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