xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 228aa992)
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 /*
37  * The IOCTL interface is designed for extensibility by embedding the
38  * structure length (argsz) and flags into structures passed between
39  * kernel and userspace.  We therefore use the _IO() macro for these
40  * defines to avoid implicitly embedding a size into the ioctl request.
41  * As structure fields are added, argsz will increase to match and flag
42  * bits will be defined to indicate additional fields with valid data.
43  * It's *always* the caller's responsibility to indicate the size of
44  * the structure passed by setting argsz appropriately.
45  */
46 
47 #define VFIO_TYPE	(';')
48 #define VFIO_BASE	100
49 
50 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
51 
52 /**
53  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
54  *
55  * Report the version of the VFIO API.  This allows us to bump the entire
56  * API version should we later need to add or change features in incompatible
57  * ways.
58  * Return: VFIO_API_VERSION
59  * Availability: Always
60  */
61 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
62 
63 /**
64  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
65  *
66  * Check whether an extension is supported.
67  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
68  * Availability: Always
69  */
70 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
71 
72 /**
73  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
74  *
75  * Set the iommu to the given type.  The type must be supported by an
76  * iommu driver as verified by calling CHECK_EXTENSION using the same
77  * type.  A group must be set to this file descriptor before this
78  * ioctl is available.  The IOMMU interfaces enabled by this call are
79  * specific to the value set.
80  * Return: 0 on success, -errno on failure
81  * Availability: When VFIO group attached
82  */
83 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
84 
85 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
86 
87 /**
88  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
89  *						struct vfio_group_status)
90  *
91  * Retrieve information about the group.  Fills in provided
92  * struct vfio_group_info.  Caller sets argsz.
93  * Return: 0 on succes, -errno on failure.
94  * Availability: Always
95  */
96 struct vfio_group_status {
97 	__u32	argsz;
98 	__u32	flags;
99 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
100 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
101 };
102 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
103 
104 /**
105  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
106  *
107  * Set the container for the VFIO group to the open VFIO file
108  * descriptor provided.  Groups may only belong to a single
109  * container.  Containers may, at their discretion, support multiple
110  * groups.  Only when a container is set are all of the interfaces
111  * of the VFIO file descriptor and the VFIO group file descriptor
112  * available to the user.
113  * Return: 0 on success, -errno on failure.
114  * Availability: Always
115  */
116 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
117 
118 /**
119  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
120  *
121  * Remove the group from the attached container.  This is the
122  * opposite of the SET_CONTAINER call and returns the group to
123  * an initial state.  All device file descriptors must be released
124  * prior to calling this interface.  When removing the last group
125  * from a container, the IOMMU will be disabled and all state lost,
126  * effectively also returning the VFIO file descriptor to an initial
127  * state.
128  * Return: 0 on success, -errno on failure.
129  * Availability: When attached to container
130  */
131 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
132 
133 /**
134  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
135  *
136  * Return a new file descriptor for the device object described by
137  * the provided string.  The string should match a device listed in
138  * the devices subdirectory of the IOMMU group sysfs entry.  The
139  * group containing the device must already be added to this context.
140  * Return: new file descriptor on success, -errno on failure.
141  * Availability: When attached to container
142  */
143 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
144 
145 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
146 
147 /**
148  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
149  *						struct vfio_device_info)
150  *
151  * Retrieve information about the device.  Fills in provided
152  * struct vfio_device_info.  Caller sets argsz.
153  * Return: 0 on success, -errno on failure.
154  */
155 struct vfio_device_info {
156 	__u32	argsz;
157 	__u32	flags;
158 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
159 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
160 	__u32	num_regions;	/* Max region index + 1 */
161 	__u32	num_irqs;	/* Max IRQ index + 1 */
162 };
163 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
164 
165 /**
166  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
167  *				       struct vfio_region_info)
168  *
169  * Retrieve information about a device region.  Caller provides
170  * struct vfio_region_info with index value set.  Caller sets argsz.
171  * Implementation of region mapping is bus driver specific.  This is
172  * intended to describe MMIO, I/O port, as well as bus specific
173  * regions (ex. PCI config space).  Zero sized regions may be used
174  * to describe unimplemented regions (ex. unimplemented PCI BARs).
175  * Return: 0 on success, -errno on failure.
176  */
177 struct vfio_region_info {
178 	__u32	argsz;
179 	__u32	flags;
180 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
181 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
182 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
183 	__u32	index;		/* Region index */
184 	__u32	resv;		/* Reserved for alignment */
185 	__u64	size;		/* Region size (bytes) */
186 	__u64	offset;		/* Region offset from start of device fd */
187 };
188 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
189 
190 /**
191  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
192  *				    struct vfio_irq_info)
193  *
194  * Retrieve information about a device IRQ.  Caller provides
195  * struct vfio_irq_info with index value set.  Caller sets argsz.
196  * Implementation of IRQ mapping is bus driver specific.  Indexes
197  * using multiple IRQs are primarily intended to support MSI-like
198  * interrupt blocks.  Zero count irq blocks may be used to describe
199  * unimplemented interrupt types.
200  *
201  * The EVENTFD flag indicates the interrupt index supports eventfd based
202  * signaling.
203  *
204  * The MASKABLE flags indicates the index supports MASK and UNMASK
205  * actions described below.
206  *
207  * AUTOMASKED indicates that after signaling, the interrupt line is
208  * automatically masked by VFIO and the user needs to unmask the line
209  * to receive new interrupts.  This is primarily intended to distinguish
210  * level triggered interrupts.
211  *
212  * The NORESIZE flag indicates that the interrupt lines within the index
213  * are setup as a set and new subindexes cannot be enabled without first
214  * disabling the entire index.  This is used for interrupts like PCI MSI
215  * and MSI-X where the driver may only use a subset of the available
216  * indexes, but VFIO needs to enable a specific number of vectors
217  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
218  * then add and unmask vectors, it's up to userspace to make the decision
219  * whether to allocate the maximum supported number of vectors or tear
220  * down setup and incrementally increase the vectors as each is enabled.
221  */
222 struct vfio_irq_info {
223 	__u32	argsz;
224 	__u32	flags;
225 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
226 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
227 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
228 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
229 	__u32	index;		/* IRQ index */
230 	__u32	count;		/* Number of IRQs within this index */
231 };
232 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
233 
234 /**
235  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
236  *
237  * Set signaling, masking, and unmasking of interrupts.  Caller provides
238  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
239  * the range of subindexes being specified.
240  *
241  * The DATA flags specify the type of data provided.  If DATA_NONE, the
242  * operation performs the specified action immediately on the specified
243  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
244  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
245  *
246  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
247  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
248  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
249  * data = {1,0,1}
250  *
251  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
252  * A value of -1 can be used to either de-assign interrupts if already
253  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
254  * to be trigger for interrupts [0,0] and [0,2]:
255  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
256  * data = {fd1, -1, fd2}
257  * If index [0,1] is previously set, two count = 1 ioctls calls would be
258  * required to set [0,0] and [0,2] without changing [0,1].
259  *
260  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
261  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
262  * from userspace (ie. simulate hardware triggering).
263  *
264  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
265  * enables the interrupt index for the device.  Individual subindex interrupts
266  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
267  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
268  *
269  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
270  * ACTION_TRIGGER specifies kernel->user signaling.
271  */
272 struct vfio_irq_set {
273 	__u32	argsz;
274 	__u32	flags;
275 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
276 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
277 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
278 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
279 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
280 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
281 	__u32	index;
282 	__u32	start;
283 	__u32	count;
284 	__u8	data[];
285 };
286 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
287 
288 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
289 					 VFIO_IRQ_SET_DATA_BOOL | \
290 					 VFIO_IRQ_SET_DATA_EVENTFD)
291 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
292 					 VFIO_IRQ_SET_ACTION_UNMASK | \
293 					 VFIO_IRQ_SET_ACTION_TRIGGER)
294 /**
295  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
296  *
297  * Reset a device.
298  */
299 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
300 
301 /*
302  * The VFIO-PCI bus driver makes use of the following fixed region and
303  * IRQ index mapping.  Unimplemented regions return a size of zero.
304  * Unimplemented IRQ types return a count of zero.
305  */
306 
307 enum {
308 	VFIO_PCI_BAR0_REGION_INDEX,
309 	VFIO_PCI_BAR1_REGION_INDEX,
310 	VFIO_PCI_BAR2_REGION_INDEX,
311 	VFIO_PCI_BAR3_REGION_INDEX,
312 	VFIO_PCI_BAR4_REGION_INDEX,
313 	VFIO_PCI_BAR5_REGION_INDEX,
314 	VFIO_PCI_ROM_REGION_INDEX,
315 	VFIO_PCI_CONFIG_REGION_INDEX,
316 	/*
317 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
318 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
319 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
320 	 * range is found at it's identity mapped offset from the region
321 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
322 	 * between described ranges are unimplemented.
323 	 */
324 	VFIO_PCI_VGA_REGION_INDEX,
325 	VFIO_PCI_NUM_REGIONS
326 };
327 
328 enum {
329 	VFIO_PCI_INTX_IRQ_INDEX,
330 	VFIO_PCI_MSI_IRQ_INDEX,
331 	VFIO_PCI_MSIX_IRQ_INDEX,
332 	VFIO_PCI_ERR_IRQ_INDEX,
333 	VFIO_PCI_NUM_IRQS
334 };
335 
336 /**
337  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
338  *					      struct vfio_pci_hot_reset_info)
339  *
340  * Return: 0 on success, -errno on failure:
341  *	-enospc = insufficient buffer, -enodev = unsupported for device.
342  */
343 struct vfio_pci_dependent_device {
344 	__u32	group_id;
345 	__u16	segment;
346 	__u8	bus;
347 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
348 };
349 
350 struct vfio_pci_hot_reset_info {
351 	__u32	argsz;
352 	__u32	flags;
353 	__u32	count;
354 	struct vfio_pci_dependent_device	devices[];
355 };
356 
357 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
358 
359 /**
360  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
361  *				    struct vfio_pci_hot_reset)
362  *
363  * Return: 0 on success, -errno on failure.
364  */
365 struct vfio_pci_hot_reset {
366 	__u32	argsz;
367 	__u32	flags;
368 	__u32	count;
369 	__s32	group_fds[];
370 };
371 
372 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
373 
374 /* -------- API for Type1 VFIO IOMMU -------- */
375 
376 /**
377  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
378  *
379  * Retrieve information about the IOMMU object. Fills in provided
380  * struct vfio_iommu_info. Caller sets argsz.
381  *
382  * XXX Should we do these by CHECK_EXTENSION too?
383  */
384 struct vfio_iommu_type1_info {
385 	__u32	argsz;
386 	__u32	flags;
387 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
388 	__u64	iova_pgsizes;		/* Bitmap of supported page sizes */
389 };
390 
391 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
392 
393 /**
394  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
395  *
396  * Map process virtual addresses to IO virtual addresses using the
397  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
398  */
399 struct vfio_iommu_type1_dma_map {
400 	__u32	argsz;
401 	__u32	flags;
402 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
403 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
404 	__u64	vaddr;				/* Process virtual address */
405 	__u64	iova;				/* IO virtual address */
406 	__u64	size;				/* Size of mapping (bytes) */
407 };
408 
409 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
410 
411 /**
412  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
413  *							struct vfio_dma_unmap)
414  *
415  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
416  * Caller sets argsz.  The actual unmapped size is returned in the size
417  * field.  No guarantee is made to the user that arbitrary unmaps of iova
418  * or size different from those used in the original mapping call will
419  * succeed.
420  */
421 struct vfio_iommu_type1_dma_unmap {
422 	__u32	argsz;
423 	__u32	flags;
424 	__u64	iova;				/* IO virtual address */
425 	__u64	size;				/* Size of mapping (bytes) */
426 };
427 
428 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
429 
430 /*
431  * IOCTLs to enable/disable IOMMU container usage.
432  * No parameters are supported.
433  */
434 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
435 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
436 
437 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
438 
439 /*
440  * The SPAPR TCE info struct provides the information about the PCI bus
441  * address ranges available for DMA, these values are programmed into
442  * the hardware so the guest has to know that information.
443  *
444  * The DMA 32 bit window start is an absolute PCI bus address.
445  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
446  * addresses too so the window works as a filter rather than an offset
447  * for IOVA addresses.
448  *
449  * A flag will need to be added if other page sizes are supported,
450  * so as defined here, it is always 4k.
451  */
452 struct vfio_iommu_spapr_tce_info {
453 	__u32 argsz;
454 	__u32 flags;			/* reserved for future use */
455 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
456 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
457 };
458 
459 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
460 
461 /*
462  * EEH PE operation struct provides ways to:
463  * - enable/disable EEH functionality;
464  * - unfreeze IO/DMA for frozen PE;
465  * - read PE state;
466  * - reset PE;
467  * - configure PE.
468  */
469 struct vfio_eeh_pe_op {
470 	__u32 argsz;
471 	__u32 flags;
472 	__u32 op;
473 };
474 
475 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
476 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
477 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
478 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
479 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
480 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
481 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
482 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
483 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
484 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
485 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
486 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
487 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
488 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
489 
490 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
491 
492 /* ***************************************************************** */
493 
494 #endif /* VFIO_H */
495