xref: /linux-6.15/include/uapi/linux/vfio.h (revision 9062ff40)
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  * The vfio-ap bus driver makes use of the following IRQ index mapping.
669  * Unimplemented IRQ types return a count of zero.
670  */
671 enum {
672 	VFIO_AP_REQ_IRQ_INDEX,
673 	VFIO_AP_NUM_IRQS
674 };
675 
676 /**
677  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12,
678  *					      struct vfio_pci_hot_reset_info)
679  *
680  * This command is used to query the affected devices in the hot reset for
681  * a given device.
682  *
683  * This command always reports the segment, bus, and devfn information for
684  * each affected device, and selectively reports the group_id or devid per
685  * the way how the calling device is opened.
686  *
687  *	- If the calling device is opened via the traditional group/container
688  *	  API, group_id is reported.  User should check if it has owned all
689  *	  the affected devices and provides a set of group fds to prove the
690  *	  ownership in VFIO_DEVICE_PCI_HOT_RESET ioctl.
691  *
692  *	- If the calling device is opened as a cdev, devid is reported.
693  *	  Flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set to indicate this
694  *	  data type.  All the affected devices should be represented in
695  *	  the dev_set, ex. bound to a vfio driver, and also be owned by
696  *	  this interface which is determined by the following conditions:
697  *	  1) Has a valid devid within the iommufd_ctx of the calling device.
698  *	     Ownership cannot be determined across separate iommufd_ctx and
699  *	     the cdev calling conventions do not support a proof-of-ownership
700  *	     model as provided in the legacy group interface.  In this case
701  *	     valid devid with value greater than zero is provided in the return
702  *	     structure.
703  *	  2) Does not have a valid devid within the iommufd_ctx of the calling
704  *	     device, but belongs to the same IOMMU group as the calling device
705  *	     or another opened device that has a valid devid within the
706  *	     iommufd_ctx of the calling device.  This provides implicit ownership
707  *	     for devices within the same DMA isolation context.  In this case
708  *	     the devid value of VFIO_PCI_DEVID_OWNED is provided in the return
709  *	     structure.
710  *
711  *	  A devid value of VFIO_PCI_DEVID_NOT_OWNED is provided in the return
712  *	  structure for affected devices where device is NOT represented in the
713  *	  dev_set or ownership is not available.  Such devices prevent the use
714  *	  of VFIO_DEVICE_PCI_HOT_RESET ioctl outside of the proof-of-ownership
715  *	  calling conventions (ie. via legacy group accessed devices).  Flag
716  *	  VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED would be set when all the
717  *	  affected devices are represented in the dev_set and also owned by
718  *	  the user.  This flag is available only when
719  *	  flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set, otherwise reserved.
720  *
721  * Return: 0 on success, -errno on failure:
722  *	-enospc = insufficient buffer, -enodev = unsupported for device.
723  */
724 struct vfio_pci_dependent_device {
725 	union {
726 		__u32   group_id;
727 		__u32	devid;
728 #define VFIO_PCI_DEVID_OWNED		0
729 #define VFIO_PCI_DEVID_NOT_OWNED	-1
730 	};
731 	__u16	segment;
732 	__u8	bus;
733 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
734 };
735 
736 struct vfio_pci_hot_reset_info {
737 	__u32	argsz;
738 	__u32	flags;
739 #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID		(1 << 0)
740 #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED	(1 << 1)
741 	__u32	count;
742 	struct vfio_pci_dependent_device	devices[];
743 };
744 
745 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
746 
747 /**
748  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
749  *				    struct vfio_pci_hot_reset)
750  *
751  * Return: 0 on success, -errno on failure.
752  */
753 struct vfio_pci_hot_reset {
754 	__u32	argsz;
755 	__u32	flags;
756 	__u32	count;
757 	__s32	group_fds[];
758 };
759 
760 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
761 
762 /**
763  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
764  *                                    struct vfio_device_query_gfx_plane)
765  *
766  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
767  *
768  * flags supported:
769  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
770  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
771  *   support for dma-buf.
772  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
773  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
774  *   support for region.
775  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
776  *   with each call to query the plane info.
777  * - Others are invalid and return -EINVAL.
778  *
779  * Note:
780  * 1. Plane could be disabled by guest. In that case, success will be
781  *    returned with zero-initialized drm_format, size, width and height
782  *    fields.
783  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
784  *
785  * Return: 0 on success, -errno on other failure.
786  */
787 struct vfio_device_gfx_plane_info {
788 	__u32 argsz;
789 	__u32 flags;
790 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
791 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
792 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
793 	/* in */
794 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
795 	/* out */
796 	__u32 drm_format;	/* drm format of plane */
797 	__u64 drm_format_mod;   /* tiled mode */
798 	__u32 width;	/* width of plane */
799 	__u32 height;	/* height of plane */
800 	__u32 stride;	/* stride of plane */
801 	__u32 size;	/* size of plane in bytes, align on page*/
802 	__u32 x_pos;	/* horizontal position of cursor plane */
803 	__u32 y_pos;	/* vertical position of cursor plane*/
804 	__u32 x_hot;    /* horizontal position of cursor hotspot */
805 	__u32 y_hot;    /* vertical position of cursor hotspot */
806 	union {
807 		__u32 region_index;	/* region index */
808 		__u32 dmabuf_id;	/* dma-buf id */
809 	};
810 };
811 
812 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
813 
814 /**
815  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
816  *
817  * Return a new dma-buf file descriptor for an exposed guest framebuffer
818  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
819  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
820  */
821 
822 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
823 
824 /**
825  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
826  *                              struct vfio_device_ioeventfd)
827  *
828  * Perform a write to the device at the specified device fd offset, with
829  * the specified data and width when the provided eventfd is triggered.
830  * vfio bus drivers may not support this for all regions, for all widths,
831  * or at all.  vfio-pci currently only enables support for BAR regions,
832  * excluding the MSI-X vector table.
833  *
834  * Return: 0 on success, -errno on failure.
835  */
836 struct vfio_device_ioeventfd {
837 	__u32	argsz;
838 	__u32	flags;
839 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
840 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
841 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
842 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
843 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
844 	__u64	offset;			/* device fd offset of write */
845 	__u64	data;			/* data to be written */
846 	__s32	fd;			/* -1 for de-assignment */
847 };
848 
849 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
850 
851 /**
852  * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
853  *			       struct vfio_device_feature)
854  *
855  * Get, set, or probe feature data of the device.  The feature is selected
856  * using the FEATURE_MASK portion of the flags field.  Support for a feature
857  * can be probed by setting both the FEATURE_MASK and PROBE bits.  A probe
858  * may optionally include the GET and/or SET bits to determine read vs write
859  * access of the feature respectively.  Probing a feature will return success
860  * if the feature is supported and all of the optionally indicated GET/SET
861  * methods are supported.  The format of the data portion of the structure is
862  * specific to the given feature.  The data portion is not required for
863  * probing.  GET and SET are mutually exclusive, except for use with PROBE.
864  *
865  * Return 0 on success, -errno on failure.
866  */
867 struct vfio_device_feature {
868 	__u32	argsz;
869 	__u32	flags;
870 #define VFIO_DEVICE_FEATURE_MASK	(0xffff) /* 16-bit feature index */
871 #define VFIO_DEVICE_FEATURE_GET		(1 << 16) /* Get feature into data[] */
872 #define VFIO_DEVICE_FEATURE_SET		(1 << 17) /* Set feature from data[] */
873 #define VFIO_DEVICE_FEATURE_PROBE	(1 << 18) /* Probe feature support */
874 	__u8	data[];
875 };
876 
877 #define VFIO_DEVICE_FEATURE		_IO(VFIO_TYPE, VFIO_BASE + 17)
878 
879 /*
880  * Provide support for setting a PCI VF Token, which is used as a shared
881  * secret between PF and VF drivers.  This feature may only be set on a
882  * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
883  * open VFs.  Data provided when setting this feature is a 16-byte array
884  * (__u8 b[16]), representing a UUID.
885  */
886 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN	(0)
887 
888 /*
889  * Indicates the device can support the migration API through
890  * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and
891  * ERROR states are always supported. Support for additional states is
892  * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be
893  * set.
894  *
895  * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and
896  * RESUMING are supported.
897  *
898  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P
899  * is supported in addition to the STOP_COPY states.
900  *
901  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that
902  * PRE_COPY is supported in addition to the STOP_COPY states.
903  *
904  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY
905  * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported
906  * in addition to the STOP_COPY states.
907  *
908  * Other combinations of flags have behavior to be defined in the future.
909  */
910 struct vfio_device_feature_migration {
911 	__aligned_u64 flags;
912 #define VFIO_MIGRATION_STOP_COPY	(1 << 0)
913 #define VFIO_MIGRATION_P2P		(1 << 1)
914 #define VFIO_MIGRATION_PRE_COPY		(1 << 2)
915 };
916 #define VFIO_DEVICE_FEATURE_MIGRATION 1
917 
918 /*
919  * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO
920  * device. The new state is supplied in device_state, see enum
921  * vfio_device_mig_state for details
922  *
923  * The kernel migration driver must fully transition the device to the new state
924  * value before the operation returns to the user.
925  *
926  * The kernel migration driver must not generate asynchronous device state
927  * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET
928  * ioctl as described above.
929  *
930  * If this function fails then current device_state may be the original
931  * operating state or some other state along the combination transition path.
932  * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt
933  * to return to the original state, or attempt to return to some other state
934  * such as RUNNING or STOP.
935  *
936  * If the new_state starts a new data transfer session then the FD associated
937  * with that session is returned in data_fd. The user is responsible to close
938  * this FD when it is finished. The user must consider the migration data stream
939  * carried over the FD to be opaque and must preserve the byte order of the
940  * stream. The user is not required to preserve buffer segmentation when writing
941  * the data stream during the RESUMING operation.
942  *
943  * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO
944  * device, data_fd will be -1.
945  */
946 struct vfio_device_feature_mig_state {
947 	__u32 device_state; /* From enum vfio_device_mig_state */
948 	__s32 data_fd;
949 };
950 #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2
951 
952 /*
953  * The device migration Finite State Machine is described by the enum
954  * vfio_device_mig_state. Some of the FSM arcs will create a migration data
955  * transfer session by returning a FD, in this case the migration data will
956  * flow over the FD using read() and write() as discussed below.
957  *
958  * There are 5 states to support VFIO_MIGRATION_STOP_COPY:
959  *  RUNNING - The device is running normally
960  *  STOP - The device does not change the internal or external state
961  *  STOP_COPY - The device internal state can be read out
962  *  RESUMING - The device is stopped and is loading a new internal state
963  *  ERROR - The device has failed and must be reset
964  *
965  * And optional states to support VFIO_MIGRATION_P2P:
966  *  RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA
967  * And VFIO_MIGRATION_PRE_COPY:
968  *  PRE_COPY - The device is running normally but tracking internal state
969  *             changes
970  * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY:
971  *  PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA
972  *
973  * The FSM takes actions on the arcs between FSM states. The driver implements
974  * the following behavior for the FSM arcs:
975  *
976  * RUNNING_P2P -> STOP
977  * STOP_COPY -> STOP
978  *   While in STOP the device must stop the operation of the device. The device
979  *   must not generate interrupts, DMA, or any other change to external state.
980  *   It must not change its internal state. When stopped the device and kernel
981  *   migration driver must accept and respond to interaction to support external
982  *   subsystems in the STOP state, for example PCI MSI-X and PCI config space.
983  *   Failure by the user to restrict device access while in STOP must not result
984  *   in error conditions outside the user context (ex. host system faults).
985  *
986  *   The STOP_COPY arc will terminate a data transfer session.
987  *
988  * RESUMING -> STOP
989  *   Leaving RESUMING terminates a data transfer session and indicates the
990  *   device should complete processing of the data delivered by write(). The
991  *   kernel migration driver should complete the incorporation of data written
992  *   to the data transfer FD into the device internal state and perform
993  *   final validity and consistency checking of the new device state. If the
994  *   user provided data is found to be incomplete, inconsistent, or otherwise
995  *   invalid, the migration driver must fail the SET_STATE ioctl and
996  *   optionally go to the ERROR state as described below.
997  *
998  *   While in STOP the device has the same behavior as other STOP states
999  *   described above.
1000  *
1001  *   To abort a RESUMING session the device must be reset.
1002  *
1003  * PRE_COPY -> RUNNING
1004  * RUNNING_P2P -> RUNNING
1005  *   While in RUNNING the device is fully operational, the device may generate
1006  *   interrupts, DMA, respond to MMIO, all vfio device regions are functional,
1007  *   and the device may advance its internal state.
1008  *
1009  *   The PRE_COPY arc will terminate a data transfer session.
1010  *
1011  * PRE_COPY_P2P -> RUNNING_P2P
1012  * RUNNING -> RUNNING_P2P
1013  * STOP -> RUNNING_P2P
1014  *   While in RUNNING_P2P the device is partially running in the P2P quiescent
1015  *   state defined below.
1016  *
1017  *   The PRE_COPY_P2P arc will terminate a data transfer session.
1018  *
1019  * RUNNING -> PRE_COPY
1020  * RUNNING_P2P -> PRE_COPY_P2P
1021  * STOP -> STOP_COPY
1022  *   PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states
1023  *   which share a data transfer session. Moving between these states alters
1024  *   what is streamed in session, but does not terminate or otherwise affect
1025  *   the associated fd.
1026  *
1027  *   These arcs begin the process of saving the device state and will return a
1028  *   new data_fd. The migration driver may perform actions such as enabling
1029  *   dirty logging of device state when entering PRE_COPY or PER_COPY_P2P.
1030  *
1031  *   Each arc does not change the device operation, the device remains
1032  *   RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below
1033  *   in PRE_COPY_P2P -> STOP_COPY.
1034  *
1035  * PRE_COPY -> PRE_COPY_P2P
1036  *   Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above.
1037  *   However, while in the PRE_COPY_P2P state, the device is partially running
1038  *   in the P2P quiescent state defined below, like RUNNING_P2P.
1039  *
1040  * PRE_COPY_P2P -> PRE_COPY
1041  *   This arc allows returning the device to a full RUNNING behavior while
1042  *   continuing all the behaviors of PRE_COPY.
1043  *
1044  * PRE_COPY_P2P -> STOP_COPY
1045  *   While in the STOP_COPY state the device has the same behavior as STOP
1046  *   with the addition that the data transfers session continues to stream the
1047  *   migration state. End of stream on the FD indicates the entire device
1048  *   state has been transferred.
1049  *
1050  *   The user should take steps to restrict access to vfio device regions while
1051  *   the device is in STOP_COPY or risk corruption of the device migration data
1052  *   stream.
1053  *
1054  * STOP -> RESUMING
1055  *   Entering the RESUMING state starts a process of restoring the device state
1056  *   and will return a new data_fd. The data stream fed into the data_fd should
1057  *   be taken from the data transfer output of a single FD during saving from
1058  *   a compatible device. The migration driver may alter/reset the internal
1059  *   device state for this arc if required to prepare the device to receive the
1060  *   migration data.
1061  *
1062  * STOP_COPY -> PRE_COPY
1063  * STOP_COPY -> PRE_COPY_P2P
1064  *   These arcs are not permitted and return error if requested. Future
1065  *   revisions of this API may define behaviors for these arcs, in this case
1066  *   support will be discoverable by a new flag in
1067  *   VFIO_DEVICE_FEATURE_MIGRATION.
1068  *
1069  * any -> ERROR
1070  *   ERROR cannot be specified as a device state, however any transition request
1071  *   can be failed with an errno return and may then move the device_state into
1072  *   ERROR. In this case the device was unable to execute the requested arc and
1073  *   was also unable to restore the device to any valid device_state.
1074  *   To recover from ERROR VFIO_DEVICE_RESET must be used to return the
1075  *   device_state back to RUNNING.
1076  *
1077  * The optional peer to peer (P2P) quiescent state is intended to be a quiescent
1078  * state for the device for the purposes of managing multiple devices within a
1079  * user context where peer-to-peer DMA between devices may be active. The
1080  * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating
1081  * any new P2P DMA transactions. If the device can identify P2P transactions
1082  * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration
1083  * driver must complete any such outstanding operations prior to completing the
1084  * FSM arc into a P2P state. For the purpose of specification the states
1085  * behave as though the device was fully running if not supported. Like while in
1086  * STOP or STOP_COPY the user must not touch the device, otherwise the state
1087  * can be exited.
1088  *
1089  * The remaining possible transitions are interpreted as combinations of the
1090  * above FSM arcs. As there are multiple paths through the FSM arcs the path
1091  * should be selected based on the following rules:
1092  *   - Select the shortest path.
1093  *   - The path cannot have saving group states as interior arcs, only
1094  *     starting/end states.
1095  * Refer to vfio_mig_get_next_state() for the result of the algorithm.
1096  *
1097  * The automatic transit through the FSM arcs that make up the combination
1098  * transition is invisible to the user. When working with combination arcs the
1099  * user may see any step along the path in the device_state if SET_STATE
1100  * fails. When handling these types of errors users should anticipate future
1101  * revisions of this protocol using new states and those states becoming
1102  * visible in this case.
1103  *
1104  * The optional states cannot be used with SET_STATE if the device does not
1105  * support them. The user can discover if these states are supported by using
1106  * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can
1107  * avoid knowing about these optional states if the kernel driver supports them.
1108  *
1109  * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY
1110  * is not present.
1111  */
1112 enum vfio_device_mig_state {
1113 	VFIO_DEVICE_STATE_ERROR = 0,
1114 	VFIO_DEVICE_STATE_STOP = 1,
1115 	VFIO_DEVICE_STATE_RUNNING = 2,
1116 	VFIO_DEVICE_STATE_STOP_COPY = 3,
1117 	VFIO_DEVICE_STATE_RESUMING = 4,
1118 	VFIO_DEVICE_STATE_RUNNING_P2P = 5,
1119 	VFIO_DEVICE_STATE_PRE_COPY = 6,
1120 	VFIO_DEVICE_STATE_PRE_COPY_P2P = 7,
1121 };
1122 
1123 /**
1124  * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21)
1125  *
1126  * This ioctl is used on the migration data FD in the precopy phase of the
1127  * migration data transfer. It returns an estimate of the current data sizes
1128  * remaining to be transferred. It allows the user to judge when it is
1129  * appropriate to leave PRE_COPY for STOP_COPY.
1130  *
1131  * This ioctl is valid only in PRE_COPY states and kernel driver should
1132  * return -EINVAL from any other migration state.
1133  *
1134  * The vfio_precopy_info data structure returned by this ioctl provides
1135  * estimates of data available from the device during the PRE_COPY states.
1136  * This estimate is split into two categories, initial_bytes and
1137  * dirty_bytes.
1138  *
1139  * The initial_bytes field indicates the amount of initial precopy
1140  * data available from the device. This field should have a non-zero initial
1141  * value and decrease as migration data is read from the device.
1142  * It is recommended to leave PRE_COPY for STOP_COPY only after this field
1143  * reaches zero. Leaving PRE_COPY earlier might make things slower.
1144  *
1145  * The dirty_bytes field tracks device state changes relative to data
1146  * previously retrieved.  This field starts at zero and may increase as
1147  * the internal device state is modified or decrease as that modified
1148  * state is read from the device.
1149  *
1150  * Userspace may use the combination of these fields to estimate the
1151  * potential data size available during the PRE_COPY phases, as well as
1152  * trends relative to the rate the device is dirtying its internal
1153  * state, but these fields are not required to have any bearing relative
1154  * to the data size available during the STOP_COPY phase.
1155  *
1156  * Drivers have a lot of flexibility in when and what they transfer during the
1157  * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO.
1158  *
1159  * During pre-copy the migration data FD has a temporary "end of stream" that is
1160  * reached when both initial_bytes and dirty_byte are zero. For instance, this
1161  * may indicate that the device is idle and not currently dirtying any internal
1162  * state. When read() is done on this temporary end of stream the kernel driver
1163  * should return ENOMSG from read(). Userspace can wait for more data (which may
1164  * never come) by using poll.
1165  *
1166  * Once in STOP_COPY the migration data FD has a permanent end of stream
1167  * signaled in the usual way by read() always returning 0 and poll always
1168  * returning readable. ENOMSG may not be returned in STOP_COPY.
1169  * Support for this ioctl is mandatory if a driver claims to support
1170  * VFIO_MIGRATION_PRE_COPY.
1171  *
1172  * Return: 0 on success, -1 and errno set on failure.
1173  */
1174 struct vfio_precopy_info {
1175 	__u32 argsz;
1176 	__u32 flags;
1177 	__aligned_u64 initial_bytes;
1178 	__aligned_u64 dirty_bytes;
1179 };
1180 
1181 #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21)
1182 
1183 /*
1184  * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power
1185  * state with the platform-based power management.  Device use of lower power
1186  * states depends on factors managed by the runtime power management core,
1187  * including system level support and coordinating support among dependent
1188  * devices.  Enabling device low power entry does not guarantee lower power
1189  * usage by the device, nor is a mechanism provided through this feature to
1190  * know the current power state of the device.  If any device access happens
1191  * (either from the host or through the vfio uAPI) when the device is in the
1192  * low power state, then the host will move the device out of the low power
1193  * state as necessary prior to the access.  Once the access is completed, the
1194  * device may re-enter the low power state.  For single shot low power support
1195  * with wake-up notification, see
1196  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below.  Access to mmap'd
1197  * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after
1198  * calling LOW_POWER_EXIT.
1199  */
1200 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3
1201 
1202 /*
1203  * This device feature has the same behavior as
1204  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user
1205  * provides an eventfd for wake-up notification.  When the device moves out of
1206  * the low power state for the wake-up, the host will not allow the device to
1207  * re-enter a low power state without a subsequent user call to one of the low
1208  * power entry device feature IOCTLs.  Access to mmap'd device regions is
1209  * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the
1210  * low power exit.  The low power exit can happen either through LOW_POWER_EXIT
1211  * or through any other access (where the wake-up notification has been
1212  * generated).  The access to mmap'd device regions will not trigger low power
1213  * exit.
1214  *
1215  * The notification through the provided eventfd will be generated only when
1216  * the device has entered and is resumed from a low power state after
1217  * calling this device feature IOCTL.  A device that has not entered low power
1218  * state, as managed through the runtime power management core, will not
1219  * generate a notification through the provided eventfd on access.  Calling the
1220  * LOW_POWER_EXIT feature is optional in the case where notification has been
1221  * signaled on the provided eventfd that a resume from low power has occurred.
1222  */
1223 struct vfio_device_low_power_entry_with_wakeup {
1224 	__s32 wakeup_eventfd;
1225 	__u32 reserved;
1226 };
1227 
1228 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4
1229 
1230 /*
1231  * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as
1232  * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or
1233  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features.
1234  * This device feature IOCTL may itself generate a wakeup eventfd notification
1235  * in the latter case if the device had previously entered a low power state.
1236  */
1237 #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5
1238 
1239 /*
1240  * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging.
1241  * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports
1242  * DMA logging.
1243  *
1244  * DMA logging allows a device to internally record what DMAs the device is
1245  * initiating and report them back to userspace. It is part of the VFIO
1246  * migration infrastructure that allows implementing dirty page tracking
1247  * during the pre copy phase of live migration. Only DMA WRITEs are logged,
1248  * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE.
1249  *
1250  * When DMA logging is started a range of IOVAs to monitor is provided and the
1251  * device can optimize its logging to cover only the IOVA range given. Each
1252  * DMA that the device initiates inside the range will be logged by the device
1253  * for later retrieval.
1254  *
1255  * page_size is an input that hints what tracking granularity the device
1256  * should try to achieve. If the device cannot do the hinted page size then
1257  * it's the driver choice which page size to pick based on its support.
1258  * On output the device will return the page size it selected.
1259  *
1260  * ranges is a pointer to an array of
1261  * struct vfio_device_feature_dma_logging_range.
1262  *
1263  * The core kernel code guarantees to support by minimum num_ranges that fit
1264  * into a single kernel page. User space can try higher values but should give
1265  * up if the above can't be achieved as of some driver limitations.
1266  *
1267  * A single call to start device DMA logging can be issued and a matching stop
1268  * should follow at the end. Another start is not allowed in the meantime.
1269  */
1270 struct vfio_device_feature_dma_logging_control {
1271 	__aligned_u64 page_size;
1272 	__u32 num_ranges;
1273 	__u32 __reserved;
1274 	__aligned_u64 ranges;
1275 };
1276 
1277 struct vfio_device_feature_dma_logging_range {
1278 	__aligned_u64 iova;
1279 	__aligned_u64 length;
1280 };
1281 
1282 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6
1283 
1284 /*
1285  * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started
1286  * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START
1287  */
1288 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7
1289 
1290 /*
1291  * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log
1292  *
1293  * Query the device's DMA log for written pages within the given IOVA range.
1294  * During querying the log is cleared for the IOVA range.
1295  *
1296  * bitmap is a pointer to an array of u64s that will hold the output bitmap
1297  * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits
1298  * is given by:
1299  *  bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64))
1300  *
1301  * The input page_size can be any power of two value and does not have to
1302  * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver
1303  * will format its internal logging to match the reporting page size, possibly
1304  * by replicating bits if the internal page size is lower than requested.
1305  *
1306  * The LOGGING_REPORT will only set bits in the bitmap and never clear or
1307  * perform any initialization of the user provided bitmap.
1308  *
1309  * If any error is returned userspace should assume that the dirty log is
1310  * corrupted. Error recovery is to consider all memory dirty and try to
1311  * restart the dirty tracking, or to abort/restart the whole migration.
1312  *
1313  * If DMA logging is not enabled, an error will be returned.
1314  *
1315  */
1316 struct vfio_device_feature_dma_logging_report {
1317 	__aligned_u64 iova;
1318 	__aligned_u64 length;
1319 	__aligned_u64 page_size;
1320 	__aligned_u64 bitmap;
1321 };
1322 
1323 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8
1324 
1325 /*
1326  * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will
1327  * be required to complete stop copy.
1328  *
1329  * Note: Can be called on each device state.
1330  */
1331 
1332 struct vfio_device_feature_mig_data_size {
1333 	__aligned_u64 stop_copy_length;
1334 };
1335 
1336 #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9
1337 
1338 /* -------- API for Type1 VFIO IOMMU -------- */
1339 
1340 /**
1341  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
1342  *
1343  * Retrieve information about the IOMMU object. Fills in provided
1344  * struct vfio_iommu_info. Caller sets argsz.
1345  *
1346  * XXX Should we do these by CHECK_EXTENSION too?
1347  */
1348 struct vfio_iommu_type1_info {
1349 	__u32	argsz;
1350 	__u32	flags;
1351 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
1352 #define VFIO_IOMMU_INFO_CAPS	(1 << 1)	/* Info supports caps */
1353 	__u64	iova_pgsizes;	/* Bitmap of supported page sizes */
1354 	__u32   cap_offset;	/* Offset within info struct of first cap */
1355 };
1356 
1357 /*
1358  * The IOVA capability allows to report the valid IOVA range(s)
1359  * excluding any non-relaxable reserved regions exposed by
1360  * devices attached to the container. Any DMA map attempt
1361  * outside the valid iova range will return error.
1362  *
1363  * The structures below define version 1 of this capability.
1364  */
1365 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE  1
1366 
1367 struct vfio_iova_range {
1368 	__u64	start;
1369 	__u64	end;
1370 };
1371 
1372 struct vfio_iommu_type1_info_cap_iova_range {
1373 	struct	vfio_info_cap_header header;
1374 	__u32	nr_iovas;
1375 	__u32	reserved;
1376 	struct	vfio_iova_range iova_ranges[];
1377 };
1378 
1379 /*
1380  * The migration capability allows to report supported features for migration.
1381  *
1382  * The structures below define version 1 of this capability.
1383  *
1384  * The existence of this capability indicates that IOMMU kernel driver supports
1385  * dirty page logging.
1386  *
1387  * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty
1388  * page logging.
1389  * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap
1390  * size in bytes that can be used by user applications when getting the dirty
1391  * bitmap.
1392  */
1393 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION  2
1394 
1395 struct vfio_iommu_type1_info_cap_migration {
1396 	struct	vfio_info_cap_header header;
1397 	__u32	flags;
1398 	__u64	pgsize_bitmap;
1399 	__u64	max_dirty_bitmap_size;		/* in bytes */
1400 };
1401 
1402 /*
1403  * The DMA available capability allows to report the current number of
1404  * simultaneously outstanding DMA mappings that are allowed.
1405  *
1406  * The structure below defines version 1 of this capability.
1407  *
1408  * avail: specifies the current number of outstanding DMA mappings allowed.
1409  */
1410 #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3
1411 
1412 struct vfio_iommu_type1_info_dma_avail {
1413 	struct	vfio_info_cap_header header;
1414 	__u32	avail;
1415 };
1416 
1417 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
1418 
1419 /**
1420  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
1421  *
1422  * Map process virtual addresses to IO virtual addresses using the
1423  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
1424  *
1425  * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr
1426  * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR.  To
1427  * maintain memory consistency within the user application, the updated vaddr
1428  * must address the same memory object as originally mapped.  Failure to do so
1429  * will result in user memory corruption and/or device misbehavior.  iova and
1430  * size must match those in the original MAP_DMA call.  Protection is not
1431  * changed, and the READ & WRITE flags must be 0.
1432  */
1433 struct vfio_iommu_type1_dma_map {
1434 	__u32	argsz;
1435 	__u32	flags;
1436 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
1437 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
1438 #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2)
1439 	__u64	vaddr;				/* Process virtual address */
1440 	__u64	iova;				/* IO virtual address */
1441 	__u64	size;				/* Size of mapping (bytes) */
1442 };
1443 
1444 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
1445 
1446 struct vfio_bitmap {
1447 	__u64        pgsize;	/* page size for bitmap in bytes */
1448 	__u64        size;	/* in bytes */
1449 	__u64 __user *data;	/* one bit per page */
1450 };
1451 
1452 /**
1453  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
1454  *							struct vfio_dma_unmap)
1455  *
1456  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
1457  * Caller sets argsz.  The actual unmapped size is returned in the size
1458  * field.  No guarantee is made to the user that arbitrary unmaps of iova
1459  * or size different from those used in the original mapping call will
1460  * succeed.
1461  *
1462  * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap
1463  * before unmapping IO virtual addresses. When this flag is set, the user must
1464  * provide a struct vfio_bitmap in data[]. User must provide zero-allocated
1465  * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field.
1466  * A bit in the bitmap represents one page, of user provided page size in
1467  * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set
1468  * indicates that the page at that offset from iova is dirty. A Bitmap of the
1469  * pages in the range of unmapped size is returned in the user-provided
1470  * vfio_bitmap.data.
1471  *
1472  * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses.  iova and size
1473  * must be 0.  This cannot be combined with the get-dirty-bitmap flag.
1474  *
1475  * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host
1476  * virtual addresses in the iova range.  DMA to already-mapped pages continues.
1477  * Groups may not be added to the container while any addresses are invalid.
1478  * This cannot be combined with the get-dirty-bitmap flag.
1479  */
1480 struct vfio_iommu_type1_dma_unmap {
1481 	__u32	argsz;
1482 	__u32	flags;
1483 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0)
1484 #define VFIO_DMA_UNMAP_FLAG_ALL		     (1 << 1)
1485 #define VFIO_DMA_UNMAP_FLAG_VADDR	     (1 << 2)
1486 	__u64	iova;				/* IO virtual address */
1487 	__u64	size;				/* Size of mapping (bytes) */
1488 	__u8    data[];
1489 };
1490 
1491 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
1492 
1493 /*
1494  * IOCTLs to enable/disable IOMMU container usage.
1495  * No parameters are supported.
1496  */
1497 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
1498 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
1499 
1500 /**
1501  * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
1502  *                                     struct vfio_iommu_type1_dirty_bitmap)
1503  * IOCTL is used for dirty pages logging.
1504  * Caller should set flag depending on which operation to perform, details as
1505  * below:
1506  *
1507  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs
1508  * the IOMMU driver to log pages that are dirtied or potentially dirtied by
1509  * the device; designed to be used when a migration is in progress. Dirty pages
1510  * are logged until logging is disabled by user application by calling the IOCTL
1511  * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag.
1512  *
1513  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs
1514  * the IOMMU driver to stop logging dirtied pages.
1515  *
1516  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set
1517  * returns the dirty pages bitmap for IOMMU container for a given IOVA range.
1518  * The user must specify the IOVA range and the pgsize through the structure
1519  * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface
1520  * supports getting a bitmap of the smallest supported pgsize only and can be
1521  * modified in future to get a bitmap of any specified supported pgsize. The
1522  * user must provide a zeroed memory area for the bitmap memory and specify its
1523  * size in bitmap.size. One bit is used to represent one page consecutively
1524  * starting from iova offset. The user should provide page size in bitmap.pgsize
1525  * field. A bit set in the bitmap indicates that the page at that offset from
1526  * iova is dirty. The caller must set argsz to a value including the size of
1527  * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the
1528  * actual bitmap. If dirty pages logging is not enabled, an error will be
1529  * returned.
1530  *
1531  * Only one of the flags _START, _STOP and _GET may be specified at a time.
1532  *
1533  */
1534 struct vfio_iommu_type1_dirty_bitmap {
1535 	__u32        argsz;
1536 	__u32        flags;
1537 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START	(1 << 0)
1538 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP	(1 << 1)
1539 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP	(1 << 2)
1540 	__u8         data[];
1541 };
1542 
1543 struct vfio_iommu_type1_dirty_bitmap_get {
1544 	__u64              iova;	/* IO virtual address */
1545 	__u64              size;	/* Size of iova range */
1546 	struct vfio_bitmap bitmap;
1547 };
1548 
1549 #define VFIO_IOMMU_DIRTY_PAGES             _IO(VFIO_TYPE, VFIO_BASE + 17)
1550 
1551 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
1552 
1553 /*
1554  * The SPAPR TCE DDW info struct provides the information about
1555  * the details of Dynamic DMA window capability.
1556  *
1557  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
1558  * @max_dynamic_windows_supported tells the maximum number of windows
1559  * which the platform can create.
1560  * @levels tells the maximum number of levels in multi-level IOMMU tables;
1561  * this allows splitting a table into smaller chunks which reduces
1562  * the amount of physically contiguous memory required for the table.
1563  */
1564 struct vfio_iommu_spapr_tce_ddw_info {
1565 	__u64 pgsizes;			/* Bitmap of supported page sizes */
1566 	__u32 max_dynamic_windows_supported;
1567 	__u32 levels;
1568 };
1569 
1570 /*
1571  * The SPAPR TCE info struct provides the information about the PCI bus
1572  * address ranges available for DMA, these values are programmed into
1573  * the hardware so the guest has to know that information.
1574  *
1575  * The DMA 32 bit window start is an absolute PCI bus address.
1576  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
1577  * addresses too so the window works as a filter rather than an offset
1578  * for IOVA addresses.
1579  *
1580  * Flags supported:
1581  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
1582  *   (DDW) support is present. @ddw is only supported when DDW is present.
1583  */
1584 struct vfio_iommu_spapr_tce_info {
1585 	__u32 argsz;
1586 	__u32 flags;
1587 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
1588 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
1589 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
1590 	struct vfio_iommu_spapr_tce_ddw_info ddw;
1591 };
1592 
1593 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
1594 
1595 /*
1596  * EEH PE operation struct provides ways to:
1597  * - enable/disable EEH functionality;
1598  * - unfreeze IO/DMA for frozen PE;
1599  * - read PE state;
1600  * - reset PE;
1601  * - configure PE;
1602  * - inject EEH error.
1603  */
1604 struct vfio_eeh_pe_err {
1605 	__u32 type;
1606 	__u32 func;
1607 	__u64 addr;
1608 	__u64 mask;
1609 };
1610 
1611 struct vfio_eeh_pe_op {
1612 	__u32 argsz;
1613 	__u32 flags;
1614 	__u32 op;
1615 	union {
1616 		struct vfio_eeh_pe_err err;
1617 	};
1618 };
1619 
1620 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
1621 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
1622 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
1623 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
1624 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
1625 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
1626 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
1627 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
1628 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
1629 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
1630 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
1631 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
1632 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
1633 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
1634 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
1635 
1636 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
1637 
1638 /**
1639  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
1640  *
1641  * Registers user space memory where DMA is allowed. It pins
1642  * user pages and does the locked memory accounting so
1643  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
1644  * get faster.
1645  */
1646 struct vfio_iommu_spapr_register_memory {
1647 	__u32	argsz;
1648 	__u32	flags;
1649 	__u64	vaddr;				/* Process virtual address */
1650 	__u64	size;				/* Size of mapping (bytes) */
1651 };
1652 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
1653 
1654 /**
1655  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
1656  *
1657  * Unregisters user space memory registered with
1658  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
1659  * Uses vfio_iommu_spapr_register_memory for parameters.
1660  */
1661 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
1662 
1663 /**
1664  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
1665  *
1666  * Creates an additional TCE table and programs it (sets a new DMA window)
1667  * to every IOMMU group in the container. It receives page shift, window
1668  * size and number of levels in the TCE table being created.
1669  *
1670  * It allocates and returns an offset on a PCI bus of the new DMA window.
1671  */
1672 struct vfio_iommu_spapr_tce_create {
1673 	__u32 argsz;
1674 	__u32 flags;
1675 	/* in */
1676 	__u32 page_shift;
1677 	__u32 __resv1;
1678 	__u64 window_size;
1679 	__u32 levels;
1680 	__u32 __resv2;
1681 	/* out */
1682 	__u64 start_addr;
1683 };
1684 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
1685 
1686 /**
1687  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
1688  *
1689  * Unprograms a TCE table from all groups in the container and destroys it.
1690  * It receives a PCI bus offset as a window id.
1691  */
1692 struct vfio_iommu_spapr_tce_remove {
1693 	__u32 argsz;
1694 	__u32 flags;
1695 	/* in */
1696 	__u64 start_addr;
1697 };
1698 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
1699 
1700 /* ***************************************************************** */
1701 
1702 #endif /* _UAPIVFIO_H */
1703