xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 878096ee)
1 /*
2  * VFIO API definition
3  *
4  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
5  *     Author: Alex Williamson <alex.williamson@redhat.com>
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #ifndef VFIO_H
12 #define VFIO_H
13 
14 #include <linux/types.h>
15 #include <linux/ioctl.h>
16 
17 #define VFIO_API_VERSION	0
18 
19 
20 /* Kernel & User level defines for VFIO IOCTLs. */
21 
22 /* Extensions */
23 
24 #define VFIO_TYPE1_IOMMU		1
25 
26 /*
27  * The IOCTL interface is designed for extensibility by embedding the
28  * structure length (argsz) and flags into structures passed between
29  * kernel and userspace.  We therefore use the _IO() macro for these
30  * defines to avoid implicitly embedding a size into the ioctl request.
31  * As structure fields are added, argsz will increase to match and flag
32  * bits will be defined to indicate additional fields with valid data.
33  * It's *always* the caller's responsibility to indicate the size of
34  * the structure passed by setting argsz appropriately.
35  */
36 
37 #define VFIO_TYPE	(';')
38 #define VFIO_BASE	100
39 
40 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
41 
42 /**
43  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
44  *
45  * Report the version of the VFIO API.  This allows us to bump the entire
46  * API version should we later need to add or change features in incompatible
47  * ways.
48  * Return: VFIO_API_VERSION
49  * Availability: Always
50  */
51 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
52 
53 /**
54  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
55  *
56  * Check whether an extension is supported.
57  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
58  * Availability: Always
59  */
60 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
61 
62 /**
63  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
64  *
65  * Set the iommu to the given type.  The type must be supported by an
66  * iommu driver as verified by calling CHECK_EXTENSION using the same
67  * type.  A group must be set to this file descriptor before this
68  * ioctl is available.  The IOMMU interfaces enabled by this call are
69  * specific to the value set.
70  * Return: 0 on success, -errno on failure
71  * Availability: When VFIO group attached
72  */
73 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
74 
75 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
76 
77 /**
78  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
79  *						struct vfio_group_status)
80  *
81  * Retrieve information about the group.  Fills in provided
82  * struct vfio_group_info.  Caller sets argsz.
83  * Return: 0 on succes, -errno on failure.
84  * Availability: Always
85  */
86 struct vfio_group_status {
87 	__u32	argsz;
88 	__u32	flags;
89 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
90 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
91 };
92 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
93 
94 /**
95  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
96  *
97  * Set the container for the VFIO group to the open VFIO file
98  * descriptor provided.  Groups may only belong to a single
99  * container.  Containers may, at their discretion, support multiple
100  * groups.  Only when a container is set are all of the interfaces
101  * of the VFIO file descriptor and the VFIO group file descriptor
102  * available to the user.
103  * Return: 0 on success, -errno on failure.
104  * Availability: Always
105  */
106 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
107 
108 /**
109  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
110  *
111  * Remove the group from the attached container.  This is the
112  * opposite of the SET_CONTAINER call and returns the group to
113  * an initial state.  All device file descriptors must be released
114  * prior to calling this interface.  When removing the last group
115  * from a container, the IOMMU will be disabled and all state lost,
116  * effectively also returning the VFIO file descriptor to an initial
117  * state.
118  * Return: 0 on success, -errno on failure.
119  * Availability: When attached to container
120  */
121 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
122 
123 /**
124  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
125  *
126  * Return a new file descriptor for the device object described by
127  * the provided string.  The string should match a device listed in
128  * the devices subdirectory of the IOMMU group sysfs entry.  The
129  * group containing the device must already be added to this context.
130  * Return: new file descriptor on success, -errno on failure.
131  * Availability: When attached to container
132  */
133 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
134 
135 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
136 
137 /**
138  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
139  *						struct vfio_device_info)
140  *
141  * Retrieve information about the device.  Fills in provided
142  * struct vfio_device_info.  Caller sets argsz.
143  * Return: 0 on success, -errno on failure.
144  */
145 struct vfio_device_info {
146 	__u32	argsz;
147 	__u32	flags;
148 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
149 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
150 	__u32	num_regions;	/* Max region index + 1 */
151 	__u32	num_irqs;	/* Max IRQ index + 1 */
152 };
153 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
154 
155 /**
156  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
157  *				       struct vfio_region_info)
158  *
159  * Retrieve information about a device region.  Caller provides
160  * struct vfio_region_info with index value set.  Caller sets argsz.
161  * Implementation of region mapping is bus driver specific.  This is
162  * intended to describe MMIO, I/O port, as well as bus specific
163  * regions (ex. PCI config space).  Zero sized regions may be used
164  * to describe unimplemented regions (ex. unimplemented PCI BARs).
165  * Return: 0 on success, -errno on failure.
166  */
167 struct vfio_region_info {
168 	__u32	argsz;
169 	__u32	flags;
170 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
171 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
172 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
173 	__u32	index;		/* Region index */
174 	__u32	resv;		/* Reserved for alignment */
175 	__u64	size;		/* Region size (bytes) */
176 	__u64	offset;		/* Region offset from start of device fd */
177 };
178 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
179 
180 /**
181  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
182  *				    struct vfio_irq_info)
183  *
184  * Retrieve information about a device IRQ.  Caller provides
185  * struct vfio_irq_info with index value set.  Caller sets argsz.
186  * Implementation of IRQ mapping is bus driver specific.  Indexes
187  * using multiple IRQs are primarily intended to support MSI-like
188  * interrupt blocks.  Zero count irq blocks may be used to describe
189  * unimplemented interrupt types.
190  *
191  * The EVENTFD flag indicates the interrupt index supports eventfd based
192  * signaling.
193  *
194  * The MASKABLE flags indicates the index supports MASK and UNMASK
195  * actions described below.
196  *
197  * AUTOMASKED indicates that after signaling, the interrupt line is
198  * automatically masked by VFIO and the user needs to unmask the line
199  * to receive new interrupts.  This is primarily intended to distinguish
200  * level triggered interrupts.
201  *
202  * The NORESIZE flag indicates that the interrupt lines within the index
203  * are setup as a set and new subindexes cannot be enabled without first
204  * disabling the entire index.  This is used for interrupts like PCI MSI
205  * and MSI-X where the driver may only use a subset of the available
206  * indexes, but VFIO needs to enable a specific number of vectors
207  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
208  * then add and unmask vectors, it's up to userspace to make the decision
209  * whether to allocate the maximum supported number of vectors or tear
210  * down setup and incrementally increase the vectors as each is enabled.
211  */
212 struct vfio_irq_info {
213 	__u32	argsz;
214 	__u32	flags;
215 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
216 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
217 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
218 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
219 	__u32	index;		/* IRQ index */
220 	__u32	count;		/* Number of IRQs within this index */
221 };
222 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
223 
224 /**
225  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
226  *
227  * Set signaling, masking, and unmasking of interrupts.  Caller provides
228  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
229  * the range of subindexes being specified.
230  *
231  * The DATA flags specify the type of data provided.  If DATA_NONE, the
232  * operation performs the specified action immediately on the specified
233  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
234  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
235  *
236  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
237  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
238  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
239  * data = {1,0,1}
240  *
241  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
242  * A value of -1 can be used to either de-assign interrupts if already
243  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
244  * to be trigger for interrupts [0,0] and [0,2]:
245  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
246  * data = {fd1, -1, fd2}
247  * If index [0,1] is previously set, two count = 1 ioctls calls would be
248  * required to set [0,0] and [0,2] without changing [0,1].
249  *
250  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
251  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
252  * from userspace (ie. simulate hardware triggering).
253  *
254  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
255  * enables the interrupt index for the device.  Individual subindex interrupts
256  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
257  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
258  *
259  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
260  * ACTION_TRIGGER specifies kernel->user signaling.
261  */
262 struct vfio_irq_set {
263 	__u32	argsz;
264 	__u32	flags;
265 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
266 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
267 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
268 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
269 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
270 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
271 	__u32	index;
272 	__u32	start;
273 	__u32	count;
274 	__u8	data[];
275 };
276 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
277 
278 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
279 					 VFIO_IRQ_SET_DATA_BOOL | \
280 					 VFIO_IRQ_SET_DATA_EVENTFD)
281 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
282 					 VFIO_IRQ_SET_ACTION_UNMASK | \
283 					 VFIO_IRQ_SET_ACTION_TRIGGER)
284 /**
285  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
286  *
287  * Reset a device.
288  */
289 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
290 
291 /*
292  * The VFIO-PCI bus driver makes use of the following fixed region and
293  * IRQ index mapping.  Unimplemented regions return a size of zero.
294  * Unimplemented IRQ types return a count of zero.
295  */
296 
297 enum {
298 	VFIO_PCI_BAR0_REGION_INDEX,
299 	VFIO_PCI_BAR1_REGION_INDEX,
300 	VFIO_PCI_BAR2_REGION_INDEX,
301 	VFIO_PCI_BAR3_REGION_INDEX,
302 	VFIO_PCI_BAR4_REGION_INDEX,
303 	VFIO_PCI_BAR5_REGION_INDEX,
304 	VFIO_PCI_ROM_REGION_INDEX,
305 	VFIO_PCI_CONFIG_REGION_INDEX,
306 	/*
307 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
308 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
309 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
310 	 * range is found at it's identity mapped offset from the region
311 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
312 	 * between described ranges are unimplemented.
313 	 */
314 	VFIO_PCI_VGA_REGION_INDEX,
315 	VFIO_PCI_NUM_REGIONS
316 };
317 
318 enum {
319 	VFIO_PCI_INTX_IRQ_INDEX,
320 	VFIO_PCI_MSI_IRQ_INDEX,
321 	VFIO_PCI_MSIX_IRQ_INDEX,
322 	VFIO_PCI_ERR_IRQ_INDEX,
323 	VFIO_PCI_NUM_IRQS
324 };
325 
326 /* -------- API for Type1 VFIO IOMMU -------- */
327 
328 /**
329  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
330  *
331  * Retrieve information about the IOMMU object. Fills in provided
332  * struct vfio_iommu_info. Caller sets argsz.
333  *
334  * XXX Should we do these by CHECK_EXTENSION too?
335  */
336 struct vfio_iommu_type1_info {
337 	__u32	argsz;
338 	__u32	flags;
339 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
340 	__u64	iova_pgsizes;		/* Bitmap of supported page sizes */
341 };
342 
343 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
344 
345 /**
346  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
347  *
348  * Map process virtual addresses to IO virtual addresses using the
349  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
350  */
351 struct vfio_iommu_type1_dma_map {
352 	__u32	argsz;
353 	__u32	flags;
354 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
355 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
356 	__u64	vaddr;				/* Process virtual address */
357 	__u64	iova;				/* IO virtual address */
358 	__u64	size;				/* Size of mapping (bytes) */
359 };
360 
361 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
362 
363 /**
364  * VFIO_IOMMU_UNMAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 14, struct vfio_dma_unmap)
365  *
366  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
367  * Caller sets argsz.
368  */
369 struct vfio_iommu_type1_dma_unmap {
370 	__u32	argsz;
371 	__u32	flags;
372 	__u64	iova;				/* IO virtual address */
373 	__u64	size;				/* Size of mapping (bytes) */
374 };
375 
376 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
377 
378 #endif /* VFIO_H */
379