4 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
5 * Author: Alex Williamson <alex.williamson@redhat.com>
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.
14 #include <linux/types.h>
15 #include <linux/ioctl.h>
17 #define VFIO_API_VERSION 0
20 /* Kernel & User level defines for VFIO IOCTLs. */
24 #define VFIO_TYPE1_IOMMU 1
25 #define VFIO_SPAPR_TCE_IOMMU 2
26 #define VFIO_TYPE1v2_IOMMU 3
28 * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping). This
29 * capability is subject to change as groups are added or removed.
31 #define VFIO_DMA_CC_IOMMU 4
33 /* Check if EEH is supported */
37 #define VFIO_TYPE1_NESTING_IOMMU 6 /* Implies v2 */
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.
50 #define VFIO_TYPE (';')
53 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
56 * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
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
61 * Return: VFIO_API_VERSION
62 * Availability: Always
64 #define VFIO_GET_API_VERSION _IO(VFIO_TYPE, VFIO_BASE + 0)
67 * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
69 * Check whether an extension is supported.
70 * Return: 0 if not supported, 1 (or some other positive integer) if supported.
71 * Availability: Always
73 #define VFIO_CHECK_EXTENSION _IO(VFIO_TYPE, VFIO_BASE + 1)
76 * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
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
86 #define VFIO_SET_IOMMU _IO(VFIO_TYPE, VFIO_BASE + 2)
88 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
91 * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
92 * struct vfio_group_status)
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
99 struct vfio_group_status
{
102 #define VFIO_GROUP_FLAGS_VIABLE (1 << 0)
103 #define VFIO_GROUP_FLAGS_CONTAINER_SET (1 << 1)
105 #define VFIO_GROUP_GET_STATUS _IO(VFIO_TYPE, VFIO_BASE + 3)
108 * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
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
119 #define VFIO_GROUP_SET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 4)
122 * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
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
131 * Return: 0 on success, -errno on failure.
132 * Availability: When attached to container
134 #define VFIO_GROUP_UNSET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 5)
137 * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
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
146 #define VFIO_GROUP_GET_DEVICE_FD _IO(VFIO_TYPE, VFIO_BASE + 6)
148 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
151 * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
152 * struct vfio_device_info)
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.
158 struct vfio_device_info
{
161 #define VFIO_DEVICE_FLAGS_RESET (1 << 0) /* Device supports reset */
162 #define VFIO_DEVICE_FLAGS_PCI (1 << 1) /* vfio-pci device */
163 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2) /* vfio-platform device */
164 #define VFIO_DEVICE_FLAGS_AMBA (1 << 3) /* vfio-amba device */
165 __u32 num_regions
; /* Max region index + 1 */
166 __u32 num_irqs
; /* Max IRQ index + 1 */
168 #define VFIO_DEVICE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 7)
171 * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
172 * struct vfio_region_info)
174 * Retrieve information about a device region. Caller provides
175 * struct vfio_region_info with index value set. Caller sets argsz.
176 * Implementation of region mapping is bus driver specific. This is
177 * intended to describe MMIO, I/O port, as well as bus specific
178 * regions (ex. PCI config space). Zero sized regions may be used
179 * to describe unimplemented regions (ex. unimplemented PCI BARs).
180 * Return: 0 on success, -errno on failure.
182 struct vfio_region_info
{
185 #define VFIO_REGION_INFO_FLAG_READ (1 << 0) /* Region supports read */
186 #define VFIO_REGION_INFO_FLAG_WRITE (1 << 1) /* Region supports write */
187 #define VFIO_REGION_INFO_FLAG_MMAP (1 << 2) /* Region supports mmap */
188 __u32 index
; /* Region index */
189 __u32 resv
; /* Reserved for alignment */
190 __u64 size
; /* Region size (bytes) */
191 __u64 offset
; /* Region offset from start of device fd */
193 #define VFIO_DEVICE_GET_REGION_INFO _IO(VFIO_TYPE, VFIO_BASE + 8)
196 * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
197 * struct vfio_irq_info)
199 * Retrieve information about a device IRQ. Caller provides
200 * struct vfio_irq_info with index value set. Caller sets argsz.
201 * Implementation of IRQ mapping is bus driver specific. Indexes
202 * using multiple IRQs are primarily intended to support MSI-like
203 * interrupt blocks. Zero count irq blocks may be used to describe
204 * unimplemented interrupt types.
206 * The EVENTFD flag indicates the interrupt index supports eventfd based
209 * The MASKABLE flags indicates the index supports MASK and UNMASK
210 * actions described below.
212 * AUTOMASKED indicates that after signaling, the interrupt line is
213 * automatically masked by VFIO and the user needs to unmask the line
214 * to receive new interrupts. This is primarily intended to distinguish
215 * level triggered interrupts.
217 * The NORESIZE flag indicates that the interrupt lines within the index
218 * are setup as a set and new subindexes cannot be enabled without first
219 * disabling the entire index. This is used for interrupts like PCI MSI
220 * and MSI-X where the driver may only use a subset of the available
221 * indexes, but VFIO needs to enable a specific number of vectors
222 * upfront. In the case of MSI-X, where the user can enable MSI-X and
223 * then add and unmask vectors, it's up to userspace to make the decision
224 * whether to allocate the maximum supported number of vectors or tear
225 * down setup and incrementally increase the vectors as each is enabled.
227 struct vfio_irq_info
{
230 #define VFIO_IRQ_INFO_EVENTFD (1 << 0)
231 #define VFIO_IRQ_INFO_MASKABLE (1 << 1)
232 #define VFIO_IRQ_INFO_AUTOMASKED (1 << 2)
233 #define VFIO_IRQ_INFO_NORESIZE (1 << 3)
234 __u32 index
; /* IRQ index */
235 __u32 count
; /* Number of IRQs within this index */
237 #define VFIO_DEVICE_GET_IRQ_INFO _IO(VFIO_TYPE, VFIO_BASE + 9)
240 * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
242 * Set signaling, masking, and unmasking of interrupts. Caller provides
243 * struct vfio_irq_set with all fields set. 'start' and 'count' indicate
244 * the range of subindexes being specified.
246 * The DATA flags specify the type of data provided. If DATA_NONE, the
247 * operation performs the specified action immediately on the specified
248 * interrupt(s). For example, to unmask AUTOMASKED interrupt [0,0]:
249 * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
251 * DATA_BOOL allows sparse support for the same on arrays of interrupts.
252 * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
253 * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
256 * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
257 * A value of -1 can be used to either de-assign interrupts if already
258 * assigned or skip un-assigned interrupts. For example, to set an eventfd
259 * to be trigger for interrupts [0,0] and [0,2]:
260 * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
261 * data = {fd1, -1, fd2}
262 * If index [0,1] is previously set, two count = 1 ioctls calls would be
263 * required to set [0,0] and [0,2] without changing [0,1].
265 * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
266 * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
267 * from userspace (ie. simulate hardware triggering).
269 * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
270 * enables the interrupt index for the device. Individual subindex interrupts
271 * can be disabled using the -1 value for DATA_EVENTFD or the index can be
272 * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
274 * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
275 * ACTION_TRIGGER specifies kernel->user signaling.
277 struct vfio_irq_set
{
280 #define VFIO_IRQ_SET_DATA_NONE (1 << 0) /* Data not present */
281 #define VFIO_IRQ_SET_DATA_BOOL (1 << 1) /* Data is bool (u8) */
282 #define VFIO_IRQ_SET_DATA_EVENTFD (1 << 2) /* Data is eventfd (s32) */
283 #define VFIO_IRQ_SET_ACTION_MASK (1 << 3) /* Mask interrupt */
284 #define VFIO_IRQ_SET_ACTION_UNMASK (1 << 4) /* Unmask interrupt */
285 #define VFIO_IRQ_SET_ACTION_TRIGGER (1 << 5) /* Trigger interrupt */
291 #define VFIO_DEVICE_SET_IRQS _IO(VFIO_TYPE, VFIO_BASE + 10)
293 #define VFIO_IRQ_SET_DATA_TYPE_MASK (VFIO_IRQ_SET_DATA_NONE | \
294 VFIO_IRQ_SET_DATA_BOOL | \
295 VFIO_IRQ_SET_DATA_EVENTFD)
296 #define VFIO_IRQ_SET_ACTION_TYPE_MASK (VFIO_IRQ_SET_ACTION_MASK | \
297 VFIO_IRQ_SET_ACTION_UNMASK | \
298 VFIO_IRQ_SET_ACTION_TRIGGER)
300 * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
304 #define VFIO_DEVICE_RESET _IO(VFIO_TYPE, VFIO_BASE + 11)
307 * The VFIO-PCI bus driver makes use of the following fixed region and
308 * IRQ index mapping. Unimplemented regions return a size of zero.
309 * Unimplemented IRQ types return a count of zero.
313 VFIO_PCI_BAR0_REGION_INDEX
,
314 VFIO_PCI_BAR1_REGION_INDEX
,
315 VFIO_PCI_BAR2_REGION_INDEX
,
316 VFIO_PCI_BAR3_REGION_INDEX
,
317 VFIO_PCI_BAR4_REGION_INDEX
,
318 VFIO_PCI_BAR5_REGION_INDEX
,
319 VFIO_PCI_ROM_REGION_INDEX
,
320 VFIO_PCI_CONFIG_REGION_INDEX
,
322 * Expose VGA regions defined for PCI base class 03, subclass 00.
323 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
324 * as well as the MMIO range 0xa0000 to 0xbffff. Each implemented
325 * range is found at it's identity mapped offset from the region
326 * offset, for example 0x3b0 is region_info.offset + 0x3b0. Areas
327 * between described ranges are unimplemented.
329 VFIO_PCI_VGA_REGION_INDEX
,
334 VFIO_PCI_INTX_IRQ_INDEX
,
335 VFIO_PCI_MSI_IRQ_INDEX
,
336 VFIO_PCI_MSIX_IRQ_INDEX
,
337 VFIO_PCI_ERR_IRQ_INDEX
,
338 VFIO_PCI_REQ_IRQ_INDEX
,
343 * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
344 * struct vfio_pci_hot_reset_info)
346 * Return: 0 on success, -errno on failure:
347 * -enospc = insufficient buffer, -enodev = unsupported for device.
349 struct vfio_pci_dependent_device
{
353 __u8 devfn
; /* Use PCI_SLOT/PCI_FUNC */
356 struct vfio_pci_hot_reset_info
{
360 struct vfio_pci_dependent_device devices
[];
363 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
366 * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
367 * struct vfio_pci_hot_reset)
369 * Return: 0 on success, -errno on failure.
371 struct vfio_pci_hot_reset
{
378 #define VFIO_DEVICE_PCI_HOT_RESET _IO(VFIO_TYPE, VFIO_BASE + 13)
380 /* -------- API for Type1 VFIO IOMMU -------- */
383 * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
385 * Retrieve information about the IOMMU object. Fills in provided
386 * struct vfio_iommu_info. Caller sets argsz.
388 * XXX Should we do these by CHECK_EXTENSION too?
390 struct vfio_iommu_type1_info
{
393 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0) /* supported page sizes info */
394 __u64 iova_pgsizes
; /* Bitmap of supported page sizes */
397 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
400 * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
402 * Map process virtual addresses to IO virtual addresses using the
403 * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
405 struct vfio_iommu_type1_dma_map
{
408 #define VFIO_DMA_MAP_FLAG_READ (1 << 0) /* readable from device */
409 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1) /* writable from device */
410 __u64 vaddr
; /* Process virtual address */
411 __u64 iova
; /* IO virtual address */
412 __u64 size
; /* Size of mapping (bytes) */
415 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
418 * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
419 * struct vfio_dma_unmap)
421 * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
422 * Caller sets argsz. The actual unmapped size is returned in the size
423 * field. No guarantee is made to the user that arbitrary unmaps of iova
424 * or size different from those used in the original mapping call will
427 struct vfio_iommu_type1_dma_unmap
{
430 __u64 iova
; /* IO virtual address */
431 __u64 size
; /* Size of mapping (bytes) */
434 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
437 * IOCTLs to enable/disable IOMMU container usage.
438 * No parameters are supported.
440 #define VFIO_IOMMU_ENABLE _IO(VFIO_TYPE, VFIO_BASE + 15)
441 #define VFIO_IOMMU_DISABLE _IO(VFIO_TYPE, VFIO_BASE + 16)
443 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
446 * The SPAPR TCE info struct provides the information about the PCI bus
447 * address ranges available for DMA, these values are programmed into
448 * the hardware so the guest has to know that information.
450 * The DMA 32 bit window start is an absolute PCI bus address.
451 * The IOVA address passed via map/unmap ioctls are absolute PCI bus
452 * addresses too so the window works as a filter rather than an offset
453 * for IOVA addresses.
455 * A flag will need to be added if other page sizes are supported,
456 * so as defined here, it is always 4k.
458 struct vfio_iommu_spapr_tce_info
{
460 __u32 flags
; /* reserved for future use */
461 __u32 dma32_window_start
; /* 32 bit window start (bytes) */
462 __u32 dma32_window_size
; /* 32 bit window size (bytes) */
465 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
468 * EEH PE operation struct provides ways to:
469 * - enable/disable EEH functionality;
470 * - unfreeze IO/DMA for frozen PE;
475 struct vfio_eeh_pe_op
{
481 #define VFIO_EEH_PE_DISABLE 0 /* Disable EEH functionality */
482 #define VFIO_EEH_PE_ENABLE 1 /* Enable EEH functionality */
483 #define VFIO_EEH_PE_UNFREEZE_IO 2 /* Enable IO for frozen PE */
484 #define VFIO_EEH_PE_UNFREEZE_DMA 3 /* Enable DMA for frozen PE */
485 #define VFIO_EEH_PE_GET_STATE 4 /* PE state retrieval */
486 #define VFIO_EEH_PE_STATE_NORMAL 0 /* PE in functional state */
487 #define VFIO_EEH_PE_STATE_RESET 1 /* PE reset in progress */
488 #define VFIO_EEH_PE_STATE_STOPPED 2 /* Stopped DMA and IO */
489 #define VFIO_EEH_PE_STATE_STOPPED_DMA 4 /* Stopped DMA only */
490 #define VFIO_EEH_PE_STATE_UNAVAIL 5 /* State unavailable */
491 #define VFIO_EEH_PE_RESET_DEACTIVATE 5 /* Deassert PE reset */
492 #define VFIO_EEH_PE_RESET_HOT 6 /* Assert hot reset */
493 #define VFIO_EEH_PE_RESET_FUNDAMENTAL 7 /* Assert fundamental reset */
494 #define VFIO_EEH_PE_CONFIGURE 8 /* PE configuration */
496 #define VFIO_EEH_PE_OP _IO(VFIO_TYPE, VFIO_BASE + 21)
498 /* ***************************************************************** */