xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 77a8257e)
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 /* Check if EEH is supported */
34 #define VFIO_EEH			5
35 
36 /* Two-stage IOMMU */
37 #define VFIO_TYPE1_NESTING_IOMMU	6	/* Implies v2 */
38 
39 /*
40  * The IOCTL interface is designed for extensibility by embedding the
41  * structure length (argsz) and flags into structures passed between
42  * kernel and userspace.  We therefore use the _IO() macro for these
43  * defines to avoid implicitly embedding a size into the ioctl request.
44  * As structure fields are added, argsz will increase to match and flag
45  * bits will be defined to indicate additional fields with valid data.
46  * It's *always* the caller's responsibility to indicate the size of
47  * the structure passed by setting argsz appropriately.
48  */
49 
50 #define VFIO_TYPE	(';')
51 #define VFIO_BASE	100
52 
53 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
54 
55 /**
56  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
57  *
58  * Report the version of the VFIO API.  This allows us to bump the entire
59  * API version should we later need to add or change features in incompatible
60  * ways.
61  * Return: VFIO_API_VERSION
62  * Availability: Always
63  */
64 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
65 
66 /**
67  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
68  *
69  * Check whether an extension is supported.
70  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
71  * Availability: Always
72  */
73 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
74 
75 /**
76  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
77  *
78  * Set the iommu to the given type.  The type must be supported by an
79  * iommu driver as verified by calling CHECK_EXTENSION using the same
80  * type.  A group must be set to this file descriptor before this
81  * ioctl is available.  The IOMMU interfaces enabled by this call are
82  * specific to the value set.
83  * Return: 0 on success, -errno on failure
84  * Availability: When VFIO group attached
85  */
86 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
87 
88 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
89 
90 /**
91  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
92  *						struct vfio_group_status)
93  *
94  * Retrieve information about the group.  Fills in provided
95  * struct vfio_group_info.  Caller sets argsz.
96  * Return: 0 on succes, -errno on failure.
97  * Availability: Always
98  */
99 struct vfio_group_status {
100 	__u32	argsz;
101 	__u32	flags;
102 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
103 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
104 };
105 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
106 
107 /**
108  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
109  *
110  * Set the container for the VFIO group to the open VFIO file
111  * descriptor provided.  Groups may only belong to a single
112  * container.  Containers may, at their discretion, support multiple
113  * groups.  Only when a container is set are all of the interfaces
114  * of the VFIO file descriptor and the VFIO group file descriptor
115  * available to the user.
116  * Return: 0 on success, -errno on failure.
117  * Availability: Always
118  */
119 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
120 
121 /**
122  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
123  *
124  * Remove the group from the attached container.  This is the
125  * opposite of the SET_CONTAINER call and returns the group to
126  * an initial state.  All device file descriptors must be released
127  * prior to calling this interface.  When removing the last group
128  * from a container, the IOMMU will be disabled and all state lost,
129  * effectively also returning the VFIO file descriptor to an initial
130  * state.
131  * Return: 0 on success, -errno on failure.
132  * Availability: When attached to container
133  */
134 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
135 
136 /**
137  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
138  *
139  * Return a new file descriptor for the device object described by
140  * the provided string.  The string should match a device listed in
141  * the devices subdirectory of the IOMMU group sysfs entry.  The
142  * group containing the device must already be added to this context.
143  * Return: new file descriptor on success, -errno on failure.
144  * Availability: When attached to container
145  */
146 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
147 
148 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
149 
150 /**
151  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
152  *						struct vfio_device_info)
153  *
154  * Retrieve information about the device.  Fills in provided
155  * struct vfio_device_info.  Caller sets argsz.
156  * Return: 0 on success, -errno on failure.
157  */
158 struct vfio_device_info {
159 	__u32	argsz;
160 	__u32	flags;
161 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
162 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
163 	__u32	num_regions;	/* Max region index + 1 */
164 	__u32	num_irqs;	/* Max IRQ index + 1 */
165 };
166 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
167 
168 /**
169  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
170  *				       struct vfio_region_info)
171  *
172  * Retrieve information about a device region.  Caller provides
173  * struct vfio_region_info with index value set.  Caller sets argsz.
174  * Implementation of region mapping is bus driver specific.  This is
175  * intended to describe MMIO, I/O port, as well as bus specific
176  * regions (ex. PCI config space).  Zero sized regions may be used
177  * to describe unimplemented regions (ex. unimplemented PCI BARs).
178  * Return: 0 on success, -errno on failure.
179  */
180 struct vfio_region_info {
181 	__u32	argsz;
182 	__u32	flags;
183 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
184 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
185 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
186 	__u32	index;		/* Region index */
187 	__u32	resv;		/* Reserved for alignment */
188 	__u64	size;		/* Region size (bytes) */
189 	__u64	offset;		/* Region offset from start of device fd */
190 };
191 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
192 
193 /**
194  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
195  *				    struct vfio_irq_info)
196  *
197  * Retrieve information about a device IRQ.  Caller provides
198  * struct vfio_irq_info with index value set.  Caller sets argsz.
199  * Implementation of IRQ mapping is bus driver specific.  Indexes
200  * using multiple IRQs are primarily intended to support MSI-like
201  * interrupt blocks.  Zero count irq blocks may be used to describe
202  * unimplemented interrupt types.
203  *
204  * The EVENTFD flag indicates the interrupt index supports eventfd based
205  * signaling.
206  *
207  * The MASKABLE flags indicates the index supports MASK and UNMASK
208  * actions described below.
209  *
210  * AUTOMASKED indicates that after signaling, the interrupt line is
211  * automatically masked by VFIO and the user needs to unmask the line
212  * to receive new interrupts.  This is primarily intended to distinguish
213  * level triggered interrupts.
214  *
215  * The NORESIZE flag indicates that the interrupt lines within the index
216  * are setup as a set and new subindexes cannot be enabled without first
217  * disabling the entire index.  This is used for interrupts like PCI MSI
218  * and MSI-X where the driver may only use a subset of the available
219  * indexes, but VFIO needs to enable a specific number of vectors
220  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
221  * then add and unmask vectors, it's up to userspace to make the decision
222  * whether to allocate the maximum supported number of vectors or tear
223  * down setup and incrementally increase the vectors as each is enabled.
224  */
225 struct vfio_irq_info {
226 	__u32	argsz;
227 	__u32	flags;
228 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
229 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
230 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
231 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
232 	__u32	index;		/* IRQ index */
233 	__u32	count;		/* Number of IRQs within this index */
234 };
235 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
236 
237 /**
238  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
239  *
240  * Set signaling, masking, and unmasking of interrupts.  Caller provides
241  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
242  * the range of subindexes being specified.
243  *
244  * The DATA flags specify the type of data provided.  If DATA_NONE, the
245  * operation performs the specified action immediately on the specified
246  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
247  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
248  *
249  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
250  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
251  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
252  * data = {1,0,1}
253  *
254  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
255  * A value of -1 can be used to either de-assign interrupts if already
256  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
257  * to be trigger for interrupts [0,0] and [0,2]:
258  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
259  * data = {fd1, -1, fd2}
260  * If index [0,1] is previously set, two count = 1 ioctls calls would be
261  * required to set [0,0] and [0,2] without changing [0,1].
262  *
263  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
264  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
265  * from userspace (ie. simulate hardware triggering).
266  *
267  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
268  * enables the interrupt index for the device.  Individual subindex interrupts
269  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
270  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
271  *
272  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
273  * ACTION_TRIGGER specifies kernel->user signaling.
274  */
275 struct vfio_irq_set {
276 	__u32	argsz;
277 	__u32	flags;
278 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
279 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
280 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
281 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
282 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
283 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
284 	__u32	index;
285 	__u32	start;
286 	__u32	count;
287 	__u8	data[];
288 };
289 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
290 
291 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
292 					 VFIO_IRQ_SET_DATA_BOOL | \
293 					 VFIO_IRQ_SET_DATA_EVENTFD)
294 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
295 					 VFIO_IRQ_SET_ACTION_UNMASK | \
296 					 VFIO_IRQ_SET_ACTION_TRIGGER)
297 /**
298  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
299  *
300  * Reset a device.
301  */
302 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
303 
304 /*
305  * The VFIO-PCI bus driver makes use of the following fixed region and
306  * IRQ index mapping.  Unimplemented regions return a size of zero.
307  * Unimplemented IRQ types return a count of zero.
308  */
309 
310 enum {
311 	VFIO_PCI_BAR0_REGION_INDEX,
312 	VFIO_PCI_BAR1_REGION_INDEX,
313 	VFIO_PCI_BAR2_REGION_INDEX,
314 	VFIO_PCI_BAR3_REGION_INDEX,
315 	VFIO_PCI_BAR4_REGION_INDEX,
316 	VFIO_PCI_BAR5_REGION_INDEX,
317 	VFIO_PCI_ROM_REGION_INDEX,
318 	VFIO_PCI_CONFIG_REGION_INDEX,
319 	/*
320 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
321 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
322 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
323 	 * range is found at it's identity mapped offset from the region
324 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
325 	 * between described ranges are unimplemented.
326 	 */
327 	VFIO_PCI_VGA_REGION_INDEX,
328 	VFIO_PCI_NUM_REGIONS
329 };
330 
331 enum {
332 	VFIO_PCI_INTX_IRQ_INDEX,
333 	VFIO_PCI_MSI_IRQ_INDEX,
334 	VFIO_PCI_MSIX_IRQ_INDEX,
335 	VFIO_PCI_ERR_IRQ_INDEX,
336 	VFIO_PCI_REQ_IRQ_INDEX,
337 	VFIO_PCI_NUM_IRQS
338 };
339 
340 /**
341  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
342  *					      struct vfio_pci_hot_reset_info)
343  *
344  * Return: 0 on success, -errno on failure:
345  *	-enospc = insufficient buffer, -enodev = unsupported for device.
346  */
347 struct vfio_pci_dependent_device {
348 	__u32	group_id;
349 	__u16	segment;
350 	__u8	bus;
351 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
352 };
353 
354 struct vfio_pci_hot_reset_info {
355 	__u32	argsz;
356 	__u32	flags;
357 	__u32	count;
358 	struct vfio_pci_dependent_device	devices[];
359 };
360 
361 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
362 
363 /**
364  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
365  *				    struct vfio_pci_hot_reset)
366  *
367  * Return: 0 on success, -errno on failure.
368  */
369 struct vfio_pci_hot_reset {
370 	__u32	argsz;
371 	__u32	flags;
372 	__u32	count;
373 	__s32	group_fds[];
374 };
375 
376 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
377 
378 /* -------- API for Type1 VFIO IOMMU -------- */
379 
380 /**
381  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
382  *
383  * Retrieve information about the IOMMU object. Fills in provided
384  * struct vfio_iommu_info. Caller sets argsz.
385  *
386  * XXX Should we do these by CHECK_EXTENSION too?
387  */
388 struct vfio_iommu_type1_info {
389 	__u32	argsz;
390 	__u32	flags;
391 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
392 	__u64	iova_pgsizes;		/* Bitmap of supported page sizes */
393 };
394 
395 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
396 
397 /**
398  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
399  *
400  * Map process virtual addresses to IO virtual addresses using the
401  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
402  */
403 struct vfio_iommu_type1_dma_map {
404 	__u32	argsz;
405 	__u32	flags;
406 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
407 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
408 	__u64	vaddr;				/* Process virtual address */
409 	__u64	iova;				/* IO virtual address */
410 	__u64	size;				/* Size of mapping (bytes) */
411 };
412 
413 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
414 
415 /**
416  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
417  *							struct vfio_dma_unmap)
418  *
419  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
420  * Caller sets argsz.  The actual unmapped size is returned in the size
421  * field.  No guarantee is made to the user that arbitrary unmaps of iova
422  * or size different from those used in the original mapping call will
423  * succeed.
424  */
425 struct vfio_iommu_type1_dma_unmap {
426 	__u32	argsz;
427 	__u32	flags;
428 	__u64	iova;				/* IO virtual address */
429 	__u64	size;				/* Size of mapping (bytes) */
430 };
431 
432 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
433 
434 /*
435  * IOCTLs to enable/disable IOMMU container usage.
436  * No parameters are supported.
437  */
438 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
439 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
440 
441 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
442 
443 /*
444  * The SPAPR TCE info struct provides the information about the PCI bus
445  * address ranges available for DMA, these values are programmed into
446  * the hardware so the guest has to know that information.
447  *
448  * The DMA 32 bit window start is an absolute PCI bus address.
449  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
450  * addresses too so the window works as a filter rather than an offset
451  * for IOVA addresses.
452  *
453  * A flag will need to be added if other page sizes are supported,
454  * so as defined here, it is always 4k.
455  */
456 struct vfio_iommu_spapr_tce_info {
457 	__u32 argsz;
458 	__u32 flags;			/* reserved for future use */
459 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
460 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
461 };
462 
463 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
464 
465 /*
466  * EEH PE operation struct provides ways to:
467  * - enable/disable EEH functionality;
468  * - unfreeze IO/DMA for frozen PE;
469  * - read PE state;
470  * - reset PE;
471  * - configure PE.
472  */
473 struct vfio_eeh_pe_op {
474 	__u32 argsz;
475 	__u32 flags;
476 	__u32 op;
477 };
478 
479 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
480 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
481 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
482 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
483 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
484 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
485 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
486 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
487 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
488 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
489 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
490 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
491 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
492 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
493 
494 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
495 
496 /* ***************************************************************** */
497 
498 #endif /* VFIO_H */
499