xref: /linux-6.15/include/uapi/linux/vfio.h (revision a5bfe22d)
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 	__u32	num_regions;	/* Max region index + 1 */
217 	__u32	num_irqs;	/* Max IRQ index + 1 */
218 	__u32   cap_offset;	/* Offset within info struct of first cap */
219 };
220 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
221 
222 /*
223  * Vendor driver using Mediated device framework should provide device_api
224  * attribute in supported type attribute groups. Device API string should be one
225  * of the following corresponding to device flags in vfio_device_info structure.
226  */
227 
228 #define VFIO_DEVICE_API_PCI_STRING		"vfio-pci"
229 #define VFIO_DEVICE_API_PLATFORM_STRING		"vfio-platform"
230 #define VFIO_DEVICE_API_AMBA_STRING		"vfio-amba"
231 #define VFIO_DEVICE_API_CCW_STRING		"vfio-ccw"
232 #define VFIO_DEVICE_API_AP_STRING		"vfio-ap"
233 
234 /*
235  * The following capabilities are unique to s390 zPCI devices.  Their contents
236  * are further-defined in vfio_zdev.h
237  */
238 #define VFIO_DEVICE_INFO_CAP_ZPCI_BASE		1
239 #define VFIO_DEVICE_INFO_CAP_ZPCI_GROUP		2
240 #define VFIO_DEVICE_INFO_CAP_ZPCI_UTIL		3
241 #define VFIO_DEVICE_INFO_CAP_ZPCI_PFIP		4
242 
243 /*
244  * The following VFIO_DEVICE_INFO capability reports support for PCIe AtomicOp
245  * completion to the root bus with supported widths provided via flags.
246  */
247 #define VFIO_DEVICE_INFO_CAP_PCI_ATOMIC_COMP	5
248 struct vfio_device_info_cap_pci_atomic_comp {
249 	struct vfio_info_cap_header header;
250 	__u32 flags;
251 #define VFIO_PCI_ATOMIC_COMP32	(1 << 0)
252 #define VFIO_PCI_ATOMIC_COMP64	(1 << 1)
253 #define VFIO_PCI_ATOMIC_COMP128	(1 << 2)
254 	__u32 reserved;
255 };
256 
257 /**
258  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
259  *				       struct vfio_region_info)
260  *
261  * Retrieve information about a device region.  Caller provides
262  * struct vfio_region_info with index value set.  Caller sets argsz.
263  * Implementation of region mapping is bus driver specific.  This is
264  * intended to describe MMIO, I/O port, as well as bus specific
265  * regions (ex. PCI config space).  Zero sized regions may be used
266  * to describe unimplemented regions (ex. unimplemented PCI BARs).
267  * Return: 0 on success, -errno on failure.
268  */
269 struct vfio_region_info {
270 	__u32	argsz;
271 	__u32	flags;
272 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
273 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
274 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
275 #define VFIO_REGION_INFO_FLAG_CAPS	(1 << 3) /* Info supports caps */
276 	__u32	index;		/* Region index */
277 	__u32	cap_offset;	/* Offset within info struct of first cap */
278 	__u64	size;		/* Region size (bytes) */
279 	__u64	offset;		/* Region offset from start of device fd */
280 };
281 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
282 
283 /*
284  * The sparse mmap capability allows finer granularity of specifying areas
285  * within a region with mmap support.  When specified, the user should only
286  * mmap the offset ranges specified by the areas array.  mmaps outside of the
287  * areas specified may fail (such as the range covering a PCI MSI-X table) or
288  * may result in improper device behavior.
289  *
290  * The structures below define version 1 of this capability.
291  */
292 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP	1
293 
294 struct vfio_region_sparse_mmap_area {
295 	__u64	offset;	/* Offset of mmap'able area within region */
296 	__u64	size;	/* Size of mmap'able area */
297 };
298 
299 struct vfio_region_info_cap_sparse_mmap {
300 	struct vfio_info_cap_header header;
301 	__u32	nr_areas;
302 	__u32	reserved;
303 	struct vfio_region_sparse_mmap_area areas[];
304 };
305 
306 /*
307  * The device specific type capability allows regions unique to a specific
308  * device or class of devices to be exposed.  This helps solve the problem for
309  * vfio bus drivers of defining which region indexes correspond to which region
310  * on the device, without needing to resort to static indexes, as done by
311  * vfio-pci.  For instance, if we were to go back in time, we might remove
312  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
313  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
314  * make a "VGA" device specific type to describe the VGA access space.  This
315  * means that non-VGA devices wouldn't need to waste this index, and thus the
316  * address space associated with it due to implementation of device file
317  * descriptor offsets in vfio-pci.
318  *
319  * The current implementation is now part of the user ABI, so we can't use this
320  * for VGA, but there are other upcoming use cases, such as opregions for Intel
321  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
322  * use this for future additions.
323  *
324  * The structure below defines version 1 of this capability.
325  */
326 #define VFIO_REGION_INFO_CAP_TYPE	2
327 
328 struct vfio_region_info_cap_type {
329 	struct vfio_info_cap_header header;
330 	__u32 type;	/* global per bus driver */
331 	__u32 subtype;	/* type specific */
332 };
333 
334 /*
335  * List of region types, global per bus driver.
336  * If you introduce a new type, please add it here.
337  */
338 
339 /* PCI region type containing a PCI vendor part */
340 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
341 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
342 #define VFIO_REGION_TYPE_GFX                    (1)
343 #define VFIO_REGION_TYPE_CCW			(2)
344 #define VFIO_REGION_TYPE_MIGRATION_DEPRECATED   (3)
345 
346 /* sub-types for VFIO_REGION_TYPE_PCI_* */
347 
348 /* 8086 vendor PCI sub-types */
349 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
350 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
351 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
352 
353 /* 10de vendor PCI sub-types */
354 /*
355  * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
356  *
357  * Deprecated, region no longer provided
358  */
359 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM	(1)
360 
361 /* 1014 vendor PCI sub-types */
362 /*
363  * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
364  * to do TLB invalidation on a GPU.
365  *
366  * Deprecated, region no longer provided
367  */
368 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD	(1)
369 
370 /* sub-types for VFIO_REGION_TYPE_GFX */
371 #define VFIO_REGION_SUBTYPE_GFX_EDID            (1)
372 
373 /**
374  * struct vfio_region_gfx_edid - EDID region layout.
375  *
376  * Set display link state and EDID blob.
377  *
378  * The EDID blob has monitor information such as brand, name, serial
379  * number, physical size, supported video modes and more.
380  *
381  * This special region allows userspace (typically qemu) set a virtual
382  * EDID for the virtual monitor, which allows a flexible display
383  * configuration.
384  *
385  * For the edid blob spec look here:
386  *    https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
387  *
388  * On linux systems you can find the EDID blob in sysfs:
389  *    /sys/class/drm/${card}/${connector}/edid
390  *
391  * You can use the edid-decode ulility (comes with xorg-x11-utils) to
392  * decode the EDID blob.
393  *
394  * @edid_offset: location of the edid blob, relative to the
395  *               start of the region (readonly).
396  * @edid_max_size: max size of the edid blob (readonly).
397  * @edid_size: actual edid size (read/write).
398  * @link_state: display link state (read/write).
399  * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
400  * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
401  * @max_xres: max display width (0 == no limitation, readonly).
402  * @max_yres: max display height (0 == no limitation, readonly).
403  *
404  * EDID update protocol:
405  *   (1) set link-state to down.
406  *   (2) update edid blob and size.
407  *   (3) set link-state to up.
408  */
409 struct vfio_region_gfx_edid {
410 	__u32 edid_offset;
411 	__u32 edid_max_size;
412 	__u32 edid_size;
413 	__u32 max_xres;
414 	__u32 max_yres;
415 	__u32 link_state;
416 #define VFIO_DEVICE_GFX_LINK_STATE_UP    1
417 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN  2
418 };
419 
420 /* sub-types for VFIO_REGION_TYPE_CCW */
421 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD	(1)
422 #define VFIO_REGION_SUBTYPE_CCW_SCHIB		(2)
423 #define VFIO_REGION_SUBTYPE_CCW_CRW		(3)
424 
425 /* sub-types for VFIO_REGION_TYPE_MIGRATION */
426 #define VFIO_REGION_SUBTYPE_MIGRATION_DEPRECATED (1)
427 
428 struct vfio_device_migration_info {
429 	__u32 device_state;         /* VFIO device state */
430 #define VFIO_DEVICE_STATE_V1_STOP      (0)
431 #define VFIO_DEVICE_STATE_V1_RUNNING   (1 << 0)
432 #define VFIO_DEVICE_STATE_V1_SAVING    (1 << 1)
433 #define VFIO_DEVICE_STATE_V1_RESUMING  (1 << 2)
434 #define VFIO_DEVICE_STATE_MASK      (VFIO_DEVICE_STATE_V1_RUNNING | \
435 				     VFIO_DEVICE_STATE_V1_SAVING |  \
436 				     VFIO_DEVICE_STATE_V1_RESUMING)
437 
438 #define VFIO_DEVICE_STATE_VALID(state) \
439 	(state & VFIO_DEVICE_STATE_V1_RESUMING ? \
440 	(state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_V1_RESUMING : 1)
441 
442 #define VFIO_DEVICE_STATE_IS_ERROR(state) \
443 	((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_V1_SAVING | \
444 					      VFIO_DEVICE_STATE_V1_RESUMING))
445 
446 #define VFIO_DEVICE_STATE_SET_ERROR(state) \
447 	((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_STATE_V1_SAVING | \
448 					     VFIO_DEVICE_STATE_V1_RESUMING)
449 
450 	__u32 reserved;
451 	__u64 pending_bytes;
452 	__u64 data_offset;
453 	__u64 data_size;
454 };
455 
456 /*
457  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
458  * which allows direct access to non-MSIX registers which happened to be within
459  * the same system page.
460  *
461  * Even though the userspace gets direct access to the MSIX data, the existing
462  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
463  */
464 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE	3
465 
466 /*
467  * Capability with compressed real address (aka SSA - small system address)
468  * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
469  * and by the userspace to associate a NVLink bridge with a GPU.
470  *
471  * Deprecated, capability no longer provided
472  */
473 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT	4
474 
475 struct vfio_region_info_cap_nvlink2_ssatgt {
476 	struct vfio_info_cap_header header;
477 	__u64 tgt;
478 };
479 
480 /*
481  * Capability with an NVLink link speed. The value is read by
482  * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
483  * property in the device tree. The value is fixed in the hardware
484  * and failing to provide the correct value results in the link
485  * not working with no indication from the driver why.
486  *
487  * Deprecated, capability no longer provided
488  */
489 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD	5
490 
491 struct vfio_region_info_cap_nvlink2_lnkspd {
492 	struct vfio_info_cap_header header;
493 	__u32 link_speed;
494 	__u32 __pad;
495 };
496 
497 /**
498  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
499  *				    struct vfio_irq_info)
500  *
501  * Retrieve information about a device IRQ.  Caller provides
502  * struct vfio_irq_info with index value set.  Caller sets argsz.
503  * Implementation of IRQ mapping is bus driver specific.  Indexes
504  * using multiple IRQs are primarily intended to support MSI-like
505  * interrupt blocks.  Zero count irq blocks may be used to describe
506  * unimplemented interrupt types.
507  *
508  * The EVENTFD flag indicates the interrupt index supports eventfd based
509  * signaling.
510  *
511  * The MASKABLE flags indicates the index supports MASK and UNMASK
512  * actions described below.
513  *
514  * AUTOMASKED indicates that after signaling, the interrupt line is
515  * automatically masked by VFIO and the user needs to unmask the line
516  * to receive new interrupts.  This is primarily intended to distinguish
517  * level triggered interrupts.
518  *
519  * The NORESIZE flag indicates that the interrupt lines within the index
520  * are setup as a set and new subindexes cannot be enabled without first
521  * disabling the entire index.  This is used for interrupts like PCI MSI
522  * and MSI-X where the driver may only use a subset of the available
523  * indexes, but VFIO needs to enable a specific number of vectors
524  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
525  * then add and unmask vectors, it's up to userspace to make the decision
526  * whether to allocate the maximum supported number of vectors or tear
527  * down setup and incrementally increase the vectors as each is enabled.
528  * Absence of the NORESIZE flag indicates that vectors can be enabled
529  * and disabled dynamically without impacting other vectors within the
530  * index.
531  */
532 struct vfio_irq_info {
533 	__u32	argsz;
534 	__u32	flags;
535 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
536 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
537 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
538 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
539 	__u32	index;		/* IRQ index */
540 	__u32	count;		/* Number of IRQs within this index */
541 };
542 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
543 
544 /**
545  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
546  *
547  * Set signaling, masking, and unmasking of interrupts.  Caller provides
548  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
549  * the range of subindexes being specified.
550  *
551  * The DATA flags specify the type of data provided.  If DATA_NONE, the
552  * operation performs the specified action immediately on the specified
553  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
554  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
555  *
556  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
557  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
558  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
559  * data = {1,0,1}
560  *
561  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
562  * A value of -1 can be used to either de-assign interrupts if already
563  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
564  * to be trigger for interrupts [0,0] and [0,2]:
565  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
566  * data = {fd1, -1, fd2}
567  * If index [0,1] is previously set, two count = 1 ioctls calls would be
568  * required to set [0,0] and [0,2] without changing [0,1].
569  *
570  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
571  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
572  * from userspace (ie. simulate hardware triggering).
573  *
574  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
575  * enables the interrupt index for the device.  Individual subindex interrupts
576  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
577  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
578  *
579  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
580  * ACTION_TRIGGER specifies kernel->user signaling.
581  */
582 struct vfio_irq_set {
583 	__u32	argsz;
584 	__u32	flags;
585 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
586 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
587 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
588 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
589 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
590 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
591 	__u32	index;
592 	__u32	start;
593 	__u32	count;
594 	__u8	data[];
595 };
596 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
597 
598 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
599 					 VFIO_IRQ_SET_DATA_BOOL | \
600 					 VFIO_IRQ_SET_DATA_EVENTFD)
601 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
602 					 VFIO_IRQ_SET_ACTION_UNMASK | \
603 					 VFIO_IRQ_SET_ACTION_TRIGGER)
604 /**
605  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
606  *
607  * Reset a device.
608  */
609 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
610 
611 /*
612  * The VFIO-PCI bus driver makes use of the following fixed region and
613  * IRQ index mapping.  Unimplemented regions return a size of zero.
614  * Unimplemented IRQ types return a count of zero.
615  */
616 
617 enum {
618 	VFIO_PCI_BAR0_REGION_INDEX,
619 	VFIO_PCI_BAR1_REGION_INDEX,
620 	VFIO_PCI_BAR2_REGION_INDEX,
621 	VFIO_PCI_BAR3_REGION_INDEX,
622 	VFIO_PCI_BAR4_REGION_INDEX,
623 	VFIO_PCI_BAR5_REGION_INDEX,
624 	VFIO_PCI_ROM_REGION_INDEX,
625 	VFIO_PCI_CONFIG_REGION_INDEX,
626 	/*
627 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
628 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
629 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
630 	 * range is found at it's identity mapped offset from the region
631 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
632 	 * between described ranges are unimplemented.
633 	 */
634 	VFIO_PCI_VGA_REGION_INDEX,
635 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
636 				 /* device specific cap to define content. */
637 };
638 
639 enum {
640 	VFIO_PCI_INTX_IRQ_INDEX,
641 	VFIO_PCI_MSI_IRQ_INDEX,
642 	VFIO_PCI_MSIX_IRQ_INDEX,
643 	VFIO_PCI_ERR_IRQ_INDEX,
644 	VFIO_PCI_REQ_IRQ_INDEX,
645 	VFIO_PCI_NUM_IRQS
646 };
647 
648 /*
649  * The vfio-ccw bus driver makes use of the following fixed region and
650  * IRQ index mapping. Unimplemented regions return a size of zero.
651  * Unimplemented IRQ types return a count of zero.
652  */
653 
654 enum {
655 	VFIO_CCW_CONFIG_REGION_INDEX,
656 	VFIO_CCW_NUM_REGIONS
657 };
658 
659 enum {
660 	VFIO_CCW_IO_IRQ_INDEX,
661 	VFIO_CCW_CRW_IRQ_INDEX,
662 	VFIO_CCW_REQ_IRQ_INDEX,
663 	VFIO_CCW_NUM_IRQS
664 };
665 
666 /**
667  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12,
668  *					      struct vfio_pci_hot_reset_info)
669  *
670  * Return: 0 on success, -errno on failure:
671  *	-enospc = insufficient buffer, -enodev = unsupported for device.
672  */
673 struct vfio_pci_dependent_device {
674 	__u32	group_id;
675 	__u16	segment;
676 	__u8	bus;
677 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
678 };
679 
680 struct vfio_pci_hot_reset_info {
681 	__u32	argsz;
682 	__u32	flags;
683 	__u32	count;
684 	struct vfio_pci_dependent_device	devices[];
685 };
686 
687 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
688 
689 /**
690  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
691  *				    struct vfio_pci_hot_reset)
692  *
693  * Return: 0 on success, -errno on failure.
694  */
695 struct vfio_pci_hot_reset {
696 	__u32	argsz;
697 	__u32	flags;
698 	__u32	count;
699 	__s32	group_fds[];
700 };
701 
702 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
703 
704 /**
705  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
706  *                                    struct vfio_device_query_gfx_plane)
707  *
708  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
709  *
710  * flags supported:
711  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
712  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
713  *   support for dma-buf.
714  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
715  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
716  *   support for region.
717  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
718  *   with each call to query the plane info.
719  * - Others are invalid and return -EINVAL.
720  *
721  * Note:
722  * 1. Plane could be disabled by guest. In that case, success will be
723  *    returned with zero-initialized drm_format, size, width and height
724  *    fields.
725  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
726  *
727  * Return: 0 on success, -errno on other failure.
728  */
729 struct vfio_device_gfx_plane_info {
730 	__u32 argsz;
731 	__u32 flags;
732 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
733 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
734 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
735 	/* in */
736 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
737 	/* out */
738 	__u32 drm_format;	/* drm format of plane */
739 	__u64 drm_format_mod;   /* tiled mode */
740 	__u32 width;	/* width of plane */
741 	__u32 height;	/* height of plane */
742 	__u32 stride;	/* stride of plane */
743 	__u32 size;	/* size of plane in bytes, align on page*/
744 	__u32 x_pos;	/* horizontal position of cursor plane */
745 	__u32 y_pos;	/* vertical position of cursor plane*/
746 	__u32 x_hot;    /* horizontal position of cursor hotspot */
747 	__u32 y_hot;    /* vertical position of cursor hotspot */
748 	union {
749 		__u32 region_index;	/* region index */
750 		__u32 dmabuf_id;	/* dma-buf id */
751 	};
752 };
753 
754 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
755 
756 /**
757  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
758  *
759  * Return a new dma-buf file descriptor for an exposed guest framebuffer
760  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
761  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
762  */
763 
764 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
765 
766 /**
767  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
768  *                              struct vfio_device_ioeventfd)
769  *
770  * Perform a write to the device at the specified device fd offset, with
771  * the specified data and width when the provided eventfd is triggered.
772  * vfio bus drivers may not support this for all regions, for all widths,
773  * or at all.  vfio-pci currently only enables support for BAR regions,
774  * excluding the MSI-X vector table.
775  *
776  * Return: 0 on success, -errno on failure.
777  */
778 struct vfio_device_ioeventfd {
779 	__u32	argsz;
780 	__u32	flags;
781 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
782 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
783 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
784 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
785 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
786 	__u64	offset;			/* device fd offset of write */
787 	__u64	data;			/* data to be written */
788 	__s32	fd;			/* -1 for de-assignment */
789 };
790 
791 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
792 
793 /**
794  * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
795  *			       struct vfio_device_feature)
796  *
797  * Get, set, or probe feature data of the device.  The feature is selected
798  * using the FEATURE_MASK portion of the flags field.  Support for a feature
799  * can be probed by setting both the FEATURE_MASK and PROBE bits.  A probe
800  * may optionally include the GET and/or SET bits to determine read vs write
801  * access of the feature respectively.  Probing a feature will return success
802  * if the feature is supported and all of the optionally indicated GET/SET
803  * methods are supported.  The format of the data portion of the structure is
804  * specific to the given feature.  The data portion is not required for
805  * probing.  GET and SET are mutually exclusive, except for use with PROBE.
806  *
807  * Return 0 on success, -errno on failure.
808  */
809 struct vfio_device_feature {
810 	__u32	argsz;
811 	__u32	flags;
812 #define VFIO_DEVICE_FEATURE_MASK	(0xffff) /* 16-bit feature index */
813 #define VFIO_DEVICE_FEATURE_GET		(1 << 16) /* Get feature into data[] */
814 #define VFIO_DEVICE_FEATURE_SET		(1 << 17) /* Set feature from data[] */
815 #define VFIO_DEVICE_FEATURE_PROBE	(1 << 18) /* Probe feature support */
816 	__u8	data[];
817 };
818 
819 #define VFIO_DEVICE_FEATURE		_IO(VFIO_TYPE, VFIO_BASE + 17)
820 
821 /*
822  * Provide support for setting a PCI VF Token, which is used as a shared
823  * secret between PF and VF drivers.  This feature may only be set on a
824  * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
825  * open VFs.  Data provided when setting this feature is a 16-byte array
826  * (__u8 b[16]), representing a UUID.
827  */
828 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN	(0)
829 
830 /*
831  * Indicates the device can support the migration API through
832  * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and
833  * ERROR states are always supported. Support for additional states is
834  * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be
835  * set.
836  *
837  * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and
838  * RESUMING are supported.
839  *
840  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P
841  * is supported in addition to the STOP_COPY states.
842  *
843  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that
844  * PRE_COPY is supported in addition to the STOP_COPY states.
845  *
846  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY
847  * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported
848  * in addition to the STOP_COPY states.
849  *
850  * Other combinations of flags have behavior to be defined in the future.
851  */
852 struct vfio_device_feature_migration {
853 	__aligned_u64 flags;
854 #define VFIO_MIGRATION_STOP_COPY	(1 << 0)
855 #define VFIO_MIGRATION_P2P		(1 << 1)
856 #define VFIO_MIGRATION_PRE_COPY		(1 << 2)
857 };
858 #define VFIO_DEVICE_FEATURE_MIGRATION 1
859 
860 /*
861  * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO
862  * device. The new state is supplied in device_state, see enum
863  * vfio_device_mig_state for details
864  *
865  * The kernel migration driver must fully transition the device to the new state
866  * value before the operation returns to the user.
867  *
868  * The kernel migration driver must not generate asynchronous device state
869  * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET
870  * ioctl as described above.
871  *
872  * If this function fails then current device_state may be the original
873  * operating state or some other state along the combination transition path.
874  * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt
875  * to return to the original state, or attempt to return to some other state
876  * such as RUNNING or STOP.
877  *
878  * If the new_state starts a new data transfer session then the FD associated
879  * with that session is returned in data_fd. The user is responsible to close
880  * this FD when it is finished. The user must consider the migration data stream
881  * carried over the FD to be opaque and must preserve the byte order of the
882  * stream. The user is not required to preserve buffer segmentation when writing
883  * the data stream during the RESUMING operation.
884  *
885  * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO
886  * device, data_fd will be -1.
887  */
888 struct vfio_device_feature_mig_state {
889 	__u32 device_state; /* From enum vfio_device_mig_state */
890 	__s32 data_fd;
891 };
892 #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2
893 
894 /*
895  * The device migration Finite State Machine is described by the enum
896  * vfio_device_mig_state. Some of the FSM arcs will create a migration data
897  * transfer session by returning a FD, in this case the migration data will
898  * flow over the FD using read() and write() as discussed below.
899  *
900  * There are 5 states to support VFIO_MIGRATION_STOP_COPY:
901  *  RUNNING - The device is running normally
902  *  STOP - The device does not change the internal or external state
903  *  STOP_COPY - The device internal state can be read out
904  *  RESUMING - The device is stopped and is loading a new internal state
905  *  ERROR - The device has failed and must be reset
906  *
907  * And optional states to support VFIO_MIGRATION_P2P:
908  *  RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA
909  * And VFIO_MIGRATION_PRE_COPY:
910  *  PRE_COPY - The device is running normally but tracking internal state
911  *             changes
912  * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY:
913  *  PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA
914  *
915  * The FSM takes actions on the arcs between FSM states. The driver implements
916  * the following behavior for the FSM arcs:
917  *
918  * RUNNING_P2P -> STOP
919  * STOP_COPY -> STOP
920  *   While in STOP the device must stop the operation of the device. The device
921  *   must not generate interrupts, DMA, or any other change to external state.
922  *   It must not change its internal state. When stopped the device and kernel
923  *   migration driver must accept and respond to interaction to support external
924  *   subsystems in the STOP state, for example PCI MSI-X and PCI config space.
925  *   Failure by the user to restrict device access while in STOP must not result
926  *   in error conditions outside the user context (ex. host system faults).
927  *
928  *   The STOP_COPY arc will terminate a data transfer session.
929  *
930  * RESUMING -> STOP
931  *   Leaving RESUMING terminates a data transfer session and indicates the
932  *   device should complete processing of the data delivered by write(). The
933  *   kernel migration driver should complete the incorporation of data written
934  *   to the data transfer FD into the device internal state and perform
935  *   final validity and consistency checking of the new device state. If the
936  *   user provided data is found to be incomplete, inconsistent, or otherwise
937  *   invalid, the migration driver must fail the SET_STATE ioctl and
938  *   optionally go to the ERROR state as described below.
939  *
940  *   While in STOP the device has the same behavior as other STOP states
941  *   described above.
942  *
943  *   To abort a RESUMING session the device must be reset.
944  *
945  * PRE_COPY -> RUNNING
946  * RUNNING_P2P -> RUNNING
947  *   While in RUNNING the device is fully operational, the device may generate
948  *   interrupts, DMA, respond to MMIO, all vfio device regions are functional,
949  *   and the device may advance its internal state.
950  *
951  *   The PRE_COPY arc will terminate a data transfer session.
952  *
953  * PRE_COPY_P2P -> RUNNING_P2P
954  * RUNNING -> RUNNING_P2P
955  * STOP -> RUNNING_P2P
956  *   While in RUNNING_P2P the device is partially running in the P2P quiescent
957  *   state defined below.
958  *
959  *   The PRE_COPY_P2P arc will terminate a data transfer session.
960  *
961  * RUNNING -> PRE_COPY
962  * RUNNING_P2P -> PRE_COPY_P2P
963  * STOP -> STOP_COPY
964  *   PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states
965  *   which share a data transfer session. Moving between these states alters
966  *   what is streamed in session, but does not terminate or otherwise affect
967  *   the associated fd.
968  *
969  *   These arcs begin the process of saving the device state and will return a
970  *   new data_fd. The migration driver may perform actions such as enabling
971  *   dirty logging of device state when entering PRE_COPY or PER_COPY_P2P.
972  *
973  *   Each arc does not change the device operation, the device remains
974  *   RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below
975  *   in PRE_COPY_P2P -> STOP_COPY.
976  *
977  * PRE_COPY -> PRE_COPY_P2P
978  *   Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above.
979  *   However, while in the PRE_COPY_P2P state, the device is partially running
980  *   in the P2P quiescent state defined below, like RUNNING_P2P.
981  *
982  * PRE_COPY_P2P -> PRE_COPY
983  *   This arc allows returning the device to a full RUNNING behavior while
984  *   continuing all the behaviors of PRE_COPY.
985  *
986  * PRE_COPY_P2P -> STOP_COPY
987  *   While in the STOP_COPY state the device has the same behavior as STOP
988  *   with the addition that the data transfers session continues to stream the
989  *   migration state. End of stream on the FD indicates the entire device
990  *   state has been transferred.
991  *
992  *   The user should take steps to restrict access to vfio device regions while
993  *   the device is in STOP_COPY or risk corruption of the device migration data
994  *   stream.
995  *
996  * STOP -> RESUMING
997  *   Entering the RESUMING state starts a process of restoring the device state
998  *   and will return a new data_fd. The data stream fed into the data_fd should
999  *   be taken from the data transfer output of a single FD during saving from
1000  *   a compatible device. The migration driver may alter/reset the internal
1001  *   device state for this arc if required to prepare the device to receive the
1002  *   migration data.
1003  *
1004  * STOP_COPY -> PRE_COPY
1005  * STOP_COPY -> PRE_COPY_P2P
1006  *   These arcs are not permitted and return error if requested. Future
1007  *   revisions of this API may define behaviors for these arcs, in this case
1008  *   support will be discoverable by a new flag in
1009  *   VFIO_DEVICE_FEATURE_MIGRATION.
1010  *
1011  * any -> ERROR
1012  *   ERROR cannot be specified as a device state, however any transition request
1013  *   can be failed with an errno return and may then move the device_state into
1014  *   ERROR. In this case the device was unable to execute the requested arc and
1015  *   was also unable to restore the device to any valid device_state.
1016  *   To recover from ERROR VFIO_DEVICE_RESET must be used to return the
1017  *   device_state back to RUNNING.
1018  *
1019  * The optional peer to peer (P2P) quiescent state is intended to be a quiescent
1020  * state for the device for the purposes of managing multiple devices within a
1021  * user context where peer-to-peer DMA between devices may be active. The
1022  * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating
1023  * any new P2P DMA transactions. If the device can identify P2P transactions
1024  * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration
1025  * driver must complete any such outstanding operations prior to completing the
1026  * FSM arc into a P2P state. For the purpose of specification the states
1027  * behave as though the device was fully running if not supported. Like while in
1028  * STOP or STOP_COPY the user must not touch the device, otherwise the state
1029  * can be exited.
1030  *
1031  * The remaining possible transitions are interpreted as combinations of the
1032  * above FSM arcs. As there are multiple paths through the FSM arcs the path
1033  * should be selected based on the following rules:
1034  *   - Select the shortest path.
1035  *   - The path cannot have saving group states as interior arcs, only
1036  *     starting/end states.
1037  * Refer to vfio_mig_get_next_state() for the result of the algorithm.
1038  *
1039  * The automatic transit through the FSM arcs that make up the combination
1040  * transition is invisible to the user. When working with combination arcs the
1041  * user may see any step along the path in the device_state if SET_STATE
1042  * fails. When handling these types of errors users should anticipate future
1043  * revisions of this protocol using new states and those states becoming
1044  * visible in this case.
1045  *
1046  * The optional states cannot be used with SET_STATE if the device does not
1047  * support them. The user can discover if these states are supported by using
1048  * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can
1049  * avoid knowing about these optional states if the kernel driver supports them.
1050  *
1051  * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY
1052  * is not present.
1053  */
1054 enum vfio_device_mig_state {
1055 	VFIO_DEVICE_STATE_ERROR = 0,
1056 	VFIO_DEVICE_STATE_STOP = 1,
1057 	VFIO_DEVICE_STATE_RUNNING = 2,
1058 	VFIO_DEVICE_STATE_STOP_COPY = 3,
1059 	VFIO_DEVICE_STATE_RESUMING = 4,
1060 	VFIO_DEVICE_STATE_RUNNING_P2P = 5,
1061 	VFIO_DEVICE_STATE_PRE_COPY = 6,
1062 	VFIO_DEVICE_STATE_PRE_COPY_P2P = 7,
1063 };
1064 
1065 /**
1066  * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21)
1067  *
1068  * This ioctl is used on the migration data FD in the precopy phase of the
1069  * migration data transfer. It returns an estimate of the current data sizes
1070  * remaining to be transferred. It allows the user to judge when it is
1071  * appropriate to leave PRE_COPY for STOP_COPY.
1072  *
1073  * This ioctl is valid only in PRE_COPY states and kernel driver should
1074  * return -EINVAL from any other migration state.
1075  *
1076  * The vfio_precopy_info data structure returned by this ioctl provides
1077  * estimates of data available from the device during the PRE_COPY states.
1078  * This estimate is split into two categories, initial_bytes and
1079  * dirty_bytes.
1080  *
1081  * The initial_bytes field indicates the amount of initial precopy
1082  * data available from the device. This field should have a non-zero initial
1083  * value and decrease as migration data is read from the device.
1084  * It is recommended to leave PRE_COPY for STOP_COPY only after this field
1085  * reaches zero. Leaving PRE_COPY earlier might make things slower.
1086  *
1087  * The dirty_bytes field tracks device state changes relative to data
1088  * previously retrieved.  This field starts at zero and may increase as
1089  * the internal device state is modified or decrease as that modified
1090  * state is read from the device.
1091  *
1092  * Userspace may use the combination of these fields to estimate the
1093  * potential data size available during the PRE_COPY phases, as well as
1094  * trends relative to the rate the device is dirtying its internal
1095  * state, but these fields are not required to have any bearing relative
1096  * to the data size available during the STOP_COPY phase.
1097  *
1098  * Drivers have a lot of flexibility in when and what they transfer during the
1099  * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO.
1100  *
1101  * During pre-copy the migration data FD has a temporary "end of stream" that is
1102  * reached when both initial_bytes and dirty_byte are zero. For instance, this
1103  * may indicate that the device is idle and not currently dirtying any internal
1104  * state. When read() is done on this temporary end of stream the kernel driver
1105  * should return ENOMSG from read(). Userspace can wait for more data (which may
1106  * never come) by using poll.
1107  *
1108  * Once in STOP_COPY the migration data FD has a permanent end of stream
1109  * signaled in the usual way by read() always returning 0 and poll always
1110  * returning readable. ENOMSG may not be returned in STOP_COPY.
1111  * Support for this ioctl is mandatory if a driver claims to support
1112  * VFIO_MIGRATION_PRE_COPY.
1113  *
1114  * Return: 0 on success, -1 and errno set on failure.
1115  */
1116 struct vfio_precopy_info {
1117 	__u32 argsz;
1118 	__u32 flags;
1119 	__aligned_u64 initial_bytes;
1120 	__aligned_u64 dirty_bytes;
1121 };
1122 
1123 #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21)
1124 
1125 /*
1126  * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power
1127  * state with the platform-based power management.  Device use of lower power
1128  * states depends on factors managed by the runtime power management core,
1129  * including system level support and coordinating support among dependent
1130  * devices.  Enabling device low power entry does not guarantee lower power
1131  * usage by the device, nor is a mechanism provided through this feature to
1132  * know the current power state of the device.  If any device access happens
1133  * (either from the host or through the vfio uAPI) when the device is in the
1134  * low power state, then the host will move the device out of the low power
1135  * state as necessary prior to the access.  Once the access is completed, the
1136  * device may re-enter the low power state.  For single shot low power support
1137  * with wake-up notification, see
1138  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below.  Access to mmap'd
1139  * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after
1140  * calling LOW_POWER_EXIT.
1141  */
1142 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3
1143 
1144 /*
1145  * This device feature has the same behavior as
1146  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user
1147  * provides an eventfd for wake-up notification.  When the device moves out of
1148  * the low power state for the wake-up, the host will not allow the device to
1149  * re-enter a low power state without a subsequent user call to one of the low
1150  * power entry device feature IOCTLs.  Access to mmap'd device regions is
1151  * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the
1152  * low power exit.  The low power exit can happen either through LOW_POWER_EXIT
1153  * or through any other access (where the wake-up notification has been
1154  * generated).  The access to mmap'd device regions will not trigger low power
1155  * exit.
1156  *
1157  * The notification through the provided eventfd will be generated only when
1158  * the device has entered and is resumed from a low power state after
1159  * calling this device feature IOCTL.  A device that has not entered low power
1160  * state, as managed through the runtime power management core, will not
1161  * generate a notification through the provided eventfd on access.  Calling the
1162  * LOW_POWER_EXIT feature is optional in the case where notification has been
1163  * signaled on the provided eventfd that a resume from low power has occurred.
1164  */
1165 struct vfio_device_low_power_entry_with_wakeup {
1166 	__s32 wakeup_eventfd;
1167 	__u32 reserved;
1168 };
1169 
1170 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4
1171 
1172 /*
1173  * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as
1174  * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or
1175  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features.
1176  * This device feature IOCTL may itself generate a wakeup eventfd notification
1177  * in the latter case if the device had previously entered a low power state.
1178  */
1179 #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5
1180 
1181 /*
1182  * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging.
1183  * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports
1184  * DMA logging.
1185  *
1186  * DMA logging allows a device to internally record what DMAs the device is
1187  * initiating and report them back to userspace. It is part of the VFIO
1188  * migration infrastructure that allows implementing dirty page tracking
1189  * during the pre copy phase of live migration. Only DMA WRITEs are logged,
1190  * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE.
1191  *
1192  * When DMA logging is started a range of IOVAs to monitor is provided and the
1193  * device can optimize its logging to cover only the IOVA range given. Each
1194  * DMA that the device initiates inside the range will be logged by the device
1195  * for later retrieval.
1196  *
1197  * page_size is an input that hints what tracking granularity the device
1198  * should try to achieve. If the device cannot do the hinted page size then
1199  * it's the driver choice which page size to pick based on its support.
1200  * On output the device will return the page size it selected.
1201  *
1202  * ranges is a pointer to an array of
1203  * struct vfio_device_feature_dma_logging_range.
1204  *
1205  * The core kernel code guarantees to support by minimum num_ranges that fit
1206  * into a single kernel page. User space can try higher values but should give
1207  * up if the above can't be achieved as of some driver limitations.
1208  *
1209  * A single call to start device DMA logging can be issued and a matching stop
1210  * should follow at the end. Another start is not allowed in the meantime.
1211  */
1212 struct vfio_device_feature_dma_logging_control {
1213 	__aligned_u64 page_size;
1214 	__u32 num_ranges;
1215 	__u32 __reserved;
1216 	__aligned_u64 ranges;
1217 };
1218 
1219 struct vfio_device_feature_dma_logging_range {
1220 	__aligned_u64 iova;
1221 	__aligned_u64 length;
1222 };
1223 
1224 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6
1225 
1226 /*
1227  * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started
1228  * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START
1229  */
1230 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7
1231 
1232 /*
1233  * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log
1234  *
1235  * Query the device's DMA log for written pages within the given IOVA range.
1236  * During querying the log is cleared for the IOVA range.
1237  *
1238  * bitmap is a pointer to an array of u64s that will hold the output bitmap
1239  * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits
1240  * is given by:
1241  *  bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64))
1242  *
1243  * The input page_size can be any power of two value and does not have to
1244  * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver
1245  * will format its internal logging to match the reporting page size, possibly
1246  * by replicating bits if the internal page size is lower than requested.
1247  *
1248  * The LOGGING_REPORT will only set bits in the bitmap and never clear or
1249  * perform any initialization of the user provided bitmap.
1250  *
1251  * If any error is returned userspace should assume that the dirty log is
1252  * corrupted. Error recovery is to consider all memory dirty and try to
1253  * restart the dirty tracking, or to abort/restart the whole migration.
1254  *
1255  * If DMA logging is not enabled, an error will be returned.
1256  *
1257  */
1258 struct vfio_device_feature_dma_logging_report {
1259 	__aligned_u64 iova;
1260 	__aligned_u64 length;
1261 	__aligned_u64 page_size;
1262 	__aligned_u64 bitmap;
1263 };
1264 
1265 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8
1266 
1267 /*
1268  * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will
1269  * be required to complete stop copy.
1270  *
1271  * Note: Can be called on each device state.
1272  */
1273 
1274 struct vfio_device_feature_mig_data_size {
1275 	__aligned_u64 stop_copy_length;
1276 };
1277 
1278 #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9
1279 
1280 /* -------- API for Type1 VFIO IOMMU -------- */
1281 
1282 /**
1283  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
1284  *
1285  * Retrieve information about the IOMMU object. Fills in provided
1286  * struct vfio_iommu_info. Caller sets argsz.
1287  *
1288  * XXX Should we do these by CHECK_EXTENSION too?
1289  */
1290 struct vfio_iommu_type1_info {
1291 	__u32	argsz;
1292 	__u32	flags;
1293 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
1294 #define VFIO_IOMMU_INFO_CAPS	(1 << 1)	/* Info supports caps */
1295 	__u64	iova_pgsizes;	/* Bitmap of supported page sizes */
1296 	__u32   cap_offset;	/* Offset within info struct of first cap */
1297 };
1298 
1299 /*
1300  * The IOVA capability allows to report the valid IOVA range(s)
1301  * excluding any non-relaxable reserved regions exposed by
1302  * devices attached to the container. Any DMA map attempt
1303  * outside the valid iova range will return error.
1304  *
1305  * The structures below define version 1 of this capability.
1306  */
1307 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE  1
1308 
1309 struct vfio_iova_range {
1310 	__u64	start;
1311 	__u64	end;
1312 };
1313 
1314 struct vfio_iommu_type1_info_cap_iova_range {
1315 	struct	vfio_info_cap_header header;
1316 	__u32	nr_iovas;
1317 	__u32	reserved;
1318 	struct	vfio_iova_range iova_ranges[];
1319 };
1320 
1321 /*
1322  * The migration capability allows to report supported features for migration.
1323  *
1324  * The structures below define version 1 of this capability.
1325  *
1326  * The existence of this capability indicates that IOMMU kernel driver supports
1327  * dirty page logging.
1328  *
1329  * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty
1330  * page logging.
1331  * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap
1332  * size in bytes that can be used by user applications when getting the dirty
1333  * bitmap.
1334  */
1335 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION  2
1336 
1337 struct vfio_iommu_type1_info_cap_migration {
1338 	struct	vfio_info_cap_header header;
1339 	__u32	flags;
1340 	__u64	pgsize_bitmap;
1341 	__u64	max_dirty_bitmap_size;		/* in bytes */
1342 };
1343 
1344 /*
1345  * The DMA available capability allows to report the current number of
1346  * simultaneously outstanding DMA mappings that are allowed.
1347  *
1348  * The structure below defines version 1 of this capability.
1349  *
1350  * avail: specifies the current number of outstanding DMA mappings allowed.
1351  */
1352 #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3
1353 
1354 struct vfio_iommu_type1_info_dma_avail {
1355 	struct	vfio_info_cap_header header;
1356 	__u32	avail;
1357 };
1358 
1359 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
1360 
1361 /**
1362  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
1363  *
1364  * Map process virtual addresses to IO virtual addresses using the
1365  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
1366  *
1367  * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr
1368  * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR.  To
1369  * maintain memory consistency within the user application, the updated vaddr
1370  * must address the same memory object as originally mapped.  Failure to do so
1371  * will result in user memory corruption and/or device misbehavior.  iova and
1372  * size must match those in the original MAP_DMA call.  Protection is not
1373  * changed, and the READ & WRITE flags must be 0.
1374  */
1375 struct vfio_iommu_type1_dma_map {
1376 	__u32	argsz;
1377 	__u32	flags;
1378 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
1379 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
1380 #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2)
1381 	__u64	vaddr;				/* Process virtual address */
1382 	__u64	iova;				/* IO virtual address */
1383 	__u64	size;				/* Size of mapping (bytes) */
1384 };
1385 
1386 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
1387 
1388 struct vfio_bitmap {
1389 	__u64        pgsize;	/* page size for bitmap in bytes */
1390 	__u64        size;	/* in bytes */
1391 	__u64 __user *data;	/* one bit per page */
1392 };
1393 
1394 /**
1395  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
1396  *							struct vfio_dma_unmap)
1397  *
1398  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
1399  * Caller sets argsz.  The actual unmapped size is returned in the size
1400  * field.  No guarantee is made to the user that arbitrary unmaps of iova
1401  * or size different from those used in the original mapping call will
1402  * succeed.
1403  *
1404  * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap
1405  * before unmapping IO virtual addresses. When this flag is set, the user must
1406  * provide a struct vfio_bitmap in data[]. User must provide zero-allocated
1407  * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field.
1408  * A bit in the bitmap represents one page, of user provided page size in
1409  * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set
1410  * indicates that the page at that offset from iova is dirty. A Bitmap of the
1411  * pages in the range of unmapped size is returned in the user-provided
1412  * vfio_bitmap.data.
1413  *
1414  * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses.  iova and size
1415  * must be 0.  This cannot be combined with the get-dirty-bitmap flag.
1416  *
1417  * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host
1418  * virtual addresses in the iova range.  DMA to already-mapped pages continues.
1419  * Groups may not be added to the container while any addresses are invalid.
1420  * This cannot be combined with the get-dirty-bitmap flag.
1421  */
1422 struct vfio_iommu_type1_dma_unmap {
1423 	__u32	argsz;
1424 	__u32	flags;
1425 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0)
1426 #define VFIO_DMA_UNMAP_FLAG_ALL		     (1 << 1)
1427 #define VFIO_DMA_UNMAP_FLAG_VADDR	     (1 << 2)
1428 	__u64	iova;				/* IO virtual address */
1429 	__u64	size;				/* Size of mapping (bytes) */
1430 	__u8    data[];
1431 };
1432 
1433 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
1434 
1435 /*
1436  * IOCTLs to enable/disable IOMMU container usage.
1437  * No parameters are supported.
1438  */
1439 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
1440 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
1441 
1442 /**
1443  * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
1444  *                                     struct vfio_iommu_type1_dirty_bitmap)
1445  * IOCTL is used for dirty pages logging.
1446  * Caller should set flag depending on which operation to perform, details as
1447  * below:
1448  *
1449  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs
1450  * the IOMMU driver to log pages that are dirtied or potentially dirtied by
1451  * the device; designed to be used when a migration is in progress. Dirty pages
1452  * are logged until logging is disabled by user application by calling the IOCTL
1453  * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag.
1454  *
1455  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs
1456  * the IOMMU driver to stop logging dirtied pages.
1457  *
1458  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set
1459  * returns the dirty pages bitmap for IOMMU container for a given IOVA range.
1460  * The user must specify the IOVA range and the pgsize through the structure
1461  * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface
1462  * supports getting a bitmap of the smallest supported pgsize only and can be
1463  * modified in future to get a bitmap of any specified supported pgsize. The
1464  * user must provide a zeroed memory area for the bitmap memory and specify its
1465  * size in bitmap.size. One bit is used to represent one page consecutively
1466  * starting from iova offset. The user should provide page size in bitmap.pgsize
1467  * field. A bit set in the bitmap indicates that the page at that offset from
1468  * iova is dirty. The caller must set argsz to a value including the size of
1469  * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the
1470  * actual bitmap. If dirty pages logging is not enabled, an error will be
1471  * returned.
1472  *
1473  * Only one of the flags _START, _STOP and _GET may be specified at a time.
1474  *
1475  */
1476 struct vfio_iommu_type1_dirty_bitmap {
1477 	__u32        argsz;
1478 	__u32        flags;
1479 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START	(1 << 0)
1480 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP	(1 << 1)
1481 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP	(1 << 2)
1482 	__u8         data[];
1483 };
1484 
1485 struct vfio_iommu_type1_dirty_bitmap_get {
1486 	__u64              iova;	/* IO virtual address */
1487 	__u64              size;	/* Size of iova range */
1488 	struct vfio_bitmap bitmap;
1489 };
1490 
1491 #define VFIO_IOMMU_DIRTY_PAGES             _IO(VFIO_TYPE, VFIO_BASE + 17)
1492 
1493 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
1494 
1495 /*
1496  * The SPAPR TCE DDW info struct provides the information about
1497  * the details of Dynamic DMA window capability.
1498  *
1499  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
1500  * @max_dynamic_windows_supported tells the maximum number of windows
1501  * which the platform can create.
1502  * @levels tells the maximum number of levels in multi-level IOMMU tables;
1503  * this allows splitting a table into smaller chunks which reduces
1504  * the amount of physically contiguous memory required for the table.
1505  */
1506 struct vfio_iommu_spapr_tce_ddw_info {
1507 	__u64 pgsizes;			/* Bitmap of supported page sizes */
1508 	__u32 max_dynamic_windows_supported;
1509 	__u32 levels;
1510 };
1511 
1512 /*
1513  * The SPAPR TCE info struct provides the information about the PCI bus
1514  * address ranges available for DMA, these values are programmed into
1515  * the hardware so the guest has to know that information.
1516  *
1517  * The DMA 32 bit window start is an absolute PCI bus address.
1518  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
1519  * addresses too so the window works as a filter rather than an offset
1520  * for IOVA addresses.
1521  *
1522  * Flags supported:
1523  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
1524  *   (DDW) support is present. @ddw is only supported when DDW is present.
1525  */
1526 struct vfio_iommu_spapr_tce_info {
1527 	__u32 argsz;
1528 	__u32 flags;
1529 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
1530 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
1531 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
1532 	struct vfio_iommu_spapr_tce_ddw_info ddw;
1533 };
1534 
1535 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
1536 
1537 /*
1538  * EEH PE operation struct provides ways to:
1539  * - enable/disable EEH functionality;
1540  * - unfreeze IO/DMA for frozen PE;
1541  * - read PE state;
1542  * - reset PE;
1543  * - configure PE;
1544  * - inject EEH error.
1545  */
1546 struct vfio_eeh_pe_err {
1547 	__u32 type;
1548 	__u32 func;
1549 	__u64 addr;
1550 	__u64 mask;
1551 };
1552 
1553 struct vfio_eeh_pe_op {
1554 	__u32 argsz;
1555 	__u32 flags;
1556 	__u32 op;
1557 	union {
1558 		struct vfio_eeh_pe_err err;
1559 	};
1560 };
1561 
1562 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
1563 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
1564 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
1565 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
1566 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
1567 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
1568 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
1569 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
1570 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
1571 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
1572 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
1573 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
1574 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
1575 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
1576 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
1577 
1578 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
1579 
1580 /**
1581  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
1582  *
1583  * Registers user space memory where DMA is allowed. It pins
1584  * user pages and does the locked memory accounting so
1585  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
1586  * get faster.
1587  */
1588 struct vfio_iommu_spapr_register_memory {
1589 	__u32	argsz;
1590 	__u32	flags;
1591 	__u64	vaddr;				/* Process virtual address */
1592 	__u64	size;				/* Size of mapping (bytes) */
1593 };
1594 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
1595 
1596 /**
1597  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
1598  *
1599  * Unregisters user space memory registered with
1600  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
1601  * Uses vfio_iommu_spapr_register_memory for parameters.
1602  */
1603 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
1604 
1605 /**
1606  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
1607  *
1608  * Creates an additional TCE table and programs it (sets a new DMA window)
1609  * to every IOMMU group in the container. It receives page shift, window
1610  * size and number of levels in the TCE table being created.
1611  *
1612  * It allocates and returns an offset on a PCI bus of the new DMA window.
1613  */
1614 struct vfio_iommu_spapr_tce_create {
1615 	__u32 argsz;
1616 	__u32 flags;
1617 	/* in */
1618 	__u32 page_shift;
1619 	__u32 __resv1;
1620 	__u64 window_size;
1621 	__u32 levels;
1622 	__u32 __resv2;
1623 	/* out */
1624 	__u64 start_addr;
1625 };
1626 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
1627 
1628 /**
1629  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
1630  *
1631  * Unprograms a TCE table from all groups in the container and destroys it.
1632  * It receives a PCI bus offset as a window id.
1633  */
1634 struct vfio_iommu_spapr_tce_remove {
1635 	__u32 argsz;
1636 	__u32 flags;
1637 	/* in */
1638 	__u64 start_addr;
1639 };
1640 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
1641 
1642 /* ***************************************************************** */
1643 
1644 #endif /* _UAPIVFIO_H */
1645