• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 _UAPIVFIO_H
13 #define _UAPIVFIO_H
14 
15 #include <linux/types.h>
16 #include <linux/ioctl.h>
17 
18 #define VFIO_API_VERSION	0
19 
20 
21 /* Kernel & User level defines for VFIO IOCTLs. */
22 
23 /* Extensions */
24 
25 #define VFIO_TYPE1_IOMMU		1
26 #define VFIO_SPAPR_TCE_IOMMU		2
27 #define VFIO_TYPE1v2_IOMMU		3
28 /*
29  * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping).  This
30  * capability is subject to change as groups are added or removed.
31  */
32 #define VFIO_DMA_CC_IOMMU		4
33 
34 /* Check if EEH is supported */
35 #define VFIO_EEH			5
36 
37 /* Two-stage IOMMU */
38 #define VFIO_TYPE1_NESTING_IOMMU	6	/* Implies v2 */
39 
40 #define VFIO_SPAPR_TCE_v2_IOMMU		7
41 
42 /*
43  * The No-IOMMU IOMMU offers no translation or isolation for devices and
44  * supports no ioctls outside of VFIO_CHECK_EXTENSION.  Use of VFIO's No-IOMMU
45  * code will taint the host kernel and should be used with extreme caution.
46  */
47 #define VFIO_NOIOMMU_IOMMU		8
48 
49 /*
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 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
299 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
300 
301 /* 8086 Vendor sub-types */
302 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
303 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
304 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
305 
306 #define VFIO_REGION_TYPE_GFX                    (1)
307 #define VFIO_REGION_SUBTYPE_GFX_EDID            (1)
308 
309 /**
310  * struct vfio_region_gfx_edid - EDID region layout.
311  *
312  * Set display link state and EDID blob.
313  *
314  * The EDID blob has monitor information such as brand, name, serial
315  * number, physical size, supported video modes and more.
316  *
317  * This special region allows userspace (typically qemu) set a virtual
318  * EDID for the virtual monitor, which allows a flexible display
319  * configuration.
320  *
321  * For the edid blob spec look here:
322  *    https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
323  *
324  * On linux systems you can find the EDID blob in sysfs:
325  *    /sys/class/drm/${card}/${connector}/edid
326  *
327  * You can use the edid-decode ulility (comes with xorg-x11-utils) to
328  * decode the EDID blob.
329  *
330  * @edid_offset: location of the edid blob, relative to the
331  *               start of the region (readonly).
332  * @edid_max_size: max size of the edid blob (readonly).
333  * @edid_size: actual edid size (read/write).
334  * @link_state: display link state (read/write).
335  * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
336  * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
337  * @max_xres: max display width (0 == no limitation, readonly).
338  * @max_yres: max display height (0 == no limitation, readonly).
339  *
340  * EDID update protocol:
341  *   (1) set link-state to down.
342  *   (2) update edid blob and size.
343  *   (3) set link-state to up.
344  */
345 struct vfio_region_gfx_edid {
346 	__u32 edid_offset;
347 	__u32 edid_max_size;
348 	__u32 edid_size;
349 	__u32 max_xres;
350 	__u32 max_yres;
351 	__u32 link_state;
352 #define VFIO_DEVICE_GFX_LINK_STATE_UP    1
353 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN  2
354 };
355 
356 /*
357  * 10de vendor sub-type
358  *
359  * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
360  */
361 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM	(1)
362 
363 /*
364  * 1014 vendor sub-type
365  *
366  * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
367  * to do TLB invalidation on a GPU.
368  */
369 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD	(1)
370 
371 /*
372  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
373  * which allows direct access to non-MSIX registers which happened to be within
374  * the same system page.
375  *
376  * Even though the userspace gets direct access to the MSIX data, the existing
377  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
378  */
379 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE	3
380 
381 /*
382  * Capability with compressed real address (aka SSA - small system address)
383  * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
384  * and by the userspace to associate a NVLink bridge with a GPU.
385  */
386 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT	4
387 
388 struct vfio_region_info_cap_nvlink2_ssatgt {
389 	struct vfio_info_cap_header header;
390 	__u64 tgt;
391 };
392 
393 /*
394  * Capability with an NVLink link speed. The value is read by
395  * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
396  * property in the device tree. The value is fixed in the hardware
397  * and failing to provide the correct value results in the link
398  * not working with no indication from the driver why.
399  */
400 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD	5
401 
402 struct vfio_region_info_cap_nvlink2_lnkspd {
403 	struct vfio_info_cap_header header;
404 	__u32 link_speed;
405 	__u32 __pad;
406 };
407 
408 /**
409  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
410  *				    struct vfio_irq_info)
411  *
412  * Retrieve information about a device IRQ.  Caller provides
413  * struct vfio_irq_info with index value set.  Caller sets argsz.
414  * Implementation of IRQ mapping is bus driver specific.  Indexes
415  * using multiple IRQs are primarily intended to support MSI-like
416  * interrupt blocks.  Zero count irq blocks may be used to describe
417  * unimplemented interrupt types.
418  *
419  * The EVENTFD flag indicates the interrupt index supports eventfd based
420  * signaling.
421  *
422  * The MASKABLE flags indicates the index supports MASK and UNMASK
423  * actions described below.
424  *
425  * AUTOMASKED indicates that after signaling, the interrupt line is
426  * automatically masked by VFIO and the user needs to unmask the line
427  * to receive new interrupts.  This is primarily intended to distinguish
428  * level triggered interrupts.
429  *
430  * The NORESIZE flag indicates that the interrupt lines within the index
431  * are setup as a set and new subindexes cannot be enabled without first
432  * disabling the entire index.  This is used for interrupts like PCI MSI
433  * and MSI-X where the driver may only use a subset of the available
434  * indexes, but VFIO needs to enable a specific number of vectors
435  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
436  * then add and unmask vectors, it's up to userspace to make the decision
437  * whether to allocate the maximum supported number of vectors or tear
438  * down setup and incrementally increase the vectors as each is enabled.
439  */
440 struct vfio_irq_info {
441 	__u32	argsz;
442 	__u32	flags;
443 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
444 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
445 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
446 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
447 	__u32	index;		/* IRQ index */
448 	__u32	count;		/* Number of IRQs within this index */
449 };
450 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
451 
452 /**
453  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
454  *
455  * Set signaling, masking, and unmasking of interrupts.  Caller provides
456  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
457  * the range of subindexes being specified.
458  *
459  * The DATA flags specify the type of data provided.  If DATA_NONE, the
460  * operation performs the specified action immediately on the specified
461  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
462  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
463  *
464  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
465  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
466  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
467  * data = {1,0,1}
468  *
469  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
470  * A value of -1 can be used to either de-assign interrupts if already
471  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
472  * to be trigger for interrupts [0,0] and [0,2]:
473  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
474  * data = {fd1, -1, fd2}
475  * If index [0,1] is previously set, two count = 1 ioctls calls would be
476  * required to set [0,0] and [0,2] without changing [0,1].
477  *
478  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
479  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
480  * from userspace (ie. simulate hardware triggering).
481  *
482  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
483  * enables the interrupt index for the device.  Individual subindex interrupts
484  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
485  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
486  *
487  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
488  * ACTION_TRIGGER specifies kernel->user signaling.
489  */
490 struct vfio_irq_set {
491 	__u32	argsz;
492 	__u32	flags;
493 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
494 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
495 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
496 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
497 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
498 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
499 	__u32	index;
500 	__u32	start;
501 	__u32	count;
502 	__u8	data[];
503 };
504 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
505 
506 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
507 					 VFIO_IRQ_SET_DATA_BOOL | \
508 					 VFIO_IRQ_SET_DATA_EVENTFD)
509 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
510 					 VFIO_IRQ_SET_ACTION_UNMASK | \
511 					 VFIO_IRQ_SET_ACTION_TRIGGER)
512 /**
513  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
514  *
515  * Reset a device.
516  */
517 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
518 
519 /*
520  * The VFIO-PCI bus driver makes use of the following fixed region and
521  * IRQ index mapping.  Unimplemented regions return a size of zero.
522  * Unimplemented IRQ types return a count of zero.
523  */
524 
525 enum {
526 	VFIO_PCI_BAR0_REGION_INDEX,
527 	VFIO_PCI_BAR1_REGION_INDEX,
528 	VFIO_PCI_BAR2_REGION_INDEX,
529 	VFIO_PCI_BAR3_REGION_INDEX,
530 	VFIO_PCI_BAR4_REGION_INDEX,
531 	VFIO_PCI_BAR5_REGION_INDEX,
532 	VFIO_PCI_ROM_REGION_INDEX,
533 	VFIO_PCI_CONFIG_REGION_INDEX,
534 	/*
535 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
536 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
537 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
538 	 * range is found at it's identity mapped offset from the region
539 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
540 	 * between described ranges are unimplemented.
541 	 */
542 	VFIO_PCI_VGA_REGION_INDEX,
543 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
544 				 /* device specific cap to define content. */
545 };
546 
547 enum {
548 	VFIO_PCI_INTX_IRQ_INDEX,
549 	VFIO_PCI_MSI_IRQ_INDEX,
550 	VFIO_PCI_MSIX_IRQ_INDEX,
551 	VFIO_PCI_ERR_IRQ_INDEX,
552 	VFIO_PCI_REQ_IRQ_INDEX,
553 	VFIO_PCI_NUM_IRQS
554 };
555 
556 /*
557  * The vfio-ccw bus driver makes use of the following fixed region and
558  * IRQ index mapping. Unimplemented regions return a size of zero.
559  * Unimplemented IRQ types return a count of zero.
560  */
561 
562 enum {
563 	VFIO_CCW_CONFIG_REGION_INDEX,
564 	VFIO_CCW_NUM_REGIONS
565 };
566 
567 enum {
568 	VFIO_CCW_IO_IRQ_INDEX,
569 	VFIO_CCW_NUM_IRQS
570 };
571 
572 /**
573  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
574  *					      struct vfio_pci_hot_reset_info)
575  *
576  * Return: 0 on success, -errno on failure:
577  *	-enospc = insufficient buffer, -enodev = unsupported for device.
578  */
579 struct vfio_pci_dependent_device {
580 	__u32	group_id;
581 	__u16	segment;
582 	__u8	bus;
583 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
584 };
585 
586 struct vfio_pci_hot_reset_info {
587 	__u32	argsz;
588 	__u32	flags;
589 	__u32	count;
590 	struct vfio_pci_dependent_device	devices[];
591 };
592 
593 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
594 
595 /**
596  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
597  *				    struct vfio_pci_hot_reset)
598  *
599  * Return: 0 on success, -errno on failure.
600  */
601 struct vfio_pci_hot_reset {
602 	__u32	argsz;
603 	__u32	flags;
604 	__u32	count;
605 	__s32	group_fds[];
606 };
607 
608 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
609 
610 /**
611  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
612  *                                    struct vfio_device_query_gfx_plane)
613  *
614  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
615  *
616  * flags supported:
617  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
618  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
619  *   support for dma-buf.
620  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
621  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
622  *   support for region.
623  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
624  *   with each call to query the plane info.
625  * - Others are invalid and return -EINVAL.
626  *
627  * Note:
628  * 1. Plane could be disabled by guest. In that case, success will be
629  *    returned with zero-initialized drm_format, size, width and height
630  *    fields.
631  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
632  *
633  * Return: 0 on success, -errno on other failure.
634  */
635 struct vfio_device_gfx_plane_info {
636 	__u32 argsz;
637 	__u32 flags;
638 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
639 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
640 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
641 	/* in */
642 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
643 	/* out */
644 	__u32 drm_format;	/* drm format of plane */
645 	__u64 drm_format_mod;   /* tiled mode */
646 	__u32 width;	/* width of plane */
647 	__u32 height;	/* height of plane */
648 	__u32 stride;	/* stride of plane */
649 	__u32 size;	/* size of plane in bytes, align on page*/
650 	__u32 x_pos;	/* horizontal position of cursor plane */
651 	__u32 y_pos;	/* vertical position of cursor plane*/
652 	__u32 x_hot;    /* horizontal position of cursor hotspot */
653 	__u32 y_hot;    /* vertical position of cursor hotspot */
654 	union {
655 		__u32 region_index;	/* region index */
656 		__u32 dmabuf_id;	/* dma-buf id */
657 	};
658 };
659 
660 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
661 
662 /**
663  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
664  *
665  * Return a new dma-buf file descriptor for an exposed guest framebuffer
666  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
667  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
668  */
669 
670 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
671 
672 /**
673  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
674  *                              struct vfio_device_ioeventfd)
675  *
676  * Perform a write to the device at the specified device fd offset, with
677  * the specified data and width when the provided eventfd is triggered.
678  * vfio bus drivers may not support this for all regions, for all widths,
679  * or at all.  vfio-pci currently only enables support for BAR regions,
680  * excluding the MSI-X vector table.
681  *
682  * Return: 0 on success, -errno on failure.
683  */
684 struct vfio_device_ioeventfd {
685 	__u32	argsz;
686 	__u32	flags;
687 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
688 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
689 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
690 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
691 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
692 	__u64	offset;			/* device fd offset of write */
693 	__u64	data;			/* data to be written */
694 	__s32	fd;			/* -1 for de-assignment */
695 };
696 
697 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
698 
699 /* -------- API for Type1 VFIO IOMMU -------- */
700 
701 /**
702  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
703  *
704  * Retrieve information about the IOMMU object. Fills in provided
705  * struct vfio_iommu_info. Caller sets argsz.
706  *
707  * XXX Should we do these by CHECK_EXTENSION too?
708  */
709 struct vfio_iommu_type1_info {
710 	__u32	argsz;
711 	__u32	flags;
712 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
713 	__u64	iova_pgsizes;		/* Bitmap of supported page sizes */
714 };
715 
716 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
717 
718 /**
719  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
720  *
721  * Map process virtual addresses to IO virtual addresses using the
722  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
723  */
724 struct vfio_iommu_type1_dma_map {
725 	__u32	argsz;
726 	__u32	flags;
727 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
728 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
729 	__u64	vaddr;				/* Process virtual address */
730 	__u64	iova;				/* IO virtual address */
731 	__u64	size;				/* Size of mapping (bytes) */
732 };
733 
734 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
735 
736 /**
737  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
738  *							struct vfio_dma_unmap)
739  *
740  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
741  * Caller sets argsz.  The actual unmapped size is returned in the size
742  * field.  No guarantee is made to the user that arbitrary unmaps of iova
743  * or size different from those used in the original mapping call will
744  * succeed.
745  */
746 struct vfio_iommu_type1_dma_unmap {
747 	__u32	argsz;
748 	__u32	flags;
749 	__u64	iova;				/* IO virtual address */
750 	__u64	size;				/* Size of mapping (bytes) */
751 };
752 
753 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
754 
755 /*
756  * IOCTLs to enable/disable IOMMU container usage.
757  * No parameters are supported.
758  */
759 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
760 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
761 
762 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
763 
764 /*
765  * The SPAPR TCE DDW info struct provides the information about
766  * the details of Dynamic DMA window capability.
767  *
768  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
769  * @max_dynamic_windows_supported tells the maximum number of windows
770  * which the platform can create.
771  * @levels tells the maximum number of levels in multi-level IOMMU tables;
772  * this allows splitting a table into smaller chunks which reduces
773  * the amount of physically contiguous memory required for the table.
774  */
775 struct vfio_iommu_spapr_tce_ddw_info {
776 	__u64 pgsizes;			/* Bitmap of supported page sizes */
777 	__u32 max_dynamic_windows_supported;
778 	__u32 levels;
779 };
780 
781 /*
782  * The SPAPR TCE info struct provides the information about the PCI bus
783  * address ranges available for DMA, these values are programmed into
784  * the hardware so the guest has to know that information.
785  *
786  * The DMA 32 bit window start is an absolute PCI bus address.
787  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
788  * addresses too so the window works as a filter rather than an offset
789  * for IOVA addresses.
790  *
791  * Flags supported:
792  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
793  *   (DDW) support is present. @ddw is only supported when DDW is present.
794  */
795 struct vfio_iommu_spapr_tce_info {
796 	__u32 argsz;
797 	__u32 flags;
798 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
799 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
800 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
801 	struct vfio_iommu_spapr_tce_ddw_info ddw;
802 };
803 
804 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
805 
806 /*
807  * EEH PE operation struct provides ways to:
808  * - enable/disable EEH functionality;
809  * - unfreeze IO/DMA for frozen PE;
810  * - read PE state;
811  * - reset PE;
812  * - configure PE;
813  * - inject EEH error.
814  */
815 struct vfio_eeh_pe_err {
816 	__u32 type;
817 	__u32 func;
818 	__u64 addr;
819 	__u64 mask;
820 };
821 
822 struct vfio_eeh_pe_op {
823 	__u32 argsz;
824 	__u32 flags;
825 	__u32 op;
826 	union {
827 		struct vfio_eeh_pe_err err;
828 	};
829 };
830 
831 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
832 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
833 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
834 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
835 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
836 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
837 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
838 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
839 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
840 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
841 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
842 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
843 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
844 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
845 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
846 
847 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
848 
849 /**
850  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
851  *
852  * Registers user space memory where DMA is allowed. It pins
853  * user pages and does the locked memory accounting so
854  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
855  * get faster.
856  */
857 struct vfio_iommu_spapr_register_memory {
858 	__u32	argsz;
859 	__u32	flags;
860 	__u64	vaddr;				/* Process virtual address */
861 	__u64	size;				/* Size of mapping (bytes) */
862 };
863 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
864 
865 /**
866  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
867  *
868  * Unregisters user space memory registered with
869  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
870  * Uses vfio_iommu_spapr_register_memory for parameters.
871  */
872 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
873 
874 /**
875  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
876  *
877  * Creates an additional TCE table and programs it (sets a new DMA window)
878  * to every IOMMU group in the container. It receives page shift, window
879  * size and number of levels in the TCE table being created.
880  *
881  * It allocates and returns an offset on a PCI bus of the new DMA window.
882  */
883 struct vfio_iommu_spapr_tce_create {
884 	__u32 argsz;
885 	__u32 flags;
886 	/* in */
887 	__u32 page_shift;
888 	__u32 __resv1;
889 	__u64 window_size;
890 	__u32 levels;
891 	__u32 __resv2;
892 	/* out */
893 	__u64 start_addr;
894 };
895 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
896 
897 /**
898  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
899  *
900  * Unprograms a TCE table from all groups in the container and destroys it.
901  * It receives a PCI bus offset as a window id.
902  */
903 struct vfio_iommu_spapr_tce_remove {
904 	__u32 argsz;
905 	__u32 flags;
906 	/* in */
907 	__u64 start_addr;
908 };
909 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
910 
911 /* ***************************************************************** */
912 
913 #endif /* _UAPIVFIO_H */
914