xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 8f1e884b)
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 #define VFIO_SPAPR_TCE_IOMMU		2
26 #define VFIO_TYPE1v2_IOMMU		3
27 /*
28  * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping).  This
29  * capability is subject to change as groups are added or removed.
30  */
31 #define VFIO_DMA_CC_IOMMU		4
32 
33 /*
34  * The IOCTL interface is designed for extensibility by embedding the
35  * structure length (argsz) and flags into structures passed between
36  * kernel and userspace.  We therefore use the _IO() macro for these
37  * defines to avoid implicitly embedding a size into the ioctl request.
38  * As structure fields are added, argsz will increase to match and flag
39  * bits will be defined to indicate additional fields with valid data.
40  * It's *always* the caller's responsibility to indicate the size of
41  * the structure passed by setting argsz appropriately.
42  */
43 
44 #define VFIO_TYPE	(';')
45 #define VFIO_BASE	100
46 
47 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
48 
49 /**
50  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
51  *
52  * Report the version of the VFIO API.  This allows us to bump the entire
53  * API version should we later need to add or change features in incompatible
54  * ways.
55  * Return: VFIO_API_VERSION
56  * Availability: Always
57  */
58 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
59 
60 /**
61  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
62  *
63  * Check whether an extension is supported.
64  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
65  * Availability: Always
66  */
67 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
68 
69 /**
70  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
71  *
72  * Set the iommu to the given type.  The type must be supported by an
73  * iommu driver as verified by calling CHECK_EXTENSION using the same
74  * type.  A group must be set to this file descriptor before this
75  * ioctl is available.  The IOMMU interfaces enabled by this call are
76  * specific to the value set.
77  * Return: 0 on success, -errno on failure
78  * Availability: When VFIO group attached
79  */
80 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
81 
82 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
83 
84 /**
85  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
86  *						struct vfio_group_status)
87  *
88  * Retrieve information about the group.  Fills in provided
89  * struct vfio_group_info.  Caller sets argsz.
90  * Return: 0 on succes, -errno on failure.
91  * Availability: Always
92  */
93 struct vfio_group_status {
94 	__u32	argsz;
95 	__u32	flags;
96 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
97 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
98 };
99 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
100 
101 /**
102  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
103  *
104  * Set the container for the VFIO group to the open VFIO file
105  * descriptor provided.  Groups may only belong to a single
106  * container.  Containers may, at their discretion, support multiple
107  * groups.  Only when a container is set are all of the interfaces
108  * of the VFIO file descriptor and the VFIO group file descriptor
109  * available to the user.
110  * Return: 0 on success, -errno on failure.
111  * Availability: Always
112  */
113 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
114 
115 /**
116  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
117  *
118  * Remove the group from the attached container.  This is the
119  * opposite of the SET_CONTAINER call and returns the group to
120  * an initial state.  All device file descriptors must be released
121  * prior to calling this interface.  When removing the last group
122  * from a container, the IOMMU will be disabled and all state lost,
123  * effectively also returning the VFIO file descriptor to an initial
124  * state.
125  * Return: 0 on success, -errno on failure.
126  * Availability: When attached to container
127  */
128 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
129 
130 /**
131  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
132  *
133  * Return a new file descriptor for the device object described by
134  * the provided string.  The string should match a device listed in
135  * the devices subdirectory of the IOMMU group sysfs entry.  The
136  * group containing the device must already be added to this context.
137  * Return: new file descriptor on success, -errno on failure.
138  * Availability: When attached to container
139  */
140 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
141 
142 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
143 
144 /**
145  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
146  *						struct vfio_device_info)
147  *
148  * Retrieve information about the device.  Fills in provided
149  * struct vfio_device_info.  Caller sets argsz.
150  * Return: 0 on success, -errno on failure.
151  */
152 struct vfio_device_info {
153 	__u32	argsz;
154 	__u32	flags;
155 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
156 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
157 	__u32	num_regions;	/* Max region index + 1 */
158 	__u32	num_irqs;	/* Max IRQ index + 1 */
159 };
160 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
161 
162 /**
163  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
164  *				       struct vfio_region_info)
165  *
166  * Retrieve information about a device region.  Caller provides
167  * struct vfio_region_info with index value set.  Caller sets argsz.
168  * Implementation of region mapping is bus driver specific.  This is
169  * intended to describe MMIO, I/O port, as well as bus specific
170  * regions (ex. PCI config space).  Zero sized regions may be used
171  * to describe unimplemented regions (ex. unimplemented PCI BARs).
172  * Return: 0 on success, -errno on failure.
173  */
174 struct vfio_region_info {
175 	__u32	argsz;
176 	__u32	flags;
177 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
178 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
179 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
180 	__u32	index;		/* Region index */
181 	__u32	resv;		/* Reserved for alignment */
182 	__u64	size;		/* Region size (bytes) */
183 	__u64	offset;		/* Region offset from start of device fd */
184 };
185 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
186 
187 /**
188  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
189  *				    struct vfio_irq_info)
190  *
191  * Retrieve information about a device IRQ.  Caller provides
192  * struct vfio_irq_info with index value set.  Caller sets argsz.
193  * Implementation of IRQ mapping is bus driver specific.  Indexes
194  * using multiple IRQs are primarily intended to support MSI-like
195  * interrupt blocks.  Zero count irq blocks may be used to describe
196  * unimplemented interrupt types.
197  *
198  * The EVENTFD flag indicates the interrupt index supports eventfd based
199  * signaling.
200  *
201  * The MASKABLE flags indicates the index supports MASK and UNMASK
202  * actions described below.
203  *
204  * AUTOMASKED indicates that after signaling, the interrupt line is
205  * automatically masked by VFIO and the user needs to unmask the line
206  * to receive new interrupts.  This is primarily intended to distinguish
207  * level triggered interrupts.
208  *
209  * The NORESIZE flag indicates that the interrupt lines within the index
210  * are setup as a set and new subindexes cannot be enabled without first
211  * disabling the entire index.  This is used for interrupts like PCI MSI
212  * and MSI-X where the driver may only use a subset of the available
213  * indexes, but VFIO needs to enable a specific number of vectors
214  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
215  * then add and unmask vectors, it's up to userspace to make the decision
216  * whether to allocate the maximum supported number of vectors or tear
217  * down setup and incrementally increase the vectors as each is enabled.
218  */
219 struct vfio_irq_info {
220 	__u32	argsz;
221 	__u32	flags;
222 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
223 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
224 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
225 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
226 	__u32	index;		/* IRQ index */
227 	__u32	count;		/* Number of IRQs within this index */
228 };
229 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
230 
231 /**
232  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
233  *
234  * Set signaling, masking, and unmasking of interrupts.  Caller provides
235  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
236  * the range of subindexes being specified.
237  *
238  * The DATA flags specify the type of data provided.  If DATA_NONE, the
239  * operation performs the specified action immediately on the specified
240  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
241  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
242  *
243  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
244  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
245  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
246  * data = {1,0,1}
247  *
248  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
249  * A value of -1 can be used to either de-assign interrupts if already
250  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
251  * to be trigger for interrupts [0,0] and [0,2]:
252  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
253  * data = {fd1, -1, fd2}
254  * If index [0,1] is previously set, two count = 1 ioctls calls would be
255  * required to set [0,0] and [0,2] without changing [0,1].
256  *
257  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
258  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
259  * from userspace (ie. simulate hardware triggering).
260  *
261  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
262  * enables the interrupt index for the device.  Individual subindex interrupts
263  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
264  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
265  *
266  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
267  * ACTION_TRIGGER specifies kernel->user signaling.
268  */
269 struct vfio_irq_set {
270 	__u32	argsz;
271 	__u32	flags;
272 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
273 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
274 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
275 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
276 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
277 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
278 	__u32	index;
279 	__u32	start;
280 	__u32	count;
281 	__u8	data[];
282 };
283 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
284 
285 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
286 					 VFIO_IRQ_SET_DATA_BOOL | \
287 					 VFIO_IRQ_SET_DATA_EVENTFD)
288 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
289 					 VFIO_IRQ_SET_ACTION_UNMASK | \
290 					 VFIO_IRQ_SET_ACTION_TRIGGER)
291 /**
292  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
293  *
294  * Reset a device.
295  */
296 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
297 
298 /*
299  * The VFIO-PCI bus driver makes use of the following fixed region and
300  * IRQ index mapping.  Unimplemented regions return a size of zero.
301  * Unimplemented IRQ types return a count of zero.
302  */
303 
304 enum {
305 	VFIO_PCI_BAR0_REGION_INDEX,
306 	VFIO_PCI_BAR1_REGION_INDEX,
307 	VFIO_PCI_BAR2_REGION_INDEX,
308 	VFIO_PCI_BAR3_REGION_INDEX,
309 	VFIO_PCI_BAR4_REGION_INDEX,
310 	VFIO_PCI_BAR5_REGION_INDEX,
311 	VFIO_PCI_ROM_REGION_INDEX,
312 	VFIO_PCI_CONFIG_REGION_INDEX,
313 	/*
314 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
315 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
316 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
317 	 * range is found at it's identity mapped offset from the region
318 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
319 	 * between described ranges are unimplemented.
320 	 */
321 	VFIO_PCI_VGA_REGION_INDEX,
322 	VFIO_PCI_NUM_REGIONS
323 };
324 
325 enum {
326 	VFIO_PCI_INTX_IRQ_INDEX,
327 	VFIO_PCI_MSI_IRQ_INDEX,
328 	VFIO_PCI_MSIX_IRQ_INDEX,
329 	VFIO_PCI_ERR_IRQ_INDEX,
330 	VFIO_PCI_NUM_IRQS
331 };
332 
333 /**
334  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
335  *					      struct vfio_pci_hot_reset_info)
336  *
337  * Return: 0 on success, -errno on failure:
338  *	-enospc = insufficient buffer, -enodev = unsupported for device.
339  */
340 struct vfio_pci_dependent_device {
341 	__u32	group_id;
342 	__u16	segment;
343 	__u8	bus;
344 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
345 };
346 
347 struct vfio_pci_hot_reset_info {
348 	__u32	argsz;
349 	__u32	flags;
350 	__u32	count;
351 	struct vfio_pci_dependent_device	devices[];
352 };
353 
354 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
355 
356 /**
357  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
358  *				    struct vfio_pci_hot_reset)
359  *
360  * Return: 0 on success, -errno on failure.
361  */
362 struct vfio_pci_hot_reset {
363 	__u32	argsz;
364 	__u32	flags;
365 	__u32	count;
366 	__s32	group_fds[];
367 };
368 
369 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
370 
371 /* -------- API for Type1 VFIO IOMMU -------- */
372 
373 /**
374  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
375  *
376  * Retrieve information about the IOMMU object. Fills in provided
377  * struct vfio_iommu_info. Caller sets argsz.
378  *
379  * XXX Should we do these by CHECK_EXTENSION too?
380  */
381 struct vfio_iommu_type1_info {
382 	__u32	argsz;
383 	__u32	flags;
384 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
385 	__u64	iova_pgsizes;		/* Bitmap of supported page sizes */
386 };
387 
388 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
389 
390 /**
391  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
392  *
393  * Map process virtual addresses to IO virtual addresses using the
394  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
395  */
396 struct vfio_iommu_type1_dma_map {
397 	__u32	argsz;
398 	__u32	flags;
399 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
400 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
401 	__u64	vaddr;				/* Process virtual address */
402 	__u64	iova;				/* IO virtual address */
403 	__u64	size;				/* Size of mapping (bytes) */
404 };
405 
406 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
407 
408 /**
409  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
410  *							struct vfio_dma_unmap)
411  *
412  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
413  * Caller sets argsz.  The actual unmapped size is returned in the size
414  * field.  No guarantee is made to the user that arbitrary unmaps of iova
415  * or size different from those used in the original mapping call will
416  * succeed.
417  */
418 struct vfio_iommu_type1_dma_unmap {
419 	__u32	argsz;
420 	__u32	flags;
421 	__u64	iova;				/* IO virtual address */
422 	__u64	size;				/* Size of mapping (bytes) */
423 };
424 
425 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
426 
427 /*
428  * IOCTLs to enable/disable IOMMU container usage.
429  * No parameters are supported.
430  */
431 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
432 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
433 
434 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
435 
436 /*
437  * The SPAPR TCE info struct provides the information about the PCI bus
438  * address ranges available for DMA, these values are programmed into
439  * the hardware so the guest has to know that information.
440  *
441  * The DMA 32 bit window start is an absolute PCI bus address.
442  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
443  * addresses too so the window works as a filter rather than an offset
444  * for IOVA addresses.
445  *
446  * A flag will need to be added if other page sizes are supported,
447  * so as defined here, it is always 4k.
448  */
449 struct vfio_iommu_spapr_tce_info {
450 	__u32 argsz;
451 	__u32 flags;			/* reserved for future use */
452 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
453 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
454 };
455 
456 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
457 
458 /* ***************************************************************** */
459 
460 #endif /* VFIO_H */
461