xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 65a117da)
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 /*
50  * The IOCTL interface is designed for extensibility by embedding the
51  * structure length (argsz) and flags into structures passed between
52  * kernel and userspace.  We therefore use the _IO() macro for these
53  * defines to avoid implicitly embedding a size into the ioctl request.
54  * As structure fields are added, argsz will increase to match and flag
55  * bits will be defined to indicate additional fields with valid data.
56  * It's *always* the caller's responsibility to indicate the size of
57  * the structure passed by setting argsz appropriately.
58  */
59 
60 #define VFIO_TYPE	(';')
61 #define VFIO_BASE	100
62 
63 /*
64  * For extension of INFO ioctls, VFIO makes use of a capability chain
65  * designed after PCI/e capabilities.  A flag bit indicates whether
66  * this capability chain is supported and a field defined in the fixed
67  * structure defines the offset of the first capability in the chain.
68  * This field is only valid when the corresponding bit in the flags
69  * bitmap is set.  This offset field is relative to the start of the
70  * INFO buffer, as is the next field within each capability header.
71  * The id within the header is a shared address space per INFO ioctl,
72  * while the version field is specific to the capability id.  The
73  * contents following the header are specific to the capability id.
74  */
75 struct vfio_info_cap_header {
76 	__u16	id;		/* Identifies capability */
77 	__u16	version;	/* Version specific to the capability ID */
78 	__u32	next;		/* Offset of next capability */
79 };
80 
81 /*
82  * Callers of INFO ioctls passing insufficiently sized buffers will see
83  * the capability chain flag bit set, a zero value for the first capability
84  * offset (if available within the provided argsz), and argsz will be
85  * updated to report the necessary buffer size.  For compatibility, the
86  * INFO ioctl will not report error in this case, but the capability chain
87  * will not be available.
88  */
89 
90 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
91 
92 /**
93  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
94  *
95  * Report the version of the VFIO API.  This allows us to bump the entire
96  * API version should we later need to add or change features in incompatible
97  * ways.
98  * Return: VFIO_API_VERSION
99  * Availability: Always
100  */
101 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
102 
103 /**
104  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
105  *
106  * Check whether an extension is supported.
107  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
108  * Availability: Always
109  */
110 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
111 
112 /**
113  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
114  *
115  * Set the iommu to the given type.  The type must be supported by an
116  * iommu driver as verified by calling CHECK_EXTENSION using the same
117  * type.  A group must be set to this file descriptor before this
118  * ioctl is available.  The IOMMU interfaces enabled by this call are
119  * specific to the value set.
120  * Return: 0 on success, -errno on failure
121  * Availability: When VFIO group attached
122  */
123 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
124 
125 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
126 
127 /**
128  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
129  *						struct vfio_group_status)
130  *
131  * Retrieve information about the group.  Fills in provided
132  * struct vfio_group_info.  Caller sets argsz.
133  * Return: 0 on succes, -errno on failure.
134  * Availability: Always
135  */
136 struct vfio_group_status {
137 	__u32	argsz;
138 	__u32	flags;
139 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
140 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
141 };
142 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
143 
144 /**
145  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
146  *
147  * Set the container for the VFIO group to the open VFIO file
148  * descriptor provided.  Groups may only belong to a single
149  * container.  Containers may, at their discretion, support multiple
150  * groups.  Only when a container is set are all of the interfaces
151  * of the VFIO file descriptor and the VFIO group file descriptor
152  * available to the user.
153  * Return: 0 on success, -errno on failure.
154  * Availability: Always
155  */
156 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
157 
158 /**
159  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
160  *
161  * Remove the group from the attached container.  This is the
162  * opposite of the SET_CONTAINER call and returns the group to
163  * an initial state.  All device file descriptors must be released
164  * prior to calling this interface.  When removing the last group
165  * from a container, the IOMMU will be disabled and all state lost,
166  * effectively also returning the VFIO file descriptor to an initial
167  * state.
168  * Return: 0 on success, -errno on failure.
169  * Availability: When attached to container
170  */
171 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
172 
173 /**
174  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
175  *
176  * Return a new file descriptor for the device object described by
177  * the provided string.  The string should match a device listed in
178  * the devices subdirectory of the IOMMU group sysfs entry.  The
179  * group containing the device must already be added to this context.
180  * Return: new file descriptor on success, -errno on failure.
181  * Availability: When attached to container
182  */
183 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
184 
185 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
186 
187 /**
188  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
189  *						struct vfio_device_info)
190  *
191  * Retrieve information about the device.  Fills in provided
192  * struct vfio_device_info.  Caller sets argsz.
193  * Return: 0 on success, -errno on failure.
194  */
195 struct vfio_device_info {
196 	__u32	argsz;
197 	__u32	flags;
198 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
199 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
200 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2)	/* vfio-platform device */
201 #define VFIO_DEVICE_FLAGS_AMBA  (1 << 3)	/* vfio-amba device */
202 #define VFIO_DEVICE_FLAGS_CCW	(1 << 4)	/* vfio-ccw device */
203 #define VFIO_DEVICE_FLAGS_AP	(1 << 5)	/* vfio-ap device */
204 	__u32	num_regions;	/* Max region index + 1 */
205 	__u32	num_irqs;	/* Max IRQ index + 1 */
206 };
207 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
208 
209 /*
210  * Vendor driver using Mediated device framework should provide device_api
211  * attribute in supported type attribute groups. Device API string should be one
212  * of the following corresponding to device flags in vfio_device_info structure.
213  */
214 
215 #define VFIO_DEVICE_API_PCI_STRING		"vfio-pci"
216 #define VFIO_DEVICE_API_PLATFORM_STRING		"vfio-platform"
217 #define VFIO_DEVICE_API_AMBA_STRING		"vfio-amba"
218 #define VFIO_DEVICE_API_CCW_STRING		"vfio-ccw"
219 #define VFIO_DEVICE_API_AP_STRING		"vfio-ap"
220 
221 /**
222  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
223  *				       struct vfio_region_info)
224  *
225  * Retrieve information about a device region.  Caller provides
226  * struct vfio_region_info with index value set.  Caller sets argsz.
227  * Implementation of region mapping is bus driver specific.  This is
228  * intended to describe MMIO, I/O port, as well as bus specific
229  * regions (ex. PCI config space).  Zero sized regions may be used
230  * to describe unimplemented regions (ex. unimplemented PCI BARs).
231  * Return: 0 on success, -errno on failure.
232  */
233 struct vfio_region_info {
234 	__u32	argsz;
235 	__u32	flags;
236 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
237 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
238 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
239 #define VFIO_REGION_INFO_FLAG_CAPS	(1 << 3) /* Info supports caps */
240 	__u32	index;		/* Region index */
241 	__u32	cap_offset;	/* Offset within info struct of first cap */
242 	__u64	size;		/* Region size (bytes) */
243 	__u64	offset;		/* Region offset from start of device fd */
244 };
245 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
246 
247 /*
248  * The sparse mmap capability allows finer granularity of specifying areas
249  * within a region with mmap support.  When specified, the user should only
250  * mmap the offset ranges specified by the areas array.  mmaps outside of the
251  * areas specified may fail (such as the range covering a PCI MSI-X table) or
252  * may result in improper device behavior.
253  *
254  * The structures below define version 1 of this capability.
255  */
256 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP	1
257 
258 struct vfio_region_sparse_mmap_area {
259 	__u64	offset;	/* Offset of mmap'able area within region */
260 	__u64	size;	/* Size of mmap'able area */
261 };
262 
263 struct vfio_region_info_cap_sparse_mmap {
264 	struct vfio_info_cap_header header;
265 	__u32	nr_areas;
266 	__u32	reserved;
267 	struct vfio_region_sparse_mmap_area areas[];
268 };
269 
270 /*
271  * The device specific type capability allows regions unique to a specific
272  * device or class of devices to be exposed.  This helps solve the problem for
273  * vfio bus drivers of defining which region indexes correspond to which region
274  * on the device, without needing to resort to static indexes, as done by
275  * vfio-pci.  For instance, if we were to go back in time, we might remove
276  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
277  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
278  * make a "VGA" device specific type to describe the VGA access space.  This
279  * means that non-VGA devices wouldn't need to waste this index, and thus the
280  * address space associated with it due to implementation of device file
281  * descriptor offsets in vfio-pci.
282  *
283  * The current implementation is now part of the user ABI, so we can't use this
284  * for VGA, but there are other upcoming use cases, such as opregions for Intel
285  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
286  * use this for future additions.
287  *
288  * The structure below defines version 1 of this capability.
289  */
290 #define VFIO_REGION_INFO_CAP_TYPE	2
291 
292 struct vfio_region_info_cap_type {
293 	struct vfio_info_cap_header header;
294 	__u32 type;	/* global per bus driver */
295 	__u32 subtype;	/* type specific */
296 };
297 
298 /*
299  * List of region types, global per bus driver.
300  * If you introduce a new type, please add it here.
301  */
302 
303 /* PCI region type containing a PCI vendor part */
304 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
305 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
306 #define VFIO_REGION_TYPE_GFX                    (1)
307 #define VFIO_REGION_TYPE_CCW			(2)
308 
309 /* sub-types for VFIO_REGION_TYPE_PCI_* */
310 
311 /* 8086 vendor PCI sub-types */
312 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
313 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
314 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
315 
316 /* 10de vendor PCI sub-types */
317 /*
318  * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
319  */
320 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM	(1)
321 
322 /* 1014 vendor PCI sub-types */
323 /*
324  * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
325  * to do TLB invalidation on a GPU.
326  */
327 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD	(1)
328 
329 /* sub-types for VFIO_REGION_TYPE_GFX */
330 #define VFIO_REGION_SUBTYPE_GFX_EDID            (1)
331 
332 /**
333  * struct vfio_region_gfx_edid - EDID region layout.
334  *
335  * Set display link state and EDID blob.
336  *
337  * The EDID blob has monitor information such as brand, name, serial
338  * number, physical size, supported video modes and more.
339  *
340  * This special region allows userspace (typically qemu) set a virtual
341  * EDID for the virtual monitor, which allows a flexible display
342  * configuration.
343  *
344  * For the edid blob spec look here:
345  *    https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
346  *
347  * On linux systems you can find the EDID blob in sysfs:
348  *    /sys/class/drm/${card}/${connector}/edid
349  *
350  * You can use the edid-decode ulility (comes with xorg-x11-utils) to
351  * decode the EDID blob.
352  *
353  * @edid_offset: location of the edid blob, relative to the
354  *               start of the region (readonly).
355  * @edid_max_size: max size of the edid blob (readonly).
356  * @edid_size: actual edid size (read/write).
357  * @link_state: display link state (read/write).
358  * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
359  * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
360  * @max_xres: max display width (0 == no limitation, readonly).
361  * @max_yres: max display height (0 == no limitation, readonly).
362  *
363  * EDID update protocol:
364  *   (1) set link-state to down.
365  *   (2) update edid blob and size.
366  *   (3) set link-state to up.
367  */
368 struct vfio_region_gfx_edid {
369 	__u32 edid_offset;
370 	__u32 edid_max_size;
371 	__u32 edid_size;
372 	__u32 max_xres;
373 	__u32 max_yres;
374 	__u32 link_state;
375 #define VFIO_DEVICE_GFX_LINK_STATE_UP    1
376 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN  2
377 };
378 
379 /* sub-types for VFIO_REGION_TYPE_CCW */
380 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD	(1)
381 
382 /*
383  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
384  * which allows direct access to non-MSIX registers which happened to be within
385  * the same system page.
386  *
387  * Even though the userspace gets direct access to the MSIX data, the existing
388  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
389  */
390 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE	3
391 
392 /*
393  * Capability with compressed real address (aka SSA - small system address)
394  * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
395  * and by the userspace to associate a NVLink bridge with a GPU.
396  */
397 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT	4
398 
399 struct vfio_region_info_cap_nvlink2_ssatgt {
400 	struct vfio_info_cap_header header;
401 	__u64 tgt;
402 };
403 
404 /*
405  * Capability with an NVLink link speed. The value is read by
406  * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
407  * property in the device tree. The value is fixed in the hardware
408  * and failing to provide the correct value results in the link
409  * not working with no indication from the driver why.
410  */
411 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD	5
412 
413 struct vfio_region_info_cap_nvlink2_lnkspd {
414 	struct vfio_info_cap_header header;
415 	__u32 link_speed;
416 	__u32 __pad;
417 };
418 
419 /**
420  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
421  *				    struct vfio_irq_info)
422  *
423  * Retrieve information about a device IRQ.  Caller provides
424  * struct vfio_irq_info with index value set.  Caller sets argsz.
425  * Implementation of IRQ mapping is bus driver specific.  Indexes
426  * using multiple IRQs are primarily intended to support MSI-like
427  * interrupt blocks.  Zero count irq blocks may be used to describe
428  * unimplemented interrupt types.
429  *
430  * The EVENTFD flag indicates the interrupt index supports eventfd based
431  * signaling.
432  *
433  * The MASKABLE flags indicates the index supports MASK and UNMASK
434  * actions described below.
435  *
436  * AUTOMASKED indicates that after signaling, the interrupt line is
437  * automatically masked by VFIO and the user needs to unmask the line
438  * to receive new interrupts.  This is primarily intended to distinguish
439  * level triggered interrupts.
440  *
441  * The NORESIZE flag indicates that the interrupt lines within the index
442  * are setup as a set and new subindexes cannot be enabled without first
443  * disabling the entire index.  This is used for interrupts like PCI MSI
444  * and MSI-X where the driver may only use a subset of the available
445  * indexes, but VFIO needs to enable a specific number of vectors
446  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
447  * then add and unmask vectors, it's up to userspace to make the decision
448  * whether to allocate the maximum supported number of vectors or tear
449  * down setup and incrementally increase the vectors as each is enabled.
450  */
451 struct vfio_irq_info {
452 	__u32	argsz;
453 	__u32	flags;
454 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
455 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
456 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
457 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
458 	__u32	index;		/* IRQ index */
459 	__u32	count;		/* Number of IRQs within this index */
460 };
461 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
462 
463 /**
464  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
465  *
466  * Set signaling, masking, and unmasking of interrupts.  Caller provides
467  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
468  * the range of subindexes being specified.
469  *
470  * The DATA flags specify the type of data provided.  If DATA_NONE, the
471  * operation performs the specified action immediately on the specified
472  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
473  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
474  *
475  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
476  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
477  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
478  * data = {1,0,1}
479  *
480  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
481  * A value of -1 can be used to either de-assign interrupts if already
482  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
483  * to be trigger for interrupts [0,0] and [0,2]:
484  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
485  * data = {fd1, -1, fd2}
486  * If index [0,1] is previously set, two count = 1 ioctls calls would be
487  * required to set [0,0] and [0,2] without changing [0,1].
488  *
489  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
490  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
491  * from userspace (ie. simulate hardware triggering).
492  *
493  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
494  * enables the interrupt index for the device.  Individual subindex interrupts
495  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
496  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
497  *
498  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
499  * ACTION_TRIGGER specifies kernel->user signaling.
500  */
501 struct vfio_irq_set {
502 	__u32	argsz;
503 	__u32	flags;
504 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
505 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
506 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
507 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
508 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
509 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
510 	__u32	index;
511 	__u32	start;
512 	__u32	count;
513 	__u8	data[];
514 };
515 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
516 
517 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
518 					 VFIO_IRQ_SET_DATA_BOOL | \
519 					 VFIO_IRQ_SET_DATA_EVENTFD)
520 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
521 					 VFIO_IRQ_SET_ACTION_UNMASK | \
522 					 VFIO_IRQ_SET_ACTION_TRIGGER)
523 /**
524  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
525  *
526  * Reset a device.
527  */
528 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
529 
530 /*
531  * The VFIO-PCI bus driver makes use of the following fixed region and
532  * IRQ index mapping.  Unimplemented regions return a size of zero.
533  * Unimplemented IRQ types return a count of zero.
534  */
535 
536 enum {
537 	VFIO_PCI_BAR0_REGION_INDEX,
538 	VFIO_PCI_BAR1_REGION_INDEX,
539 	VFIO_PCI_BAR2_REGION_INDEX,
540 	VFIO_PCI_BAR3_REGION_INDEX,
541 	VFIO_PCI_BAR4_REGION_INDEX,
542 	VFIO_PCI_BAR5_REGION_INDEX,
543 	VFIO_PCI_ROM_REGION_INDEX,
544 	VFIO_PCI_CONFIG_REGION_INDEX,
545 	/*
546 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
547 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
548 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
549 	 * range is found at it's identity mapped offset from the region
550 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
551 	 * between described ranges are unimplemented.
552 	 */
553 	VFIO_PCI_VGA_REGION_INDEX,
554 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
555 				 /* device specific cap to define content. */
556 };
557 
558 enum {
559 	VFIO_PCI_INTX_IRQ_INDEX,
560 	VFIO_PCI_MSI_IRQ_INDEX,
561 	VFIO_PCI_MSIX_IRQ_INDEX,
562 	VFIO_PCI_ERR_IRQ_INDEX,
563 	VFIO_PCI_REQ_IRQ_INDEX,
564 	VFIO_PCI_NUM_IRQS
565 };
566 
567 /*
568  * The vfio-ccw bus driver makes use of the following fixed region and
569  * IRQ index mapping. Unimplemented regions return a size of zero.
570  * Unimplemented IRQ types return a count of zero.
571  */
572 
573 enum {
574 	VFIO_CCW_CONFIG_REGION_INDEX,
575 	VFIO_CCW_NUM_REGIONS
576 };
577 
578 enum {
579 	VFIO_CCW_IO_IRQ_INDEX,
580 	VFIO_CCW_NUM_IRQS
581 };
582 
583 /**
584  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
585  *					      struct vfio_pci_hot_reset_info)
586  *
587  * Return: 0 on success, -errno on failure:
588  *	-enospc = insufficient buffer, -enodev = unsupported for device.
589  */
590 struct vfio_pci_dependent_device {
591 	__u32	group_id;
592 	__u16	segment;
593 	__u8	bus;
594 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
595 };
596 
597 struct vfio_pci_hot_reset_info {
598 	__u32	argsz;
599 	__u32	flags;
600 	__u32	count;
601 	struct vfio_pci_dependent_device	devices[];
602 };
603 
604 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
605 
606 /**
607  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
608  *				    struct vfio_pci_hot_reset)
609  *
610  * Return: 0 on success, -errno on failure.
611  */
612 struct vfio_pci_hot_reset {
613 	__u32	argsz;
614 	__u32	flags;
615 	__u32	count;
616 	__s32	group_fds[];
617 };
618 
619 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
620 
621 /**
622  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
623  *                                    struct vfio_device_query_gfx_plane)
624  *
625  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
626  *
627  * flags supported:
628  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
629  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
630  *   support for dma-buf.
631  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
632  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
633  *   support for region.
634  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
635  *   with each call to query the plane info.
636  * - Others are invalid and return -EINVAL.
637  *
638  * Note:
639  * 1. Plane could be disabled by guest. In that case, success will be
640  *    returned with zero-initialized drm_format, size, width and height
641  *    fields.
642  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
643  *
644  * Return: 0 on success, -errno on other failure.
645  */
646 struct vfio_device_gfx_plane_info {
647 	__u32 argsz;
648 	__u32 flags;
649 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
650 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
651 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
652 	/* in */
653 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
654 	/* out */
655 	__u32 drm_format;	/* drm format of plane */
656 	__u64 drm_format_mod;   /* tiled mode */
657 	__u32 width;	/* width of plane */
658 	__u32 height;	/* height of plane */
659 	__u32 stride;	/* stride of plane */
660 	__u32 size;	/* size of plane in bytes, align on page*/
661 	__u32 x_pos;	/* horizontal position of cursor plane */
662 	__u32 y_pos;	/* vertical position of cursor plane*/
663 	__u32 x_hot;    /* horizontal position of cursor hotspot */
664 	__u32 y_hot;    /* vertical position of cursor hotspot */
665 	union {
666 		__u32 region_index;	/* region index */
667 		__u32 dmabuf_id;	/* dma-buf id */
668 	};
669 };
670 
671 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
672 
673 /**
674  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
675  *
676  * Return a new dma-buf file descriptor for an exposed guest framebuffer
677  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
678  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
679  */
680 
681 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
682 
683 /**
684  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
685  *                              struct vfio_device_ioeventfd)
686  *
687  * Perform a write to the device at the specified device fd offset, with
688  * the specified data and width when the provided eventfd is triggered.
689  * vfio bus drivers may not support this for all regions, for all widths,
690  * or at all.  vfio-pci currently only enables support for BAR regions,
691  * excluding the MSI-X vector table.
692  *
693  * Return: 0 on success, -errno on failure.
694  */
695 struct vfio_device_ioeventfd {
696 	__u32	argsz;
697 	__u32	flags;
698 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
699 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
700 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
701 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
702 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
703 	__u64	offset;			/* device fd offset of write */
704 	__u64	data;			/* data to be written */
705 	__s32	fd;			/* -1 for de-assignment */
706 };
707 
708 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
709 
710 /**
711  * VFIO_DEVICE_FEATURE - _IORW(VFIO_TYPE, VFIO_BASE + 17,
712  *			       struct vfio_device_feature)
713  *
714  * Get, set, or probe feature data of the device.  The feature is selected
715  * using the FEATURE_MASK portion of the flags field.  Support for a feature
716  * can be probed by setting both the FEATURE_MASK and PROBE bits.  A probe
717  * may optionally include the GET and/or SET bits to determine read vs write
718  * access of the feature respectively.  Probing a feature will return success
719  * if the feature is supported and all of the optionally indicated GET/SET
720  * methods are supported.  The format of the data portion of the structure is
721  * specific to the given feature.  The data portion is not required for
722  * probing.  GET and SET are mutually exclusive, except for use with PROBE.
723  *
724  * Return 0 on success, -errno on failure.
725  */
726 struct vfio_device_feature {
727 	__u32	argsz;
728 	__u32	flags;
729 #define VFIO_DEVICE_FEATURE_MASK	(0xffff) /* 16-bit feature index */
730 #define VFIO_DEVICE_FEATURE_GET		(1 << 16) /* Get feature into data[] */
731 #define VFIO_DEVICE_FEATURE_SET		(1 << 17) /* Set feature from data[] */
732 #define VFIO_DEVICE_FEATURE_PROBE	(1 << 18) /* Probe feature support */
733 	__u8	data[];
734 };
735 
736 #define VFIO_DEVICE_FEATURE		_IO(VFIO_TYPE, VFIO_BASE + 17)
737 
738 /*
739  * Provide support for setting a PCI VF Token, which is used as a shared
740  * secret between PF and VF drivers.  This feature may only be set on a
741  * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
742  * open VFs.  Data provided when setting this feature is a 16-byte array
743  * (__u8 b[16]), representing a UUID.
744  */
745 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN	(0)
746 
747 /* -------- API for Type1 VFIO IOMMU -------- */
748 
749 /**
750  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
751  *
752  * Retrieve information about the IOMMU object. Fills in provided
753  * struct vfio_iommu_info. Caller sets argsz.
754  *
755  * XXX Should we do these by CHECK_EXTENSION too?
756  */
757 struct vfio_iommu_type1_info {
758 	__u32	argsz;
759 	__u32	flags;
760 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
761 #define VFIO_IOMMU_INFO_CAPS	(1 << 1)	/* Info supports caps */
762 	__u64	iova_pgsizes;	/* Bitmap of supported page sizes */
763 	__u32   cap_offset;	/* Offset within info struct of first cap */
764 };
765 
766 /*
767  * The IOVA capability allows to report the valid IOVA range(s)
768  * excluding any non-relaxable reserved regions exposed by
769  * devices attached to the container. Any DMA map attempt
770  * outside the valid iova range will return error.
771  *
772  * The structures below define version 1 of this capability.
773  */
774 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE  1
775 
776 struct vfio_iova_range {
777 	__u64	start;
778 	__u64	end;
779 };
780 
781 struct vfio_iommu_type1_info_cap_iova_range {
782 	struct	vfio_info_cap_header header;
783 	__u32	nr_iovas;
784 	__u32	reserved;
785 	struct	vfio_iova_range iova_ranges[];
786 };
787 
788 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
789 
790 /**
791  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
792  *
793  * Map process virtual addresses to IO virtual addresses using the
794  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
795  */
796 struct vfio_iommu_type1_dma_map {
797 	__u32	argsz;
798 	__u32	flags;
799 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
800 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
801 	__u64	vaddr;				/* Process virtual address */
802 	__u64	iova;				/* IO virtual address */
803 	__u64	size;				/* Size of mapping (bytes) */
804 };
805 
806 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
807 
808 /**
809  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
810  *							struct vfio_dma_unmap)
811  *
812  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
813  * Caller sets argsz.  The actual unmapped size is returned in the size
814  * field.  No guarantee is made to the user that arbitrary unmaps of iova
815  * or size different from those used in the original mapping call will
816  * succeed.
817  */
818 struct vfio_iommu_type1_dma_unmap {
819 	__u32	argsz;
820 	__u32	flags;
821 	__u64	iova;				/* IO virtual address */
822 	__u64	size;				/* Size of mapping (bytes) */
823 };
824 
825 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
826 
827 /*
828  * IOCTLs to enable/disable IOMMU container usage.
829  * No parameters are supported.
830  */
831 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
832 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
833 
834 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
835 
836 /*
837  * The SPAPR TCE DDW info struct provides the information about
838  * the details of Dynamic DMA window capability.
839  *
840  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
841  * @max_dynamic_windows_supported tells the maximum number of windows
842  * which the platform can create.
843  * @levels tells the maximum number of levels in multi-level IOMMU tables;
844  * this allows splitting a table into smaller chunks which reduces
845  * the amount of physically contiguous memory required for the table.
846  */
847 struct vfio_iommu_spapr_tce_ddw_info {
848 	__u64 pgsizes;			/* Bitmap of supported page sizes */
849 	__u32 max_dynamic_windows_supported;
850 	__u32 levels;
851 };
852 
853 /*
854  * The SPAPR TCE info struct provides the information about the PCI bus
855  * address ranges available for DMA, these values are programmed into
856  * the hardware so the guest has to know that information.
857  *
858  * The DMA 32 bit window start is an absolute PCI bus address.
859  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
860  * addresses too so the window works as a filter rather than an offset
861  * for IOVA addresses.
862  *
863  * Flags supported:
864  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
865  *   (DDW) support is present. @ddw is only supported when DDW is present.
866  */
867 struct vfio_iommu_spapr_tce_info {
868 	__u32 argsz;
869 	__u32 flags;
870 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
871 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
872 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
873 	struct vfio_iommu_spapr_tce_ddw_info ddw;
874 };
875 
876 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
877 
878 /*
879  * EEH PE operation struct provides ways to:
880  * - enable/disable EEH functionality;
881  * - unfreeze IO/DMA for frozen PE;
882  * - read PE state;
883  * - reset PE;
884  * - configure PE;
885  * - inject EEH error.
886  */
887 struct vfio_eeh_pe_err {
888 	__u32 type;
889 	__u32 func;
890 	__u64 addr;
891 	__u64 mask;
892 };
893 
894 struct vfio_eeh_pe_op {
895 	__u32 argsz;
896 	__u32 flags;
897 	__u32 op;
898 	union {
899 		struct vfio_eeh_pe_err err;
900 	};
901 };
902 
903 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
904 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
905 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
906 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
907 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
908 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
909 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
910 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
911 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
912 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
913 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
914 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
915 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
916 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
917 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
918 
919 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
920 
921 /**
922  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
923  *
924  * Registers user space memory where DMA is allowed. It pins
925  * user pages and does the locked memory accounting so
926  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
927  * get faster.
928  */
929 struct vfio_iommu_spapr_register_memory {
930 	__u32	argsz;
931 	__u32	flags;
932 	__u64	vaddr;				/* Process virtual address */
933 	__u64	size;				/* Size of mapping (bytes) */
934 };
935 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
936 
937 /**
938  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
939  *
940  * Unregisters user space memory registered with
941  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
942  * Uses vfio_iommu_spapr_register_memory for parameters.
943  */
944 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
945 
946 /**
947  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
948  *
949  * Creates an additional TCE table and programs it (sets a new DMA window)
950  * to every IOMMU group in the container. It receives page shift, window
951  * size and number of levels in the TCE table being created.
952  *
953  * It allocates and returns an offset on a PCI bus of the new DMA window.
954  */
955 struct vfio_iommu_spapr_tce_create {
956 	__u32 argsz;
957 	__u32 flags;
958 	/* in */
959 	__u32 page_shift;
960 	__u32 __resv1;
961 	__u64 window_size;
962 	__u32 levels;
963 	__u32 __resv2;
964 	/* out */
965 	__u64 start_addr;
966 };
967 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
968 
969 /**
970  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
971  *
972  * Unprograms a TCE table from all groups in the container and destroys it.
973  * It receives a PCI bus offset as a window id.
974  */
975 struct vfio_iommu_spapr_tce_remove {
976 	__u32 argsz;
977 	__u32 flags;
978 	/* in */
979 	__u64 start_addr;
980 };
981 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
982 
983 /* ***************************************************************** */
984 
985 #endif /* VFIO_H */
986