0, perform the following steps. 562 * b. Read data_offset, indicating that the vendor driver should make data 563 * available through the data section. The vendor driver should return this 564 * read operation only after data is available from (region + data_offset) 565 * to (region + data_offset + data_size). 566 * c. Read data_size, which is the amount of data in bytes available through 567 * the migration region. 568 * Read on data_offset and data_size should return the offset and size of 569 * the current buffer if the user application reads data_offset and 570 * data_size more than once here. 571 * d. Read data_size bytes of data from (region + data_offset) from the 572 * migration region. 573 * e. Process the data. 574 * f. Read pending_bytes, which indicates that the data from the previous 575 * iteration has been read. If pending_bytes > 0, go to step b. 576 * 577 * The user application can transition from the _SAVING|_RUNNING 578 * (pre-copy state) to the _SAVING (stop-and-copy) state regardless of the 579 * number of pending bytes. The user application should iterate in _SAVING 580 * (stop-and-copy) until pending_bytes is 0. 581 * 582 * The sequence to be followed while _RESUMING device state is as follows: 583 * While data for this device is available, repeat the following steps: 584 * a. Read data_offset from where the user application should write data. 585 * b. Write migration data starting at the migration region + data_offset for 586 * the length determined by data_size from the migration source. 587 * c. Write data_size, which indicates to the vendor driver that data is 588 * written in the migration region. Vendor driver must return this write 589 * operations on consuming data. Vendor driver should apply the 590 * user-provided migration region data to the device resume state. 591 * 592 * If an error occurs during the above sequences, the vendor driver can return 593 * an error code for next read() or write() operation, which will terminate the 594 * loop. The user application should then take the next necessary action, for 595 * example, failing migration or terminating the user application. 596 * 597 * For the user application, data is opaque. The user application should write 598 * data in the same order as the data is received and the data should be of 599 * same transaction size at the source. 600 */ 601 602 struct vfio_device_migration_info { 603 __u32 device_state; /* VFIO device state */ 604 #define VFIO_DEVICE_STATE_STOP (0) 605 #define VFIO_DEVICE_STATE_RUNNING (1 << 0) 606 #define VFIO_DEVICE_STATE_SAVING (1 << 1) 607 #define VFIO_DEVICE_STATE_RESUMING (1 << 2) 608 #define VFIO_DEVICE_STATE_MASK (VFIO_DEVICE_STATE_RUNNING | \ 609 VFIO_DEVICE_STATE_SAVING | \ 610 VFIO_DEVICE_STATE_RESUMING) 611 612 #define VFIO_DEVICE_STATE_VALID(state) \ 613 (state & VFIO_DEVICE_STATE_RESUMING ? \ 614 (state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_RESUMING : 1) 615 616 #define VFIO_DEVICE_STATE_IS_ERROR(state) \ 617 ((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_SAVING | \ 618 VFIO_DEVICE_STATE_RESUMING)) 619 620 #define VFIO_DEVICE_STATE_SET_ERROR(state) \ 621 ((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_SATE_SAVING | \ 622 VFIO_DEVICE_STATE_RESUMING) 623 624 __u32 reserved; 625 __u64 pending_bytes; 626 __u64 data_offset; 627 __u64 data_size; 628 }; 629 630 /* 631 * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped 632 * which allows direct access to non-MSIX registers which happened to be within 633 * the same system page. 634 * 635 * Even though the userspace gets direct access to the MSIX data, the existing 636 * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration. 637 */ 638 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE 3 639 640 /* 641 * Capability with compressed real address (aka SSA - small system address) 642 * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing 643 * and by the userspace to associate a NVLink bridge with a GPU. 644 */ 645 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT 4 646 647 struct vfio_region_info_cap_nvlink2_ssatgt { 648 struct vfio_info_cap_header header; 649 __u64 tgt; 650 }; 651 652 /* 653 * Capability with an NVLink link speed. The value is read by 654 * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed" 655 * property in the device tree. The value is fixed in the hardware 656 * and failing to provide the correct value results in the link 657 * not working with no indication from the driver why. 658 */ 659 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD 5 660 661 struct vfio_region_info_cap_nvlink2_lnkspd { 662 struct vfio_info_cap_header header; 663 __u32 link_speed; 664 __u32 __pad; 665 }; 666 667 /** 668 * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9, 669 * struct vfio_irq_info) 670 * 671 * Retrieve information about a device IRQ. Caller provides 672 * struct vfio_irq_info with index value set. Caller sets argsz. 673 * Implementation of IRQ mapping is bus driver specific. Indexes 674 * using multiple IRQs are primarily intended to support MSI-like 675 * interrupt blocks. Zero count irq blocks may be used to describe 676 * unimplemented interrupt types. 677 * 678 * The EVENTFD flag indicates the interrupt index supports eventfd based 679 * signaling. 680 * 681 * The MASKABLE flags indicates the index supports MASK and UNMASK 682 * actions described below. 683 * 684 * AUTOMASKED indicates that after signaling, the interrupt line is 685 * automatically masked by VFIO and the user needs to unmask the line 686 * to receive new interrupts. This is primarily intended to distinguish 687 * level triggered interrupts. 688 * 689 * The NORESIZE flag indicates that the interrupt lines within the index 690 * are setup as a set and new subindexes cannot be enabled without first 691 * disabling the entire index. This is used for interrupts like PCI MSI 692 * and MSI-X where the driver may only use a subset of the available 693 * indexes, but VFIO needs to enable a specific number of vectors 694 * upfront. In the case of MSI-X, where the user can enable MSI-X and 695 * then add and unmask vectors, it's up to userspace to make the decision 696 * whether to allocate the maximum supported number of vectors or tear 697 * down setup and incrementally increase the vectors as each is enabled. 698 */ 699 struct vfio_irq_info { 700 __u32 argsz; 701 __u32 flags; 702 #define VFIO_IRQ_INFO_EVENTFD (1 << 0) 703 #define VFIO_IRQ_INFO_MASKABLE (1 << 1) 704 #define VFIO_IRQ_INFO_AUTOMASKED (1 << 2) 705 #define VFIO_IRQ_INFO_NORESIZE (1 << 3) 706 __u32 index; /* IRQ index */ 707 __u32 count; /* Number of IRQs within this index */ 708 }; 709 #define VFIO_DEVICE_GET_IRQ_INFO _IO(VFIO_TYPE, VFIO_BASE + 9) 710 711 /** 712 * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set) 713 * 714 * Set signaling, masking, and unmasking of interrupts. Caller provides 715 * struct vfio_irq_set with all fields set. 'start' and 'count' indicate 716 * the range of subindexes being specified. 717 * 718 * The DATA flags specify the type of data provided. If DATA_NONE, the 719 * operation performs the specified action immediately on the specified 720 * interrupt(s). For example, to unmask AUTOMASKED interrupt [0,0]: 721 * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1. 722 * 723 * DATA_BOOL allows sparse support for the same on arrays of interrupts. 724 * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]): 725 * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3, 726 * data = {1,0,1} 727 * 728 * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd. 729 * A value of -1 can be used to either de-assign interrupts if already 730 * assigned or skip un-assigned interrupts. For example, to set an eventfd 731 * to be trigger for interrupts [0,0] and [0,2]: 732 * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3, 733 * data = {fd1, -1, fd2} 734 * If index [0,1] is previously set, two count = 1 ioctls calls would be 735 * required to set [0,0] and [0,2] without changing [0,1]. 736 * 737 * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used 738 * with ACTION_TRIGGER to perform kernel level interrupt loopback testing 739 * from userspace (ie. simulate hardware triggering). 740 * 741 * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER 742 * enables the interrupt index for the device. Individual subindex interrupts 743 * can be disabled using the -1 value for DATA_EVENTFD or the index can be 744 * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0. 745 * 746 * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while 747 * ACTION_TRIGGER specifies kernel->user signaling. 748 */ 749 struct vfio_irq_set { 750 __u32 argsz; 751 __u32 flags; 752 #define VFIO_IRQ_SET_DATA_NONE (1 << 0) /* Data not present */ 753 #define VFIO_IRQ_SET_DATA_BOOL (1 << 1) /* Data is bool (u8) */ 754 #define VFIO_IRQ_SET_DATA_EVENTFD (1 << 2) /* Data is eventfd (s32) */ 755 #define VFIO_IRQ_SET_ACTION_MASK (1 << 3) /* Mask interrupt */ 756 #define VFIO_IRQ_SET_ACTION_UNMASK (1 << 4) /* Unmask interrupt */ 757 #define VFIO_IRQ_SET_ACTION_TRIGGER (1 << 5) /* Trigger interrupt */ 758 __u32 index; 759 __u32 start; 760 __u32 count; 761 __u8 data[]; 762 }; 763 #define VFIO_DEVICE_SET_IRQS _IO(VFIO_TYPE, VFIO_BASE + 10) 764 765 #define VFIO_IRQ_SET_DATA_TYPE_MASK (VFIO_IRQ_SET_DATA_NONE | \ 766 VFIO_IRQ_SET_DATA_BOOL | \ 767 VFIO_IRQ_SET_DATA_EVENTFD) 768 #define VFIO_IRQ_SET_ACTION_TYPE_MASK (VFIO_IRQ_SET_ACTION_MASK | \ 769 VFIO_IRQ_SET_ACTION_UNMASK | \ 770 VFIO_IRQ_SET_ACTION_TRIGGER) 771 /** 772 * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11) 773 * 774 * Reset a device. 775 */ 776 #define VFIO_DEVICE_RESET _IO(VFIO_TYPE, VFIO_BASE + 11) 777 778 /* 779 * The VFIO-PCI bus driver makes use of the following fixed region and 780 * IRQ index mapping. Unimplemented regions return a size of zero. 781 * Unimplemented IRQ types return a count of zero. 782 */ 783 784 enum { 785 VFIO_PCI_BAR0_REGION_INDEX, 786 VFIO_PCI_BAR1_REGION_INDEX, 787 VFIO_PCI_BAR2_REGION_INDEX, 788 VFIO_PCI_BAR3_REGION_INDEX, 789 VFIO_PCI_BAR4_REGION_INDEX, 790 VFIO_PCI_BAR5_REGION_INDEX, 791 VFIO_PCI_ROM_REGION_INDEX, 792 VFIO_PCI_CONFIG_REGION_INDEX, 793 /* 794 * Expose VGA regions defined for PCI base class 03, subclass 00. 795 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df 796 * as well as the MMIO range 0xa0000 to 0xbffff. Each implemented 797 * range is found at it's identity mapped offset from the region 798 * offset, for example 0x3b0 is region_info.offset + 0x3b0. Areas 799 * between described ranges are unimplemented. 800 */ 801 VFIO_PCI_VGA_REGION_INDEX, 802 VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */ 803 /* device specific cap to define content. */ 804 }; 805 806 enum { 807 VFIO_PCI_INTX_IRQ_INDEX, 808 VFIO_PCI_MSI_IRQ_INDEX, 809 VFIO_PCI_MSIX_IRQ_INDEX, 810 VFIO_PCI_ERR_IRQ_INDEX, 811 VFIO_PCI_REQ_IRQ_INDEX, 812 VFIO_PCI_NUM_IRQS 813 }; 814 815 /* 816 * The vfio-ccw bus driver makes use of the following fixed region and 817 * IRQ index mapping. Unimplemented regions return a size of zero. 818 * Unimplemented IRQ types return a count of zero. 819 */ 820 821 enum { 822 VFIO_CCW_CONFIG_REGION_INDEX, 823 VFIO_CCW_NUM_REGIONS 824 }; 825 826 enum { 827 VFIO_CCW_IO_IRQ_INDEX, 828 VFIO_CCW_CRW_IRQ_INDEX, 829 VFIO_CCW_REQ_IRQ_INDEX, 830 VFIO_CCW_NUM_IRQS 831 }; 832 833 /** 834 * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12, 835 * struct vfio_pci_hot_reset_info) 836 * 837 * Return: 0 on success, -errno on failure: 838 * -enospc = insufficient buffer, -enodev = unsupported for device. 839 */ 840 struct vfio_pci_dependent_device { 841 __u32 group_id; 842 __u16 segment; 843 __u8 bus; 844 __u8 devfn; /* Use PCI_SLOT/PCI_FUNC */ 845 }; 846 847 struct vfio_pci_hot_reset_info { 848 __u32 argsz; 849 __u32 flags; 850 __u32 count; 851 struct vfio_pci_dependent_device devices[]; 852 }; 853 854 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 855 856 /** 857 * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13, 858 * struct vfio_pci_hot_reset) 859 * 860 * Return: 0 on success, -errno on failure. 861 */ 862 struct vfio_pci_hot_reset { 863 __u32 argsz; 864 __u32 flags; 865 __u32 count; 866 __s32 group_fds[]; 867 }; 868 869 #define VFIO_DEVICE_PCI_HOT_RESET _IO(VFIO_TYPE, VFIO_BASE + 13) 870 871 /** 872 * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14, 873 * struct vfio_device_query_gfx_plane) 874 * 875 * Set the drm_plane_type and flags, then retrieve the gfx plane info. 876 * 877 * flags supported: 878 * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set 879 * to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no 880 * support for dma-buf. 881 * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set 882 * to ask if the mdev supports region. 0 on support, -EINVAL on no 883 * support for region. 884 * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set 885 * with each call to query the plane info. 886 * - Others are invalid and return -EINVAL. 887 * 888 * Note: 889 * 1. Plane could be disabled by guest. In that case, success will be 890 * returned with zero-initialized drm_format, size, width and height 891 * fields. 892 * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available 893 * 894 * Return: 0 on success, -errno on other failure. 895 */ 896 struct vfio_device_gfx_plane_info { 897 __u32 argsz; 898 __u32 flags; 899 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0) 900 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1) 901 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2) 902 /* in */ 903 __u32 drm_plane_type; /* type of plane: DRM_PLANE_TYPE_* */ 904 /* out */ 905 __u32 drm_format; /* drm format of plane */ 906 __u64 drm_format_mod; /* tiled mode */ 907 __u32 width; /* width of plane */ 908 __u32 height; /* height of plane */ 909 __u32 stride; /* stride of plane */ 910 __u32 size; /* size of plane in bytes, align on page*/ 911 __u32 x_pos; /* horizontal position of cursor plane */ 912 __u32 y_pos; /* vertical position of cursor plane*/ 913 __u32 x_hot; /* horizontal position of cursor hotspot */ 914 __u32 y_hot; /* vertical position of cursor hotspot */ 915 union { 916 __u32 region_index; /* region index */ 917 __u32 dmabuf_id; /* dma-buf id */ 918 }; 919 }; 920 921 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14) 922 923 /** 924 * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32) 925 * 926 * Return a new dma-buf file descriptor for an exposed guest framebuffer 927 * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_ 928 * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer. 929 */ 930 931 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15) 932 933 /** 934 * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16, 935 * struct vfio_device_ioeventfd) 936 * 937 * Perform a write to the device at the specified device fd offset, with 938 * the specified data and width when the provided eventfd is triggered. 939 * vfio bus drivers may not support this for all regions, for all widths, 940 * or at all. vfio-pci currently only enables support for BAR regions, 941 * excluding the MSI-X vector table. 942 * 943 * Return: 0 on success, -errno on failure. 944 */ 945 struct vfio_device_ioeventfd { 946 __u32 argsz; 947 __u32 flags; 948 #define VFIO_DEVICE_IOEVENTFD_8 (1 << 0) /* 1-byte write */ 949 #define VFIO_DEVICE_IOEVENTFD_16 (1 << 1) /* 2-byte write */ 950 #define VFIO_DEVICE_IOEVENTFD_32 (1 << 2) /* 4-byte write */ 951 #define VFIO_DEVICE_IOEVENTFD_64 (1 << 3) /* 8-byte write */ 952 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK (0xf) 953 __u64 offset; /* device fd offset of write */ 954 __u64 data; /* data to be written */ 955 __s32 fd; /* -1 for de-assignment */ 956 }; 957 958 #define VFIO_DEVICE_IOEVENTFD _IO(VFIO_TYPE, VFIO_BASE + 16) 959 960 /** 961 * VFIO_DEVICE_FEATURE - _IORW(VFIO_TYPE, VFIO_BASE + 17, 962 * struct vfio_device_feature) 963 * 964 * Get, set, or probe feature data of the device. The feature is selected 965 * using the FEATURE_MASK portion of the flags field. Support for a feature 966 * can be probed by setting both the FEATURE_MASK and PROBE bits. A probe 967 * may optionally include the GET and/or SET bits to determine read vs write 968 * access of the feature respectively. Probing a feature will return success 969 * if the feature is supported and all of the optionally indicated GET/SET 970 * methods are supported. The format of the data portion of the structure is 971 * specific to the given feature. The data portion is not required for 972 * probing. GET and SET are mutually exclusive, except for use with PROBE. 973 * 974 * Return 0 on success, -errno on failure. 975 */ 976 struct vfio_device_feature { 977 __u32 argsz; 978 __u32 flags; 979 #define VFIO_DEVICE_FEATURE_MASK (0xffff) /* 16-bit feature index */ 980 #define VFIO_DEVICE_FEATURE_GET (1 << 16) /* Get feature into data[] */ 981 #define VFIO_DEVICE_FEATURE_SET (1 << 17) /* Set feature from data[] */ 982 #define VFIO_DEVICE_FEATURE_PROBE (1 << 18) /* Probe feature support */ 983 __u8 data[]; 984 }; 985 986 #define VFIO_DEVICE_FEATURE _IO(VFIO_TYPE, VFIO_BASE + 17) 987 988 /* 989 * Provide support for setting a PCI VF Token, which is used as a shared 990 * secret between PF and VF drivers. This feature may only be set on a 991 * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing 992 * open VFs. Data provided when setting this feature is a 16-byte array 993 * (__u8 b[16]), representing a UUID. 994 */ 995 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN (0) 996 997 /* -------- API for Type1 VFIO IOMMU -------- */ 998 999 /** 1000 * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info) 1001 * 1002 * Retrieve information about the IOMMU object. Fills in provided 1003 * struct vfio_iommu_info. Caller sets argsz. 1004 * 1005 * XXX Should we do these by CHECK_EXTENSION too? 1006 */ 1007 struct vfio_iommu_type1_info { 1008 __u32 argsz; 1009 __u32 flags; 1010 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0) /* supported page sizes info */ 1011 #define VFIO_IOMMU_INFO_CAPS (1 << 1) /* Info supports caps */ 1012 __u64 iova_pgsizes; /* Bitmap of supported page sizes */ 1013 __u32 cap_offset; /* Offset within info struct of first cap */ 1014 }; 1015 1016 /* 1017 * The IOVA capability allows to report the valid IOVA range(s) 1018 * excluding any non-relaxable reserved regions exposed by 1019 * devices attached to the container. Any DMA map attempt 1020 * outside the valid iova range will return error. 1021 * 1022 * The structures below define version 1 of this capability. 1023 */ 1024 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE 1 1025 1026 struct vfio_iova_range { 1027 __u64 start; 1028 __u64 end; 1029 }; 1030 1031 struct vfio_iommu_type1_info_cap_iova_range { 1032 struct vfio_info_cap_header header; 1033 __u32 nr_iovas; 1034 __u32 reserved; 1035 struct vfio_iova_range iova_ranges[]; 1036 }; 1037 1038 /* 1039 * The migration capability allows to report supported features for migration. 1040 * 1041 * The structures below define version 1 of this capability. 1042 * 1043 * The existence of this capability indicates that IOMMU kernel driver supports 1044 * dirty page logging. 1045 * 1046 * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty 1047 * page logging. 1048 * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap 1049 * size in bytes that can be used by user applications when getting the dirty 1050 * bitmap. 1051 */ 1052 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION 2 1053 1054 struct vfio_iommu_type1_info_cap_migration { 1055 struct vfio_info_cap_header header; 1056 __u32 flags; 1057 __u64 pgsize_bitmap; 1058 __u64 max_dirty_bitmap_size; /* in bytes */ 1059 }; 1060 1061 /* 1062 * The DMA available capability allows to report the current number of 1063 * simultaneously outstanding DMA mappings that are allowed. 1064 * 1065 * The structure below defines version 1 of this capability. 1066 * 1067 * avail: specifies the current number of outstanding DMA mappings allowed. 1068 */ 1069 #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3 1070 1071 struct vfio_iommu_type1_info_dma_avail { 1072 struct vfio_info_cap_header header; 1073 __u32 avail; 1074 }; 1075 1076 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 1077 1078 /** 1079 * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map) 1080 * 1081 * Map process virtual addresses to IO virtual addresses using the 1082 * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required. 1083 * 1084 * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova, and 1085 * unblock translation of host virtual addresses in the iova range. The vaddr 1086 * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR. To 1087 * maintain memory consistency within the user application, the updated vaddr 1088 * must address the same memory object as originally mapped. Failure to do so 1089 * will result in user memory corruption and/or device misbehavior. iova and 1090 * size must match those in the original MAP_DMA call. Protection is not 1091 * changed, and the READ & WRITE flags must be 0. 1092 */ 1093 struct vfio_iommu_type1_dma_map { 1094 __u32 argsz; 1095 __u32 flags; 1096 #define VFIO_DMA_MAP_FLAG_READ (1 << 0) /* readable from device */ 1097 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1) /* writable from device */ 1098 #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2) 1099 __u64 vaddr; /* Process virtual address */ 1100 __u64 iova; /* IO virtual address */ 1101 __u64 size; /* Size of mapping (bytes) */ 1102 }; 1103 1104 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13) 1105 1106 struct vfio_bitmap { 1107 __u64 pgsize; /* page size for bitmap in bytes */ 1108 __u64 size; /* in bytes */ 1109 __u64 __user *data; /* one bit per page */ 1110 }; 1111 1112 /** 1113 * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14, 1114 * struct vfio_dma_unmap) 1115 * 1116 * Unmap IO virtual addresses using the provided struct vfio_dma_unmap. 1117 * Caller sets argsz. The actual unmapped size is returned in the size 1118 * field. No guarantee is made to the user that arbitrary unmaps of iova 1119 * or size different from those used in the original mapping call will 1120 * succeed. 1121 * 1122 * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap 1123 * before unmapping IO virtual addresses. When this flag is set, the user must 1124 * provide a struct vfio_bitmap in data[]. User must provide zero-allocated 1125 * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field. 1126 * A bit in the bitmap represents one page, of user provided page size in 1127 * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set 1128 * indicates that the page at that offset from iova is dirty. A Bitmap of the 1129 * pages in the range of unmapped size is returned in the user-provided 1130 * vfio_bitmap.data. 1131 * 1132 * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses. iova and size 1133 * must be 0. This cannot be combined with the get-dirty-bitmap flag. 1134 * 1135 * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host 1136 * virtual addresses in the iova range. Tasks that attempt to translate an 1137 * iova's vaddr will block. DMA to already-mapped pages continues. This 1138 * cannot be combined with the get-dirty-bitmap flag. 1139 */ 1140 struct vfio_iommu_type1_dma_unmap { 1141 __u32 argsz; 1142 __u32 flags; 1143 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0) 1144 #define VFIO_DMA_UNMAP_FLAG_ALL (1 << 1) 1145 #define VFIO_DMA_UNMAP_FLAG_VADDR (1 << 2) 1146 __u64 iova; /* IO virtual address */ 1147 __u64 size; /* Size of mapping (bytes) */ 1148 __u8 data[]; 1149 }; 1150 1151 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14) 1152 1153 /* 1154 * IOCTLs to enable/disable IOMMU container usage. 1155 * No parameters are supported. 1156 */ 1157 #define VFIO_IOMMU_ENABLE _IO(VFIO_TYPE, VFIO_BASE + 15) 1158 #define VFIO_IOMMU_DISABLE _IO(VFIO_TYPE, VFIO_BASE + 16) 1159 1160 /** 1161 * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17, 1162 * struct vfio_iommu_type1_dirty_bitmap) 1163 * IOCTL is used for dirty pages logging. 1164 * Caller should set flag depending on which operation to perform, details as 1165 * below: 1166 * 1167 * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs 1168 * the IOMMU driver to log pages that are dirtied or potentially dirtied by 1169 * the device; designed to be used when a migration is in progress. Dirty pages 1170 * are logged until logging is disabled by user application by calling the IOCTL 1171 * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag. 1172 * 1173 * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs 1174 * the IOMMU driver to stop logging dirtied pages. 1175 * 1176 * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set 1177 * returns the dirty pages bitmap for IOMMU container for a given IOVA range. 1178 * The user must specify the IOVA range and the pgsize through the structure 1179 * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface 1180 * supports getting a bitmap of the smallest supported pgsize only and can be 1181 * modified in future to get a bitmap of any specified supported pgsize. The 1182 * user must provide a zeroed memory area for the bitmap memory and specify its 1183 * size in bitmap.size. One bit is used to represent one page consecutively 1184 * starting from iova offset. The user should provide page size in bitmap.pgsize 1185 * field. A bit set in the bitmap indicates that the page at that offset from 1186 * iova is dirty. The caller must set argsz to a value including the size of 1187 * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the 1188 * actual bitmap. If dirty pages logging is not enabled, an error will be 1189 * returned. 1190 * 1191 * Only one of the flags _START, _STOP and _GET may be specified at a time. 1192 * 1193 */ 1194 struct vfio_iommu_type1_dirty_bitmap { 1195 __u32 argsz; 1196 __u32 flags; 1197 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START (1 << 0) 1198 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP (1 << 1) 1199 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP (1 << 2) 1200 __u8 data[]; 1201 }; 1202 1203 struct vfio_iommu_type1_dirty_bitmap_get { 1204 __u64 iova; /* IO virtual address */ 1205 __u64 size; /* Size of iova range */ 1206 struct vfio_bitmap bitmap; 1207 }; 1208 1209 #define VFIO_IOMMU_DIRTY_PAGES _IO(VFIO_TYPE, VFIO_BASE + 17) 1210 1211 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */ 1212 1213 /* 1214 * The SPAPR TCE DDW info struct provides the information about 1215 * the details of Dynamic DMA window capability. 1216 * 1217 * @pgsizes contains a page size bitmask, 4K/64K/16M are supported. 1218 * @max_dynamic_windows_supported tells the maximum number of windows 1219 * which the platform can create. 1220 * @levels tells the maximum number of levels in multi-level IOMMU tables; 1221 * this allows splitting a table into smaller chunks which reduces 1222 * the amount of physically contiguous memory required for the table. 1223 */ 1224 struct vfio_iommu_spapr_tce_ddw_info { 1225 __u64 pgsizes; /* Bitmap of supported page sizes */ 1226 __u32 max_dynamic_windows_supported; 1227 __u32 levels; 1228 }; 1229 1230 /* 1231 * The SPAPR TCE info struct provides the information about the PCI bus 1232 * address ranges available for DMA, these values are programmed into 1233 * the hardware so the guest has to know that information. 1234 * 1235 * The DMA 32 bit window start is an absolute PCI bus address. 1236 * The IOVA address passed via map/unmap ioctls are absolute PCI bus 1237 * addresses too so the window works as a filter rather than an offset 1238 * for IOVA addresses. 1239 * 1240 * Flags supported: 1241 * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows 1242 * (DDW) support is present. @ddw is only supported when DDW is present. 1243 */ 1244 struct vfio_iommu_spapr_tce_info { 1245 __u32 argsz; 1246 __u32 flags; 1247 #define VFIO_IOMMU_SPAPR_INFO_DDW (1 << 0) /* DDW supported */ 1248 __u32 dma32_window_start; /* 32 bit window start (bytes) */ 1249 __u32 dma32_window_size; /* 32 bit window size (bytes) */ 1250 struct vfio_iommu_spapr_tce_ddw_info ddw; 1251 }; 1252 1253 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 1254 1255 /* 1256 * EEH PE operation struct provides ways to: 1257 * - enable/disable EEH functionality; 1258 * - unfreeze IO/DMA for frozen PE; 1259 * - read PE state; 1260 * - reset PE; 1261 * - configure PE; 1262 * - inject EEH error. 1263 */ 1264 struct vfio_eeh_pe_err { 1265 __u32 type; 1266 __u32 func; 1267 __u64 addr; 1268 __u64 mask; 1269 }; 1270 1271 struct vfio_eeh_pe_op { 1272 __u32 argsz; 1273 __u32 flags; 1274 __u32 op; 1275 union { 1276 struct vfio_eeh_pe_err err; 1277 }; 1278 }; 1279 1280 #define VFIO_EEH_PE_DISABLE 0 /* Disable EEH functionality */ 1281 #define VFIO_EEH_PE_ENABLE 1 /* Enable EEH functionality */ 1282 #define VFIO_EEH_PE_UNFREEZE_IO 2 /* Enable IO for frozen PE */ 1283 #define VFIO_EEH_PE_UNFREEZE_DMA 3 /* Enable DMA for frozen PE */ 1284 #define VFIO_EEH_PE_GET_STATE 4 /* PE state retrieval */ 1285 #define VFIO_EEH_PE_STATE_NORMAL 0 /* PE in functional state */ 1286 #define VFIO_EEH_PE_STATE_RESET 1 /* PE reset in progress */ 1287 #define VFIO_EEH_PE_STATE_STOPPED 2 /* Stopped DMA and IO */ 1288 #define VFIO_EEH_PE_STATE_STOPPED_DMA 4 /* Stopped DMA only */ 1289 #define VFIO_EEH_PE_STATE_UNAVAIL 5 /* State unavailable */ 1290 #define VFIO_EEH_PE_RESET_DEACTIVATE 5 /* Deassert PE reset */ 1291 #define VFIO_EEH_PE_RESET_HOT 6 /* Assert hot reset */ 1292 #define VFIO_EEH_PE_RESET_FUNDAMENTAL 7 /* Assert fundamental reset */ 1293 #define VFIO_EEH_PE_CONFIGURE 8 /* PE configuration */ 1294 #define VFIO_EEH_PE_INJECT_ERR 9 /* Inject EEH error */ 1295 1296 #define VFIO_EEH_PE_OP _IO(VFIO_TYPE, VFIO_BASE + 21) 1297 1298 /** 1299 * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory) 1300 * 1301 * Registers user space memory where DMA is allowed. It pins 1302 * user pages and does the locked memory accounting so 1303 * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls 1304 * get faster. 1305 */ 1306 struct vfio_iommu_spapr_register_memory { 1307 __u32 argsz; 1308 __u32 flags; 1309 __u64 vaddr; /* Process virtual address */ 1310 __u64 size; /* Size of mapping (bytes) */ 1311 }; 1312 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 17) 1313 1314 /** 1315 * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory) 1316 * 1317 * Unregisters user space memory registered with 1318 * VFIO_IOMMU_SPAPR_REGISTER_MEMORY. 1319 * Uses vfio_iommu_spapr_register_memory for parameters. 1320 */ 1321 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 18) 1322 1323 /** 1324 * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create) 1325 * 1326 * Creates an additional TCE table and programs it (sets a new DMA window) 1327 * to every IOMMU group in the container. It receives page shift, window 1328 * size and number of levels in the TCE table being created. 1329 * 1330 * It allocates and returns an offset on a PCI bus of the new DMA window. 1331 */ 1332 struct vfio_iommu_spapr_tce_create { 1333 __u32 argsz; 1334 __u32 flags; 1335 /* in */ 1336 __u32 page_shift; 1337 __u32 __resv1; 1338 __u64 window_size; 1339 __u32 levels; 1340 __u32 __resv2; 1341 /* out */ 1342 __u64 start_addr; 1343 }; 1344 #define VFIO_IOMMU_SPAPR_TCE_CREATE _IO(VFIO_TYPE, VFIO_BASE + 19) 1345 1346 /** 1347 * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove) 1348 * 1349 * Unprograms a TCE table from all groups in the container and destroys it. 1350 * It receives a PCI bus offset as a window id. 1351 */ 1352 struct vfio_iommu_spapr_tce_remove { 1353 __u32 argsz; 1354 __u32 flags; 1355 /* in */ 1356 __u64 start_addr; 1357 }; 1358 #define VFIO_IOMMU_SPAPR_TCE_REMOVE _IO(VFIO_TYPE, VFIO_BASE + 20) 1359 1360 /* ***************************************************************** */ 1361 1362 #endif /* _UAPIVFIO_H */ 1363