xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 39164c13)
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 #define VFIO_SPAPR_TCE_v2_IOMMU		7
40 
41 /*
42  * The No-IOMMU IOMMU offers no translation or isolation for devices and
43  * supports no ioctls outside of VFIO_CHECK_EXTENSION.  Use of VFIO's No-IOMMU
44  * code will taint the host kernel and should be used with extreme caution.
45  */
46 #define VFIO_NOIOMMU_IOMMU		8
47 
48 /*
49  * The IOCTL interface is designed for extensibility by embedding the
50  * structure length (argsz) and flags into structures passed between
51  * kernel and userspace.  We therefore use the _IO() macro for these
52  * defines to avoid implicitly embedding a size into the ioctl request.
53  * As structure fields are added, argsz will increase to match and flag
54  * bits will be defined to indicate additional fields with valid data.
55  * It's *always* the caller's responsibility to indicate the size of
56  * the structure passed by setting argsz appropriately.
57  */
58 
59 #define VFIO_TYPE	(';')
60 #define VFIO_BASE	100
61 
62 /*
63  * For extension of INFO ioctls, VFIO makes use of a capability chain
64  * designed after PCI/e capabilities.  A flag bit indicates whether
65  * this capability chain is supported and a field defined in the fixed
66  * structure defines the offset of the first capability in the chain.
67  * This field is only valid when the corresponding bit in the flags
68  * bitmap is set.  This offset field is relative to the start of the
69  * INFO buffer, as is the next field within each capability header.
70  * The id within the header is a shared address space per INFO ioctl,
71  * while the version field is specific to the capability id.  The
72  * contents following the header are specific to the capability id.
73  */
74 struct vfio_info_cap_header {
75 	__u16	id;		/* Identifies capability */
76 	__u16	version;	/* Version specific to the capability ID */
77 	__u32	next;		/* Offset of next capability */
78 };
79 
80 /*
81  * Callers of INFO ioctls passing insufficiently sized buffers will see
82  * the capability chain flag bit set, a zero value for the first capability
83  * offset (if available within the provided argsz), and argsz will be
84  * updated to report the necessary buffer size.  For compatibility, the
85  * INFO ioctl will not report error in this case, but the capability chain
86  * will not be available.
87  */
88 
89 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
90 
91 /**
92  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
93  *
94  * Report the version of the VFIO API.  This allows us to bump the entire
95  * API version should we later need to add or change features in incompatible
96  * ways.
97  * Return: VFIO_API_VERSION
98  * Availability: Always
99  */
100 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
101 
102 /**
103  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
104  *
105  * Check whether an extension is supported.
106  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
107  * Availability: Always
108  */
109 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
110 
111 /**
112  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
113  *
114  * Set the iommu to the given type.  The type must be supported by an
115  * iommu driver as verified by calling CHECK_EXTENSION using the same
116  * type.  A group must be set to this file descriptor before this
117  * ioctl is available.  The IOMMU interfaces enabled by this call are
118  * specific to the value set.
119  * Return: 0 on success, -errno on failure
120  * Availability: When VFIO group attached
121  */
122 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
123 
124 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
125 
126 /**
127  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
128  *						struct vfio_group_status)
129  *
130  * Retrieve information about the group.  Fills in provided
131  * struct vfio_group_info.  Caller sets argsz.
132  * Return: 0 on succes, -errno on failure.
133  * Availability: Always
134  */
135 struct vfio_group_status {
136 	__u32	argsz;
137 	__u32	flags;
138 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
139 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
140 };
141 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
142 
143 /**
144  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
145  *
146  * Set the container for the VFIO group to the open VFIO file
147  * descriptor provided.  Groups may only belong to a single
148  * container.  Containers may, at their discretion, support multiple
149  * groups.  Only when a container is set are all of the interfaces
150  * of the VFIO file descriptor and the VFIO group file descriptor
151  * available to the user.
152  * Return: 0 on success, -errno on failure.
153  * Availability: Always
154  */
155 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
156 
157 /**
158  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
159  *
160  * Remove the group from the attached container.  This is the
161  * opposite of the SET_CONTAINER call and returns the group to
162  * an initial state.  All device file descriptors must be released
163  * prior to calling this interface.  When removing the last group
164  * from a container, the IOMMU will be disabled and all state lost,
165  * effectively also returning the VFIO file descriptor to an initial
166  * state.
167  * Return: 0 on success, -errno on failure.
168  * Availability: When attached to container
169  */
170 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
171 
172 /**
173  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
174  *
175  * Return a new file descriptor for the device object described by
176  * the provided string.  The string should match a device listed in
177  * the devices subdirectory of the IOMMU group sysfs entry.  The
178  * group containing the device must already be added to this context.
179  * Return: new file descriptor on success, -errno on failure.
180  * Availability: When attached to container
181  */
182 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
183 
184 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
185 
186 /**
187  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
188  *						struct vfio_device_info)
189  *
190  * Retrieve information about the device.  Fills in provided
191  * struct vfio_device_info.  Caller sets argsz.
192  * Return: 0 on success, -errno on failure.
193  */
194 struct vfio_device_info {
195 	__u32	argsz;
196 	__u32	flags;
197 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
198 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
199 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2)	/* vfio-platform device */
200 #define VFIO_DEVICE_FLAGS_AMBA  (1 << 3)	/* vfio-amba device */
201 	__u32	num_regions;	/* Max region index + 1 */
202 	__u32	num_irqs;	/* Max IRQ index + 1 */
203 };
204 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
205 
206 /*
207  * Vendor driver using Mediated device framework should provide device_api
208  * attribute in supported type attribute groups. Device API string should be one
209  * of the following corresponding to device flags in vfio_device_info structure.
210  */
211 
212 #define VFIO_DEVICE_API_PCI_STRING		"vfio-pci"
213 #define VFIO_DEVICE_API_PLATFORM_STRING		"vfio-platform"
214 #define VFIO_DEVICE_API_AMBA_STRING		"vfio-amba"
215 
216 /**
217  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
218  *				       struct vfio_region_info)
219  *
220  * Retrieve information about a device region.  Caller provides
221  * struct vfio_region_info with index value set.  Caller sets argsz.
222  * Implementation of region mapping is bus driver specific.  This is
223  * intended to describe MMIO, I/O port, as well as bus specific
224  * regions (ex. PCI config space).  Zero sized regions may be used
225  * to describe unimplemented regions (ex. unimplemented PCI BARs).
226  * Return: 0 on success, -errno on failure.
227  */
228 struct vfio_region_info {
229 	__u32	argsz;
230 	__u32	flags;
231 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
232 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
233 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
234 #define VFIO_REGION_INFO_FLAG_CAPS	(1 << 3) /* Info supports caps */
235 	__u32	index;		/* Region index */
236 	__u32	cap_offset;	/* Offset within info struct of first cap */
237 	__u64	size;		/* Region size (bytes) */
238 	__u64	offset;		/* Region offset from start of device fd */
239 };
240 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
241 
242 /*
243  * The sparse mmap capability allows finer granularity of specifying areas
244  * within a region with mmap support.  When specified, the user should only
245  * mmap the offset ranges specified by the areas array.  mmaps outside of the
246  * areas specified may fail (such as the range covering a PCI MSI-X table) or
247  * may result in improper device behavior.
248  *
249  * The structures below define version 1 of this capability.
250  */
251 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP	1
252 
253 struct vfio_region_sparse_mmap_area {
254 	__u64	offset;	/* Offset of mmap'able area within region */
255 	__u64	size;	/* Size of mmap'able area */
256 };
257 
258 struct vfio_region_info_cap_sparse_mmap {
259 	struct vfio_info_cap_header header;
260 	__u32	nr_areas;
261 	__u32	reserved;
262 	struct vfio_region_sparse_mmap_area areas[];
263 };
264 
265 /*
266  * The device specific type capability allows regions unique to a specific
267  * device or class of devices to be exposed.  This helps solve the problem for
268  * vfio bus drivers of defining which region indexes correspond to which region
269  * on the device, without needing to resort to static indexes, as done by
270  * vfio-pci.  For instance, if we were to go back in time, we might remove
271  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
272  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
273  * make a "VGA" device specific type to describe the VGA access space.  This
274  * means that non-VGA devices wouldn't need to waste this index, and thus the
275  * address space associated with it due to implementation of device file
276  * descriptor offsets in vfio-pci.
277  *
278  * The current implementation is now part of the user ABI, so we can't use this
279  * for VGA, but there are other upcoming use cases, such as opregions for Intel
280  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
281  * use this for future additions.
282  *
283  * The structure below defines version 1 of this capability.
284  */
285 #define VFIO_REGION_INFO_CAP_TYPE	2
286 
287 struct vfio_region_info_cap_type {
288 	struct vfio_info_cap_header header;
289 	__u32 type;	/* global per bus driver */
290 	__u32 subtype;	/* type specific */
291 };
292 
293 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
294 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
295 
296 /* 8086 Vendor sub-types */
297 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
298 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
299 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
300 
301 /**
302  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
303  *				    struct vfio_irq_info)
304  *
305  * Retrieve information about a device IRQ.  Caller provides
306  * struct vfio_irq_info with index value set.  Caller sets argsz.
307  * Implementation of IRQ mapping is bus driver specific.  Indexes
308  * using multiple IRQs are primarily intended to support MSI-like
309  * interrupt blocks.  Zero count irq blocks may be used to describe
310  * unimplemented interrupt types.
311  *
312  * The EVENTFD flag indicates the interrupt index supports eventfd based
313  * signaling.
314  *
315  * The MASKABLE flags indicates the index supports MASK and UNMASK
316  * actions described below.
317  *
318  * AUTOMASKED indicates that after signaling, the interrupt line is
319  * automatically masked by VFIO and the user needs to unmask the line
320  * to receive new interrupts.  This is primarily intended to distinguish
321  * level triggered interrupts.
322  *
323  * The NORESIZE flag indicates that the interrupt lines within the index
324  * are setup as a set and new subindexes cannot be enabled without first
325  * disabling the entire index.  This is used for interrupts like PCI MSI
326  * and MSI-X where the driver may only use a subset of the available
327  * indexes, but VFIO needs to enable a specific number of vectors
328  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
329  * then add and unmask vectors, it's up to userspace to make the decision
330  * whether to allocate the maximum supported number of vectors or tear
331  * down setup and incrementally increase the vectors as each is enabled.
332  */
333 struct vfio_irq_info {
334 	__u32	argsz;
335 	__u32	flags;
336 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
337 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
338 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
339 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
340 	__u32	index;		/* IRQ index */
341 	__u32	count;		/* Number of IRQs within this index */
342 };
343 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
344 
345 /**
346  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
347  *
348  * Set signaling, masking, and unmasking of interrupts.  Caller provides
349  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
350  * the range of subindexes being specified.
351  *
352  * The DATA flags specify the type of data provided.  If DATA_NONE, the
353  * operation performs the specified action immediately on the specified
354  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
355  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
356  *
357  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
358  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
359  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
360  * data = {1,0,1}
361  *
362  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
363  * A value of -1 can be used to either de-assign interrupts if already
364  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
365  * to be trigger for interrupts [0,0] and [0,2]:
366  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
367  * data = {fd1, -1, fd2}
368  * If index [0,1] is previously set, two count = 1 ioctls calls would be
369  * required to set [0,0] and [0,2] without changing [0,1].
370  *
371  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
372  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
373  * from userspace (ie. simulate hardware triggering).
374  *
375  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
376  * enables the interrupt index for the device.  Individual subindex interrupts
377  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
378  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
379  *
380  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
381  * ACTION_TRIGGER specifies kernel->user signaling.
382  */
383 struct vfio_irq_set {
384 	__u32	argsz;
385 	__u32	flags;
386 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
387 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
388 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
389 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
390 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
391 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
392 	__u32	index;
393 	__u32	start;
394 	__u32	count;
395 	__u8	data[];
396 };
397 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
398 
399 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
400 					 VFIO_IRQ_SET_DATA_BOOL | \
401 					 VFIO_IRQ_SET_DATA_EVENTFD)
402 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
403 					 VFIO_IRQ_SET_ACTION_UNMASK | \
404 					 VFIO_IRQ_SET_ACTION_TRIGGER)
405 /**
406  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
407  *
408  * Reset a device.
409  */
410 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
411 
412 /*
413  * The VFIO-PCI bus driver makes use of the following fixed region and
414  * IRQ index mapping.  Unimplemented regions return a size of zero.
415  * Unimplemented IRQ types return a count of zero.
416  */
417 
418 enum {
419 	VFIO_PCI_BAR0_REGION_INDEX,
420 	VFIO_PCI_BAR1_REGION_INDEX,
421 	VFIO_PCI_BAR2_REGION_INDEX,
422 	VFIO_PCI_BAR3_REGION_INDEX,
423 	VFIO_PCI_BAR4_REGION_INDEX,
424 	VFIO_PCI_BAR5_REGION_INDEX,
425 	VFIO_PCI_ROM_REGION_INDEX,
426 	VFIO_PCI_CONFIG_REGION_INDEX,
427 	/*
428 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
429 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
430 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
431 	 * range is found at it's identity mapped offset from the region
432 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
433 	 * between described ranges are unimplemented.
434 	 */
435 	VFIO_PCI_VGA_REGION_INDEX,
436 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
437 				 /* device specific cap to define content. */
438 };
439 
440 enum {
441 	VFIO_PCI_INTX_IRQ_INDEX,
442 	VFIO_PCI_MSI_IRQ_INDEX,
443 	VFIO_PCI_MSIX_IRQ_INDEX,
444 	VFIO_PCI_ERR_IRQ_INDEX,
445 	VFIO_PCI_REQ_IRQ_INDEX,
446 	VFIO_PCI_NUM_IRQS
447 };
448 
449 /**
450  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
451  *					      struct vfio_pci_hot_reset_info)
452  *
453  * Return: 0 on success, -errno on failure:
454  *	-enospc = insufficient buffer, -enodev = unsupported for device.
455  */
456 struct vfio_pci_dependent_device {
457 	__u32	group_id;
458 	__u16	segment;
459 	__u8	bus;
460 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
461 };
462 
463 struct vfio_pci_hot_reset_info {
464 	__u32	argsz;
465 	__u32	flags;
466 	__u32	count;
467 	struct vfio_pci_dependent_device	devices[];
468 };
469 
470 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
471 
472 /**
473  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
474  *				    struct vfio_pci_hot_reset)
475  *
476  * Return: 0 on success, -errno on failure.
477  */
478 struct vfio_pci_hot_reset {
479 	__u32	argsz;
480 	__u32	flags;
481 	__u32	count;
482 	__s32	group_fds[];
483 };
484 
485 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
486 
487 /* -------- API for Type1 VFIO IOMMU -------- */
488 
489 /**
490  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
491  *
492  * Retrieve information about the IOMMU object. Fills in provided
493  * struct vfio_iommu_info. Caller sets argsz.
494  *
495  * XXX Should we do these by CHECK_EXTENSION too?
496  */
497 struct vfio_iommu_type1_info {
498 	__u32	argsz;
499 	__u32	flags;
500 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
501 	__u64	iova_pgsizes;		/* Bitmap of supported page sizes */
502 };
503 
504 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
505 
506 /**
507  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
508  *
509  * Map process virtual addresses to IO virtual addresses using the
510  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
511  */
512 struct vfio_iommu_type1_dma_map {
513 	__u32	argsz;
514 	__u32	flags;
515 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
516 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
517 	__u64	vaddr;				/* Process virtual address */
518 	__u64	iova;				/* IO virtual address */
519 	__u64	size;				/* Size of mapping (bytes) */
520 };
521 
522 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
523 
524 /**
525  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
526  *							struct vfio_dma_unmap)
527  *
528  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
529  * Caller sets argsz.  The actual unmapped size is returned in the size
530  * field.  No guarantee is made to the user that arbitrary unmaps of iova
531  * or size different from those used in the original mapping call will
532  * succeed.
533  */
534 struct vfio_iommu_type1_dma_unmap {
535 	__u32	argsz;
536 	__u32	flags;
537 	__u64	iova;				/* IO virtual address */
538 	__u64	size;				/* Size of mapping (bytes) */
539 };
540 
541 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
542 
543 /*
544  * IOCTLs to enable/disable IOMMU container usage.
545  * No parameters are supported.
546  */
547 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
548 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
549 
550 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
551 
552 /*
553  * The SPAPR TCE DDW info struct provides the information about
554  * the details of Dynamic DMA window capability.
555  *
556  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
557  * @max_dynamic_windows_supported tells the maximum number of windows
558  * which the platform can create.
559  * @levels tells the maximum number of levels in multi-level IOMMU tables;
560  * this allows splitting a table into smaller chunks which reduces
561  * the amount of physically contiguous memory required for the table.
562  */
563 struct vfio_iommu_spapr_tce_ddw_info {
564 	__u64 pgsizes;			/* Bitmap of supported page sizes */
565 	__u32 max_dynamic_windows_supported;
566 	__u32 levels;
567 };
568 
569 /*
570  * The SPAPR TCE info struct provides the information about the PCI bus
571  * address ranges available for DMA, these values are programmed into
572  * the hardware so the guest has to know that information.
573  *
574  * The DMA 32 bit window start is an absolute PCI bus address.
575  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
576  * addresses too so the window works as a filter rather than an offset
577  * for IOVA addresses.
578  *
579  * Flags supported:
580  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
581  *   (DDW) support is present. @ddw is only supported when DDW is present.
582  */
583 struct vfio_iommu_spapr_tce_info {
584 	__u32 argsz;
585 	__u32 flags;
586 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
587 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
588 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
589 	struct vfio_iommu_spapr_tce_ddw_info ddw;
590 };
591 
592 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
593 
594 /*
595  * EEH PE operation struct provides ways to:
596  * - enable/disable EEH functionality;
597  * - unfreeze IO/DMA for frozen PE;
598  * - read PE state;
599  * - reset PE;
600  * - configure PE;
601  * - inject EEH error.
602  */
603 struct vfio_eeh_pe_err {
604 	__u32 type;
605 	__u32 func;
606 	__u64 addr;
607 	__u64 mask;
608 };
609 
610 struct vfio_eeh_pe_op {
611 	__u32 argsz;
612 	__u32 flags;
613 	__u32 op;
614 	union {
615 		struct vfio_eeh_pe_err err;
616 	};
617 };
618 
619 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
620 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
621 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
622 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
623 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
624 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
625 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
626 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
627 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
628 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
629 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
630 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
631 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
632 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
633 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
634 
635 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
636 
637 /**
638  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
639  *
640  * Registers user space memory where DMA is allowed. It pins
641  * user pages and does the locked memory accounting so
642  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
643  * get faster.
644  */
645 struct vfio_iommu_spapr_register_memory {
646 	__u32	argsz;
647 	__u32	flags;
648 	__u64	vaddr;				/* Process virtual address */
649 	__u64	size;				/* Size of mapping (bytes) */
650 };
651 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
652 
653 /**
654  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
655  *
656  * Unregisters user space memory registered with
657  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
658  * Uses vfio_iommu_spapr_register_memory for parameters.
659  */
660 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
661 
662 /**
663  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
664  *
665  * Creates an additional TCE table and programs it (sets a new DMA window)
666  * to every IOMMU group in the container. It receives page shift, window
667  * size and number of levels in the TCE table being created.
668  *
669  * It allocates and returns an offset on a PCI bus of the new DMA window.
670  */
671 struct vfio_iommu_spapr_tce_create {
672 	__u32 argsz;
673 	__u32 flags;
674 	/* in */
675 	__u32 page_shift;
676 	__u32 __resv1;
677 	__u64 window_size;
678 	__u32 levels;
679 	__u32 __resv2;
680 	/* out */
681 	__u64 start_addr;
682 };
683 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
684 
685 /**
686  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
687  *
688  * Unprograms a TCE table from all groups in the container and destroys it.
689  * It receives a PCI bus offset as a window id.
690  */
691 struct vfio_iommu_spapr_tce_remove {
692 	__u32 argsz;
693 	__u32 flags;
694 	/* in */
695 	__u64 start_addr;
696 };
697 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
698 
699 /* ***************************************************************** */
700 
701 #endif /* VFIO_H */
702