xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 06831001)
1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2 /*
3  * VFIO API definition
4  *
5  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
6  *     Author: Alex Williamson <alex.williamson@redhat.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 #ifndef VFIO_H
13 #define VFIO_H
14 
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  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
245  *				       struct vfio_region_info)
246  *
247  * Retrieve information about a device region.  Caller provides
248  * struct vfio_region_info with index value set.  Caller sets argsz.
249  * Implementation of region mapping is bus driver specific.  This is
250  * intended to describe MMIO, I/O port, as well as bus specific
251  * regions (ex. PCI config space).  Zero sized regions may be used
252  * to describe unimplemented regions (ex. unimplemented PCI BARs).
253  * Return: 0 on success, -errno on failure.
254  */
255 struct vfio_region_info {
256 	__u32	argsz;
257 	__u32	flags;
258 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
259 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
260 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
261 #define VFIO_REGION_INFO_FLAG_CAPS	(1 << 3) /* Info supports caps */
262 	__u32	index;		/* Region index */
263 	__u32	cap_offset;	/* Offset within info struct of first cap */
264 	__u64	size;		/* Region size (bytes) */
265 	__u64	offset;		/* Region offset from start of device fd */
266 };
267 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
268 
269 /*
270  * The sparse mmap capability allows finer granularity of specifying areas
271  * within a region with mmap support.  When specified, the user should only
272  * mmap the offset ranges specified by the areas array.  mmaps outside of the
273  * areas specified may fail (such as the range covering a PCI MSI-X table) or
274  * may result in improper device behavior.
275  *
276  * The structures below define version 1 of this capability.
277  */
278 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP	1
279 
280 struct vfio_region_sparse_mmap_area {
281 	__u64	offset;	/* Offset of mmap'able area within region */
282 	__u64	size;	/* Size of mmap'able area */
283 };
284 
285 struct vfio_region_info_cap_sparse_mmap {
286 	struct vfio_info_cap_header header;
287 	__u32	nr_areas;
288 	__u32	reserved;
289 	struct vfio_region_sparse_mmap_area areas[];
290 };
291 
292 /*
293  * The device specific type capability allows regions unique to a specific
294  * device or class of devices to be exposed.  This helps solve the problem for
295  * vfio bus drivers of defining which region indexes correspond to which region
296  * on the device, without needing to resort to static indexes, as done by
297  * vfio-pci.  For instance, if we were to go back in time, we might remove
298  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
299  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
300  * make a "VGA" device specific type to describe the VGA access space.  This
301  * means that non-VGA devices wouldn't need to waste this index, and thus the
302  * address space associated with it due to implementation of device file
303  * descriptor offsets in vfio-pci.
304  *
305  * The current implementation is now part of the user ABI, so we can't use this
306  * for VGA, but there are other upcoming use cases, such as opregions for Intel
307  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
308  * use this for future additions.
309  *
310  * The structure below defines version 1 of this capability.
311  */
312 #define VFIO_REGION_INFO_CAP_TYPE	2
313 
314 struct vfio_region_info_cap_type {
315 	struct vfio_info_cap_header header;
316 	__u32 type;	/* global per bus driver */
317 	__u32 subtype;	/* type specific */
318 };
319 
320 /*
321  * List of region types, global per bus driver.
322  * If you introduce a new type, please add it here.
323  */
324 
325 /* PCI region type containing a PCI vendor part */
326 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
327 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
328 #define VFIO_REGION_TYPE_GFX                    (1)
329 #define VFIO_REGION_TYPE_CCW			(2)
330 #define VFIO_REGION_TYPE_MIGRATION_DEPRECATED   (3)
331 
332 /* sub-types for VFIO_REGION_TYPE_PCI_* */
333 
334 /* 8086 vendor PCI sub-types */
335 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
336 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
337 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
338 
339 /* 10de vendor PCI sub-types */
340 /*
341  * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
342  *
343  * Deprecated, region no longer provided
344  */
345 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM	(1)
346 
347 /* 1014 vendor PCI sub-types */
348 /*
349  * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
350  * to do TLB invalidation on a GPU.
351  *
352  * Deprecated, region no longer provided
353  */
354 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD	(1)
355 
356 /* sub-types for VFIO_REGION_TYPE_GFX */
357 #define VFIO_REGION_SUBTYPE_GFX_EDID            (1)
358 
359 /**
360  * struct vfio_region_gfx_edid - EDID region layout.
361  *
362  * Set display link state and EDID blob.
363  *
364  * The EDID blob has monitor information such as brand, name, serial
365  * number, physical size, supported video modes and more.
366  *
367  * This special region allows userspace (typically qemu) set a virtual
368  * EDID for the virtual monitor, which allows a flexible display
369  * configuration.
370  *
371  * For the edid blob spec look here:
372  *    https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
373  *
374  * On linux systems you can find the EDID blob in sysfs:
375  *    /sys/class/drm/${card}/${connector}/edid
376  *
377  * You can use the edid-decode ulility (comes with xorg-x11-utils) to
378  * decode the EDID blob.
379  *
380  * @edid_offset: location of the edid blob, relative to the
381  *               start of the region (readonly).
382  * @edid_max_size: max size of the edid blob (readonly).
383  * @edid_size: actual edid size (read/write).
384  * @link_state: display link state (read/write).
385  * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
386  * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
387  * @max_xres: max display width (0 == no limitation, readonly).
388  * @max_yres: max display height (0 == no limitation, readonly).
389  *
390  * EDID update protocol:
391  *   (1) set link-state to down.
392  *   (2) update edid blob and size.
393  *   (3) set link-state to up.
394  */
395 struct vfio_region_gfx_edid {
396 	__u32 edid_offset;
397 	__u32 edid_max_size;
398 	__u32 edid_size;
399 	__u32 max_xres;
400 	__u32 max_yres;
401 	__u32 link_state;
402 #define VFIO_DEVICE_GFX_LINK_STATE_UP    1
403 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN  2
404 };
405 
406 /* sub-types for VFIO_REGION_TYPE_CCW */
407 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD	(1)
408 #define VFIO_REGION_SUBTYPE_CCW_SCHIB		(2)
409 #define VFIO_REGION_SUBTYPE_CCW_CRW		(3)
410 
411 /* sub-types for VFIO_REGION_TYPE_MIGRATION */
412 #define VFIO_REGION_SUBTYPE_MIGRATION_DEPRECATED (1)
413 
414 struct vfio_device_migration_info {
415 	__u32 device_state;         /* VFIO device state */
416 #define VFIO_DEVICE_STATE_V1_STOP      (0)
417 #define VFIO_DEVICE_STATE_V1_RUNNING   (1 << 0)
418 #define VFIO_DEVICE_STATE_V1_SAVING    (1 << 1)
419 #define VFIO_DEVICE_STATE_V1_RESUMING  (1 << 2)
420 #define VFIO_DEVICE_STATE_MASK      (VFIO_DEVICE_STATE_V1_RUNNING | \
421 				     VFIO_DEVICE_STATE_V1_SAVING |  \
422 				     VFIO_DEVICE_STATE_V1_RESUMING)
423 
424 #define VFIO_DEVICE_STATE_VALID(state) \
425 	(state & VFIO_DEVICE_STATE_V1_RESUMING ? \
426 	(state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_V1_RESUMING : 1)
427 
428 #define VFIO_DEVICE_STATE_IS_ERROR(state) \
429 	((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_V1_SAVING | \
430 					      VFIO_DEVICE_STATE_V1_RESUMING))
431 
432 #define VFIO_DEVICE_STATE_SET_ERROR(state) \
433 	((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_STATE_V1_SAVING | \
434 					     VFIO_DEVICE_STATE_V1_RESUMING)
435 
436 	__u32 reserved;
437 	__u64 pending_bytes;
438 	__u64 data_offset;
439 	__u64 data_size;
440 };
441 
442 /*
443  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
444  * which allows direct access to non-MSIX registers which happened to be within
445  * the same system page.
446  *
447  * Even though the userspace gets direct access to the MSIX data, the existing
448  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
449  */
450 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE	3
451 
452 /*
453  * Capability with compressed real address (aka SSA - small system address)
454  * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
455  * and by the userspace to associate a NVLink bridge with a GPU.
456  *
457  * Deprecated, capability no longer provided
458  */
459 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT	4
460 
461 struct vfio_region_info_cap_nvlink2_ssatgt {
462 	struct vfio_info_cap_header header;
463 	__u64 tgt;
464 };
465 
466 /*
467  * Capability with an NVLink link speed. The value is read by
468  * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
469  * property in the device tree. The value is fixed in the hardware
470  * and failing to provide the correct value results in the link
471  * not working with no indication from the driver why.
472  *
473  * Deprecated, capability no longer provided
474  */
475 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD	5
476 
477 struct vfio_region_info_cap_nvlink2_lnkspd {
478 	struct vfio_info_cap_header header;
479 	__u32 link_speed;
480 	__u32 __pad;
481 };
482 
483 /**
484  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
485  *				    struct vfio_irq_info)
486  *
487  * Retrieve information about a device IRQ.  Caller provides
488  * struct vfio_irq_info with index value set.  Caller sets argsz.
489  * Implementation of IRQ mapping is bus driver specific.  Indexes
490  * using multiple IRQs are primarily intended to support MSI-like
491  * interrupt blocks.  Zero count irq blocks may be used to describe
492  * unimplemented interrupt types.
493  *
494  * The EVENTFD flag indicates the interrupt index supports eventfd based
495  * signaling.
496  *
497  * The MASKABLE flags indicates the index supports MASK and UNMASK
498  * actions described below.
499  *
500  * AUTOMASKED indicates that after signaling, the interrupt line is
501  * automatically masked by VFIO and the user needs to unmask the line
502  * to receive new interrupts.  This is primarily intended to distinguish
503  * level triggered interrupts.
504  *
505  * The NORESIZE flag indicates that the interrupt lines within the index
506  * are setup as a set and new subindexes cannot be enabled without first
507  * disabling the entire index.  This is used for interrupts like PCI MSI
508  * and MSI-X where the driver may only use a subset of the available
509  * indexes, but VFIO needs to enable a specific number of vectors
510  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
511  * then add and unmask vectors, it's up to userspace to make the decision
512  * whether to allocate the maximum supported number of vectors or tear
513  * down setup and incrementally increase the vectors as each is enabled.
514  */
515 struct vfio_irq_info {
516 	__u32	argsz;
517 	__u32	flags;
518 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
519 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
520 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
521 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
522 	__u32	index;		/* IRQ index */
523 	__u32	count;		/* Number of IRQs within this index */
524 };
525 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
526 
527 /**
528  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
529  *
530  * Set signaling, masking, and unmasking of interrupts.  Caller provides
531  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
532  * the range of subindexes being specified.
533  *
534  * The DATA flags specify the type of data provided.  If DATA_NONE, the
535  * operation performs the specified action immediately on the specified
536  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
537  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
538  *
539  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
540  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
541  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
542  * data = {1,0,1}
543  *
544  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
545  * A value of -1 can be used to either de-assign interrupts if already
546  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
547  * to be trigger for interrupts [0,0] and [0,2]:
548  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
549  * data = {fd1, -1, fd2}
550  * If index [0,1] is previously set, two count = 1 ioctls calls would be
551  * required to set [0,0] and [0,2] without changing [0,1].
552  *
553  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
554  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
555  * from userspace (ie. simulate hardware triggering).
556  *
557  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
558  * enables the interrupt index for the device.  Individual subindex interrupts
559  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
560  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
561  *
562  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
563  * ACTION_TRIGGER specifies kernel->user signaling.
564  */
565 struct vfio_irq_set {
566 	__u32	argsz;
567 	__u32	flags;
568 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
569 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
570 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
571 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
572 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
573 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
574 	__u32	index;
575 	__u32	start;
576 	__u32	count;
577 	__u8	data[];
578 };
579 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
580 
581 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
582 					 VFIO_IRQ_SET_DATA_BOOL | \
583 					 VFIO_IRQ_SET_DATA_EVENTFD)
584 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
585 					 VFIO_IRQ_SET_ACTION_UNMASK | \
586 					 VFIO_IRQ_SET_ACTION_TRIGGER)
587 /**
588  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
589  *
590  * Reset a device.
591  */
592 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
593 
594 /*
595  * The VFIO-PCI bus driver makes use of the following fixed region and
596  * IRQ index mapping.  Unimplemented regions return a size of zero.
597  * Unimplemented IRQ types return a count of zero.
598  */
599 
600 enum {
601 	VFIO_PCI_BAR0_REGION_INDEX,
602 	VFIO_PCI_BAR1_REGION_INDEX,
603 	VFIO_PCI_BAR2_REGION_INDEX,
604 	VFIO_PCI_BAR3_REGION_INDEX,
605 	VFIO_PCI_BAR4_REGION_INDEX,
606 	VFIO_PCI_BAR5_REGION_INDEX,
607 	VFIO_PCI_ROM_REGION_INDEX,
608 	VFIO_PCI_CONFIG_REGION_INDEX,
609 	/*
610 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
611 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
612 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
613 	 * range is found at it's identity mapped offset from the region
614 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
615 	 * between described ranges are unimplemented.
616 	 */
617 	VFIO_PCI_VGA_REGION_INDEX,
618 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
619 				 /* device specific cap to define content. */
620 };
621 
622 enum {
623 	VFIO_PCI_INTX_IRQ_INDEX,
624 	VFIO_PCI_MSI_IRQ_INDEX,
625 	VFIO_PCI_MSIX_IRQ_INDEX,
626 	VFIO_PCI_ERR_IRQ_INDEX,
627 	VFIO_PCI_REQ_IRQ_INDEX,
628 	VFIO_PCI_NUM_IRQS
629 };
630 
631 /*
632  * The vfio-ccw bus driver makes use of the following fixed region and
633  * IRQ index mapping. Unimplemented regions return a size of zero.
634  * Unimplemented IRQ types return a count of zero.
635  */
636 
637 enum {
638 	VFIO_CCW_CONFIG_REGION_INDEX,
639 	VFIO_CCW_NUM_REGIONS
640 };
641 
642 enum {
643 	VFIO_CCW_IO_IRQ_INDEX,
644 	VFIO_CCW_CRW_IRQ_INDEX,
645 	VFIO_CCW_REQ_IRQ_INDEX,
646 	VFIO_CCW_NUM_IRQS
647 };
648 
649 /**
650  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12,
651  *					      struct vfio_pci_hot_reset_info)
652  *
653  * Return: 0 on success, -errno on failure:
654  *	-enospc = insufficient buffer, -enodev = unsupported for device.
655  */
656 struct vfio_pci_dependent_device {
657 	__u32	group_id;
658 	__u16	segment;
659 	__u8	bus;
660 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
661 };
662 
663 struct vfio_pci_hot_reset_info {
664 	__u32	argsz;
665 	__u32	flags;
666 	__u32	count;
667 	struct vfio_pci_dependent_device	devices[];
668 };
669 
670 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
671 
672 /**
673  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
674  *				    struct vfio_pci_hot_reset)
675  *
676  * Return: 0 on success, -errno on failure.
677  */
678 struct vfio_pci_hot_reset {
679 	__u32	argsz;
680 	__u32	flags;
681 	__u32	count;
682 	__s32	group_fds[];
683 };
684 
685 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
686 
687 /**
688  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
689  *                                    struct vfio_device_query_gfx_plane)
690  *
691  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
692  *
693  * flags supported:
694  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
695  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
696  *   support for dma-buf.
697  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
698  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
699  *   support for region.
700  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
701  *   with each call to query the plane info.
702  * - Others are invalid and return -EINVAL.
703  *
704  * Note:
705  * 1. Plane could be disabled by guest. In that case, success will be
706  *    returned with zero-initialized drm_format, size, width and height
707  *    fields.
708  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
709  *
710  * Return: 0 on success, -errno on other failure.
711  */
712 struct vfio_device_gfx_plane_info {
713 	__u32 argsz;
714 	__u32 flags;
715 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
716 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
717 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
718 	/* in */
719 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
720 	/* out */
721 	__u32 drm_format;	/* drm format of plane */
722 	__u64 drm_format_mod;   /* tiled mode */
723 	__u32 width;	/* width of plane */
724 	__u32 height;	/* height of plane */
725 	__u32 stride;	/* stride of plane */
726 	__u32 size;	/* size of plane in bytes, align on page*/
727 	__u32 x_pos;	/* horizontal position of cursor plane */
728 	__u32 y_pos;	/* vertical position of cursor plane*/
729 	__u32 x_hot;    /* horizontal position of cursor hotspot */
730 	__u32 y_hot;    /* vertical position of cursor hotspot */
731 	union {
732 		__u32 region_index;	/* region index */
733 		__u32 dmabuf_id;	/* dma-buf id */
734 	};
735 };
736 
737 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
738 
739 /**
740  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
741  *
742  * Return a new dma-buf file descriptor for an exposed guest framebuffer
743  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
744  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
745  */
746 
747 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
748 
749 /**
750  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
751  *                              struct vfio_device_ioeventfd)
752  *
753  * Perform a write to the device at the specified device fd offset, with
754  * the specified data and width when the provided eventfd is triggered.
755  * vfio bus drivers may not support this for all regions, for all widths,
756  * or at all.  vfio-pci currently only enables support for BAR regions,
757  * excluding the MSI-X vector table.
758  *
759  * Return: 0 on success, -errno on failure.
760  */
761 struct vfio_device_ioeventfd {
762 	__u32	argsz;
763 	__u32	flags;
764 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
765 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
766 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
767 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
768 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
769 	__u64	offset;			/* device fd offset of write */
770 	__u64	data;			/* data to be written */
771 	__s32	fd;			/* -1 for de-assignment */
772 };
773 
774 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
775 
776 /**
777  * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
778  *			       struct vfio_device_feature)
779  *
780  * Get, set, or probe feature data of the device.  The feature is selected
781  * using the FEATURE_MASK portion of the flags field.  Support for a feature
782  * can be probed by setting both the FEATURE_MASK and PROBE bits.  A probe
783  * may optionally include the GET and/or SET bits to determine read vs write
784  * access of the feature respectively.  Probing a feature will return success
785  * if the feature is supported and all of the optionally indicated GET/SET
786  * methods are supported.  The format of the data portion of the structure is
787  * specific to the given feature.  The data portion is not required for
788  * probing.  GET and SET are mutually exclusive, except for use with PROBE.
789  *
790  * Return 0 on success, -errno on failure.
791  */
792 struct vfio_device_feature {
793 	__u32	argsz;
794 	__u32	flags;
795 #define VFIO_DEVICE_FEATURE_MASK	(0xffff) /* 16-bit feature index */
796 #define VFIO_DEVICE_FEATURE_GET		(1 << 16) /* Get feature into data[] */
797 #define VFIO_DEVICE_FEATURE_SET		(1 << 17) /* Set feature from data[] */
798 #define VFIO_DEVICE_FEATURE_PROBE	(1 << 18) /* Probe feature support */
799 	__u8	data[];
800 };
801 
802 #define VFIO_DEVICE_FEATURE		_IO(VFIO_TYPE, VFIO_BASE + 17)
803 
804 /*
805  * Provide support for setting a PCI VF Token, which is used as a shared
806  * secret between PF and VF drivers.  This feature may only be set on a
807  * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
808  * open VFs.  Data provided when setting this feature is a 16-byte array
809  * (__u8 b[16]), representing a UUID.
810  */
811 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN	(0)
812 
813 /*
814  * Indicates the device can support the migration API through
815  * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and
816  * ERROR states are always supported. Support for additional states is
817  * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be
818  * set.
819  *
820  * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and
821  * RESUMING are supported.
822  *
823  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P
824  * is supported in addition to the STOP_COPY states.
825  *
826  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that
827  * PRE_COPY is supported in addition to the STOP_COPY states.
828  *
829  * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY
830  * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported
831  * in addition to the STOP_COPY states.
832  *
833  * Other combinations of flags have behavior to be defined in the future.
834  */
835 struct vfio_device_feature_migration {
836 	__aligned_u64 flags;
837 #define VFIO_MIGRATION_STOP_COPY	(1 << 0)
838 #define VFIO_MIGRATION_P2P		(1 << 1)
839 #define VFIO_MIGRATION_PRE_COPY		(1 << 2)
840 };
841 #define VFIO_DEVICE_FEATURE_MIGRATION 1
842 
843 /*
844  * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO
845  * device. The new state is supplied in device_state, see enum
846  * vfio_device_mig_state for details
847  *
848  * The kernel migration driver must fully transition the device to the new state
849  * value before the operation returns to the user.
850  *
851  * The kernel migration driver must not generate asynchronous device state
852  * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET
853  * ioctl as described above.
854  *
855  * If this function fails then current device_state may be the original
856  * operating state or some other state along the combination transition path.
857  * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt
858  * to return to the original state, or attempt to return to some other state
859  * such as RUNNING or STOP.
860  *
861  * If the new_state starts a new data transfer session then the FD associated
862  * with that session is returned in data_fd. The user is responsible to close
863  * this FD when it is finished. The user must consider the migration data stream
864  * carried over the FD to be opaque and must preserve the byte order of the
865  * stream. The user is not required to preserve buffer segmentation when writing
866  * the data stream during the RESUMING operation.
867  *
868  * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO
869  * device, data_fd will be -1.
870  */
871 struct vfio_device_feature_mig_state {
872 	__u32 device_state; /* From enum vfio_device_mig_state */
873 	__s32 data_fd;
874 };
875 #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2
876 
877 /*
878  * The device migration Finite State Machine is described by the enum
879  * vfio_device_mig_state. Some of the FSM arcs will create a migration data
880  * transfer session by returning a FD, in this case the migration data will
881  * flow over the FD using read() and write() as discussed below.
882  *
883  * There are 5 states to support VFIO_MIGRATION_STOP_COPY:
884  *  RUNNING - The device is running normally
885  *  STOP - The device does not change the internal or external state
886  *  STOP_COPY - The device internal state can be read out
887  *  RESUMING - The device is stopped and is loading a new internal state
888  *  ERROR - The device has failed and must be reset
889  *
890  * And optional states to support VFIO_MIGRATION_P2P:
891  *  RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA
892  * And VFIO_MIGRATION_PRE_COPY:
893  *  PRE_COPY - The device is running normally but tracking internal state
894  *             changes
895  * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY:
896  *  PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA
897  *
898  * The FSM takes actions on the arcs between FSM states. The driver implements
899  * the following behavior for the FSM arcs:
900  *
901  * RUNNING_P2P -> STOP
902  * STOP_COPY -> STOP
903  *   While in STOP the device must stop the operation of the device. The device
904  *   must not generate interrupts, DMA, or any other change to external state.
905  *   It must not change its internal state. When stopped the device and kernel
906  *   migration driver must accept and respond to interaction to support external
907  *   subsystems in the STOP state, for example PCI MSI-X and PCI config space.
908  *   Failure by the user to restrict device access while in STOP must not result
909  *   in error conditions outside the user context (ex. host system faults).
910  *
911  *   The STOP_COPY arc will terminate a data transfer session.
912  *
913  * RESUMING -> STOP
914  *   Leaving RESUMING terminates a data transfer session and indicates the
915  *   device should complete processing of the data delivered by write(). The
916  *   kernel migration driver should complete the incorporation of data written
917  *   to the data transfer FD into the device internal state and perform
918  *   final validity and consistency checking of the new device state. If the
919  *   user provided data is found to be incomplete, inconsistent, or otherwise
920  *   invalid, the migration driver must fail the SET_STATE ioctl and
921  *   optionally go to the ERROR state as described below.
922  *
923  *   While in STOP the device has the same behavior as other STOP states
924  *   described above.
925  *
926  *   To abort a RESUMING session the device must be reset.
927  *
928  * PRE_COPY -> RUNNING
929  * RUNNING_P2P -> RUNNING
930  *   While in RUNNING the device is fully operational, the device may generate
931  *   interrupts, DMA, respond to MMIO, all vfio device regions are functional,
932  *   and the device may advance its internal state.
933  *
934  *   The PRE_COPY arc will terminate a data transfer session.
935  *
936  * PRE_COPY_P2P -> RUNNING_P2P
937  * RUNNING -> RUNNING_P2P
938  * STOP -> RUNNING_P2P
939  *   While in RUNNING_P2P the device is partially running in the P2P quiescent
940  *   state defined below.
941  *
942  *   The PRE_COPY_P2P arc will terminate a data transfer session.
943  *
944  * RUNNING -> PRE_COPY
945  * RUNNING_P2P -> PRE_COPY_P2P
946  * STOP -> STOP_COPY
947  *   PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states
948  *   which share a data transfer session. Moving between these states alters
949  *   what is streamed in session, but does not terminate or otherwise affect
950  *   the associated fd.
951  *
952  *   These arcs begin the process of saving the device state and will return a
953  *   new data_fd. The migration driver may perform actions such as enabling
954  *   dirty logging of device state when entering PRE_COPY or PER_COPY_P2P.
955  *
956  *   Each arc does not change the device operation, the device remains
957  *   RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below
958  *   in PRE_COPY_P2P -> STOP_COPY.
959  *
960  * PRE_COPY -> PRE_COPY_P2P
961  *   Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above.
962  *   However, while in the PRE_COPY_P2P state, the device is partially running
963  *   in the P2P quiescent state defined below, like RUNNING_P2P.
964  *
965  * PRE_COPY_P2P -> PRE_COPY
966  *   This arc allows returning the device to a full RUNNING behavior while
967  *   continuing all the behaviors of PRE_COPY.
968  *
969  * PRE_COPY_P2P -> STOP_COPY
970  *   While in the STOP_COPY state the device has the same behavior as STOP
971  *   with the addition that the data transfers session continues to stream the
972  *   migration state. End of stream on the FD indicates the entire device
973  *   state has been transferred.
974  *
975  *   The user should take steps to restrict access to vfio device regions while
976  *   the device is in STOP_COPY or risk corruption of the device migration data
977  *   stream.
978  *
979  * STOP -> RESUMING
980  *   Entering the RESUMING state starts a process of restoring the device state
981  *   and will return a new data_fd. The data stream fed into the data_fd should
982  *   be taken from the data transfer output of a single FD during saving from
983  *   a compatible device. The migration driver may alter/reset the internal
984  *   device state for this arc if required to prepare the device to receive the
985  *   migration data.
986  *
987  * STOP_COPY -> PRE_COPY
988  * STOP_COPY -> PRE_COPY_P2P
989  *   These arcs are not permitted and return error if requested. Future
990  *   revisions of this API may define behaviors for these arcs, in this case
991  *   support will be discoverable by a new flag in
992  *   VFIO_DEVICE_FEATURE_MIGRATION.
993  *
994  * any -> ERROR
995  *   ERROR cannot be specified as a device state, however any transition request
996  *   can be failed with an errno return and may then move the device_state into
997  *   ERROR. In this case the device was unable to execute the requested arc and
998  *   was also unable to restore the device to any valid device_state.
999  *   To recover from ERROR VFIO_DEVICE_RESET must be used to return the
1000  *   device_state back to RUNNING.
1001  *
1002  * The optional peer to peer (P2P) quiescent state is intended to be a quiescent
1003  * state for the device for the purposes of managing multiple devices within a
1004  * user context where peer-to-peer DMA between devices may be active. The
1005  * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating
1006  * any new P2P DMA transactions. If the device can identify P2P transactions
1007  * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration
1008  * driver must complete any such outstanding operations prior to completing the
1009  * FSM arc into a P2P state. For the purpose of specification the states
1010  * behave as though the device was fully running if not supported. Like while in
1011  * STOP or STOP_COPY the user must not touch the device, otherwise the state
1012  * can be exited.
1013  *
1014  * The remaining possible transitions are interpreted as combinations of the
1015  * above FSM arcs. As there are multiple paths through the FSM arcs the path
1016  * should be selected based on the following rules:
1017  *   - Select the shortest path.
1018  *   - The path cannot have saving group states as interior arcs, only
1019  *     starting/end states.
1020  * Refer to vfio_mig_get_next_state() for the result of the algorithm.
1021  *
1022  * The automatic transit through the FSM arcs that make up the combination
1023  * transition is invisible to the user. When working with combination arcs the
1024  * user may see any step along the path in the device_state if SET_STATE
1025  * fails. When handling these types of errors users should anticipate future
1026  * revisions of this protocol using new states and those states becoming
1027  * visible in this case.
1028  *
1029  * The optional states cannot be used with SET_STATE if the device does not
1030  * support them. The user can discover if these states are supported by using
1031  * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can
1032  * avoid knowing about these optional states if the kernel driver supports them.
1033  *
1034  * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY
1035  * is not present.
1036  */
1037 enum vfio_device_mig_state {
1038 	VFIO_DEVICE_STATE_ERROR = 0,
1039 	VFIO_DEVICE_STATE_STOP = 1,
1040 	VFIO_DEVICE_STATE_RUNNING = 2,
1041 	VFIO_DEVICE_STATE_STOP_COPY = 3,
1042 	VFIO_DEVICE_STATE_RESUMING = 4,
1043 	VFIO_DEVICE_STATE_RUNNING_P2P = 5,
1044 	VFIO_DEVICE_STATE_PRE_COPY = 6,
1045 	VFIO_DEVICE_STATE_PRE_COPY_P2P = 7,
1046 };
1047 
1048 /**
1049  * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21)
1050  *
1051  * This ioctl is used on the migration data FD in the precopy phase of the
1052  * migration data transfer. It returns an estimate of the current data sizes
1053  * remaining to be transferred. It allows the user to judge when it is
1054  * appropriate to leave PRE_COPY for STOP_COPY.
1055  *
1056  * This ioctl is valid only in PRE_COPY states and kernel driver should
1057  * return -EINVAL from any other migration state.
1058  *
1059  * The vfio_precopy_info data structure returned by this ioctl provides
1060  * estimates of data available from the device during the PRE_COPY states.
1061  * This estimate is split into two categories, initial_bytes and
1062  * dirty_bytes.
1063  *
1064  * The initial_bytes field indicates the amount of initial precopy
1065  * data available from the device. This field should have a non-zero initial
1066  * value and decrease as migration data is read from the device.
1067  * It is recommended to leave PRE_COPY for STOP_COPY only after this field
1068  * reaches zero. Leaving PRE_COPY earlier might make things slower.
1069  *
1070  * The dirty_bytes field tracks device state changes relative to data
1071  * previously retrieved.  This field starts at zero and may increase as
1072  * the internal device state is modified or decrease as that modified
1073  * state is read from the device.
1074  *
1075  * Userspace may use the combination of these fields to estimate the
1076  * potential data size available during the PRE_COPY phases, as well as
1077  * trends relative to the rate the device is dirtying its internal
1078  * state, but these fields are not required to have any bearing relative
1079  * to the data size available during the STOP_COPY phase.
1080  *
1081  * Drivers have a lot of flexibility in when and what they transfer during the
1082  * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO.
1083  *
1084  * During pre-copy the migration data FD has a temporary "end of stream" that is
1085  * reached when both initial_bytes and dirty_byte are zero. For instance, this
1086  * may indicate that the device is idle and not currently dirtying any internal
1087  * state. When read() is done on this temporary end of stream the kernel driver
1088  * should return ENOMSG from read(). Userspace can wait for more data (which may
1089  * never come) by using poll.
1090  *
1091  * Once in STOP_COPY the migration data FD has a permanent end of stream
1092  * signaled in the usual way by read() always returning 0 and poll always
1093  * returning readable. ENOMSG may not be returned in STOP_COPY.
1094  * Support for this ioctl is mandatory if a driver claims to support
1095  * VFIO_MIGRATION_PRE_COPY.
1096  *
1097  * Return: 0 on success, -1 and errno set on failure.
1098  */
1099 struct vfio_precopy_info {
1100 	__u32 argsz;
1101 	__u32 flags;
1102 	__aligned_u64 initial_bytes;
1103 	__aligned_u64 dirty_bytes;
1104 };
1105 
1106 #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21)
1107 
1108 /*
1109  * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power
1110  * state with the platform-based power management.  Device use of lower power
1111  * states depends on factors managed by the runtime power management core,
1112  * including system level support and coordinating support among dependent
1113  * devices.  Enabling device low power entry does not guarantee lower power
1114  * usage by the device, nor is a mechanism provided through this feature to
1115  * know the current power state of the device.  If any device access happens
1116  * (either from the host or through the vfio uAPI) when the device is in the
1117  * low power state, then the host will move the device out of the low power
1118  * state as necessary prior to the access.  Once the access is completed, the
1119  * device may re-enter the low power state.  For single shot low power support
1120  * with wake-up notification, see
1121  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below.  Access to mmap'd
1122  * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after
1123  * calling LOW_POWER_EXIT.
1124  */
1125 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3
1126 
1127 /*
1128  * This device feature has the same behavior as
1129  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user
1130  * provides an eventfd for wake-up notification.  When the device moves out of
1131  * the low power state for the wake-up, the host will not allow the device to
1132  * re-enter a low power state without a subsequent user call to one of the low
1133  * power entry device feature IOCTLs.  Access to mmap'd device regions is
1134  * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the
1135  * low power exit.  The low power exit can happen either through LOW_POWER_EXIT
1136  * or through any other access (where the wake-up notification has been
1137  * generated).  The access to mmap'd device regions will not trigger low power
1138  * exit.
1139  *
1140  * The notification through the provided eventfd will be generated only when
1141  * the device has entered and is resumed from a low power state after
1142  * calling this device feature IOCTL.  A device that has not entered low power
1143  * state, as managed through the runtime power management core, will not
1144  * generate a notification through the provided eventfd on access.  Calling the
1145  * LOW_POWER_EXIT feature is optional in the case where notification has been
1146  * signaled on the provided eventfd that a resume from low power has occurred.
1147  */
1148 struct vfio_device_low_power_entry_with_wakeup {
1149 	__s32 wakeup_eventfd;
1150 	__u32 reserved;
1151 };
1152 
1153 #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4
1154 
1155 /*
1156  * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as
1157  * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or
1158  * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features.
1159  * This device feature IOCTL may itself generate a wakeup eventfd notification
1160  * in the latter case if the device had previously entered a low power state.
1161  */
1162 #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5
1163 
1164 /*
1165  * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging.
1166  * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports
1167  * DMA logging.
1168  *
1169  * DMA logging allows a device to internally record what DMAs the device is
1170  * initiating and report them back to userspace. It is part of the VFIO
1171  * migration infrastructure that allows implementing dirty page tracking
1172  * during the pre copy phase of live migration. Only DMA WRITEs are logged,
1173  * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE.
1174  *
1175  * When DMA logging is started a range of IOVAs to monitor is provided and the
1176  * device can optimize its logging to cover only the IOVA range given. Each
1177  * DMA that the device initiates inside the range will be logged by the device
1178  * for later retrieval.
1179  *
1180  * page_size is an input that hints what tracking granularity the device
1181  * should try to achieve. If the device cannot do the hinted page size then
1182  * it's the driver choice which page size to pick based on its support.
1183  * On output the device will return the page size it selected.
1184  *
1185  * ranges is a pointer to an array of
1186  * struct vfio_device_feature_dma_logging_range.
1187  *
1188  * The core kernel code guarantees to support by minimum num_ranges that fit
1189  * into a single kernel page. User space can try higher values but should give
1190  * up if the above can't be achieved as of some driver limitations.
1191  *
1192  * A single call to start device DMA logging can be issued and a matching stop
1193  * should follow at the end. Another start is not allowed in the meantime.
1194  */
1195 struct vfio_device_feature_dma_logging_control {
1196 	__aligned_u64 page_size;
1197 	__u32 num_ranges;
1198 	__u32 __reserved;
1199 	__aligned_u64 ranges;
1200 };
1201 
1202 struct vfio_device_feature_dma_logging_range {
1203 	__aligned_u64 iova;
1204 	__aligned_u64 length;
1205 };
1206 
1207 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6
1208 
1209 /*
1210  * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started
1211  * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START
1212  */
1213 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7
1214 
1215 /*
1216  * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log
1217  *
1218  * Query the device's DMA log for written pages within the given IOVA range.
1219  * During querying the log is cleared for the IOVA range.
1220  *
1221  * bitmap is a pointer to an array of u64s that will hold the output bitmap
1222  * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits
1223  * is given by:
1224  *  bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64))
1225  *
1226  * The input page_size can be any power of two value and does not have to
1227  * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver
1228  * will format its internal logging to match the reporting page size, possibly
1229  * by replicating bits if the internal page size is lower than requested.
1230  *
1231  * The LOGGING_REPORT will only set bits in the bitmap and never clear or
1232  * perform any initialization of the user provided bitmap.
1233  *
1234  * If any error is returned userspace should assume that the dirty log is
1235  * corrupted. Error recovery is to consider all memory dirty and try to
1236  * restart the dirty tracking, or to abort/restart the whole migration.
1237  *
1238  * If DMA logging is not enabled, an error will be returned.
1239  *
1240  */
1241 struct vfio_device_feature_dma_logging_report {
1242 	__aligned_u64 iova;
1243 	__aligned_u64 length;
1244 	__aligned_u64 page_size;
1245 	__aligned_u64 bitmap;
1246 };
1247 
1248 #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8
1249 
1250 /*
1251  * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will
1252  * be required to complete stop copy.
1253  *
1254  * Note: Can be called on each device state.
1255  */
1256 
1257 struct vfio_device_feature_mig_data_size {
1258 	__aligned_u64 stop_copy_length;
1259 };
1260 
1261 #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9
1262 
1263 /* -------- API for Type1 VFIO IOMMU -------- */
1264 
1265 /**
1266  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
1267  *
1268  * Retrieve information about the IOMMU object. Fills in provided
1269  * struct vfio_iommu_info. Caller sets argsz.
1270  *
1271  * XXX Should we do these by CHECK_EXTENSION too?
1272  */
1273 struct vfio_iommu_type1_info {
1274 	__u32	argsz;
1275 	__u32	flags;
1276 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
1277 #define VFIO_IOMMU_INFO_CAPS	(1 << 1)	/* Info supports caps */
1278 	__u64	iova_pgsizes;	/* Bitmap of supported page sizes */
1279 	__u32   cap_offset;	/* Offset within info struct of first cap */
1280 };
1281 
1282 /*
1283  * The IOVA capability allows to report the valid IOVA range(s)
1284  * excluding any non-relaxable reserved regions exposed by
1285  * devices attached to the container. Any DMA map attempt
1286  * outside the valid iova range will return error.
1287  *
1288  * The structures below define version 1 of this capability.
1289  */
1290 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE  1
1291 
1292 struct vfio_iova_range {
1293 	__u64	start;
1294 	__u64	end;
1295 };
1296 
1297 struct vfio_iommu_type1_info_cap_iova_range {
1298 	struct	vfio_info_cap_header header;
1299 	__u32	nr_iovas;
1300 	__u32	reserved;
1301 	struct	vfio_iova_range iova_ranges[];
1302 };
1303 
1304 /*
1305  * The migration capability allows to report supported features for migration.
1306  *
1307  * The structures below define version 1 of this capability.
1308  *
1309  * The existence of this capability indicates that IOMMU kernel driver supports
1310  * dirty page logging.
1311  *
1312  * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty
1313  * page logging.
1314  * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap
1315  * size in bytes that can be used by user applications when getting the dirty
1316  * bitmap.
1317  */
1318 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION  2
1319 
1320 struct vfio_iommu_type1_info_cap_migration {
1321 	struct	vfio_info_cap_header header;
1322 	__u32	flags;
1323 	__u64	pgsize_bitmap;
1324 	__u64	max_dirty_bitmap_size;		/* in bytes */
1325 };
1326 
1327 /*
1328  * The DMA available capability allows to report the current number of
1329  * simultaneously outstanding DMA mappings that are allowed.
1330  *
1331  * The structure below defines version 1 of this capability.
1332  *
1333  * avail: specifies the current number of outstanding DMA mappings allowed.
1334  */
1335 #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3
1336 
1337 struct vfio_iommu_type1_info_dma_avail {
1338 	struct	vfio_info_cap_header header;
1339 	__u32	avail;
1340 };
1341 
1342 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
1343 
1344 /**
1345  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
1346  *
1347  * Map process virtual addresses to IO virtual addresses using the
1348  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
1349  *
1350  * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr
1351  * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR.  To
1352  * maintain memory consistency within the user application, the updated vaddr
1353  * must address the same memory object as originally mapped.  Failure to do so
1354  * will result in user memory corruption and/or device misbehavior.  iova and
1355  * size must match those in the original MAP_DMA call.  Protection is not
1356  * changed, and the READ & WRITE flags must be 0.
1357  */
1358 struct vfio_iommu_type1_dma_map {
1359 	__u32	argsz;
1360 	__u32	flags;
1361 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
1362 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
1363 #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2)
1364 	__u64	vaddr;				/* Process virtual address */
1365 	__u64	iova;				/* IO virtual address */
1366 	__u64	size;				/* Size of mapping (bytes) */
1367 };
1368 
1369 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
1370 
1371 struct vfio_bitmap {
1372 	__u64        pgsize;	/* page size for bitmap in bytes */
1373 	__u64        size;	/* in bytes */
1374 	__u64 *data;	/* one bit per page */
1375 };
1376 
1377 /**
1378  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
1379  *							struct vfio_dma_unmap)
1380  *
1381  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
1382  * Caller sets argsz.  The actual unmapped size is returned in the size
1383  * field.  No guarantee is made to the user that arbitrary unmaps of iova
1384  * or size different from those used in the original mapping call will
1385  * succeed.
1386  *
1387  * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap
1388  * before unmapping IO virtual addresses. When this flag is set, the user must
1389  * provide a struct vfio_bitmap in data[]. User must provide zero-allocated
1390  * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field.
1391  * A bit in the bitmap represents one page, of user provided page size in
1392  * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set
1393  * indicates that the page at that offset from iova is dirty. A Bitmap of the
1394  * pages in the range of unmapped size is returned in the user-provided
1395  * vfio_bitmap.data.
1396  *
1397  * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses.  iova and size
1398  * must be 0.  This cannot be combined with the get-dirty-bitmap flag.
1399  *
1400  * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host
1401  * virtual addresses in the iova range.  DMA to already-mapped pages continues.
1402  * Groups may not be added to the container while any addresses are invalid.
1403  * This cannot be combined with the get-dirty-bitmap flag.
1404  */
1405 struct vfio_iommu_type1_dma_unmap {
1406 	__u32	argsz;
1407 	__u32	flags;
1408 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0)
1409 #define VFIO_DMA_UNMAP_FLAG_ALL		     (1 << 1)
1410 #define VFIO_DMA_UNMAP_FLAG_VADDR	     (1 << 2)
1411 	__u64	iova;				/* IO virtual address */
1412 	__u64	size;				/* Size of mapping (bytes) */
1413 	__u8    data[];
1414 };
1415 
1416 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
1417 
1418 /*
1419  * IOCTLs to enable/disable IOMMU container usage.
1420  * No parameters are supported.
1421  */
1422 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
1423 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
1424 
1425 /**
1426  * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
1427  *                                     struct vfio_iommu_type1_dirty_bitmap)
1428  * IOCTL is used for dirty pages logging.
1429  * Caller should set flag depending on which operation to perform, details as
1430  * below:
1431  *
1432  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs
1433  * the IOMMU driver to log pages that are dirtied or potentially dirtied by
1434  * the device; designed to be used when a migration is in progress. Dirty pages
1435  * are logged until logging is disabled by user application by calling the IOCTL
1436  * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag.
1437  *
1438  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs
1439  * the IOMMU driver to stop logging dirtied pages.
1440  *
1441  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set
1442  * returns the dirty pages bitmap for IOMMU container for a given IOVA range.
1443  * The user must specify the IOVA range and the pgsize through the structure
1444  * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface
1445  * supports getting a bitmap of the smallest supported pgsize only and can be
1446  * modified in future to get a bitmap of any specified supported pgsize. The
1447  * user must provide a zeroed memory area for the bitmap memory and specify its
1448  * size in bitmap.size. One bit is used to represent one page consecutively
1449  * starting from iova offset. The user should provide page size in bitmap.pgsize
1450  * field. A bit set in the bitmap indicates that the page at that offset from
1451  * iova is dirty. The caller must set argsz to a value including the size of
1452  * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the
1453  * actual bitmap. If dirty pages logging is not enabled, an error will be
1454  * returned.
1455  *
1456  * Only one of the flags _START, _STOP and _GET may be specified at a time.
1457  *
1458  */
1459 struct vfio_iommu_type1_dirty_bitmap {
1460 	__u32        argsz;
1461 	__u32        flags;
1462 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START	(1 << 0)
1463 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP	(1 << 1)
1464 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP	(1 << 2)
1465 	__u8         data[];
1466 };
1467 
1468 struct vfio_iommu_type1_dirty_bitmap_get {
1469 	__u64              iova;	/* IO virtual address */
1470 	__u64              size;	/* Size of iova range */
1471 	struct vfio_bitmap bitmap;
1472 };
1473 
1474 #define VFIO_IOMMU_DIRTY_PAGES             _IO(VFIO_TYPE, VFIO_BASE + 17)
1475 
1476 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
1477 
1478 /*
1479  * The SPAPR TCE DDW info struct provides the information about
1480  * the details of Dynamic DMA window capability.
1481  *
1482  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
1483  * @max_dynamic_windows_supported tells the maximum number of windows
1484  * which the platform can create.
1485  * @levels tells the maximum number of levels in multi-level IOMMU tables;
1486  * this allows splitting a table into smaller chunks which reduces
1487  * the amount of physically contiguous memory required for the table.
1488  */
1489 struct vfio_iommu_spapr_tce_ddw_info {
1490 	__u64 pgsizes;			/* Bitmap of supported page sizes */
1491 	__u32 max_dynamic_windows_supported;
1492 	__u32 levels;
1493 };
1494 
1495 /*
1496  * The SPAPR TCE info struct provides the information about the PCI bus
1497  * address ranges available for DMA, these values are programmed into
1498  * the hardware so the guest has to know that information.
1499  *
1500  * The DMA 32 bit window start is an absolute PCI bus address.
1501  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
1502  * addresses too so the window works as a filter rather than an offset
1503  * for IOVA addresses.
1504  *
1505  * Flags supported:
1506  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
1507  *   (DDW) support is present. @ddw is only supported when DDW is present.
1508  */
1509 struct vfio_iommu_spapr_tce_info {
1510 	__u32 argsz;
1511 	__u32 flags;
1512 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
1513 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
1514 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
1515 	struct vfio_iommu_spapr_tce_ddw_info ddw;
1516 };
1517 
1518 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
1519 
1520 /*
1521  * EEH PE operation struct provides ways to:
1522  * - enable/disable EEH functionality;
1523  * - unfreeze IO/DMA for frozen PE;
1524  * - read PE state;
1525  * - reset PE;
1526  * - configure PE;
1527  * - inject EEH error.
1528  */
1529 struct vfio_eeh_pe_err {
1530 	__u32 type;
1531 	__u32 func;
1532 	__u64 addr;
1533 	__u64 mask;
1534 };
1535 
1536 struct vfio_eeh_pe_op {
1537 	__u32 argsz;
1538 	__u32 flags;
1539 	__u32 op;
1540 	union {
1541 		struct vfio_eeh_pe_err err;
1542 	};
1543 };
1544 
1545 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
1546 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
1547 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
1548 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
1549 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
1550 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
1551 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
1552 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
1553 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
1554 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
1555 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
1556 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
1557 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
1558 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
1559 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
1560 
1561 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
1562 
1563 /**
1564  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
1565  *
1566  * Registers user space memory where DMA is allowed. It pins
1567  * user pages and does the locked memory accounting so
1568  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
1569  * get faster.
1570  */
1571 struct vfio_iommu_spapr_register_memory {
1572 	__u32	argsz;
1573 	__u32	flags;
1574 	__u64	vaddr;				/* Process virtual address */
1575 	__u64	size;				/* Size of mapping (bytes) */
1576 };
1577 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
1578 
1579 /**
1580  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
1581  *
1582  * Unregisters user space memory registered with
1583  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
1584  * Uses vfio_iommu_spapr_register_memory for parameters.
1585  */
1586 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
1587 
1588 /**
1589  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
1590  *
1591  * Creates an additional TCE table and programs it (sets a new DMA window)
1592  * to every IOMMU group in the container. It receives page shift, window
1593  * size and number of levels in the TCE table being created.
1594  *
1595  * It allocates and returns an offset on a PCI bus of the new DMA window.
1596  */
1597 struct vfio_iommu_spapr_tce_create {
1598 	__u32 argsz;
1599 	__u32 flags;
1600 	/* in */
1601 	__u32 page_shift;
1602 	__u32 __resv1;
1603 	__u64 window_size;
1604 	__u32 levels;
1605 	__u32 __resv2;
1606 	/* out */
1607 	__u64 start_addr;
1608 };
1609 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
1610 
1611 /**
1612  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
1613  *
1614  * Unprograms a TCE table from all groups in the container and destroys it.
1615  * It receives a PCI bus offset as a window id.
1616  */
1617 struct vfio_iommu_spapr_tce_remove {
1618 	__u32 argsz;
1619 	__u32 flags;
1620 	/* in */
1621 	__u64 start_addr;
1622 };
1623 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
1624 
1625 /* ***************************************************************** */
1626 
1627 #endif /* VFIO_H */
1628