logo

oasis-root

Compiled tree of Oasis Linux based on own branch at <https://hacktivis.me/git/oasis/> git clone https://anongit.hacktivis.me/git/oasis-root.git

vfio.h (71862B)


  1. /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
  2. /*
  3. * VFIO API definition
  4. *
  5. * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
  6. * Author: Alex Williamson <alex.williamson@redhat.com>
  7. *
  8. * This program is free software; you can redistribute it and/or modify
  9. * it under the terms of the GNU General Public License version 2 as
  10. * published by the Free Software Foundation.
  11. */
  12. #ifndef VFIO_H
  13. #define VFIO_H
  14. #include <linux/types.h>
  15. #include <linux/ioctl.h>
  16. #define VFIO_API_VERSION 0
  17. /* Kernel & User level defines for VFIO IOCTLs. */
  18. /* Extensions */
  19. #define VFIO_TYPE1_IOMMU 1
  20. #define VFIO_SPAPR_TCE_IOMMU 2
  21. #define VFIO_TYPE1v2_IOMMU 3
  22. /*
  23. * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping). This
  24. * capability is subject to change as groups are added or removed.
  25. */
  26. #define VFIO_DMA_CC_IOMMU 4
  27. /* Check if EEH is supported */
  28. #define VFIO_EEH 5
  29. /* Two-stage IOMMU */
  30. #define VFIO_TYPE1_NESTING_IOMMU 6 /* Implies v2 */
  31. #define VFIO_SPAPR_TCE_v2_IOMMU 7
  32. /*
  33. * The No-IOMMU IOMMU offers no translation or isolation for devices and
  34. * supports no ioctls outside of VFIO_CHECK_EXTENSION. Use of VFIO's No-IOMMU
  35. * code will taint the host kernel and should be used with extreme caution.
  36. */
  37. #define VFIO_NOIOMMU_IOMMU 8
  38. /* Supports VFIO_DMA_UNMAP_FLAG_ALL */
  39. #define VFIO_UNMAP_ALL 9
  40. /*
  41. * Supports the vaddr flag for DMA map and unmap. Not supported for mediated
  42. * devices, so this capability is subject to change as groups are added or
  43. * removed.
  44. */
  45. #define VFIO_UPDATE_VADDR 10
  46. /*
  47. * The IOCTL interface is designed for extensibility by embedding the
  48. * structure length (argsz) and flags into structures passed between
  49. * kernel and userspace. We therefore use the _IO() macro for these
  50. * defines to avoid implicitly embedding a size into the ioctl request.
  51. * As structure fields are added, argsz will increase to match and flag
  52. * bits will be defined to indicate additional fields with valid data.
  53. * It's *always* the caller's responsibility to indicate the size of
  54. * the structure passed by setting argsz appropriately.
  55. */
  56. #define VFIO_TYPE (';')
  57. #define VFIO_BASE 100
  58. /*
  59. * For extension of INFO ioctls, VFIO makes use of a capability chain
  60. * designed after PCI/e capabilities. A flag bit indicates whether
  61. * this capability chain is supported and a field defined in the fixed
  62. * structure defines the offset of the first capability in the chain.
  63. * This field is only valid when the corresponding bit in the flags
  64. * bitmap is set. This offset field is relative to the start of the
  65. * INFO buffer, as is the next field within each capability header.
  66. * The id within the header is a shared address space per INFO ioctl,
  67. * while the version field is specific to the capability id. The
  68. * contents following the header are specific to the capability id.
  69. */
  70. struct vfio_info_cap_header {
  71. __u16 id; /* Identifies capability */
  72. __u16 version; /* Version specific to the capability ID */
  73. __u32 next; /* Offset of next capability */
  74. };
  75. /*
  76. * Callers of INFO ioctls passing insufficiently sized buffers will see
  77. * the capability chain flag bit set, a zero value for the first capability
  78. * offset (if available within the provided argsz), and argsz will be
  79. * updated to report the necessary buffer size. For compatibility, the
  80. * INFO ioctl will not report error in this case, but the capability chain
  81. * will not be available.
  82. */
  83. /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
  84. /**
  85. * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
  86. *
  87. * Report the version of the VFIO API. This allows us to bump the entire
  88. * API version should we later need to add or change features in incompatible
  89. * ways.
  90. * Return: VFIO_API_VERSION
  91. * Availability: Always
  92. */
  93. #define VFIO_GET_API_VERSION _IO(VFIO_TYPE, VFIO_BASE + 0)
  94. /**
  95. * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
  96. *
  97. * Check whether an extension is supported.
  98. * Return: 0 if not supported, 1 (or some other positive integer) if supported.
  99. * Availability: Always
  100. */
  101. #define VFIO_CHECK_EXTENSION _IO(VFIO_TYPE, VFIO_BASE + 1)
  102. /**
  103. * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
  104. *
  105. * Set the iommu to the given type. The type must be supported by an
  106. * iommu driver as verified by calling CHECK_EXTENSION using the same
  107. * type. A group must be set to this file descriptor before this
  108. * ioctl is available. The IOMMU interfaces enabled by this call are
  109. * specific to the value set.
  110. * Return: 0 on success, -errno on failure
  111. * Availability: When VFIO group attached
  112. */
  113. #define VFIO_SET_IOMMU _IO(VFIO_TYPE, VFIO_BASE + 2)
  114. /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
  115. /**
  116. * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
  117. * struct vfio_group_status)
  118. *
  119. * Retrieve information about the group. Fills in provided
  120. * struct vfio_group_info. Caller sets argsz.
  121. * Return: 0 on succes, -errno on failure.
  122. * Availability: Always
  123. */
  124. struct vfio_group_status {
  125. __u32 argsz;
  126. __u32 flags;
  127. #define VFIO_GROUP_FLAGS_VIABLE (1 << 0)
  128. #define VFIO_GROUP_FLAGS_CONTAINER_SET (1 << 1)
  129. };
  130. #define VFIO_GROUP_GET_STATUS _IO(VFIO_TYPE, VFIO_BASE + 3)
  131. /**
  132. * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
  133. *
  134. * Set the container for the VFIO group to the open VFIO file
  135. * descriptor provided. Groups may only belong to a single
  136. * container. Containers may, at their discretion, support multiple
  137. * groups. Only when a container is set are all of the interfaces
  138. * of the VFIO file descriptor and the VFIO group file descriptor
  139. * available to the user.
  140. * Return: 0 on success, -errno on failure.
  141. * Availability: Always
  142. */
  143. #define VFIO_GROUP_SET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 4)
  144. /**
  145. * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
  146. *
  147. * Remove the group from the attached container. This is the
  148. * opposite of the SET_CONTAINER call and returns the group to
  149. * an initial state. All device file descriptors must be released
  150. * prior to calling this interface. When removing the last group
  151. * from a container, the IOMMU will be disabled and all state lost,
  152. * effectively also returning the VFIO file descriptor to an initial
  153. * state.
  154. * Return: 0 on success, -errno on failure.
  155. * Availability: When attached to container
  156. */
  157. #define VFIO_GROUP_UNSET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 5)
  158. /**
  159. * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
  160. *
  161. * Return a new file descriptor for the device object described by
  162. * the provided string. The string should match a device listed in
  163. * the devices subdirectory of the IOMMU group sysfs entry. The
  164. * group containing the device must already be added to this context.
  165. * Return: new file descriptor on success, -errno on failure.
  166. * Availability: When attached to container
  167. */
  168. #define VFIO_GROUP_GET_DEVICE_FD _IO(VFIO_TYPE, VFIO_BASE + 6)
  169. /* --------------- IOCTLs for DEVICE file descriptors --------------- */
  170. /**
  171. * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
  172. * struct vfio_device_info)
  173. *
  174. * Retrieve information about the device. Fills in provided
  175. * struct vfio_device_info. Caller sets argsz.
  176. * Return: 0 on success, -errno on failure.
  177. */
  178. struct vfio_device_info {
  179. __u32 argsz;
  180. __u32 flags;
  181. #define VFIO_DEVICE_FLAGS_RESET (1 << 0) /* Device supports reset */
  182. #define VFIO_DEVICE_FLAGS_PCI (1 << 1) /* vfio-pci device */
  183. #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2) /* vfio-platform device */
  184. #define VFIO_DEVICE_FLAGS_AMBA (1 << 3) /* vfio-amba device */
  185. #define VFIO_DEVICE_FLAGS_CCW (1 << 4) /* vfio-ccw device */
  186. #define VFIO_DEVICE_FLAGS_AP (1 << 5) /* vfio-ap device */
  187. #define VFIO_DEVICE_FLAGS_FSL_MC (1 << 6) /* vfio-fsl-mc device */
  188. #define VFIO_DEVICE_FLAGS_CAPS (1 << 7) /* Info supports caps */
  189. #define VFIO_DEVICE_FLAGS_CDX (1 << 8) /* vfio-cdx device */
  190. __u32 num_regions; /* Max region index + 1 */
  191. __u32 num_irqs; /* Max IRQ index + 1 */
  192. __u32 cap_offset; /* Offset within info struct of first cap */
  193. __u32 pad;
  194. };
  195. #define VFIO_DEVICE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 7)
  196. /*
  197. * Vendor driver using Mediated device framework should provide device_api
  198. * attribute in supported type attribute groups. Device API string should be one
  199. * of the following corresponding to device flags in vfio_device_info structure.
  200. */
  201. #define VFIO_DEVICE_API_PCI_STRING "vfio-pci"
  202. #define VFIO_DEVICE_API_PLATFORM_STRING "vfio-platform"
  203. #define VFIO_DEVICE_API_AMBA_STRING "vfio-amba"
  204. #define VFIO_DEVICE_API_CCW_STRING "vfio-ccw"
  205. #define VFIO_DEVICE_API_AP_STRING "vfio-ap"
  206. /*
  207. * The following capabilities are unique to s390 zPCI devices. Their contents
  208. * are further-defined in vfio_zdev.h
  209. */
  210. #define VFIO_DEVICE_INFO_CAP_ZPCI_BASE 1
  211. #define VFIO_DEVICE_INFO_CAP_ZPCI_GROUP 2
  212. #define VFIO_DEVICE_INFO_CAP_ZPCI_UTIL 3
  213. #define VFIO_DEVICE_INFO_CAP_ZPCI_PFIP 4
  214. /*
  215. * The following VFIO_DEVICE_INFO capability reports support for PCIe AtomicOp
  216. * completion to the root bus with supported widths provided via flags.
  217. */
  218. #define VFIO_DEVICE_INFO_CAP_PCI_ATOMIC_COMP 5
  219. struct vfio_device_info_cap_pci_atomic_comp {
  220. struct vfio_info_cap_header header;
  221. __u32 flags;
  222. #define VFIO_PCI_ATOMIC_COMP32 (1 << 0)
  223. #define VFIO_PCI_ATOMIC_COMP64 (1 << 1)
  224. #define VFIO_PCI_ATOMIC_COMP128 (1 << 2)
  225. __u32 reserved;
  226. };
  227. /**
  228. * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
  229. * struct vfio_region_info)
  230. *
  231. * Retrieve information about a device region. Caller provides
  232. * struct vfio_region_info with index value set. Caller sets argsz.
  233. * Implementation of region mapping is bus driver specific. This is
  234. * intended to describe MMIO, I/O port, as well as bus specific
  235. * regions (ex. PCI config space). Zero sized regions may be used
  236. * to describe unimplemented regions (ex. unimplemented PCI BARs).
  237. * Return: 0 on success, -errno on failure.
  238. */
  239. struct vfio_region_info {
  240. __u32 argsz;
  241. __u32 flags;
  242. #define VFIO_REGION_INFO_FLAG_READ (1 << 0) /* Region supports read */
  243. #define VFIO_REGION_INFO_FLAG_WRITE (1 << 1) /* Region supports write */
  244. #define VFIO_REGION_INFO_FLAG_MMAP (1 << 2) /* Region supports mmap */
  245. #define VFIO_REGION_INFO_FLAG_CAPS (1 << 3) /* Info supports caps */
  246. __u32 index; /* Region index */
  247. __u32 cap_offset; /* Offset within info struct of first cap */
  248. __aligned_u64 size; /* Region size (bytes) */
  249. __aligned_u64 offset; /* Region offset from start of device fd */
  250. };
  251. #define VFIO_DEVICE_GET_REGION_INFO _IO(VFIO_TYPE, VFIO_BASE + 8)
  252. /*
  253. * The sparse mmap capability allows finer granularity of specifying areas
  254. * within a region with mmap support. When specified, the user should only
  255. * mmap the offset ranges specified by the areas array. mmaps outside of the
  256. * areas specified may fail (such as the range covering a PCI MSI-X table) or
  257. * may result in improper device behavior.
  258. *
  259. * The structures below define version 1 of this capability.
  260. */
  261. #define VFIO_REGION_INFO_CAP_SPARSE_MMAP 1
  262. struct vfio_region_sparse_mmap_area {
  263. __aligned_u64 offset; /* Offset of mmap'able area within region */
  264. __aligned_u64 size; /* Size of mmap'able area */
  265. };
  266. struct vfio_region_info_cap_sparse_mmap {
  267. struct vfio_info_cap_header header;
  268. __u32 nr_areas;
  269. __u32 reserved;
  270. struct vfio_region_sparse_mmap_area areas[];
  271. };
  272. /*
  273. * The device specific type capability allows regions unique to a specific
  274. * device or class of devices to be exposed. This helps solve the problem for
  275. * vfio bus drivers of defining which region indexes correspond to which region
  276. * on the device, without needing to resort to static indexes, as done by
  277. * vfio-pci. For instance, if we were to go back in time, we might remove
  278. * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
  279. * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
  280. * make a "VGA" device specific type to describe the VGA access space. This
  281. * means that non-VGA devices wouldn't need to waste this index, and thus the
  282. * address space associated with it due to implementation of device file
  283. * descriptor offsets in vfio-pci.
  284. *
  285. * The current implementation is now part of the user ABI, so we can't use this
  286. * for VGA, but there are other upcoming use cases, such as opregions for Intel
  287. * IGD devices and framebuffers for vGPU devices. We missed VGA, but we'll
  288. * use this for future additions.
  289. *
  290. * The structure below defines version 1 of this capability.
  291. */
  292. #define VFIO_REGION_INFO_CAP_TYPE 2
  293. struct vfio_region_info_cap_type {
  294. struct vfio_info_cap_header header;
  295. __u32 type; /* global per bus driver */
  296. __u32 subtype; /* type specific */
  297. };
  298. /*
  299. * List of region types, global per bus driver.
  300. * If you introduce a new type, please add it here.
  301. */
  302. /* PCI region type containing a PCI vendor part */
  303. #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE (1 << 31)
  304. #define VFIO_REGION_TYPE_PCI_VENDOR_MASK (0xffff)
  305. #define VFIO_REGION_TYPE_GFX (1)
  306. #define VFIO_REGION_TYPE_CCW (2)
  307. #define VFIO_REGION_TYPE_MIGRATION_DEPRECATED (3)
  308. /* sub-types for VFIO_REGION_TYPE_PCI_* */
  309. /* 8086 vendor PCI sub-types */
  310. #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION (1)
  311. #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG (2)
  312. #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG (3)
  313. /* 10de vendor PCI sub-types */
  314. /*
  315. * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
  316. *
  317. * Deprecated, region no longer provided
  318. */
  319. #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM (1)
  320. /* 1014 vendor PCI sub-types */
  321. /*
  322. * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
  323. * to do TLB invalidation on a GPU.
  324. *
  325. * Deprecated, region no longer provided
  326. */
  327. #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD (1)
  328. /* sub-types for VFIO_REGION_TYPE_GFX */
  329. #define VFIO_REGION_SUBTYPE_GFX_EDID (1)
  330. /**
  331. * struct vfio_region_gfx_edid - EDID region layout.
  332. *
  333. * Set display link state and EDID blob.
  334. *
  335. * The EDID blob has monitor information such as brand, name, serial
  336. * number, physical size, supported video modes and more.
  337. *
  338. * This special region allows userspace (typically qemu) set a virtual
  339. * EDID for the virtual monitor, which allows a flexible display
  340. * configuration.
  341. *
  342. * For the edid blob spec look here:
  343. * https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
  344. *
  345. * On linux systems you can find the EDID blob in sysfs:
  346. * /sys/class/drm/${card}/${connector}/edid
  347. *
  348. * You can use the edid-decode ulility (comes with xorg-x11-utils) to
  349. * decode the EDID blob.
  350. *
  351. * @edid_offset: location of the edid blob, relative to the
  352. * start of the region (readonly).
  353. * @edid_max_size: max size of the edid blob (readonly).
  354. * @edid_size: actual edid size (read/write).
  355. * @link_state: display link state (read/write).
  356. * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
  357. * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
  358. * @max_xres: max display width (0 == no limitation, readonly).
  359. * @max_yres: max display height (0 == no limitation, readonly).
  360. *
  361. * EDID update protocol:
  362. * (1) set link-state to down.
  363. * (2) update edid blob and size.
  364. * (3) set link-state to up.
  365. */
  366. struct vfio_region_gfx_edid {
  367. __u32 edid_offset;
  368. __u32 edid_max_size;
  369. __u32 edid_size;
  370. __u32 max_xres;
  371. __u32 max_yres;
  372. __u32 link_state;
  373. #define VFIO_DEVICE_GFX_LINK_STATE_UP 1
  374. #define VFIO_DEVICE_GFX_LINK_STATE_DOWN 2
  375. };
  376. /* sub-types for VFIO_REGION_TYPE_CCW */
  377. #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD (1)
  378. #define VFIO_REGION_SUBTYPE_CCW_SCHIB (2)
  379. #define VFIO_REGION_SUBTYPE_CCW_CRW (3)
  380. /* sub-types for VFIO_REGION_TYPE_MIGRATION */
  381. #define VFIO_REGION_SUBTYPE_MIGRATION_DEPRECATED (1)
  382. struct vfio_device_migration_info {
  383. __u32 device_state; /* VFIO device state */
  384. #define VFIO_DEVICE_STATE_V1_STOP (0)
  385. #define VFIO_DEVICE_STATE_V1_RUNNING (1 << 0)
  386. #define VFIO_DEVICE_STATE_V1_SAVING (1 << 1)
  387. #define VFIO_DEVICE_STATE_V1_RESUMING (1 << 2)
  388. #define VFIO_DEVICE_STATE_MASK (VFIO_DEVICE_STATE_V1_RUNNING | \
  389. VFIO_DEVICE_STATE_V1_SAVING | \
  390. VFIO_DEVICE_STATE_V1_RESUMING)
  391. #define VFIO_DEVICE_STATE_VALID(state) \
  392. (state & VFIO_DEVICE_STATE_V1_RESUMING ? \
  393. (state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_V1_RESUMING : 1)
  394. #define VFIO_DEVICE_STATE_IS_ERROR(state) \
  395. ((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_V1_SAVING | \
  396. VFIO_DEVICE_STATE_V1_RESUMING))
  397. #define VFIO_DEVICE_STATE_SET_ERROR(state) \
  398. ((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_STATE_V1_SAVING | \
  399. VFIO_DEVICE_STATE_V1_RESUMING)
  400. __u32 reserved;
  401. __aligned_u64 pending_bytes;
  402. __aligned_u64 data_offset;
  403. __aligned_u64 data_size;
  404. };
  405. /*
  406. * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
  407. * which allows direct access to non-MSIX registers which happened to be within
  408. * the same system page.
  409. *
  410. * Even though the userspace gets direct access to the MSIX data, the existing
  411. * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
  412. */
  413. #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE 3
  414. /*
  415. * Capability with compressed real address (aka SSA - small system address)
  416. * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
  417. * and by the userspace to associate a NVLink bridge with a GPU.
  418. *
  419. * Deprecated, capability no longer provided
  420. */
  421. #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT 4
  422. struct vfio_region_info_cap_nvlink2_ssatgt {
  423. struct vfio_info_cap_header header;
  424. __aligned_u64 tgt;
  425. };
  426. /*
  427. * Capability with an NVLink link speed. The value is read by
  428. * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
  429. * property in the device tree. The value is fixed in the hardware
  430. * and failing to provide the correct value results in the link
  431. * not working with no indication from the driver why.
  432. *
  433. * Deprecated, capability no longer provided
  434. */
  435. #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD 5
  436. struct vfio_region_info_cap_nvlink2_lnkspd {
  437. struct vfio_info_cap_header header;
  438. __u32 link_speed;
  439. __u32 __pad;
  440. };
  441. /**
  442. * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
  443. * struct vfio_irq_info)
  444. *
  445. * Retrieve information about a device IRQ. Caller provides
  446. * struct vfio_irq_info with index value set. Caller sets argsz.
  447. * Implementation of IRQ mapping is bus driver specific. Indexes
  448. * using multiple IRQs are primarily intended to support MSI-like
  449. * interrupt blocks. Zero count irq blocks may be used to describe
  450. * unimplemented interrupt types.
  451. *
  452. * The EVENTFD flag indicates the interrupt index supports eventfd based
  453. * signaling.
  454. *
  455. * The MASKABLE flags indicates the index supports MASK and UNMASK
  456. * actions described below.
  457. *
  458. * AUTOMASKED indicates that after signaling, the interrupt line is
  459. * automatically masked by VFIO and the user needs to unmask the line
  460. * to receive new interrupts. This is primarily intended to distinguish
  461. * level triggered interrupts.
  462. *
  463. * The NORESIZE flag indicates that the interrupt lines within the index
  464. * are setup as a set and new subindexes cannot be enabled without first
  465. * disabling the entire index. This is used for interrupts like PCI MSI
  466. * and MSI-X where the driver may only use a subset of the available
  467. * indexes, but VFIO needs to enable a specific number of vectors
  468. * upfront. In the case of MSI-X, where the user can enable MSI-X and
  469. * then add and unmask vectors, it's up to userspace to make the decision
  470. * whether to allocate the maximum supported number of vectors or tear
  471. * down setup and incrementally increase the vectors as each is enabled.
  472. * Absence of the NORESIZE flag indicates that vectors can be enabled
  473. * and disabled dynamically without impacting other vectors within the
  474. * index.
  475. */
  476. struct vfio_irq_info {
  477. __u32 argsz;
  478. __u32 flags;
  479. #define VFIO_IRQ_INFO_EVENTFD (1 << 0)
  480. #define VFIO_IRQ_INFO_MASKABLE (1 << 1)
  481. #define VFIO_IRQ_INFO_AUTOMASKED (1 << 2)
  482. #define VFIO_IRQ_INFO_NORESIZE (1 << 3)
  483. __u32 index; /* IRQ index */
  484. __u32 count; /* Number of IRQs within this index */
  485. };
  486. #define VFIO_DEVICE_GET_IRQ_INFO _IO(VFIO_TYPE, VFIO_BASE + 9)
  487. /**
  488. * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
  489. *
  490. * Set signaling, masking, and unmasking of interrupts. Caller provides
  491. * struct vfio_irq_set with all fields set. 'start' and 'count' indicate
  492. * the range of subindexes being specified.
  493. *
  494. * The DATA flags specify the type of data provided. If DATA_NONE, the
  495. * operation performs the specified action immediately on the specified
  496. * interrupt(s). For example, to unmask AUTOMASKED interrupt [0,0]:
  497. * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
  498. *
  499. * DATA_BOOL allows sparse support for the same on arrays of interrupts.
  500. * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
  501. * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
  502. * data = {1,0,1}
  503. *
  504. * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
  505. * A value of -1 can be used to either de-assign interrupts if already
  506. * assigned or skip un-assigned interrupts. For example, to set an eventfd
  507. * to be trigger for interrupts [0,0] and [0,2]:
  508. * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
  509. * data = {fd1, -1, fd2}
  510. * If index [0,1] is previously set, two count = 1 ioctls calls would be
  511. * required to set [0,0] and [0,2] without changing [0,1].
  512. *
  513. * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
  514. * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
  515. * from userspace (ie. simulate hardware triggering).
  516. *
  517. * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
  518. * enables the interrupt index for the device. Individual subindex interrupts
  519. * can be disabled using the -1 value for DATA_EVENTFD or the index can be
  520. * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
  521. *
  522. * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
  523. * ACTION_TRIGGER specifies kernel->user signaling.
  524. */
  525. struct vfio_irq_set {
  526. __u32 argsz;
  527. __u32 flags;
  528. #define VFIO_IRQ_SET_DATA_NONE (1 << 0) /* Data not present */
  529. #define VFIO_IRQ_SET_DATA_BOOL (1 << 1) /* Data is bool (u8) */
  530. #define VFIO_IRQ_SET_DATA_EVENTFD (1 << 2) /* Data is eventfd (s32) */
  531. #define VFIO_IRQ_SET_ACTION_MASK (1 << 3) /* Mask interrupt */
  532. #define VFIO_IRQ_SET_ACTION_UNMASK (1 << 4) /* Unmask interrupt */
  533. #define VFIO_IRQ_SET_ACTION_TRIGGER (1 << 5) /* Trigger interrupt */
  534. __u32 index;
  535. __u32 start;
  536. __u32 count;
  537. __u8 data[];
  538. };
  539. #define VFIO_DEVICE_SET_IRQS _IO(VFIO_TYPE, VFIO_BASE + 10)
  540. #define VFIO_IRQ_SET_DATA_TYPE_MASK (VFIO_IRQ_SET_DATA_NONE | \
  541. VFIO_IRQ_SET_DATA_BOOL | \
  542. VFIO_IRQ_SET_DATA_EVENTFD)
  543. #define VFIO_IRQ_SET_ACTION_TYPE_MASK (VFIO_IRQ_SET_ACTION_MASK | \
  544. VFIO_IRQ_SET_ACTION_UNMASK | \
  545. VFIO_IRQ_SET_ACTION_TRIGGER)
  546. /**
  547. * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
  548. *
  549. * Reset a device.
  550. */
  551. #define VFIO_DEVICE_RESET _IO(VFIO_TYPE, VFIO_BASE + 11)
  552. /*
  553. * The VFIO-PCI bus driver makes use of the following fixed region and
  554. * IRQ index mapping. Unimplemented regions return a size of zero.
  555. * Unimplemented IRQ types return a count of zero.
  556. */
  557. enum {
  558. VFIO_PCI_BAR0_REGION_INDEX,
  559. VFIO_PCI_BAR1_REGION_INDEX,
  560. VFIO_PCI_BAR2_REGION_INDEX,
  561. VFIO_PCI_BAR3_REGION_INDEX,
  562. VFIO_PCI_BAR4_REGION_INDEX,
  563. VFIO_PCI_BAR5_REGION_INDEX,
  564. VFIO_PCI_ROM_REGION_INDEX,
  565. VFIO_PCI_CONFIG_REGION_INDEX,
  566. /*
  567. * Expose VGA regions defined for PCI base class 03, subclass 00.
  568. * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
  569. * as well as the MMIO range 0xa0000 to 0xbffff. Each implemented
  570. * range is found at it's identity mapped offset from the region
  571. * offset, for example 0x3b0 is region_info.offset + 0x3b0. Areas
  572. * between described ranges are unimplemented.
  573. */
  574. VFIO_PCI_VGA_REGION_INDEX,
  575. VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
  576. /* device specific cap to define content. */
  577. };
  578. enum {
  579. VFIO_PCI_INTX_IRQ_INDEX,
  580. VFIO_PCI_MSI_IRQ_INDEX,
  581. VFIO_PCI_MSIX_IRQ_INDEX,
  582. VFIO_PCI_ERR_IRQ_INDEX,
  583. VFIO_PCI_REQ_IRQ_INDEX,
  584. VFIO_PCI_NUM_IRQS
  585. };
  586. /*
  587. * The vfio-ccw bus driver makes use of the following fixed region and
  588. * IRQ index mapping. Unimplemented regions return a size of zero.
  589. * Unimplemented IRQ types return a count of zero.
  590. */
  591. enum {
  592. VFIO_CCW_CONFIG_REGION_INDEX,
  593. VFIO_CCW_NUM_REGIONS
  594. };
  595. enum {
  596. VFIO_CCW_IO_IRQ_INDEX,
  597. VFIO_CCW_CRW_IRQ_INDEX,
  598. VFIO_CCW_REQ_IRQ_INDEX,
  599. VFIO_CCW_NUM_IRQS
  600. };
  601. /*
  602. * The vfio-ap bus driver makes use of the following IRQ index mapping.
  603. * Unimplemented IRQ types return a count of zero.
  604. */
  605. enum {
  606. VFIO_AP_REQ_IRQ_INDEX,
  607. VFIO_AP_NUM_IRQS
  608. };
  609. /**
  610. * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12,
  611. * struct vfio_pci_hot_reset_info)
  612. *
  613. * This command is used to query the affected devices in the hot reset for
  614. * a given device.
  615. *
  616. * This command always reports the segment, bus, and devfn information for
  617. * each affected device, and selectively reports the group_id or devid per
  618. * the way how the calling device is opened.
  619. *
  620. * - If the calling device is opened via the traditional group/container
  621. * API, group_id is reported. User should check if it has owned all
  622. * the affected devices and provides a set of group fds to prove the
  623. * ownership in VFIO_DEVICE_PCI_HOT_RESET ioctl.
  624. *
  625. * - If the calling device is opened as a cdev, devid is reported.
  626. * Flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set to indicate this
  627. * data type. All the affected devices should be represented in
  628. * the dev_set, ex. bound to a vfio driver, and also be owned by
  629. * this interface which is determined by the following conditions:
  630. * 1) Has a valid devid within the iommufd_ctx of the calling device.
  631. * Ownership cannot be determined across separate iommufd_ctx and
  632. * the cdev calling conventions do not support a proof-of-ownership
  633. * model as provided in the legacy group interface. In this case
  634. * valid devid with value greater than zero is provided in the return
  635. * structure.
  636. * 2) Does not have a valid devid within the iommufd_ctx of the calling
  637. * device, but belongs to the same IOMMU group as the calling device
  638. * or another opened device that has a valid devid within the
  639. * iommufd_ctx of the calling device. This provides implicit ownership
  640. * for devices within the same DMA isolation context. In this case
  641. * the devid value of VFIO_PCI_DEVID_OWNED is provided in the return
  642. * structure.
  643. *
  644. * A devid value of VFIO_PCI_DEVID_NOT_OWNED is provided in the return
  645. * structure for affected devices where device is NOT represented in the
  646. * dev_set or ownership is not available. Such devices prevent the use
  647. * of VFIO_DEVICE_PCI_HOT_RESET ioctl outside of the proof-of-ownership
  648. * calling conventions (ie. via legacy group accessed devices). Flag
  649. * VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED would be set when all the
  650. * affected devices are represented in the dev_set and also owned by
  651. * the user. This flag is available only when
  652. * flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set, otherwise reserved.
  653. * When set, user could invoke VFIO_DEVICE_PCI_HOT_RESET with a zero
  654. * length fd array on the calling device as the ownership is validated
  655. * by iommufd_ctx.
  656. *
  657. * Return: 0 on success, -errno on failure:
  658. * -enospc = insufficient buffer, -enodev = unsupported for device.
  659. */
  660. struct vfio_pci_dependent_device {
  661. union {
  662. __u32 group_id;
  663. __u32 devid;
  664. #define VFIO_PCI_DEVID_OWNED 0
  665. #define VFIO_PCI_DEVID_NOT_OWNED -1
  666. };
  667. __u16 segment;
  668. __u8 bus;
  669. __u8 devfn; /* Use PCI_SLOT/PCI_FUNC */
  670. };
  671. struct vfio_pci_hot_reset_info {
  672. __u32 argsz;
  673. __u32 flags;
  674. #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID (1 << 0)
  675. #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED (1 << 1)
  676. __u32 count;
  677. struct vfio_pci_dependent_device devices[];
  678. };
  679. #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
  680. /**
  681. * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
  682. * struct vfio_pci_hot_reset)
  683. *
  684. * A PCI hot reset results in either a bus or slot reset which may affect
  685. * other devices sharing the bus/slot. The calling user must have
  686. * ownership of the full set of affected devices as determined by the
  687. * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO ioctl.
  688. *
  689. * When called on a device file descriptor acquired through the vfio
  690. * group interface, the user is required to provide proof of ownership
  691. * of those affected devices via the group_fds array in struct
  692. * vfio_pci_hot_reset.
  693. *
  694. * When called on a direct cdev opened vfio device, the flags field of
  695. * struct vfio_pci_hot_reset_info reports the ownership status of the
  696. * affected devices and this ioctl must be called with an empty group_fds
  697. * array. See above INFO ioctl definition for ownership requirements.
  698. *
  699. * Mixed usage of legacy groups and cdevs across the set of affected
  700. * devices is not supported.
  701. *
  702. * Return: 0 on success, -errno on failure.
  703. */
  704. struct vfio_pci_hot_reset {
  705. __u32 argsz;
  706. __u32 flags;
  707. __u32 count;
  708. __s32 group_fds[];
  709. };
  710. #define VFIO_DEVICE_PCI_HOT_RESET _IO(VFIO_TYPE, VFIO_BASE + 13)
  711. /**
  712. * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
  713. * struct vfio_device_query_gfx_plane)
  714. *
  715. * Set the drm_plane_type and flags, then retrieve the gfx plane info.
  716. *
  717. * flags supported:
  718. * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
  719. * to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
  720. * support for dma-buf.
  721. * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
  722. * to ask if the mdev supports region. 0 on support, -EINVAL on no
  723. * support for region.
  724. * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
  725. * with each call to query the plane info.
  726. * - Others are invalid and return -EINVAL.
  727. *
  728. * Note:
  729. * 1. Plane could be disabled by guest. In that case, success will be
  730. * returned with zero-initialized drm_format, size, width and height
  731. * fields.
  732. * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
  733. *
  734. * Return: 0 on success, -errno on other failure.
  735. */
  736. struct vfio_device_gfx_plane_info {
  737. __u32 argsz;
  738. __u32 flags;
  739. #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
  740. #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
  741. #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
  742. /* in */
  743. __u32 drm_plane_type; /* type of plane: DRM_PLANE_TYPE_* */
  744. /* out */
  745. __u32 drm_format; /* drm format of plane */
  746. __aligned_u64 drm_format_mod; /* tiled mode */
  747. __u32 width; /* width of plane */
  748. __u32 height; /* height of plane */
  749. __u32 stride; /* stride of plane */
  750. __u32 size; /* size of plane in bytes, align on page*/
  751. __u32 x_pos; /* horizontal position of cursor plane */
  752. __u32 y_pos; /* vertical position of cursor plane*/
  753. __u32 x_hot; /* horizontal position of cursor hotspot */
  754. __u32 y_hot; /* vertical position of cursor hotspot */
  755. union {
  756. __u32 region_index; /* region index */
  757. __u32 dmabuf_id; /* dma-buf id */
  758. };
  759. __u32 reserved;
  760. };
  761. #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
  762. /**
  763. * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
  764. *
  765. * Return a new dma-buf file descriptor for an exposed guest framebuffer
  766. * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
  767. * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
  768. */
  769. #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
  770. /**
  771. * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
  772. * struct vfio_device_ioeventfd)
  773. *
  774. * Perform a write to the device at the specified device fd offset, with
  775. * the specified data and width when the provided eventfd is triggered.
  776. * vfio bus drivers may not support this for all regions, for all widths,
  777. * or at all. vfio-pci currently only enables support for BAR regions,
  778. * excluding the MSI-X vector table.
  779. *
  780. * Return: 0 on success, -errno on failure.
  781. */
  782. struct vfio_device_ioeventfd {
  783. __u32 argsz;
  784. __u32 flags;
  785. #define VFIO_DEVICE_IOEVENTFD_8 (1 << 0) /* 1-byte write */
  786. #define VFIO_DEVICE_IOEVENTFD_16 (1 << 1) /* 2-byte write */
  787. #define VFIO_DEVICE_IOEVENTFD_32 (1 << 2) /* 4-byte write */
  788. #define VFIO_DEVICE_IOEVENTFD_64 (1 << 3) /* 8-byte write */
  789. #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK (0xf)
  790. __aligned_u64 offset; /* device fd offset of write */
  791. __aligned_u64 data; /* data to be written */
  792. __s32 fd; /* -1 for de-assignment */
  793. __u32 reserved;
  794. };
  795. #define VFIO_DEVICE_IOEVENTFD _IO(VFIO_TYPE, VFIO_BASE + 16)
  796. /**
  797. * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
  798. * struct vfio_device_feature)
  799. *
  800. * Get, set, or probe feature data of the device. The feature is selected
  801. * using the FEATURE_MASK portion of the flags field. Support for a feature
  802. * can be probed by setting both the FEATURE_MASK and PROBE bits. A probe
  803. * may optionally include the GET and/or SET bits to determine read vs write
  804. * access of the feature respectively. Probing a feature will return success
  805. * if the feature is supported and all of the optionally indicated GET/SET
  806. * methods are supported. The format of the data portion of the structure is
  807. * specific to the given feature. The data portion is not required for
  808. * probing. GET and SET are mutually exclusive, except for use with PROBE.
  809. *
  810. * Return 0 on success, -errno on failure.
  811. */
  812. struct vfio_device_feature {
  813. __u32 argsz;
  814. __u32 flags;
  815. #define VFIO_DEVICE_FEATURE_MASK (0xffff) /* 16-bit feature index */
  816. #define VFIO_DEVICE_FEATURE_GET (1 << 16) /* Get feature into data[] */
  817. #define VFIO_DEVICE_FEATURE_SET (1 << 17) /* Set feature from data[] */
  818. #define VFIO_DEVICE_FEATURE_PROBE (1 << 18) /* Probe feature support */
  819. __u8 data[];
  820. };
  821. #define VFIO_DEVICE_FEATURE _IO(VFIO_TYPE, VFIO_BASE + 17)
  822. /*
  823. * VFIO_DEVICE_BIND_IOMMUFD - _IOR(VFIO_TYPE, VFIO_BASE + 18,
  824. * struct vfio_device_bind_iommufd)
  825. * @argsz: User filled size of this data.
  826. * @flags: Must be 0.
  827. * @iommufd: iommufd to bind.
  828. * @out_devid: The device id generated by this bind. devid is a handle for
  829. * this device/iommufd bond and can be used in IOMMUFD commands.
  830. *
  831. * Bind a vfio_device to the specified iommufd.
  832. *
  833. * User is restricted from accessing the device before the binding operation
  834. * is completed. Only allowed on cdev fds.
  835. *
  836. * Unbind is automatically conducted when device fd is closed.
  837. *
  838. * Return: 0 on success, -errno on failure.
  839. */
  840. struct vfio_device_bind_iommufd {
  841. __u32 argsz;
  842. __u32 flags;
  843. __s32 iommufd;
  844. __u32 out_devid;
  845. };
  846. #define VFIO_DEVICE_BIND_IOMMUFD _IO(VFIO_TYPE, VFIO_BASE + 18)
  847. /*
  848. * VFIO_DEVICE_ATTACH_IOMMUFD_PT - _IOW(VFIO_TYPE, VFIO_BASE + 19,
  849. * struct vfio_device_attach_iommufd_pt)
  850. * @argsz: User filled size of this data.
  851. * @flags: Must be 0.
  852. * @pt_id: Input the target id which can represent an ioas or a hwpt
  853. * allocated via iommufd subsystem.
  854. * Output the input ioas id or the attached hwpt id which could
  855. * be the specified hwpt itself or a hwpt automatically created
  856. * for the specified ioas by kernel during the attachment.
  857. *
  858. * Associate the device with an address space within the bound iommufd.
  859. * Undo by VFIO_DEVICE_DETACH_IOMMUFD_PT or device fd close. This is only
  860. * allowed on cdev fds.
  861. *
  862. * If a vfio device is currently attached to a valid hw_pagetable, without doing
  863. * a VFIO_DEVICE_DETACH_IOMMUFD_PT, a second VFIO_DEVICE_ATTACH_IOMMUFD_PT ioctl
  864. * passing in another hw_pagetable (hwpt) id is allowed. This action, also known
  865. * as a hw_pagetable replacement, will replace the device's currently attached
  866. * hw_pagetable with a new hw_pagetable corresponding to the given pt_id.
  867. *
  868. * Return: 0 on success, -errno on failure.
  869. */
  870. struct vfio_device_attach_iommufd_pt {
  871. __u32 argsz;
  872. __u32 flags;
  873. __u32 pt_id;
  874. };
  875. #define VFIO_DEVICE_ATTACH_IOMMUFD_PT _IO(VFIO_TYPE, VFIO_BASE + 19)
  876. /*
  877. * VFIO_DEVICE_DETACH_IOMMUFD_PT - _IOW(VFIO_TYPE, VFIO_BASE + 20,
  878. * struct vfio_device_detach_iommufd_pt)
  879. * @argsz: User filled size of this data.
  880. * @flags: Must be 0.
  881. *
  882. * Remove the association of the device and its current associated address
  883. * space. After it, the device should be in a blocking DMA state. This is only
  884. * allowed on cdev fds.
  885. *
  886. * Return: 0 on success, -errno on failure.
  887. */
  888. struct vfio_device_detach_iommufd_pt {
  889. __u32 argsz;
  890. __u32 flags;
  891. };
  892. #define VFIO_DEVICE_DETACH_IOMMUFD_PT _IO(VFIO_TYPE, VFIO_BASE + 20)
  893. /*
  894. * Provide support for setting a PCI VF Token, which is used as a shared
  895. * secret between PF and VF drivers. This feature may only be set on a
  896. * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
  897. * open VFs. Data provided when setting this feature is a 16-byte array
  898. * (__u8 b[16]), representing a UUID.
  899. */
  900. #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN (0)
  901. /*
  902. * Indicates the device can support the migration API through
  903. * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and
  904. * ERROR states are always supported. Support for additional states is
  905. * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be
  906. * set.
  907. *
  908. * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and
  909. * RESUMING are supported.
  910. *
  911. * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P
  912. * is supported in addition to the STOP_COPY states.
  913. *
  914. * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that
  915. * PRE_COPY is supported in addition to the STOP_COPY states.
  916. *
  917. * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY
  918. * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported
  919. * in addition to the STOP_COPY states.
  920. *
  921. * Other combinations of flags have behavior to be defined in the future.
  922. */
  923. struct vfio_device_feature_migration {
  924. __aligned_u64 flags;
  925. #define VFIO_MIGRATION_STOP_COPY (1 << 0)
  926. #define VFIO_MIGRATION_P2P (1 << 1)
  927. #define VFIO_MIGRATION_PRE_COPY (1 << 2)
  928. };
  929. #define VFIO_DEVICE_FEATURE_MIGRATION 1
  930. /*
  931. * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO
  932. * device. The new state is supplied in device_state, see enum
  933. * vfio_device_mig_state for details
  934. *
  935. * The kernel migration driver must fully transition the device to the new state
  936. * value before the operation returns to the user.
  937. *
  938. * The kernel migration driver must not generate asynchronous device state
  939. * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET
  940. * ioctl as described above.
  941. *
  942. * If this function fails then current device_state may be the original
  943. * operating state or some other state along the combination transition path.
  944. * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt
  945. * to return to the original state, or attempt to return to some other state
  946. * such as RUNNING or STOP.
  947. *
  948. * If the new_state starts a new data transfer session then the FD associated
  949. * with that session is returned in data_fd. The user is responsible to close
  950. * this FD when it is finished. The user must consider the migration data stream
  951. * carried over the FD to be opaque and must preserve the byte order of the
  952. * stream. The user is not required to preserve buffer segmentation when writing
  953. * the data stream during the RESUMING operation.
  954. *
  955. * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO
  956. * device, data_fd will be -1.
  957. */
  958. struct vfio_device_feature_mig_state {
  959. __u32 device_state; /* From enum vfio_device_mig_state */
  960. __s32 data_fd;
  961. };
  962. #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2
  963. /*
  964. * The device migration Finite State Machine is described by the enum
  965. * vfio_device_mig_state. Some of the FSM arcs will create a migration data
  966. * transfer session by returning a FD, in this case the migration data will
  967. * flow over the FD using read() and write() as discussed below.
  968. *
  969. * There are 5 states to support VFIO_MIGRATION_STOP_COPY:
  970. * RUNNING - The device is running normally
  971. * STOP - The device does not change the internal or external state
  972. * STOP_COPY - The device internal state can be read out
  973. * RESUMING - The device is stopped and is loading a new internal state
  974. * ERROR - The device has failed and must be reset
  975. *
  976. * And optional states to support VFIO_MIGRATION_P2P:
  977. * RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA
  978. * And VFIO_MIGRATION_PRE_COPY:
  979. * PRE_COPY - The device is running normally but tracking internal state
  980. * changes
  981. * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY:
  982. * PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA
  983. *
  984. * The FSM takes actions on the arcs between FSM states. The driver implements
  985. * the following behavior for the FSM arcs:
  986. *
  987. * RUNNING_P2P -> STOP
  988. * STOP_COPY -> STOP
  989. * While in STOP the device must stop the operation of the device. The device
  990. * must not generate interrupts, DMA, or any other change to external state.
  991. * It must not change its internal state. When stopped the device and kernel
  992. * migration driver must accept and respond to interaction to support external
  993. * subsystems in the STOP state, for example PCI MSI-X and PCI config space.
  994. * Failure by the user to restrict device access while in STOP must not result
  995. * in error conditions outside the user context (ex. host system faults).
  996. *
  997. * The STOP_COPY arc will terminate a data transfer session.
  998. *
  999. * RESUMING -> STOP
  1000. * Leaving RESUMING terminates a data transfer session and indicates the
  1001. * device should complete processing of the data delivered by write(). The
  1002. * kernel migration driver should complete the incorporation of data written
  1003. * to the data transfer FD into the device internal state and perform
  1004. * final validity and consistency checking of the new device state. If the
  1005. * user provided data is found to be incomplete, inconsistent, or otherwise
  1006. * invalid, the migration driver must fail the SET_STATE ioctl and
  1007. * optionally go to the ERROR state as described below.
  1008. *
  1009. * While in STOP the device has the same behavior as other STOP states
  1010. * described above.
  1011. *
  1012. * To abort a RESUMING session the device must be reset.
  1013. *
  1014. * PRE_COPY -> RUNNING
  1015. * RUNNING_P2P -> RUNNING
  1016. * While in RUNNING the device is fully operational, the device may generate
  1017. * interrupts, DMA, respond to MMIO, all vfio device regions are functional,
  1018. * and the device may advance its internal state.
  1019. *
  1020. * The PRE_COPY arc will terminate a data transfer session.
  1021. *
  1022. * PRE_COPY_P2P -> RUNNING_P2P
  1023. * RUNNING -> RUNNING_P2P
  1024. * STOP -> RUNNING_P2P
  1025. * While in RUNNING_P2P the device is partially running in the P2P quiescent
  1026. * state defined below.
  1027. *
  1028. * The PRE_COPY_P2P arc will terminate a data transfer session.
  1029. *
  1030. * RUNNING -> PRE_COPY
  1031. * RUNNING_P2P -> PRE_COPY_P2P
  1032. * STOP -> STOP_COPY
  1033. * PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states
  1034. * which share a data transfer session. Moving between these states alters
  1035. * what is streamed in session, but does not terminate or otherwise affect
  1036. * the associated fd.
  1037. *
  1038. * These arcs begin the process of saving the device state and will return a
  1039. * new data_fd. The migration driver may perform actions such as enabling
  1040. * dirty logging of device state when entering PRE_COPY or PER_COPY_P2P.
  1041. *
  1042. * Each arc does not change the device operation, the device remains
  1043. * RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below
  1044. * in PRE_COPY_P2P -> STOP_COPY.
  1045. *
  1046. * PRE_COPY -> PRE_COPY_P2P
  1047. * Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above.
  1048. * However, while in the PRE_COPY_P2P state, the device is partially running
  1049. * in the P2P quiescent state defined below, like RUNNING_P2P.
  1050. *
  1051. * PRE_COPY_P2P -> PRE_COPY
  1052. * This arc allows returning the device to a full RUNNING behavior while
  1053. * continuing all the behaviors of PRE_COPY.
  1054. *
  1055. * PRE_COPY_P2P -> STOP_COPY
  1056. * While in the STOP_COPY state the device has the same behavior as STOP
  1057. * with the addition that the data transfers session continues to stream the
  1058. * migration state. End of stream on the FD indicates the entire device
  1059. * state has been transferred.
  1060. *
  1061. * The user should take steps to restrict access to vfio device regions while
  1062. * the device is in STOP_COPY or risk corruption of the device migration data
  1063. * stream.
  1064. *
  1065. * STOP -> RESUMING
  1066. * Entering the RESUMING state starts a process of restoring the device state
  1067. * and will return a new data_fd. The data stream fed into the data_fd should
  1068. * be taken from the data transfer output of a single FD during saving from
  1069. * a compatible device. The migration driver may alter/reset the internal
  1070. * device state for this arc if required to prepare the device to receive the
  1071. * migration data.
  1072. *
  1073. * STOP_COPY -> PRE_COPY
  1074. * STOP_COPY -> PRE_COPY_P2P
  1075. * These arcs are not permitted and return error if requested. Future
  1076. * revisions of this API may define behaviors for these arcs, in this case
  1077. * support will be discoverable by a new flag in
  1078. * VFIO_DEVICE_FEATURE_MIGRATION.
  1079. *
  1080. * any -> ERROR
  1081. * ERROR cannot be specified as a device state, however any transition request
  1082. * can be failed with an errno return and may then move the device_state into
  1083. * ERROR. In this case the device was unable to execute the requested arc and
  1084. * was also unable to restore the device to any valid device_state.
  1085. * To recover from ERROR VFIO_DEVICE_RESET must be used to return the
  1086. * device_state back to RUNNING.
  1087. *
  1088. * The optional peer to peer (P2P) quiescent state is intended to be a quiescent
  1089. * state for the device for the purposes of managing multiple devices within a
  1090. * user context where peer-to-peer DMA between devices may be active. The
  1091. * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating
  1092. * any new P2P DMA transactions. If the device can identify P2P transactions
  1093. * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration
  1094. * driver must complete any such outstanding operations prior to completing the
  1095. * FSM arc into a P2P state. For the purpose of specification the states
  1096. * behave as though the device was fully running if not supported. Like while in
  1097. * STOP or STOP_COPY the user must not touch the device, otherwise the state
  1098. * can be exited.
  1099. *
  1100. * The remaining possible transitions are interpreted as combinations of the
  1101. * above FSM arcs. As there are multiple paths through the FSM arcs the path
  1102. * should be selected based on the following rules:
  1103. * - Select the shortest path.
  1104. * - The path cannot have saving group states as interior arcs, only
  1105. * starting/end states.
  1106. * Refer to vfio_mig_get_next_state() for the result of the algorithm.
  1107. *
  1108. * The automatic transit through the FSM arcs that make up the combination
  1109. * transition is invisible to the user. When working with combination arcs the
  1110. * user may see any step along the path in the device_state if SET_STATE
  1111. * fails. When handling these types of errors users should anticipate future
  1112. * revisions of this protocol using new states and those states becoming
  1113. * visible in this case.
  1114. *
  1115. * The optional states cannot be used with SET_STATE if the device does not
  1116. * support them. The user can discover if these states are supported by using
  1117. * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can
  1118. * avoid knowing about these optional states if the kernel driver supports them.
  1119. *
  1120. * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY
  1121. * is not present.
  1122. */
  1123. enum vfio_device_mig_state {
  1124. VFIO_DEVICE_STATE_ERROR = 0,
  1125. VFIO_DEVICE_STATE_STOP = 1,
  1126. VFIO_DEVICE_STATE_RUNNING = 2,
  1127. VFIO_DEVICE_STATE_STOP_COPY = 3,
  1128. VFIO_DEVICE_STATE_RESUMING = 4,
  1129. VFIO_DEVICE_STATE_RUNNING_P2P = 5,
  1130. VFIO_DEVICE_STATE_PRE_COPY = 6,
  1131. VFIO_DEVICE_STATE_PRE_COPY_P2P = 7,
  1132. VFIO_DEVICE_STATE_NR,
  1133. };
  1134. /**
  1135. * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21)
  1136. *
  1137. * This ioctl is used on the migration data FD in the precopy phase of the
  1138. * migration data transfer. It returns an estimate of the current data sizes
  1139. * remaining to be transferred. It allows the user to judge when it is
  1140. * appropriate to leave PRE_COPY for STOP_COPY.
  1141. *
  1142. * This ioctl is valid only in PRE_COPY states and kernel driver should
  1143. * return -EINVAL from any other migration state.
  1144. *
  1145. * The vfio_precopy_info data structure returned by this ioctl provides
  1146. * estimates of data available from the device during the PRE_COPY states.
  1147. * This estimate is split into two categories, initial_bytes and
  1148. * dirty_bytes.
  1149. *
  1150. * The initial_bytes field indicates the amount of initial precopy
  1151. * data available from the device. This field should have a non-zero initial
  1152. * value and decrease as migration data is read from the device.
  1153. * It is recommended to leave PRE_COPY for STOP_COPY only after this field
  1154. * reaches zero. Leaving PRE_COPY earlier might make things slower.
  1155. *
  1156. * The dirty_bytes field tracks device state changes relative to data
  1157. * previously retrieved. This field starts at zero and may increase as
  1158. * the internal device state is modified or decrease as that modified
  1159. * state is read from the device.
  1160. *
  1161. * Userspace may use the combination of these fields to estimate the
  1162. * potential data size available during the PRE_COPY phases, as well as
  1163. * trends relative to the rate the device is dirtying its internal
  1164. * state, but these fields are not required to have any bearing relative
  1165. * to the data size available during the STOP_COPY phase.
  1166. *
  1167. * Drivers have a lot of flexibility in when and what they transfer during the
  1168. * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO.
  1169. *
  1170. * During pre-copy the migration data FD has a temporary "end of stream" that is
  1171. * reached when both initial_bytes and dirty_byte are zero. For instance, this
  1172. * may indicate that the device is idle and not currently dirtying any internal
  1173. * state. When read() is done on this temporary end of stream the kernel driver
  1174. * should return ENOMSG from read(). Userspace can wait for more data (which may
  1175. * never come) by using poll.
  1176. *
  1177. * Once in STOP_COPY the migration data FD has a permanent end of stream
  1178. * signaled in the usual way by read() always returning 0 and poll always
  1179. * returning readable. ENOMSG may not be returned in STOP_COPY.
  1180. * Support for this ioctl is mandatory if a driver claims to support
  1181. * VFIO_MIGRATION_PRE_COPY.
  1182. *
  1183. * Return: 0 on success, -1 and errno set on failure.
  1184. */
  1185. struct vfio_precopy_info {
  1186. __u32 argsz;
  1187. __u32 flags;
  1188. __aligned_u64 initial_bytes;
  1189. __aligned_u64 dirty_bytes;
  1190. };
  1191. #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21)
  1192. /*
  1193. * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power
  1194. * state with the platform-based power management. Device use of lower power
  1195. * states depends on factors managed by the runtime power management core,
  1196. * including system level support and coordinating support among dependent
  1197. * devices. Enabling device low power entry does not guarantee lower power
  1198. * usage by the device, nor is a mechanism provided through this feature to
  1199. * know the current power state of the device. If any device access happens
  1200. * (either from the host or through the vfio uAPI) when the device is in the
  1201. * low power state, then the host will move the device out of the low power
  1202. * state as necessary prior to the access. Once the access is completed, the
  1203. * device may re-enter the low power state. For single shot low power support
  1204. * with wake-up notification, see
  1205. * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below. Access to mmap'd
  1206. * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after
  1207. * calling LOW_POWER_EXIT.
  1208. */
  1209. #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3
  1210. /*
  1211. * This device feature has the same behavior as
  1212. * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user
  1213. * provides an eventfd for wake-up notification. When the device moves out of
  1214. * the low power state for the wake-up, the host will not allow the device to
  1215. * re-enter a low power state without a subsequent user call to one of the low
  1216. * power entry device feature IOCTLs. Access to mmap'd device regions is
  1217. * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the
  1218. * low power exit. The low power exit can happen either through LOW_POWER_EXIT
  1219. * or through any other access (where the wake-up notification has been
  1220. * generated). The access to mmap'd device regions will not trigger low power
  1221. * exit.
  1222. *
  1223. * The notification through the provided eventfd will be generated only when
  1224. * the device has entered and is resumed from a low power state after
  1225. * calling this device feature IOCTL. A device that has not entered low power
  1226. * state, as managed through the runtime power management core, will not
  1227. * generate a notification through the provided eventfd on access. Calling the
  1228. * LOW_POWER_EXIT feature is optional in the case where notification has been
  1229. * signaled on the provided eventfd that a resume from low power has occurred.
  1230. */
  1231. struct vfio_device_low_power_entry_with_wakeup {
  1232. __s32 wakeup_eventfd;
  1233. __u32 reserved;
  1234. };
  1235. #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4
  1236. /*
  1237. * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as
  1238. * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or
  1239. * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features.
  1240. * This device feature IOCTL may itself generate a wakeup eventfd notification
  1241. * in the latter case if the device had previously entered a low power state.
  1242. */
  1243. #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5
  1244. /*
  1245. * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging.
  1246. * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports
  1247. * DMA logging.
  1248. *
  1249. * DMA logging allows a device to internally record what DMAs the device is
  1250. * initiating and report them back to userspace. It is part of the VFIO
  1251. * migration infrastructure that allows implementing dirty page tracking
  1252. * during the pre copy phase of live migration. Only DMA WRITEs are logged,
  1253. * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE.
  1254. *
  1255. * When DMA logging is started a range of IOVAs to monitor is provided and the
  1256. * device can optimize its logging to cover only the IOVA range given. Each
  1257. * DMA that the device initiates inside the range will be logged by the device
  1258. * for later retrieval.
  1259. *
  1260. * page_size is an input that hints what tracking granularity the device
  1261. * should try to achieve. If the device cannot do the hinted page size then
  1262. * it's the driver choice which page size to pick based on its support.
  1263. * On output the device will return the page size it selected.
  1264. *
  1265. * ranges is a pointer to an array of
  1266. * struct vfio_device_feature_dma_logging_range.
  1267. *
  1268. * The core kernel code guarantees to support by minimum num_ranges that fit
  1269. * into a single kernel page. User space can try higher values but should give
  1270. * up if the above can't be achieved as of some driver limitations.
  1271. *
  1272. * A single call to start device DMA logging can be issued and a matching stop
  1273. * should follow at the end. Another start is not allowed in the meantime.
  1274. */
  1275. struct vfio_device_feature_dma_logging_control {
  1276. __aligned_u64 page_size;
  1277. __u32 num_ranges;
  1278. __u32 __reserved;
  1279. __aligned_u64 ranges;
  1280. };
  1281. struct vfio_device_feature_dma_logging_range {
  1282. __aligned_u64 iova;
  1283. __aligned_u64 length;
  1284. };
  1285. #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6
  1286. /*
  1287. * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started
  1288. * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START
  1289. */
  1290. #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7
  1291. /*
  1292. * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log
  1293. *
  1294. * Query the device's DMA log for written pages within the given IOVA range.
  1295. * During querying the log is cleared for the IOVA range.
  1296. *
  1297. * bitmap is a pointer to an array of u64s that will hold the output bitmap
  1298. * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits
  1299. * is given by:
  1300. * bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64))
  1301. *
  1302. * The input page_size can be any power of two value and does not have to
  1303. * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver
  1304. * will format its internal logging to match the reporting page size, possibly
  1305. * by replicating bits if the internal page size is lower than requested.
  1306. *
  1307. * The LOGGING_REPORT will only set bits in the bitmap and never clear or
  1308. * perform any initialization of the user provided bitmap.
  1309. *
  1310. * If any error is returned userspace should assume that the dirty log is
  1311. * corrupted. Error recovery is to consider all memory dirty and try to
  1312. * restart the dirty tracking, or to abort/restart the whole migration.
  1313. *
  1314. * If DMA logging is not enabled, an error will be returned.
  1315. *
  1316. */
  1317. struct vfio_device_feature_dma_logging_report {
  1318. __aligned_u64 iova;
  1319. __aligned_u64 length;
  1320. __aligned_u64 page_size;
  1321. __aligned_u64 bitmap;
  1322. };
  1323. #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8
  1324. /*
  1325. * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will
  1326. * be required to complete stop copy.
  1327. *
  1328. * Note: Can be called on each device state.
  1329. */
  1330. struct vfio_device_feature_mig_data_size {
  1331. __aligned_u64 stop_copy_length;
  1332. };
  1333. #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9
  1334. /**
  1335. * Upon VFIO_DEVICE_FEATURE_SET, set or clear the BUS mastering for the device
  1336. * based on the operation specified in op flag.
  1337. *
  1338. * The functionality is incorporated for devices that needs bus master control,
  1339. * but the in-band device interface lacks the support. Consequently, it is not
  1340. * applicable to PCI devices, as bus master control for PCI devices is managed
  1341. * in-band through the configuration space. At present, this feature is supported
  1342. * only for CDX devices.
  1343. * When the device's BUS MASTER setting is configured as CLEAR, it will result in
  1344. * blocking all incoming DMA requests from the device. On the other hand, configuring
  1345. * the device's BUS MASTER setting as SET (enable) will grant the device the
  1346. * capability to perform DMA to the host memory.
  1347. */
  1348. struct vfio_device_feature_bus_master {
  1349. __u32 op;
  1350. #define VFIO_DEVICE_FEATURE_CLEAR_MASTER 0 /* Clear Bus Master */
  1351. #define VFIO_DEVICE_FEATURE_SET_MASTER 1 /* Set Bus Master */
  1352. };
  1353. #define VFIO_DEVICE_FEATURE_BUS_MASTER 10
  1354. /* -------- API for Type1 VFIO IOMMU -------- */
  1355. /**
  1356. * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
  1357. *
  1358. * Retrieve information about the IOMMU object. Fills in provided
  1359. * struct vfio_iommu_info. Caller sets argsz.
  1360. *
  1361. * XXX Should we do these by CHECK_EXTENSION too?
  1362. */
  1363. struct vfio_iommu_type1_info {
  1364. __u32 argsz;
  1365. __u32 flags;
  1366. #define VFIO_IOMMU_INFO_PGSIZES (1 << 0) /* supported page sizes info */
  1367. #define VFIO_IOMMU_INFO_CAPS (1 << 1) /* Info supports caps */
  1368. __aligned_u64 iova_pgsizes; /* Bitmap of supported page sizes */
  1369. __u32 cap_offset; /* Offset within info struct of first cap */
  1370. __u32 pad;
  1371. };
  1372. /*
  1373. * The IOVA capability allows to report the valid IOVA range(s)
  1374. * excluding any non-relaxable reserved regions exposed by
  1375. * devices attached to the container. Any DMA map attempt
  1376. * outside the valid iova range will return error.
  1377. *
  1378. * The structures below define version 1 of this capability.
  1379. */
  1380. #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE 1
  1381. struct vfio_iova_range {
  1382. __u64 start;
  1383. __u64 end;
  1384. };
  1385. struct vfio_iommu_type1_info_cap_iova_range {
  1386. struct vfio_info_cap_header header;
  1387. __u32 nr_iovas;
  1388. __u32 reserved;
  1389. struct vfio_iova_range iova_ranges[];
  1390. };
  1391. /*
  1392. * The migration capability allows to report supported features for migration.
  1393. *
  1394. * The structures below define version 1 of this capability.
  1395. *
  1396. * The existence of this capability indicates that IOMMU kernel driver supports
  1397. * dirty page logging.
  1398. *
  1399. * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty
  1400. * page logging.
  1401. * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap
  1402. * size in bytes that can be used by user applications when getting the dirty
  1403. * bitmap.
  1404. */
  1405. #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION 2
  1406. struct vfio_iommu_type1_info_cap_migration {
  1407. struct vfio_info_cap_header header;
  1408. __u32 flags;
  1409. __u64 pgsize_bitmap;
  1410. __u64 max_dirty_bitmap_size; /* in bytes */
  1411. };
  1412. /*
  1413. * The DMA available capability allows to report the current number of
  1414. * simultaneously outstanding DMA mappings that are allowed.
  1415. *
  1416. * The structure below defines version 1 of this capability.
  1417. *
  1418. * avail: specifies the current number of outstanding DMA mappings allowed.
  1419. */
  1420. #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3
  1421. struct vfio_iommu_type1_info_dma_avail {
  1422. struct vfio_info_cap_header header;
  1423. __u32 avail;
  1424. };
  1425. #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
  1426. /**
  1427. * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
  1428. *
  1429. * Map process virtual addresses to IO virtual addresses using the
  1430. * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
  1431. *
  1432. * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr
  1433. * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR. To
  1434. * maintain memory consistency within the user application, the updated vaddr
  1435. * must address the same memory object as originally mapped. Failure to do so
  1436. * will result in user memory corruption and/or device misbehavior. iova and
  1437. * size must match those in the original MAP_DMA call. Protection is not
  1438. * changed, and the READ & WRITE flags must be 0.
  1439. */
  1440. struct vfio_iommu_type1_dma_map {
  1441. __u32 argsz;
  1442. __u32 flags;
  1443. #define VFIO_DMA_MAP_FLAG_READ (1 << 0) /* readable from device */
  1444. #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1) /* writable from device */
  1445. #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2)
  1446. __u64 vaddr; /* Process virtual address */
  1447. __u64 iova; /* IO virtual address */
  1448. __u64 size; /* Size of mapping (bytes) */
  1449. };
  1450. #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
  1451. struct vfio_bitmap {
  1452. __u64 pgsize; /* page size for bitmap in bytes */
  1453. __u64 size; /* in bytes */
  1454. __u64 *data; /* one bit per page */
  1455. };
  1456. /**
  1457. * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
  1458. * struct vfio_dma_unmap)
  1459. *
  1460. * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
  1461. * Caller sets argsz. The actual unmapped size is returned in the size
  1462. * field. No guarantee is made to the user that arbitrary unmaps of iova
  1463. * or size different from those used in the original mapping call will
  1464. * succeed.
  1465. *
  1466. * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap
  1467. * before unmapping IO virtual addresses. When this flag is set, the user must
  1468. * provide a struct vfio_bitmap in data[]. User must provide zero-allocated
  1469. * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field.
  1470. * A bit in the bitmap represents one page, of user provided page size in
  1471. * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set
  1472. * indicates that the page at that offset from iova is dirty. A Bitmap of the
  1473. * pages in the range of unmapped size is returned in the user-provided
  1474. * vfio_bitmap.data.
  1475. *
  1476. * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses. iova and size
  1477. * must be 0. This cannot be combined with the get-dirty-bitmap flag.
  1478. *
  1479. * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host
  1480. * virtual addresses in the iova range. DMA to already-mapped pages continues.
  1481. * Groups may not be added to the container while any addresses are invalid.
  1482. * This cannot be combined with the get-dirty-bitmap flag.
  1483. */
  1484. struct vfio_iommu_type1_dma_unmap {
  1485. __u32 argsz;
  1486. __u32 flags;
  1487. #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0)
  1488. #define VFIO_DMA_UNMAP_FLAG_ALL (1 << 1)
  1489. #define VFIO_DMA_UNMAP_FLAG_VADDR (1 << 2)
  1490. __u64 iova; /* IO virtual address */
  1491. __u64 size; /* Size of mapping (bytes) */
  1492. __u8 data[];
  1493. };
  1494. #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
  1495. /*
  1496. * IOCTLs to enable/disable IOMMU container usage.
  1497. * No parameters are supported.
  1498. */
  1499. #define VFIO_IOMMU_ENABLE _IO(VFIO_TYPE, VFIO_BASE + 15)
  1500. #define VFIO_IOMMU_DISABLE _IO(VFIO_TYPE, VFIO_BASE + 16)
  1501. /**
  1502. * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
  1503. * struct vfio_iommu_type1_dirty_bitmap)
  1504. * IOCTL is used for dirty pages logging.
  1505. * Caller should set flag depending on which operation to perform, details as
  1506. * below:
  1507. *
  1508. * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs
  1509. * the IOMMU driver to log pages that are dirtied or potentially dirtied by
  1510. * the device; designed to be used when a migration is in progress. Dirty pages
  1511. * are logged until logging is disabled by user application by calling the IOCTL
  1512. * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag.
  1513. *
  1514. * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs
  1515. * the IOMMU driver to stop logging dirtied pages.
  1516. *
  1517. * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set
  1518. * returns the dirty pages bitmap for IOMMU container for a given IOVA range.
  1519. * The user must specify the IOVA range and the pgsize through the structure
  1520. * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface
  1521. * supports getting a bitmap of the smallest supported pgsize only and can be
  1522. * modified in future to get a bitmap of any specified supported pgsize. The
  1523. * user must provide a zeroed memory area for the bitmap memory and specify its
  1524. * size in bitmap.size. One bit is used to represent one page consecutively
  1525. * starting from iova offset. The user should provide page size in bitmap.pgsize
  1526. * field. A bit set in the bitmap indicates that the page at that offset from
  1527. * iova is dirty. The caller must set argsz to a value including the size of
  1528. * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the
  1529. * actual bitmap. If dirty pages logging is not enabled, an error will be
  1530. * returned.
  1531. *
  1532. * Only one of the flags _START, _STOP and _GET may be specified at a time.
  1533. *
  1534. */
  1535. struct vfio_iommu_type1_dirty_bitmap {
  1536. __u32 argsz;
  1537. __u32 flags;
  1538. #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START (1 << 0)
  1539. #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP (1 << 1)
  1540. #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP (1 << 2)
  1541. __u8 data[];
  1542. };
  1543. struct vfio_iommu_type1_dirty_bitmap_get {
  1544. __u64 iova; /* IO virtual address */
  1545. __u64 size; /* Size of iova range */
  1546. struct vfio_bitmap bitmap;
  1547. };
  1548. #define VFIO_IOMMU_DIRTY_PAGES _IO(VFIO_TYPE, VFIO_BASE + 17)
  1549. /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
  1550. /*
  1551. * The SPAPR TCE DDW info struct provides the information about
  1552. * the details of Dynamic DMA window capability.
  1553. *
  1554. * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
  1555. * @max_dynamic_windows_supported tells the maximum number of windows
  1556. * which the platform can create.
  1557. * @levels tells the maximum number of levels in multi-level IOMMU tables;
  1558. * this allows splitting a table into smaller chunks which reduces
  1559. * the amount of physically contiguous memory required for the table.
  1560. */
  1561. struct vfio_iommu_spapr_tce_ddw_info {
  1562. __u64 pgsizes; /* Bitmap of supported page sizes */
  1563. __u32 max_dynamic_windows_supported;
  1564. __u32 levels;
  1565. };
  1566. /*
  1567. * The SPAPR TCE info struct provides the information about the PCI bus
  1568. * address ranges available for DMA, these values are programmed into
  1569. * the hardware so the guest has to know that information.
  1570. *
  1571. * The DMA 32 bit window start is an absolute PCI bus address.
  1572. * The IOVA address passed via map/unmap ioctls are absolute PCI bus
  1573. * addresses too so the window works as a filter rather than an offset
  1574. * for IOVA addresses.
  1575. *
  1576. * Flags supported:
  1577. * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
  1578. * (DDW) support is present. @ddw is only supported when DDW is present.
  1579. */
  1580. struct vfio_iommu_spapr_tce_info {
  1581. __u32 argsz;
  1582. __u32 flags;
  1583. #define VFIO_IOMMU_SPAPR_INFO_DDW (1 << 0) /* DDW supported */
  1584. __u32 dma32_window_start; /* 32 bit window start (bytes) */
  1585. __u32 dma32_window_size; /* 32 bit window size (bytes) */
  1586. struct vfio_iommu_spapr_tce_ddw_info ddw;
  1587. };
  1588. #define VFIO_IOMMU_SPAPR_TCE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
  1589. /*
  1590. * EEH PE operation struct provides ways to:
  1591. * - enable/disable EEH functionality;
  1592. * - unfreeze IO/DMA for frozen PE;
  1593. * - read PE state;
  1594. * - reset PE;
  1595. * - configure PE;
  1596. * - inject EEH error.
  1597. */
  1598. struct vfio_eeh_pe_err {
  1599. __u32 type;
  1600. __u32 func;
  1601. __u64 addr;
  1602. __u64 mask;
  1603. };
  1604. struct vfio_eeh_pe_op {
  1605. __u32 argsz;
  1606. __u32 flags;
  1607. __u32 op;
  1608. union {
  1609. struct vfio_eeh_pe_err err;
  1610. };
  1611. };
  1612. #define VFIO_EEH_PE_DISABLE 0 /* Disable EEH functionality */
  1613. #define VFIO_EEH_PE_ENABLE 1 /* Enable EEH functionality */
  1614. #define VFIO_EEH_PE_UNFREEZE_IO 2 /* Enable IO for frozen PE */
  1615. #define VFIO_EEH_PE_UNFREEZE_DMA 3 /* Enable DMA for frozen PE */
  1616. #define VFIO_EEH_PE_GET_STATE 4 /* PE state retrieval */
  1617. #define VFIO_EEH_PE_STATE_NORMAL 0 /* PE in functional state */
  1618. #define VFIO_EEH_PE_STATE_RESET 1 /* PE reset in progress */
  1619. #define VFIO_EEH_PE_STATE_STOPPED 2 /* Stopped DMA and IO */
  1620. #define VFIO_EEH_PE_STATE_STOPPED_DMA 4 /* Stopped DMA only */
  1621. #define VFIO_EEH_PE_STATE_UNAVAIL 5 /* State unavailable */
  1622. #define VFIO_EEH_PE_RESET_DEACTIVATE 5 /* Deassert PE reset */
  1623. #define VFIO_EEH_PE_RESET_HOT 6 /* Assert hot reset */
  1624. #define VFIO_EEH_PE_RESET_FUNDAMENTAL 7 /* Assert fundamental reset */
  1625. #define VFIO_EEH_PE_CONFIGURE 8 /* PE configuration */
  1626. #define VFIO_EEH_PE_INJECT_ERR 9 /* Inject EEH error */
  1627. #define VFIO_EEH_PE_OP _IO(VFIO_TYPE, VFIO_BASE + 21)
  1628. /**
  1629. * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
  1630. *
  1631. * Registers user space memory where DMA is allowed. It pins
  1632. * user pages and does the locked memory accounting so
  1633. * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
  1634. * get faster.
  1635. */
  1636. struct vfio_iommu_spapr_register_memory {
  1637. __u32 argsz;
  1638. __u32 flags;
  1639. __u64 vaddr; /* Process virtual address */
  1640. __u64 size; /* Size of mapping (bytes) */
  1641. };
  1642. #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 17)
  1643. /**
  1644. * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
  1645. *
  1646. * Unregisters user space memory registered with
  1647. * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
  1648. * Uses vfio_iommu_spapr_register_memory for parameters.
  1649. */
  1650. #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 18)
  1651. /**
  1652. * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
  1653. *
  1654. * Creates an additional TCE table and programs it (sets a new DMA window)
  1655. * to every IOMMU group in the container. It receives page shift, window
  1656. * size and number of levels in the TCE table being created.
  1657. *
  1658. * It allocates and returns an offset on a PCI bus of the new DMA window.
  1659. */
  1660. struct vfio_iommu_spapr_tce_create {
  1661. __u32 argsz;
  1662. __u32 flags;
  1663. /* in */
  1664. __u32 page_shift;
  1665. __u32 __resv1;
  1666. __u64 window_size;
  1667. __u32 levels;
  1668. __u32 __resv2;
  1669. /* out */
  1670. __u64 start_addr;
  1671. };
  1672. #define VFIO_IOMMU_SPAPR_TCE_CREATE _IO(VFIO_TYPE, VFIO_BASE + 19)
  1673. /**
  1674. * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
  1675. *
  1676. * Unprograms a TCE table from all groups in the container and destroys it.
  1677. * It receives a PCI bus offset as a window id.
  1678. */
  1679. struct vfio_iommu_spapr_tce_remove {
  1680. __u32 argsz;
  1681. __u32 flags;
  1682. /* in */
  1683. __u64 start_addr;
  1684. };
  1685. #define VFIO_IOMMU_SPAPR_TCE_REMOVE _IO(VFIO_TYPE, VFIO_BASE + 20)
  1686. /* ***************************************************************** */
  1687. #endif /* VFIO_H */