xref: /linux-6.15/include/uapi/linux/vfio.h (revision ff598081)
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 <[email protected]>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 #ifndef _UAPIVFIO_H
13 #define _UAPIVFIO_H
14 
15 #include <linux/types.h>
16 #include <linux/ioctl.h>
17 
18 #define VFIO_API_VERSION	0
19 
20 
21 /* Kernel & User level defines for VFIO IOCTLs. */
22 
23 /* Extensions */
24 
25 #define VFIO_TYPE1_IOMMU		1
26 #define VFIO_SPAPR_TCE_IOMMU		2
27 #define VFIO_TYPE1v2_IOMMU		3
28 /*
29  * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping).  This
30  * capability is subject to change as groups are added or removed.
31  */
32 #define VFIO_DMA_CC_IOMMU		4
33 
34 /* Check if EEH is supported */
35 #define VFIO_EEH			5
36 
37 /* Two-stage IOMMU */
38 #define VFIO_TYPE1_NESTING_IOMMU	6	/* Implies v2 */
39 
40 #define VFIO_SPAPR_TCE_v2_IOMMU		7
41 
42 /*
43  * The No-IOMMU IOMMU offers no translation or isolation for devices and
44  * supports no ioctls outside of VFIO_CHECK_EXTENSION.  Use of VFIO's No-IOMMU
45  * code will taint the host kernel and should be used with extreme caution.
46  */
47 #define VFIO_NOIOMMU_IOMMU		8
48 
49 /* Supports VFIO_DMA_UNMAP_FLAG_ALL */
50 #define VFIO_UNMAP_ALL			9
51 
52 /*
53  * Supports the vaddr flag for DMA map and unmap.  Not supported for mediated
54  * devices, so this capability is subject to change as groups are added or
55  * removed.
56  */
57 #define VFIO_UPDATE_VADDR		10
58 
59 /*
60  * The IOCTL interface is designed for extensibility by embedding the
61  * structure length (argsz) and flags into structures passed between
62  * kernel and userspace.  We therefore use the _IO() macro for these
63  * defines to avoid implicitly embedding a size into the ioctl request.
64  * As structure fields are added, argsz will increase to match and flag
65  * bits will be defined to indicate additional fields with valid data.
66  * It's *always* the caller's responsibility to indicate the size of
67  * the structure passed by setting argsz appropriately.
68  */
69 
70 #define VFIO_TYPE	(';')
71 #define VFIO_BASE	100
72 
73 /*
74  * For extension of INFO ioctls, VFIO makes use of a capability chain
75  * designed after PCI/e capabilities.  A flag bit indicates whether
76  * this capability chain is supported and a field defined in the fixed
77  * structure defines the offset of the first capability in the chain.
78  * This field is only valid when the corresponding bit in the flags
79  * bitmap is set.  This offset field is relative to the start of the
80  * INFO buffer, as is the next field within each capability header.
81  * The id within the header is a shared address space per INFO ioctl,
82  * while the version field is specific to the capability id.  The
83  * contents following the header are specific to the capability id.
84  */
85 struct vfio_info_cap_header {
86 	__u16	id;		/* Identifies capability */
87 	__u16	version;	/* Version specific to the capability ID */
88 	__u32	next;		/* Offset of next capability */
89 };
90 
91 /*
92  * Callers of INFO ioctls passing insufficiently sized buffers will see
93  * the capability chain flag bit set, a zero value for the first capability
94  * offset (if available within the provided argsz), and argsz will be
95  * updated to report the necessary buffer size.  For compatibility, the
96  * INFO ioctl will not report error in this case, but the capability chain
97  * will not be available.
98  */
99 
100 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
101 
102 /**
103  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
104  *
105  * Report the version of the VFIO API.  This allows us to bump the entire
106  * API version should we later need to add or change features in incompatible
107  * ways.
108  * Return: VFIO_API_VERSION
109  * Availability: Always
110  */
111 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
112 
113 /**
114  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
115  *
116  * Check whether an extension is supported.
117  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
118  * Availability: Always
119  */
120 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
121 
122 /**
123  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
124  *
125  * Set the iommu to the given type.  The type must be supported by an
126  * iommu driver as verified by calling CHECK_EXTENSION using the same
127  * type.  A group must be set to this file descriptor before this
128  * ioctl is available.  The IOMMU interfaces enabled by this call are
129  * specific to the value set.
130  * Return: 0 on success, -errno on failure
131  * Availability: When VFIO group attached
132  */
133 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
134 
135 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
136 
137 /**
138  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
139  *						struct vfio_group_status)
140  *
141  * Retrieve information about the group.  Fills in provided
142  * struct vfio_group_info.  Caller sets argsz.
143  * Return: 0 on succes, -errno on failure.
144  * Availability: Always
145  */
146 struct vfio_group_status {
147 	__u32	argsz;
148 	__u32	flags;
149 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
150 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
151 };
152 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
153 
154 /**
155  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
156  *
157  * Set the container for the VFIO group to the open VFIO file
158  * descriptor provided.  Groups may only belong to a single
159  * container.  Containers may, at their discretion, support multiple
160  * groups.  Only when a container is set are all of the interfaces
161  * of the VFIO file descriptor and the VFIO group file descriptor
162  * available to the user.
163  * Return: 0 on success, -errno on failure.
164  * Availability: Always
165  */
166 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
167 
168 /**
169  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
170  *
171  * Remove the group from the attached container.  This is the
172  * opposite of the SET_CONTAINER call and returns the group to
173  * an initial state.  All device file descriptors must be released
174  * prior to calling this interface.  When removing the last group
175  * from a container, the IOMMU will be disabled and all state lost,
176  * effectively also returning the VFIO file descriptor to an initial
177  * state.
178  * Return: 0 on success, -errno on failure.
179  * Availability: When attached to container
180  */
181 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
182 
183 /**
184  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
185  *
186  * Return a new file descriptor for the device object described by
187  * the provided string.  The string should match a device listed in
188  * the devices subdirectory of the IOMMU group sysfs entry.  The
189  * group containing the device must already be added to this context.
190  * Return: new file descriptor on success, -errno on failure.
191  * Availability: When attached to container
192  */
193 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
194 
195 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
196 
197 /**
198  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
199  *						struct vfio_device_info)
200  *
201  * Retrieve information about the device.  Fills in provided
202  * struct vfio_device_info.  Caller sets argsz.
203  * Return: 0 on success, -errno on failure.
204  */
205 struct vfio_device_info {
206 	__u32	argsz;
207 	__u32	flags;
208 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
209 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
210 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2)	/* vfio-platform device */
211 #define VFIO_DEVICE_FLAGS_AMBA  (1 << 3)	/* vfio-amba device */
212 #define VFIO_DEVICE_FLAGS_CCW	(1 << 4)	/* vfio-ccw device */
213 #define VFIO_DEVICE_FLAGS_AP	(1 << 5)	/* vfio-ap device */
214 #define VFIO_DEVICE_FLAGS_FSL_MC (1 << 6)	/* vfio-fsl-mc device */
215 #define VFIO_DEVICE_FLAGS_CAPS	(1 << 7)	/* Info supports caps */
216 #define VFIO_DEVICE_FLAGS_CDX	(1 << 8)	/* vfio-cdx device */
217 	__u32	num_regions;	/* Max region index + 1 */
218 	__u32	num_irqs;	/* Max IRQ index + 1 */
219 	__u32   cap_offset;	/* Offset within info struct of first cap */
220 };
221 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
222 
223 /*
224  * Vendor driver using Mediated device framework should provide device_api
225  * attribute in supported type attribute groups. Device API string should be one
226  * of the following corresponding to device flags in vfio_device_info structure.
227  */
228 
229 #define VFIO_DEVICE_API_PCI_STRING		"vfio-pci"
230 #define VFIO_DEVICE_API_PLATFORM_STRING		"vfio-platform"
231 #define VFIO_DEVICE_API_AMBA_STRING		"vfio-amba"
232 #define VFIO_DEVICE_API_CCW_STRING		"vfio-ccw"
233 #define VFIO_DEVICE_API_AP_STRING		"vfio-ap"
234 
235 /*
236  * The following capabilities are unique to s390 zPCI devices.  Their contents
237  * are further-defined in vfio_zdev.h
238  */
239 #define VFIO_DEVICE_INFO_CAP_ZPCI_BASE		1
240 #define VFIO_DEVICE_INFO_CAP_ZPCI_GROUP		2
241 #define VFIO_DEVICE_INFO_CAP_ZPCI_UTIL		3
242 #define VFIO_DEVICE_INFO_CAP_ZPCI_PFIP		4
243 
244 /*
245  * The following VFIO_DEVICE_INFO capability reports support for PCIe AtomicOp
246  * completion to the root bus with supported widths provided via flags.
247  */
248 #define VFIO_DEVICE_INFO_CAP_PCI_ATOMIC_COMP	5
249 struct vfio_device_info_cap_pci_atomic_comp {
250 	struct vfio_info_cap_header header;
251 	__u32 flags;
252 #define VFIO_PCI_ATOMIC_COMP32	(1 << 0)
253 #define VFIO_PCI_ATOMIC_COMP64	(1 << 1)
254 #define VFIO_PCI_ATOMIC_COMP128	(1 << 2)
255 	__u32 reserved;
256 };
257 
258 /**
259  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
260  *				       struct vfio_region_info)
261  *
262  * Retrieve information about a device region.  Caller provides
263  * struct vfio_region_info with index value set.  Caller sets argsz.
264  * Implementation of region mapping is bus driver specific.  This is
265  * intended to describe MMIO, I/O port, as well as bus specific
266  * regions (ex. PCI config space).  Zero sized regions may be used
267  * to describe unimplemented regions (ex. unimplemented PCI BARs).
268  * Return: 0 on success, -errno on failure.
269  */
270 struct vfio_region_info {
271 	__u32	argsz;
272 	__u32	flags;
273 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
274 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
275 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
276 #define VFIO_REGION_INFO_FLAG_CAPS	(1 << 3) /* Info supports caps */
277 	__u32	index;		/* Region index */
278 	__u32	cap_offset;	/* Offset within info struct of first cap */
279 	__u64	size;		/* Region size (bytes) */
280 	__u64	offset;		/* Region offset from start of device fd */
281 };
282 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
283 
284 /*
285  * The sparse mmap capability allows finer granularity of specifying areas
286  * within a region with mmap support.  When specified, the user should only
287  * mmap the offset ranges specified by the areas array.  mmaps outside of the
288  * areas specified may fail (such as the range covering a PCI MSI-X table) or
289  * may result in improper device behavior.
290  *
291  * The structures below define version 1 of this capability.
292  */
293 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP	1
294 
295 struct vfio_region_sparse_mmap_area {
296 	__u64	offset;	/* Offset of mmap'able area within region */
297 	__u64	size;	/* Size of mmap'able area */
298 };
299 
300 struct vfio_region_info_cap_sparse_mmap {
301 	struct vfio_info_cap_header header;
302 	__u32	nr_areas;
303 	__u32	reserved;
304 	struct vfio_region_sparse_mmap_area areas[];
305 };
306 
307 /*
308  * The device specific type capability allows regions unique to a specific
309  * device or class of devices to be exposed.  This helps solve the problem for
310  * vfio bus drivers of defining which region indexes correspond to which region
311  * on the device, without needing to resort to static indexes, as done by
312  * vfio-pci.  For instance, if we were to go back in time, we might remove
313  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
314  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
315  * make a "VGA" device specific type to describe the VGA access space.  This
316  * means that non-VGA devices wouldn't need to waste this index, and thus the
317  * address space associated with it due to implementation of device file
318  * descriptor offsets in vfio-pci.
319  *
320  * The current implementation is now part of the user ABI, so we can't use this
321  * for VGA, but there are other upcoming use cases, such as opregions for Intel
322  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
323  * use this for future additions.
324  *
325  * The structure below defines version 1 of this capability.
326  */
327 #define VFIO_REGION_INFO_CAP_TYPE	2
328 
329 struct vfio_region_info_cap_type {
330 	struct vfio_info_cap_header header;
331 	__u32 type;	/* global per bus driver */
332 	__u32 subtype;	/* type specific */
333 };
334 
335 /*
336  * List of region types, global per bus driver.
337  * If you introduce a new type, please add it here.
338  */
339 
340 /* PCI region type containing a PCI vendor part */
341 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
342 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
343 #define VFIO_REGION_TYPE_GFX                    (1)
344 #define VFIO_REGION_TYPE_CCW			(2)
345 #define VFIO_REGION_TYPE_MIGRATION_DEPRECATED   (3)
346 
347 /* sub-types for VFIO_REGION_TYPE_PCI_* */
348 
349 /* 8086 vendor PCI sub-types */
350 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
351 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
352 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
353 
354 /* 10de vendor PCI sub-types */
355 /*
356  * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
357  *
358  * Deprecated, region no longer provided
359  */
360 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM	(1)
361 
362 /* 1014 vendor PCI sub-types */
363 /*
364  * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
365  * to do TLB invalidation on a GPU.
366  *
367  * Deprecated, region no longer provided
368  */
369 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD	(1)
370 
371 /* sub-types for VFIO_REGION_TYPE_GFX */
372 #define VFIO_REGION_SUBTYPE_GFX_EDID            (1)
373 
374 /**
375  * struct vfio_region_gfx_edid - EDID region layout.
376  *
377  * Set display link state and EDID blob.
378  *
379  * The EDID blob has monitor information such as brand, name, serial
380  * number, physical size, supported video modes and more.
381  *
382  * This special region allows userspace (typically qemu) set a virtual
383  * EDID for the virtual monitor, which allows a flexible display
384  * configuration.
385  *
386  * For the edid blob spec look here:
387  *    https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
388  *
389  * On linux systems you can find the EDID blob in sysfs:
390  *    /sys/class/drm/${card}/${connector}/edid
391  *
392  * You can use the edid-decode ulility (comes with xorg-x11-utils) to
393  * decode the EDID blob.
394  *
395  * @edid_offset: location of the edid blob, relative to the
396  *               start of the region (readonly).
397  * @edid_max_size: max size of the edid blob (readonly).
398  * @edid_size: actual edid size (read/write).
399  * @link_state: display link state (read/write).
400  * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
401  * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
402  * @max_xres: max display width (0 == no limitation, readonly).
403  * @max_yres: max display height (0 == no limitation, readonly).
404  *
405  * EDID update protocol:
406  *   (1) set link-state to down.
407  *   (2) update edid blob and size.
408  *   (3) set link-state to up.
409  */
410 struct vfio_region_gfx_edid {
411 	__u32 edid_offset;
412 	__u32 edid_max_size;
413 	__u32 edid_size;
414 	__u32 max_xres;
415 	__u32 max_yres;
416 	__u32 link_state;
417 #define VFIO_DEVICE_GFX_LINK_STATE_UP    1
418 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN  2
419 };
420 
421 /* sub-types for VFIO_REGION_TYPE_CCW */
422 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD	(1)
423 #define VFIO_REGION_SUBTYPE_CCW_SCHIB		(2)
424 #define VFIO_REGION_SUBTYPE_CCW_CRW		(3)
425 
426 /* sub-types for VFIO_REGION_TYPE_MIGRATION */
427 #define VFIO_REGION_SUBTYPE_MIGRATION_DEPRECATED (1)
428 
429 struct vfio_device_migration_info {
430 	__u32 device_state;         /* VFIO device state */
431 #define VFIO_DEVICE_STATE_V1_STOP      (0)
432 #define VFIO_DEVICE_STATE_V1_RUNNING   (1 << 0)
433 #define VFIO_DEVICE_STATE_V1_SAVING    (1 << 1)
434 #define VFIO_DEVICE_STATE_V1_RESUMING  (1 << 2)
435 #define VFIO_DEVICE_STATE_MASK      (VFIO_DEVICE_STATE_V1_RUNNING | \
436 				     VFIO_DEVICE_STATE_V1_SAVING |  \
437 				     VFIO_DEVICE_STATE_V1_RESUMING)
438 
439 #define VFIO_DEVICE_STATE_VALID(state) \
440 	(state & VFIO_DEVICE_STATE_V1_RESUMING ? \
441 	(state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_V1_RESUMING : 1)
442 
443 #define VFIO_DEVICE_STATE_IS_ERROR(state) \
444 	((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_V1_SAVING | \
445 					      VFIO_DEVICE_STATE_V1_RESUMING))
446 
447 #define VFIO_DEVICE_STATE_SET_ERROR(state) \
448 	((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_STATE_V1_SAVING | \
449 					     VFIO_DEVICE_STATE_V1_RESUMING)
450 
451 	__u32 reserved;
452 	__u64 pending_bytes;
453 	__u64 data_offset;
454 	__u64 data_size;
455 };
456 
457 /*
458  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
459  * which allows direct access to non-MSIX registers which happened to be within
460  * the same system page.
461  *
462  * Even though the userspace gets direct access to the MSIX data, the existing
463  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
464  */
465 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE	3
466 
467 /*
468  * Capability with compressed real address (aka SSA - small system address)
469  * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
470  * and by the userspace to associate a NVLink bridge with a GPU.
471  *
472  * Deprecated, capability no longer provided
473  */
474 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT	4
475 
476 struct vfio_region_info_cap_nvlink2_ssatgt {
477 	struct vfio_info_cap_header header;
478 	__u64 tgt;
479 };
480 
481 /*
482  * Capability with an NVLink link speed. The value is read by
483  * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
484  * property in the device tree. The value is fixed in the hardware
485  * and failing to provide the correct value results in the link
486  * not working with no indication from the driver why.
487  *
488  * Deprecated, capability no longer provided
489  */
490 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD	5
491 
492 struct vfio_region_info_cap_nvlink2_lnkspd {
493 	struct vfio_info_cap_header header;
494 	__u32 link_speed;
495 	__u32 __pad;
496 };
497 
498 /**
499  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
500  *				    struct vfio_irq_info)
501  *
502  * Retrieve information about a device IRQ.  Caller provides
503  * struct vfio_irq_info with index value set.  Caller sets argsz.
504  * Implementation of IRQ mapping is bus driver specific.  Indexes
505  * using multiple IRQs are primarily intended to support MSI-like
506  * interrupt blocks.  Zero count irq blocks may be used to describe
507  * unimplemented interrupt types.
508  *
509  * The EVENTFD flag indicates the interrupt index supports eventfd based
510  * signaling.
511  *
512  * The MASKABLE flags indicates the index supports MASK and UNMASK
513  * actions described below.
514  *
515  * AUTOMASKED indicates that after signaling, the interrupt line is
516  * automatically masked by VFIO and the user needs to unmask the line
517  * to receive new interrupts.  This is primarily intended to distinguish
518  * level triggered interrupts.
519  *
520  * The NORESIZE flag indicates that the interrupt lines within the index
521  * are setup as a set and new subindexes cannot be enabled without first
522  * disabling the entire index.  This is used for interrupts like PCI MSI
523  * and MSI-X where the driver may only use a subset of the available
524  * indexes, but VFIO needs to enable a specific number of vectors
525  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
526  * then add and unmask vectors, it's up to userspace to make the decision
527  * whether to allocate the maximum supported number of vectors or tear
528  * down setup and incrementally increase the vectors as each is enabled.
529  * Absence of the NORESIZE flag indicates that vectors can be enabled
530  * and disabled dynamically without impacting other vectors within the
531  * index.
532  */
533 struct vfio_irq_info {
534 	__u32	argsz;
535 	__u32	flags;
536 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
537 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
538 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
539 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
540 	__u32	index;		/* IRQ index */
541 	__u32	count;		/* Number of IRQs within this index */
542 };
543 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
544 
545 /**
546  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
547  *
548  * Set signaling, masking, and unmasking of interrupts.  Caller provides
549  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
550  * the range of subindexes being specified.
551  *
552  * The DATA flags specify the type of data provided.  If DATA_NONE, the
553  * operation performs the specified action immediately on the specified
554  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
555  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
556  *
557  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
558  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
559  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
560  * data = {1,0,1}
561  *
562  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
563  * A value of -1 can be used to either de-assign interrupts if already
564  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
565  * to be trigger for interrupts [0,0] and [0,2]:
566  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
567  * data = {fd1, -1, fd2}
568  * If index [0,1] is previously set, two count = 1 ioctls calls would be
569  * required to set [0,0] and [0,2] without changing [0,1].
570  *
571  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
572  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
573  * from userspace (ie. simulate hardware triggering).
574  *
575  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
576  * enables the interrupt index for the device.  Individual subindex interrupts
577  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
578  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
579  *
580  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
581  * ACTION_TRIGGER specifies kernel->user signaling.
582  */
583 struct vfio_irq_set {
584 	__u32	argsz;
585 	__u32	flags;
586 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
587 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
588 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
589 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
590 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
591 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
592 	__u32	index;
593 	__u32	start;
594 	__u32	count;
595 	__u8	data[];
596 };
597 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
598 
599 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
600 					 VFIO_IRQ_SET_DATA_BOOL | \
601 					 VFIO_IRQ_SET_DATA_EVENTFD)
602 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
603 					 VFIO_IRQ_SET_ACTION_UNMASK | \
604 					 VFIO_IRQ_SET_ACTION_TRIGGER)
605 /**
606  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
607  *
608  * Reset a device.
609  */
610 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
611 
612 /*
613  * The VFIO-PCI bus driver makes use of the following fixed region and
614  * IRQ index mapping.  Unimplemented regions return a size of zero.
615  * Unimplemented IRQ types return a count of zero.
616  */
617 
618 enum {
619 	VFIO_PCI_BAR0_REGION_INDEX,
620 	VFIO_PCI_BAR1_REGION_INDEX,
621 	VFIO_PCI_BAR2_REGION_INDEX,
622 	VFIO_PCI_BAR3_REGION_INDEX,
623 	VFIO_PCI_BAR4_REGION_INDEX,
624 	VFIO_PCI_BAR5_REGION_INDEX,
625 	VFIO_PCI_ROM_REGION_INDEX,
626 	VFIO_PCI_CONFIG_REGION_INDEX,
627 	/*
628 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
629 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
630 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
631 	 * range is found at it's identity mapped offset from the region
632 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
633 	 * between described ranges are unimplemented.
634 	 */
635 	VFIO_PCI_VGA_REGION_INDEX,
636 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
637 				 /* device specific cap to define content. */
638 };
639 
640 enum {
641 	VFIO_PCI_INTX_IRQ_INDEX,
642 	VFIO_PCI_MSI_IRQ_INDEX,
643 	VFIO_PCI_MSIX_IRQ_INDEX,
644 	VFIO_PCI_ERR_IRQ_INDEX,
645 	VFIO_PCI_REQ_IRQ_INDEX,
646 	VFIO_PCI_NUM_IRQS
647 };
648 
649 /*
650  * The vfio-ccw bus driver makes use of the following fixed region and
651  * IRQ index mapping. Unimplemented regions return a size of zero.
652  * Unimplemented IRQ types return a count of zero.
653  */
654 
655 enum {
656 	VFIO_CCW_CONFIG_REGION_INDEX,
657 	VFIO_CCW_NUM_REGIONS
658 };
659 
660 enum {
661 	VFIO_CCW_IO_IRQ_INDEX,
662 	VFIO_CCW_CRW_IRQ_INDEX,
663 	VFIO_CCW_REQ_IRQ_INDEX,
664 	VFIO_CCW_NUM_IRQS
665 };
666 
667 /**
668  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12,
669  *					      struct vfio_pci_hot_reset_info)
670  *
671  * Return: 0 on success, -errno on failure:
672  *	-enospc = insufficient buffer, -enodev = unsupported for device.
673  */
674 struct vfio_pci_dependent_device {
675 	__u32	group_id;
676 	__u16	segment;
677 	__u8	bus;
678 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
679 };
680 
681 struct vfio_pci_hot_reset_info {
682 	__u32	argsz;
683 	__u32	flags;
684 	__u32	count;
685 	struct vfio_pci_dependent_device	devices[];
686 };
687 
688 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
689 
690 /**
691  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
692  *				    struct vfio_pci_hot_reset)
693  *
694  * Return: 0 on success, -errno on failure.
695  */
696 struct vfio_pci_hot_reset {
697 	__u32	argsz;
698 	__u32	flags;
699 	__u32	count;
700 	__s32	group_fds[];
701 };
702 
703 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
704 
705 /**
706  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
707  *                                    struct vfio_device_query_gfx_plane)
708  *
709  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
710  *
711  * flags supported:
712  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
713  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
714  *   support for dma-buf.
715  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
716  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
717  *   support for region.
718  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
719  *   with each call to query the plane info.
720  * - Others are invalid and return -EINVAL.
721  *
722  * Note:
723  * 1. Plane could be disabled by guest. In that case, success will be
724  *    returned with zero-initialized drm_format, size, width and height
725  *    fields.
726  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
727  *
728  * Return: 0 on success, -errno on other failure.
729  */
730 struct vfio_device_gfx_plane_info {
731 	__u32 argsz;
732 	__u32 flags;
733 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
734 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
735 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
736 	/* in */
737 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
738 	/* out */
739 	__u32 drm_format;	/* drm format of plane */
740 	__u64 drm_format_mod;   /* tiled mode */
741 	__u32 width;	/* width of plane */
742 	__u32 height;	/* height of plane */
743 	__u32 stride;	/* stride of plane */
744 	__u32 size;	/* size of plane in bytes, align on page*/
745 	__u32 x_pos;	/* horizontal position of cursor plane */
746 	__u32 y_pos;	/* vertical position of cursor plane*/
747 	__u32 x_hot;    /* horizontal position of cursor hotspot */
748 	__u32 y_hot;    /* vertical position of cursor hotspot */
749 	union {
750 		__u32 region_index;	/* region index */
751 		__u32 dmabuf_id;	/* dma-buf id */
752 	};
753 };
754 
755 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
756 
757 /**
758  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
759  *
760  * Return a new dma-buf file descriptor for an exposed guest framebuffer
761  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
762  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
763  */
764 
765 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
766 
767 /**
768  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
769  *                              struct vfio_device_ioeventfd)
770  *
771  * Perform a write to the device at the specified device fd offset, with
772  * the specified data and width when the provided eventfd is triggered.
773  * vfio bus drivers may not support this for all regions, for all widths,
774  * or at all.  vfio-pci currently only enables support for BAR regions,
775  * excluding the MSI-X vector table.
776  *
777  * Return: 0 on success, -errno on failure.
778  */
779 struct vfio_device_ioeventfd {
780 	__u32	argsz;
781 	__u32	flags;
782 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
783 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
784 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
785 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
786 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
787 	__u64	offset;			/* device fd offset of write */
788 	__u64	data;			/* data to be written */
789 	__s32	fd;			/* -1 for de-assignment */
790 };
791 
792 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
793 
794 /**
795  * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
796  *			       struct vfio_device_feature)
797  *
798  * Get, set, or probe feature data of the device.  The feature is selected
799  * using the FEATURE_MASK portion of the flags field.  Support for a feature
800  * can be probed by setting both the FEATURE_MASK and PROBE bits.  A probe
801  * may optionally include the GET and/or SET bits to determine read vs write
802  * access of the feature respectively.  Probing a feature will return success
803  * if the feature is supported and all of the optionally indicated GET/SET
804  * methods are supported.  The format of the data portion of the structure is
805  * specific to the given feature.  The data portion is not required for
806  * probing.  GET and SET are mutually exclusive, except for use with PROBE.
807  *
808  * Return 0 on success, -errno on failure.
809  */
810 struct vfio_device_feature {
811 	__u32	argsz;
812 	__u32	flags;
813 #define VFIO_DEVICE_FEATURE_MASK	(0xffff) /* 16-bit feature index */
814 #define VFIO_DEVICE_FEATURE_GET		(1 << 16) /* Get feature into data[] */
815 #define VFIO_DEVICE_FEATURE_SET		(1 << 17) /* Set feature from data[] */
816 #define VFIO_DEVICE_FEATURE_PROBE	(1 << 18) /* Probe feature support */
817 	__u8	data[];
818 };
819 
820 #define VFIO_DEVICE_FEATURE		_IO(VFIO_TYPE, VFIO_BASE + 17)
821 
822 /*
823  * Provide support for setting a PCI VF Token, which is used as a shared
824  * secret between PF and VF drivers.  This feature may only be set on a
825  * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
826  * open VFs.  Data provided when setting this feature is a 16-byte array
827  * (__u8 b[16]), representing a UUID.
828  */
829 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN	(0)
830 
831 /*
832  * Indicates the device can support the migration API through
833  * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and
834  * ERROR states are always supported. Support for additional states is
835  * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be
836  * set.
837  *
838  * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and
839  * RESUMING are supported.
840  *
841  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P
842  * is supported in addition to the STOP_COPY states.
843  *
844  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that
845  * PRE_COPY is supported in addition to the STOP_COPY states.
846  *
847  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY
848  * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported
849  * in addition to the STOP_COPY states.
850  *
851  * Other combinations of flags have behavior to be defined in the future.
852  */
853 struct vfio_device_feature_migration {
854 	__aligned_u64 flags;
855 #define VFIO_MIGRATION_STOP_COPY	(1 << 0)
856 #define VFIO_MIGRATION_P2P		(1 << 1)
857 #define VFIO_MIGRATION_PRE_COPY		(1 << 2)
858 };
859 #define VFIO_DEVICE_FEATURE_MIGRATION 1
860 
861 /*
862  * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO
863  * device. The new state is supplied in device_state, see enum
864  * vfio_device_mig_state for details
865  *
866  * The kernel migration driver must fully transition the device to the new state
867  * value before the operation returns to the user.
868  *
869  * The kernel migration driver must not generate asynchronous device state
870  * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET
871  * ioctl as described above.
872  *
873  * If this function fails then current device_state may be the original
874  * operating state or some other state along the combination transition path.
875  * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt
876  * to return to the original state, or attempt to return to some other state
877  * such as RUNNING or STOP.
878  *
879  * If the new_state starts a new data transfer session then the FD associated
880  * with that session is returned in data_fd. The user is responsible to close
881  * this FD when it is finished. The user must consider the migration data stream
882  * carried over the FD to be opaque and must preserve the byte order of the
883  * stream. The user is not required to preserve buffer segmentation when writing
884  * the data stream during the RESUMING operation.
885  *
886  * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO
887  * device, data_fd will be -1.
888  */
889 struct vfio_device_feature_mig_state {
890 	__u32 device_state; /* From enum vfio_device_mig_state */
891 	__s32 data_fd;
892 };
893 #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2
894 
895 /*
896  * The device migration Finite State Machine is described by the enum
897  * vfio_device_mig_state. Some of the FSM arcs will create a migration data
898  * transfer session by returning a FD, in this case the migration data will
899  * flow over the FD using read() and write() as discussed below.
900  *
901  * There are 5 states to support VFIO_MIGRATION_STOP_COPY:
902  *  RUNNING - The device is running normally
903  *  STOP - The device does not change the internal or external state
904  *  STOP_COPY - The device internal state can be read out
905  *  RESUMING - The device is stopped and is loading a new internal state
906  *  ERROR - The device has failed and must be reset
907  *
908  * And optional states to support VFIO_MIGRATION_P2P:
909  *  RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA
910  * And VFIO_MIGRATION_PRE_COPY:
911  *  PRE_COPY - The device is running normally but tracking internal state
912  *             changes
913  * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY:
914  *  PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA
915  *
916  * The FSM takes actions on the arcs between FSM states. The driver implements
917  * the following behavior for the FSM arcs:
918  *
919  * RUNNING_P2P -> STOP
920  * STOP_COPY -> STOP
921  *   While in STOP the device must stop the operation of the device. The device
922  *   must not generate interrupts, DMA, or any other change to external state.
923  *   It must not change its internal state. When stopped the device and kernel
924  *   migration driver must accept and respond to interaction to support external
925  *   subsystems in the STOP state, for example PCI MSI-X and PCI config space.
926  *   Failure by the user to restrict device access while in STOP must not result
927  *   in error conditions outside the user context (ex. host system faults).
928  *
929  *   The STOP_COPY arc will terminate a data transfer session.
930  *
931  * RESUMING -> STOP
932  *   Leaving RESUMING terminates a data transfer session and indicates the
933  *   device should complete processing of the data delivered by write(). The
934  *   kernel migration driver should complete the incorporation of data written
935  *   to the data transfer FD into the device internal state and perform
936  *   final validity and consistency checking of the new device state. If the
937  *   user provided data is found to be incomplete, inconsistent, or otherwise
938  *   invalid, the migration driver must fail the SET_STATE ioctl and
939  *   optionally go to the ERROR state as described below.
940  *
941  *   While in STOP the device has the same behavior as other STOP states
942  *   described above.
943  *
944  *   To abort a RESUMING session the device must be reset.
945  *
946  * PRE_COPY -> RUNNING
947  * RUNNING_P2P -> RUNNING
948  *   While in RUNNING the device is fully operational, the device may generate
949  *   interrupts, DMA, respond to MMIO, all vfio device regions are functional,
950  *   and the device may advance its internal state.
951  *
952  *   The PRE_COPY arc will terminate a data transfer session.
953  *
954  * PRE_COPY_P2P -> RUNNING_P2P
955  * RUNNING -> RUNNING_P2P
956  * STOP -> RUNNING_P2P
957  *   While in RUNNING_P2P the device is partially running in the P2P quiescent
958  *   state defined below.
959  *
960  *   The PRE_COPY_P2P arc will terminate a data transfer session.
961  *
962  * RUNNING -> PRE_COPY
963  * RUNNING_P2P -> PRE_COPY_P2P
964  * STOP -> STOP_COPY
965  *   PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states
966  *   which share a data transfer session. Moving between these states alters
967  *   what is streamed in session, but does not terminate or otherwise affect
968  *   the associated fd.
969  *
970  *   These arcs begin the process of saving the device state and will return a
971  *   new data_fd. The migration driver may perform actions such as enabling
972  *   dirty logging of device state when entering PRE_COPY or PER_COPY_P2P.
973  *
974  *   Each arc does not change the device operation, the device remains
975  *   RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below
976  *   in PRE_COPY_P2P -> STOP_COPY.
977  *
978  * PRE_COPY -> PRE_COPY_P2P
979  *   Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above.
980  *   However, while in the PRE_COPY_P2P state, the device is partially running
981  *   in the P2P quiescent state defined below, like RUNNING_P2P.
982  *
983  * PRE_COPY_P2P -> PRE_COPY
984  *   This arc allows returning the device to a full RUNNING behavior while
985  *   continuing all the behaviors of PRE_COPY.
986  *
987  * PRE_COPY_P2P -> STOP_COPY
988  *   While in the STOP_COPY state the device has the same behavior as STOP
989  *   with the addition that the data transfers session continues to stream the
990  *   migration state. End of stream on the FD indicates the entire device
991  *   state has been transferred.
992  *
993  *   The user should take steps to restrict access to vfio device regions while
994  *   the device is in STOP_COPY or risk corruption of the device migration data
995  *   stream.
996  *
997  * STOP -> RESUMING
998  *   Entering the RESUMING state starts a process of restoring the device state
999  *   and will return a new data_fd. The data stream fed into the data_fd should
1000  *   be taken from the data transfer output of a single FD during saving from
1001  *   a compatible device. The migration driver may alter/reset the internal
1002  *   device state for this arc if required to prepare the device to receive the
1003  *   migration data.
1004  *
1005  * STOP_COPY -> PRE_COPY
1006  * STOP_COPY -> PRE_COPY_P2P
1007  *   These arcs are not permitted and return error if requested. Future
1008  *   revisions of this API may define behaviors for these arcs, in this case
1009  *   support will be discoverable by a new flag in
1010  *   VFIO_DEVICE_FEATURE_MIGRATION.
1011  *
1012  * any -> ERROR
1013  *   ERROR cannot be specified as a device state, however any transition request
1014  *   can be failed with an errno return and may then move the device_state into
1015  *   ERROR. In this case the device was unable to execute the requested arc and
1016  *   was also unable to restore the device to any valid device_state.
1017  *   To recover from ERROR VFIO_DEVICE_RESET must be used to return the
1018  *   device_state back to RUNNING.
1019  *
1020  * The optional peer to peer (P2P) quiescent state is intended to be a quiescent
1021  * state for the device for the purposes of managing multiple devices within a
1022  * user context where peer-to-peer DMA between devices may be active. The
1023  * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating
1024  * any new P2P DMA transactions. If the device can identify P2P transactions
1025  * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration
1026  * driver must complete any such outstanding operations prior to completing the
1027  * FSM arc into a P2P state. For the purpose of specification the states
1028  * behave as though the device was fully running if not supported. Like while in
1029  * STOP or STOP_COPY the user must not touch the device, otherwise the state
1030  * can be exited.
1031  *
1032  * The remaining possible transitions are interpreted as combinations of the
1033  * above FSM arcs. As there are multiple paths through the FSM arcs the path
1034  * should be selected based on the following rules:
1035  *   - Select the shortest path.
1036  *   - The path cannot have saving group states as interior arcs, only
1037  *     starting/end states.
1038  * Refer to vfio_mig_get_next_state() for the result of the algorithm.
1039  *
1040  * The automatic transit through the FSM arcs that make up the combination
1041  * transition is invisible to the user. When working with combination arcs the
1042  * user may see any step along the path in the device_state if SET_STATE
1043  * fails. When handling these types of errors users should anticipate future
1044  * revisions of this protocol using new states and those states becoming
1045  * visible in this case.
1046  *
1047  * The optional states cannot be used with SET_STATE if the device does not
1048  * support them. The user can discover if these states are supported by using
1049  * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can
1050  * avoid knowing about these optional states if the kernel driver supports them.
1051  *
1052  * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY
1053  * is not present.
1054  */
1055 enum vfio_device_mig_state {
1056 	VFIO_DEVICE_STATE_ERROR = 0,
1057 	VFIO_DEVICE_STATE_STOP = 1,
1058 	VFIO_DEVICE_STATE_RUNNING = 2,
1059 	VFIO_DEVICE_STATE_STOP_COPY = 3,
1060 	VFIO_DEVICE_STATE_RESUMING = 4,
1061 	VFIO_DEVICE_STATE_RUNNING_P2P = 5,
1062 	VFIO_DEVICE_STATE_PRE_COPY = 6,
1063 	VFIO_DEVICE_STATE_PRE_COPY_P2P = 7,
1064 };
1065 
1066 /**
1067  * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21)
1068  *
1069  * This ioctl is used on the migration data FD in the precopy phase of the
1070  * migration data transfer. It returns an estimate of the current data sizes
1071  * remaining to be transferred. It allows the user to judge when it is
1072  * appropriate to leave PRE_COPY for STOP_COPY.
1073  *
1074  * This ioctl is valid only in PRE_COPY states and kernel driver should
1075  * return -EINVAL from any other migration state.
1076  *
1077  * The vfio_precopy_info data structure returned by this ioctl provides
1078  * estimates of data available from the device during the PRE_COPY states.
1079  * This estimate is split into two categories, initial_bytes and
1080  * dirty_bytes.
1081  *
1082  * The initial_bytes field indicates the amount of initial precopy
1083  * data available from the device. This field should have a non-zero initial
1084  * value and decrease as migration data is read from the device.
1085  * It is recommended to leave PRE_COPY for STOP_COPY only after this field
1086  * reaches zero. Leaving PRE_COPY earlier might make things slower.
1087  *
1088  * The dirty_bytes field tracks device state changes relative to data
1089  * previously retrieved.  This field starts at zero and may increase as
1090  * the internal device state is modified or decrease as that modified
1091  * state is read from the device.
1092  *
1093  * Userspace may use the combination of these fields to estimate the
1094  * potential data size available during the PRE_COPY phases, as well as
1095  * trends relative to the rate the device is dirtying its internal
1096  * state, but these fields are not required to have any bearing relative
1097  * to the data size available during the STOP_COPY phase.
1098  *
1099  * Drivers have a lot of flexibility in when and what they transfer during the
1100  * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO.
1101  *
1102  * During pre-copy the migration data FD has a temporary "end of stream" that is
1103  * reached when both initial_bytes and dirty_byte are zero. For instance, this
1104  * may indicate that the device is idle and not currently dirtying any internal
1105  * state. When read() is done on this temporary end of stream the kernel driver
1106  * should return ENOMSG from read(). Userspace can wait for more data (which may
1107  * never come) by using poll.
1108  *
1109  * Once in STOP_COPY the migration data FD has a permanent end of stream
1110  * signaled in the usual way by read() always returning 0 and poll always
1111  * returning readable. ENOMSG may not be returned in STOP_COPY.
1112  * Support for this ioctl is mandatory if a driver claims to support
1113  * VFIO_MIGRATION_PRE_COPY.
1114  *
1115  * Return: 0 on success, -1 and errno set on failure.
1116  */
1117 struct vfio_precopy_info {
1118 	__u32 argsz;
1119 	__u32 flags;
1120 	__aligned_u64 initial_bytes;
1121 	__aligned_u64 dirty_bytes;
1122 };
1123 
1124 #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21)
1125 
1126 /*
1127  * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power
1128  * state with the platform-based power management.  Device use of lower power
1129  * states depends on factors managed by the runtime power management core,
1130  * including system level support and coordinating support among dependent
1131  * devices.  Enabling device low power entry does not guarantee lower power
1132  * usage by the device, nor is a mechanism provided through this feature to
1133  * know the current power state of the device.  If any device access happens
1134  * (either from the host or through the vfio uAPI) when the device is in the
1135  * low power state, then the host will move the device out of the low power
1136  * state as necessary prior to the access.  Once the access is completed, the
1137  * device may re-enter the low power state.  For single shot low power support
1138  * with wake-up notification, see
1139  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below.  Access to mmap'd
1140  * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after
1141  * calling LOW_POWER_EXIT.
1142  */
1143 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3
1144 
1145 /*
1146  * This device feature has the same behavior as
1147  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user
1148  * provides an eventfd for wake-up notification.  When the device moves out of
1149  * the low power state for the wake-up, the host will not allow the device to
1150  * re-enter a low power state without a subsequent user call to one of the low
1151  * power entry device feature IOCTLs.  Access to mmap'd device regions is
1152  * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the
1153  * low power exit.  The low power exit can happen either through LOW_POWER_EXIT
1154  * or through any other access (where the wake-up notification has been
1155  * generated).  The access to mmap'd device regions will not trigger low power
1156  * exit.
1157  *
1158  * The notification through the provided eventfd will be generated only when
1159  * the device has entered and is resumed from a low power state after
1160  * calling this device feature IOCTL.  A device that has not entered low power
1161  * state, as managed through the runtime power management core, will not
1162  * generate a notification through the provided eventfd on access.  Calling the
1163  * LOW_POWER_EXIT feature is optional in the case where notification has been
1164  * signaled on the provided eventfd that a resume from low power has occurred.
1165  */
1166 struct vfio_device_low_power_entry_with_wakeup {
1167 	__s32 wakeup_eventfd;
1168 	__u32 reserved;
1169 };
1170 
1171 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4
1172 
1173 /*
1174  * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as
1175  * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or
1176  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features.
1177  * This device feature IOCTL may itself generate a wakeup eventfd notification
1178  * in the latter case if the device had previously entered a low power state.
1179  */
1180 #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5
1181 
1182 /*
1183  * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging.
1184  * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports
1185  * DMA logging.
1186  *
1187  * DMA logging allows a device to internally record what DMAs the device is
1188  * initiating and report them back to userspace. It is part of the VFIO
1189  * migration infrastructure that allows implementing dirty page tracking
1190  * during the pre copy phase of live migration. Only DMA WRITEs are logged,
1191  * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE.
1192  *
1193  * When DMA logging is started a range of IOVAs to monitor is provided and the
1194  * device can optimize its logging to cover only the IOVA range given. Each
1195  * DMA that the device initiates inside the range will be logged by the device
1196  * for later retrieval.
1197  *
1198  * page_size is an input that hints what tracking granularity the device
1199  * should try to achieve. If the device cannot do the hinted page size then
1200  * it's the driver choice which page size to pick based on its support.
1201  * On output the device will return the page size it selected.
1202  *
1203  * ranges is a pointer to an array of
1204  * struct vfio_device_feature_dma_logging_range.
1205  *
1206  * The core kernel code guarantees to support by minimum num_ranges that fit
1207  * into a single kernel page. User space can try higher values but should give
1208  * up if the above can't be achieved as of some driver limitations.
1209  *
1210  * A single call to start device DMA logging can be issued and a matching stop
1211  * should follow at the end. Another start is not allowed in the meantime.
1212  */
1213 struct vfio_device_feature_dma_logging_control {
1214 	__aligned_u64 page_size;
1215 	__u32 num_ranges;
1216 	__u32 __reserved;
1217 	__aligned_u64 ranges;
1218 };
1219 
1220 struct vfio_device_feature_dma_logging_range {
1221 	__aligned_u64 iova;
1222 	__aligned_u64 length;
1223 };
1224 
1225 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6
1226 
1227 /*
1228  * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started
1229  * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START
1230  */
1231 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7
1232 
1233 /*
1234  * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log
1235  *
1236  * Query the device's DMA log for written pages within the given IOVA range.
1237  * During querying the log is cleared for the IOVA range.
1238  *
1239  * bitmap is a pointer to an array of u64s that will hold the output bitmap
1240  * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits
1241  * is given by:
1242  *  bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64))
1243  *
1244  * The input page_size can be any power of two value and does not have to
1245  * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver
1246  * will format its internal logging to match the reporting page size, possibly
1247  * by replicating bits if the internal page size is lower than requested.
1248  *
1249  * The LOGGING_REPORT will only set bits in the bitmap and never clear or
1250  * perform any initialization of the user provided bitmap.
1251  *
1252  * If any error is returned userspace should assume that the dirty log is
1253  * corrupted. Error recovery is to consider all memory dirty and try to
1254  * restart the dirty tracking, or to abort/restart the whole migration.
1255  *
1256  * If DMA logging is not enabled, an error will be returned.
1257  *
1258  */
1259 struct vfio_device_feature_dma_logging_report {
1260 	__aligned_u64 iova;
1261 	__aligned_u64 length;
1262 	__aligned_u64 page_size;
1263 	__aligned_u64 bitmap;
1264 };
1265 
1266 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8
1267 
1268 /*
1269  * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will
1270  * be required to complete stop copy.
1271  *
1272  * Note: Can be called on each device state.
1273  */
1274 
1275 struct vfio_device_feature_mig_data_size {
1276 	__aligned_u64 stop_copy_length;
1277 };
1278 
1279 #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9
1280 
1281 /* -------- API for Type1 VFIO IOMMU -------- */
1282 
1283 /**
1284  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
1285  *
1286  * Retrieve information about the IOMMU object. Fills in provided
1287  * struct vfio_iommu_info. Caller sets argsz.
1288  *
1289  * XXX Should we do these by CHECK_EXTENSION too?
1290  */
1291 struct vfio_iommu_type1_info {
1292 	__u32	argsz;
1293 	__u32	flags;
1294 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
1295 #define VFIO_IOMMU_INFO_CAPS	(1 << 1)	/* Info supports caps */
1296 	__u64	iova_pgsizes;	/* Bitmap of supported page sizes */
1297 	__u32   cap_offset;	/* Offset within info struct of first cap */
1298 };
1299 
1300 /*
1301  * The IOVA capability allows to report the valid IOVA range(s)
1302  * excluding any non-relaxable reserved regions exposed by
1303  * devices attached to the container. Any DMA map attempt
1304  * outside the valid iova range will return error.
1305  *
1306  * The structures below define version 1 of this capability.
1307  */
1308 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE  1
1309 
1310 struct vfio_iova_range {
1311 	__u64	start;
1312 	__u64	end;
1313 };
1314 
1315 struct vfio_iommu_type1_info_cap_iova_range {
1316 	struct	vfio_info_cap_header header;
1317 	__u32	nr_iovas;
1318 	__u32	reserved;
1319 	struct	vfio_iova_range iova_ranges[];
1320 };
1321 
1322 /*
1323  * The migration capability allows to report supported features for migration.
1324  *
1325  * The structures below define version 1 of this capability.
1326  *
1327  * The existence of this capability indicates that IOMMU kernel driver supports
1328  * dirty page logging.
1329  *
1330  * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty
1331  * page logging.
1332  * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap
1333  * size in bytes that can be used by user applications when getting the dirty
1334  * bitmap.
1335  */
1336 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION  2
1337 
1338 struct vfio_iommu_type1_info_cap_migration {
1339 	struct	vfio_info_cap_header header;
1340 	__u32	flags;
1341 	__u64	pgsize_bitmap;
1342 	__u64	max_dirty_bitmap_size;		/* in bytes */
1343 };
1344 
1345 /*
1346  * The DMA available capability allows to report the current number of
1347  * simultaneously outstanding DMA mappings that are allowed.
1348  *
1349  * The structure below defines version 1 of this capability.
1350  *
1351  * avail: specifies the current number of outstanding DMA mappings allowed.
1352  */
1353 #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3
1354 
1355 struct vfio_iommu_type1_info_dma_avail {
1356 	struct	vfio_info_cap_header header;
1357 	__u32	avail;
1358 };
1359 
1360 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
1361 
1362 /**
1363  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
1364  *
1365  * Map process virtual addresses to IO virtual addresses using the
1366  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
1367  *
1368  * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr
1369  * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR.  To
1370  * maintain memory consistency within the user application, the updated vaddr
1371  * must address the same memory object as originally mapped.  Failure to do so
1372  * will result in user memory corruption and/or device misbehavior.  iova and
1373  * size must match those in the original MAP_DMA call.  Protection is not
1374  * changed, and the READ & WRITE flags must be 0.
1375  */
1376 struct vfio_iommu_type1_dma_map {
1377 	__u32	argsz;
1378 	__u32	flags;
1379 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
1380 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
1381 #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2)
1382 	__u64	vaddr;				/* Process virtual address */
1383 	__u64	iova;				/* IO virtual address */
1384 	__u64	size;				/* Size of mapping (bytes) */
1385 };
1386 
1387 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
1388 
1389 struct vfio_bitmap {
1390 	__u64        pgsize;	/* page size for bitmap in bytes */
1391 	__u64        size;	/* in bytes */
1392 	__u64 __user *data;	/* one bit per page */
1393 };
1394 
1395 /**
1396  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
1397  *							struct vfio_dma_unmap)
1398  *
1399  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
1400  * Caller sets argsz.  The actual unmapped size is returned in the size
1401  * field.  No guarantee is made to the user that arbitrary unmaps of iova
1402  * or size different from those used in the original mapping call will
1403  * succeed.
1404  *
1405  * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap
1406  * before unmapping IO virtual addresses. When this flag is set, the user must
1407  * provide a struct vfio_bitmap in data[]. User must provide zero-allocated
1408  * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field.
1409  * A bit in the bitmap represents one page, of user provided page size in
1410  * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set
1411  * indicates that the page at that offset from iova is dirty. A Bitmap of the
1412  * pages in the range of unmapped size is returned in the user-provided
1413  * vfio_bitmap.data.
1414  *
1415  * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses.  iova and size
1416  * must be 0.  This cannot be combined with the get-dirty-bitmap flag.
1417  *
1418  * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host
1419  * virtual addresses in the iova range.  DMA to already-mapped pages continues.
1420  * Groups may not be added to the container while any addresses are invalid.
1421  * This cannot be combined with the get-dirty-bitmap flag.
1422  */
1423 struct vfio_iommu_type1_dma_unmap {
1424 	__u32	argsz;
1425 	__u32	flags;
1426 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0)
1427 #define VFIO_DMA_UNMAP_FLAG_ALL		     (1 << 1)
1428 #define VFIO_DMA_UNMAP_FLAG_VADDR	     (1 << 2)
1429 	__u64	iova;				/* IO virtual address */
1430 	__u64	size;				/* Size of mapping (bytes) */
1431 	__u8    data[];
1432 };
1433 
1434 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
1435 
1436 /*
1437  * IOCTLs to enable/disable IOMMU container usage.
1438  * No parameters are supported.
1439  */
1440 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
1441 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
1442 
1443 /**
1444  * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
1445  *                                     struct vfio_iommu_type1_dirty_bitmap)
1446  * IOCTL is used for dirty pages logging.
1447  * Caller should set flag depending on which operation to perform, details as
1448  * below:
1449  *
1450  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs
1451  * the IOMMU driver to log pages that are dirtied or potentially dirtied by
1452  * the device; designed to be used when a migration is in progress. Dirty pages
1453  * are logged until logging is disabled by user application by calling the IOCTL
1454  * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag.
1455  *
1456  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs
1457  * the IOMMU driver to stop logging dirtied pages.
1458  *
1459  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set
1460  * returns the dirty pages bitmap for IOMMU container for a given IOVA range.
1461  * The user must specify the IOVA range and the pgsize through the structure
1462  * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface
1463  * supports getting a bitmap of the smallest supported pgsize only and can be
1464  * modified in future to get a bitmap of any specified supported pgsize. The
1465  * user must provide a zeroed memory area for the bitmap memory and specify its
1466  * size in bitmap.size. One bit is used to represent one page consecutively
1467  * starting from iova offset. The user should provide page size in bitmap.pgsize
1468  * field. A bit set in the bitmap indicates that the page at that offset from
1469  * iova is dirty. The caller must set argsz to a value including the size of
1470  * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the
1471  * actual bitmap. If dirty pages logging is not enabled, an error will be
1472  * returned.
1473  *
1474  * Only one of the flags _START, _STOP and _GET may be specified at a time.
1475  *
1476  */
1477 struct vfio_iommu_type1_dirty_bitmap {
1478 	__u32        argsz;
1479 	__u32        flags;
1480 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START	(1 << 0)
1481 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP	(1 << 1)
1482 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP	(1 << 2)
1483 	__u8         data[];
1484 };
1485 
1486 struct vfio_iommu_type1_dirty_bitmap_get {
1487 	__u64              iova;	/* IO virtual address */
1488 	__u64              size;	/* Size of iova range */
1489 	struct vfio_bitmap bitmap;
1490 };
1491 
1492 #define VFIO_IOMMU_DIRTY_PAGES             _IO(VFIO_TYPE, VFIO_BASE + 17)
1493 
1494 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
1495 
1496 /*
1497  * The SPAPR TCE DDW info struct provides the information about
1498  * the details of Dynamic DMA window capability.
1499  *
1500  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
1501  * @max_dynamic_windows_supported tells the maximum number of windows
1502  * which the platform can create.
1503  * @levels tells the maximum number of levels in multi-level IOMMU tables;
1504  * this allows splitting a table into smaller chunks which reduces
1505  * the amount of physically contiguous memory required for the table.
1506  */
1507 struct vfio_iommu_spapr_tce_ddw_info {
1508 	__u64 pgsizes;			/* Bitmap of supported page sizes */
1509 	__u32 max_dynamic_windows_supported;
1510 	__u32 levels;
1511 };
1512 
1513 /*
1514  * The SPAPR TCE info struct provides the information about the PCI bus
1515  * address ranges available for DMA, these values are programmed into
1516  * the hardware so the guest has to know that information.
1517  *
1518  * The DMA 32 bit window start is an absolute PCI bus address.
1519  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
1520  * addresses too so the window works as a filter rather than an offset
1521  * for IOVA addresses.
1522  *
1523  * Flags supported:
1524  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
1525  *   (DDW) support is present. @ddw is only supported when DDW is present.
1526  */
1527 struct vfio_iommu_spapr_tce_info {
1528 	__u32 argsz;
1529 	__u32 flags;
1530 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
1531 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
1532 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
1533 	struct vfio_iommu_spapr_tce_ddw_info ddw;
1534 };
1535 
1536 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
1537 
1538 /*
1539  * EEH PE operation struct provides ways to:
1540  * - enable/disable EEH functionality;
1541  * - unfreeze IO/DMA for frozen PE;
1542  * - read PE state;
1543  * - reset PE;
1544  * - configure PE;
1545  * - inject EEH error.
1546  */
1547 struct vfio_eeh_pe_err {
1548 	__u32 type;
1549 	__u32 func;
1550 	__u64 addr;
1551 	__u64 mask;
1552 };
1553 
1554 struct vfio_eeh_pe_op {
1555 	__u32 argsz;
1556 	__u32 flags;
1557 	__u32 op;
1558 	union {
1559 		struct vfio_eeh_pe_err err;
1560 	};
1561 };
1562 
1563 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
1564 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
1565 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
1566 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
1567 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
1568 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
1569 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
1570 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
1571 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
1572 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
1573 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
1574 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
1575 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
1576 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
1577 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
1578 
1579 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
1580 
1581 /**
1582  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
1583  *
1584  * Registers user space memory where DMA is allowed. It pins
1585  * user pages and does the locked memory accounting so
1586  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
1587  * get faster.
1588  */
1589 struct vfio_iommu_spapr_register_memory {
1590 	__u32	argsz;
1591 	__u32	flags;
1592 	__u64	vaddr;				/* Process virtual address */
1593 	__u64	size;				/* Size of mapping (bytes) */
1594 };
1595 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
1596 
1597 /**
1598  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
1599  *
1600  * Unregisters user space memory registered with
1601  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
1602  * Uses vfio_iommu_spapr_register_memory for parameters.
1603  */
1604 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
1605 
1606 /**
1607  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
1608  *
1609  * Creates an additional TCE table and programs it (sets a new DMA window)
1610  * to every IOMMU group in the container. It receives page shift, window
1611  * size and number of levels in the TCE table being created.
1612  *
1613  * It allocates and returns an offset on a PCI bus of the new DMA window.
1614  */
1615 struct vfio_iommu_spapr_tce_create {
1616 	__u32 argsz;
1617 	__u32 flags;
1618 	/* in */
1619 	__u32 page_shift;
1620 	__u32 __resv1;
1621 	__u64 window_size;
1622 	__u32 levels;
1623 	__u32 __resv2;
1624 	/* out */
1625 	__u64 start_addr;
1626 };
1627 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
1628 
1629 /**
1630  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
1631  *
1632  * Unprograms a TCE table from all groups in the container and destroys it.
1633  * It receives a PCI bus offset as a window id.
1634  */
1635 struct vfio_iommu_spapr_tce_remove {
1636 	__u32 argsz;
1637 	__u32 flags;
1638 	/* in */
1639 	__u64 start_addr;
1640 };
1641 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
1642 
1643 /* ***************************************************************** */
1644 
1645 #endif /* _UAPIVFIO_H */
1646