xref: /openbmc/qemu/linux-headers/linux/vfio.h (revision 740b1759)
1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
2 /*
3  * VFIO API definition
4  *
5  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
6  *     Author: Alex Williamson <alex.williamson@redhat.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 #ifndef VFIO_H
13 #define VFIO_H
14 
15 #include <linux/types.h>
16 #include <linux/ioctl.h>
17 
18 #define VFIO_API_VERSION	0
19 
20 
21 /* Kernel & User level defines for VFIO IOCTLs. */
22 
23 /* Extensions */
24 
25 #define VFIO_TYPE1_IOMMU		1
26 #define VFIO_SPAPR_TCE_IOMMU		2
27 #define VFIO_TYPE1v2_IOMMU		3
28 /*
29  * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping).  This
30  * capability is subject to change as groups are added or removed.
31  */
32 #define VFIO_DMA_CC_IOMMU		4
33 
34 /* Check if EEH is supported */
35 #define VFIO_EEH			5
36 
37 /* Two-stage IOMMU */
38 #define VFIO_TYPE1_NESTING_IOMMU	6	/* Implies v2 */
39 
40 #define VFIO_SPAPR_TCE_v2_IOMMU		7
41 
42 /*
43  * The No-IOMMU IOMMU offers no translation or isolation for devices and
44  * supports no ioctls outside of VFIO_CHECK_EXTENSION.  Use of VFIO's No-IOMMU
45  * code will taint the host kernel and should be used with extreme caution.
46  */
47 #define VFIO_NOIOMMU_IOMMU		8
48 
49 /*
50  * The IOCTL interface is designed for extensibility by embedding the
51  * structure length (argsz) and flags into structures passed between
52  * kernel and userspace.  We therefore use the _IO() macro for these
53  * defines to avoid implicitly embedding a size into the ioctl request.
54  * As structure fields are added, argsz will increase to match and flag
55  * bits will be defined to indicate additional fields with valid data.
56  * It's *always* the caller's responsibility to indicate the size of
57  * the structure passed by setting argsz appropriately.
58  */
59 
60 #define VFIO_TYPE	(';')
61 #define VFIO_BASE	100
62 
63 /*
64  * For extension of INFO ioctls, VFIO makes use of a capability chain
65  * designed after PCI/e capabilities.  A flag bit indicates whether
66  * this capability chain is supported and a field defined in the fixed
67  * structure defines the offset of the first capability in the chain.
68  * This field is only valid when the corresponding bit in the flags
69  * bitmap is set.  This offset field is relative to the start of the
70  * INFO buffer, as is the next field within each capability header.
71  * The id within the header is a shared address space per INFO ioctl,
72  * while the version field is specific to the capability id.  The
73  * contents following the header are specific to the capability id.
74  */
75 struct vfio_info_cap_header {
76 	__u16	id;		/* Identifies capability */
77 	__u16	version;	/* Version specific to the capability ID */
78 	__u32	next;		/* Offset of next capability */
79 };
80 
81 /*
82  * Callers of INFO ioctls passing insufficiently sized buffers will see
83  * the capability chain flag bit set, a zero value for the first capability
84  * offset (if available within the provided argsz), and argsz will be
85  * updated to report the necessary buffer size.  For compatibility, the
86  * INFO ioctl will not report error in this case, but the capability chain
87  * will not be available.
88  */
89 
90 /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */
91 
92 /**
93  * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0)
94  *
95  * Report the version of the VFIO API.  This allows us to bump the entire
96  * API version should we later need to add or change features in incompatible
97  * ways.
98  * Return: VFIO_API_VERSION
99  * Availability: Always
100  */
101 #define VFIO_GET_API_VERSION		_IO(VFIO_TYPE, VFIO_BASE + 0)
102 
103 /**
104  * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32)
105  *
106  * Check whether an extension is supported.
107  * Return: 0 if not supported, 1 (or some other positive integer) if supported.
108  * Availability: Always
109  */
110 #define VFIO_CHECK_EXTENSION		_IO(VFIO_TYPE, VFIO_BASE + 1)
111 
112 /**
113  * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32)
114  *
115  * Set the iommu to the given type.  The type must be supported by an
116  * iommu driver as verified by calling CHECK_EXTENSION using the same
117  * type.  A group must be set to this file descriptor before this
118  * ioctl is available.  The IOMMU interfaces enabled by this call are
119  * specific to the value set.
120  * Return: 0 on success, -errno on failure
121  * Availability: When VFIO group attached
122  */
123 #define VFIO_SET_IOMMU			_IO(VFIO_TYPE, VFIO_BASE + 2)
124 
125 /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */
126 
127 /**
128  * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3,
129  *						struct vfio_group_status)
130  *
131  * Retrieve information about the group.  Fills in provided
132  * struct vfio_group_info.  Caller sets argsz.
133  * Return: 0 on succes, -errno on failure.
134  * Availability: Always
135  */
136 struct vfio_group_status {
137 	__u32	argsz;
138 	__u32	flags;
139 #define VFIO_GROUP_FLAGS_VIABLE		(1 << 0)
140 #define VFIO_GROUP_FLAGS_CONTAINER_SET	(1 << 1)
141 };
142 #define VFIO_GROUP_GET_STATUS		_IO(VFIO_TYPE, VFIO_BASE + 3)
143 
144 /**
145  * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32)
146  *
147  * Set the container for the VFIO group to the open VFIO file
148  * descriptor provided.  Groups may only belong to a single
149  * container.  Containers may, at their discretion, support multiple
150  * groups.  Only when a container is set are all of the interfaces
151  * of the VFIO file descriptor and the VFIO group file descriptor
152  * available to the user.
153  * Return: 0 on success, -errno on failure.
154  * Availability: Always
155  */
156 #define VFIO_GROUP_SET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 4)
157 
158 /**
159  * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5)
160  *
161  * Remove the group from the attached container.  This is the
162  * opposite of the SET_CONTAINER call and returns the group to
163  * an initial state.  All device file descriptors must be released
164  * prior to calling this interface.  When removing the last group
165  * from a container, the IOMMU will be disabled and all state lost,
166  * effectively also returning the VFIO file descriptor to an initial
167  * state.
168  * Return: 0 on success, -errno on failure.
169  * Availability: When attached to container
170  */
171 #define VFIO_GROUP_UNSET_CONTAINER	_IO(VFIO_TYPE, VFIO_BASE + 5)
172 
173 /**
174  * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char)
175  *
176  * Return a new file descriptor for the device object described by
177  * the provided string.  The string should match a device listed in
178  * the devices subdirectory of the IOMMU group sysfs entry.  The
179  * group containing the device must already be added to this context.
180  * Return: new file descriptor on success, -errno on failure.
181  * Availability: When attached to container
182  */
183 #define VFIO_GROUP_GET_DEVICE_FD	_IO(VFIO_TYPE, VFIO_BASE + 6)
184 
185 /* --------------- IOCTLs for DEVICE file descriptors --------------- */
186 
187 /**
188  * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7,
189  *						struct vfio_device_info)
190  *
191  * Retrieve information about the device.  Fills in provided
192  * struct vfio_device_info.  Caller sets argsz.
193  * Return: 0 on success, -errno on failure.
194  */
195 struct vfio_device_info {
196 	__u32	argsz;
197 	__u32	flags;
198 #define VFIO_DEVICE_FLAGS_RESET	(1 << 0)	/* Device supports reset */
199 #define VFIO_DEVICE_FLAGS_PCI	(1 << 1)	/* vfio-pci device */
200 #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2)	/* vfio-platform device */
201 #define VFIO_DEVICE_FLAGS_AMBA  (1 << 3)	/* vfio-amba device */
202 #define VFIO_DEVICE_FLAGS_CCW	(1 << 4)	/* vfio-ccw device */
203 #define VFIO_DEVICE_FLAGS_AP	(1 << 5)	/* vfio-ap device */
204 	__u32	num_regions;	/* Max region index + 1 */
205 	__u32	num_irqs;	/* Max IRQ index + 1 */
206 };
207 #define VFIO_DEVICE_GET_INFO		_IO(VFIO_TYPE, VFIO_BASE + 7)
208 
209 /*
210  * Vendor driver using Mediated device framework should provide device_api
211  * attribute in supported type attribute groups. Device API string should be one
212  * of the following corresponding to device flags in vfio_device_info structure.
213  */
214 
215 #define VFIO_DEVICE_API_PCI_STRING		"vfio-pci"
216 #define VFIO_DEVICE_API_PLATFORM_STRING		"vfio-platform"
217 #define VFIO_DEVICE_API_AMBA_STRING		"vfio-amba"
218 #define VFIO_DEVICE_API_CCW_STRING		"vfio-ccw"
219 #define VFIO_DEVICE_API_AP_STRING		"vfio-ap"
220 
221 /**
222  * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8,
223  *				       struct vfio_region_info)
224  *
225  * Retrieve information about a device region.  Caller provides
226  * struct vfio_region_info with index value set.  Caller sets argsz.
227  * Implementation of region mapping is bus driver specific.  This is
228  * intended to describe MMIO, I/O port, as well as bus specific
229  * regions (ex. PCI config space).  Zero sized regions may be used
230  * to describe unimplemented regions (ex. unimplemented PCI BARs).
231  * Return: 0 on success, -errno on failure.
232  */
233 struct vfio_region_info {
234 	__u32	argsz;
235 	__u32	flags;
236 #define VFIO_REGION_INFO_FLAG_READ	(1 << 0) /* Region supports read */
237 #define VFIO_REGION_INFO_FLAG_WRITE	(1 << 1) /* Region supports write */
238 #define VFIO_REGION_INFO_FLAG_MMAP	(1 << 2) /* Region supports mmap */
239 #define VFIO_REGION_INFO_FLAG_CAPS	(1 << 3) /* Info supports caps */
240 	__u32	index;		/* Region index */
241 	__u32	cap_offset;	/* Offset within info struct of first cap */
242 	__u64	size;		/* Region size (bytes) */
243 	__u64	offset;		/* Region offset from start of device fd */
244 };
245 #define VFIO_DEVICE_GET_REGION_INFO	_IO(VFIO_TYPE, VFIO_BASE + 8)
246 
247 /*
248  * The sparse mmap capability allows finer granularity of specifying areas
249  * within a region with mmap support.  When specified, the user should only
250  * mmap the offset ranges specified by the areas array.  mmaps outside of the
251  * areas specified may fail (such as the range covering a PCI MSI-X table) or
252  * may result in improper device behavior.
253  *
254  * The structures below define version 1 of this capability.
255  */
256 #define VFIO_REGION_INFO_CAP_SPARSE_MMAP	1
257 
258 struct vfio_region_sparse_mmap_area {
259 	__u64	offset;	/* Offset of mmap'able area within region */
260 	__u64	size;	/* Size of mmap'able area */
261 };
262 
263 struct vfio_region_info_cap_sparse_mmap {
264 	struct vfio_info_cap_header header;
265 	__u32	nr_areas;
266 	__u32	reserved;
267 	struct vfio_region_sparse_mmap_area areas[];
268 };
269 
270 /*
271  * The device specific type capability allows regions unique to a specific
272  * device or class of devices to be exposed.  This helps solve the problem for
273  * vfio bus drivers of defining which region indexes correspond to which region
274  * on the device, without needing to resort to static indexes, as done by
275  * vfio-pci.  For instance, if we were to go back in time, we might remove
276  * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes
277  * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd
278  * make a "VGA" device specific type to describe the VGA access space.  This
279  * means that non-VGA devices wouldn't need to waste this index, and thus the
280  * address space associated with it due to implementation of device file
281  * descriptor offsets in vfio-pci.
282  *
283  * The current implementation is now part of the user ABI, so we can't use this
284  * for VGA, but there are other upcoming use cases, such as opregions for Intel
285  * IGD devices and framebuffers for vGPU devices.  We missed VGA, but we'll
286  * use this for future additions.
287  *
288  * The structure below defines version 1 of this capability.
289  */
290 #define VFIO_REGION_INFO_CAP_TYPE	2
291 
292 struct vfio_region_info_cap_type {
293 	struct vfio_info_cap_header header;
294 	__u32 type;	/* global per bus driver */
295 	__u32 subtype;	/* type specific */
296 };
297 
298 /*
299  * List of region types, global per bus driver.
300  * If you introduce a new type, please add it here.
301  */
302 
303 /* PCI region type containing a PCI vendor part */
304 #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE	(1 << 31)
305 #define VFIO_REGION_TYPE_PCI_VENDOR_MASK	(0xffff)
306 #define VFIO_REGION_TYPE_GFX                    (1)
307 #define VFIO_REGION_TYPE_CCW			(2)
308 #define VFIO_REGION_TYPE_MIGRATION              (3)
309 
310 /* sub-types for VFIO_REGION_TYPE_PCI_* */
311 
312 /* 8086 vendor PCI sub-types */
313 #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION	(1)
314 #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG	(2)
315 #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG	(3)
316 
317 /* 10de vendor PCI sub-types */
318 /*
319  * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space.
320  */
321 #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM	(1)
322 
323 /* 1014 vendor PCI sub-types */
324 /*
325  * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU
326  * to do TLB invalidation on a GPU.
327  */
328 #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD	(1)
329 
330 /* sub-types for VFIO_REGION_TYPE_GFX */
331 #define VFIO_REGION_SUBTYPE_GFX_EDID            (1)
332 
333 /**
334  * struct vfio_region_gfx_edid - EDID region layout.
335  *
336  * Set display link state and EDID blob.
337  *
338  * The EDID blob has monitor information such as brand, name, serial
339  * number, physical size, supported video modes and more.
340  *
341  * This special region allows userspace (typically qemu) set a virtual
342  * EDID for the virtual monitor, which allows a flexible display
343  * configuration.
344  *
345  * For the edid blob spec look here:
346  *    https://en.wikipedia.org/wiki/Extended_Display_Identification_Data
347  *
348  * On linux systems you can find the EDID blob in sysfs:
349  *    /sys/class/drm/${card}/${connector}/edid
350  *
351  * You can use the edid-decode ulility (comes with xorg-x11-utils) to
352  * decode the EDID blob.
353  *
354  * @edid_offset: location of the edid blob, relative to the
355  *               start of the region (readonly).
356  * @edid_max_size: max size of the edid blob (readonly).
357  * @edid_size: actual edid size (read/write).
358  * @link_state: display link state (read/write).
359  * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on.
360  * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off.
361  * @max_xres: max display width (0 == no limitation, readonly).
362  * @max_yres: max display height (0 == no limitation, readonly).
363  *
364  * EDID update protocol:
365  *   (1) set link-state to down.
366  *   (2) update edid blob and size.
367  *   (3) set link-state to up.
368  */
369 struct vfio_region_gfx_edid {
370 	__u32 edid_offset;
371 	__u32 edid_max_size;
372 	__u32 edid_size;
373 	__u32 max_xres;
374 	__u32 max_yres;
375 	__u32 link_state;
376 #define VFIO_DEVICE_GFX_LINK_STATE_UP    1
377 #define VFIO_DEVICE_GFX_LINK_STATE_DOWN  2
378 };
379 
380 /* sub-types for VFIO_REGION_TYPE_CCW */
381 #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD	(1)
382 #define VFIO_REGION_SUBTYPE_CCW_SCHIB		(2)
383 #define VFIO_REGION_SUBTYPE_CCW_CRW		(3)
384 
385 /* sub-types for VFIO_REGION_TYPE_MIGRATION */
386 #define VFIO_REGION_SUBTYPE_MIGRATION           (1)
387 
388 /*
389  * The structure vfio_device_migration_info is placed at the 0th offset of
390  * the VFIO_REGION_SUBTYPE_MIGRATION region to get and set VFIO device related
391  * migration information. Field accesses from this structure are only supported
392  * at their native width and alignment. Otherwise, the result is undefined and
393  * vendor drivers should return an error.
394  *
395  * device_state: (read/write)
396  *      - The user application writes to this field to inform the vendor driver
397  *        about the device state to be transitioned to.
398  *      - The vendor driver should take the necessary actions to change the
399  *        device state. After successful transition to a given state, the
400  *        vendor driver should return success on write(device_state, state)
401  *        system call. If the device state transition fails, the vendor driver
402  *        should return an appropriate -errno for the fault condition.
403  *      - On the user application side, if the device state transition fails,
404  *	  that is, if write(device_state, state) returns an error, read
405  *	  device_state again to determine the current state of the device from
406  *	  the vendor driver.
407  *      - The vendor driver should return previous state of the device unless
408  *        the vendor driver has encountered an internal error, in which case
409  *        the vendor driver may report the device_state VFIO_DEVICE_STATE_ERROR.
410  *      - The user application must use the device reset ioctl to recover the
411  *        device from VFIO_DEVICE_STATE_ERROR state. If the device is
412  *        indicated to be in a valid device state by reading device_state, the
413  *        user application may attempt to transition the device to any valid
414  *        state reachable from the current state or terminate itself.
415  *
416  *      device_state consists of 3 bits:
417  *      - If bit 0 is set, it indicates the _RUNNING state. If bit 0 is clear,
418  *        it indicates the _STOP state. When the device state is changed to
419  *        _STOP, driver should stop the device before write() returns.
420  *      - If bit 1 is set, it indicates the _SAVING state, which means that the
421  *        driver should start gathering device state information that will be
422  *        provided to the VFIO user application to save the device's state.
423  *      - If bit 2 is set, it indicates the _RESUMING state, which means that
424  *        the driver should prepare to resume the device. Data provided through
425  *        the migration region should be used to resume the device.
426  *      Bits 3 - 31 are reserved for future use. To preserve them, the user
427  *      application should perform a read-modify-write operation on this
428  *      field when modifying the specified bits.
429  *
430  *  +------- _RESUMING
431  *  |+------ _SAVING
432  *  ||+----- _RUNNING
433  *  |||
434  *  000b => Device Stopped, not saving or resuming
435  *  001b => Device running, which is the default state
436  *  010b => Stop the device & save the device state, stop-and-copy state
437  *  011b => Device running and save the device state, pre-copy state
438  *  100b => Device stopped and the device state is resuming
439  *  101b => Invalid state
440  *  110b => Error state
441  *  111b => Invalid state
442  *
443  * State transitions:
444  *
445  *              _RESUMING  _RUNNING    Pre-copy    Stop-and-copy   _STOP
446  *                (100b)     (001b)     (011b)        (010b)       (000b)
447  * 0. Running or default state
448  *                             |
449  *
450  * 1. Normal Shutdown (optional)
451  *                             |------------------------------------->|
452  *
453  * 2. Save the state or suspend
454  *                             |------------------------->|---------->|
455  *
456  * 3. Save the state during live migration
457  *                             |----------->|------------>|---------->|
458  *
459  * 4. Resuming
460  *                  |<---------|
461  *
462  * 5. Resumed
463  *                  |--------->|
464  *
465  * 0. Default state of VFIO device is _RUNNNG when the user application starts.
466  * 1. During normal shutdown of the user application, the user application may
467  *    optionally change the VFIO device state from _RUNNING to _STOP. This
468  *    transition is optional. The vendor driver must support this transition but
469  *    must not require it.
470  * 2. When the user application saves state or suspends the application, the
471  *    device state transitions from _RUNNING to stop-and-copy and then to _STOP.
472  *    On state transition from _RUNNING to stop-and-copy, driver must stop the
473  *    device, save the device state and send it to the application through the
474  *    migration region. The sequence to be followed for such transition is given
475  *    below.
476  * 3. In live migration of user application, the state transitions from _RUNNING
477  *    to pre-copy, to stop-and-copy, and to _STOP.
478  *    On state transition from _RUNNING to pre-copy, the driver should start
479  *    gathering the device state while the application is still running and send
480  *    the device state data to application through the migration region.
481  *    On state transition from pre-copy to stop-and-copy, the driver must stop
482  *    the device, save the device state and send it to the user application
483  *    through the migration region.
484  *    Vendor drivers must support the pre-copy state even for implementations
485  *    where no data is provided to the user before the stop-and-copy state. The
486  *    user must not be required to consume all migration data before the device
487  *    transitions to a new state, including the stop-and-copy state.
488  *    The sequence to be followed for above two transitions is given below.
489  * 4. To start the resuming phase, the device state should be transitioned from
490  *    the _RUNNING to the _RESUMING state.
491  *    In the _RESUMING state, the driver should use the device state data
492  *    received through the migration region to resume the device.
493  * 5. After providing saved device data to the driver, the application should
494  *    change the state from _RESUMING to _RUNNING.
495  *
496  * reserved:
497  *      Reads on this field return zero and writes are ignored.
498  *
499  * pending_bytes: (read only)
500  *      The number of pending bytes still to be migrated from the vendor driver.
501  *
502  * data_offset: (read only)
503  *      The user application should read data_offset field from the migration
504  *      region. The user application should read the device data from this
505  *      offset within the migration region during the _SAVING state or write
506  *      the device data during the _RESUMING state. See below for details of
507  *      sequence to be followed.
508  *
509  * data_size: (read/write)
510  *      The user application should read data_size to get the size in bytes of
511  *      the data copied in the migration region during the _SAVING state and
512  *      write the size in bytes of the data copied in the migration region
513  *      during the _RESUMING state.
514  *
515  * The format of the migration region is as follows:
516  *  ------------------------------------------------------------------
517  * |vfio_device_migration_info|    data section                      |
518  * |                          |     ///////////////////////////////  |
519  * ------------------------------------------------------------------
520  *   ^                              ^
521  *  offset 0-trapped part        data_offset
522  *
523  * The structure vfio_device_migration_info is always followed by the data
524  * section in the region, so data_offset will always be nonzero. The offset
525  * from where the data is copied is decided by the kernel driver. The data
526  * section can be trapped, mmapped, or partitioned, depending on how the kernel
527  * driver defines the data section. The data section partition can be defined
528  * as mapped by the sparse mmap capability. If mmapped, data_offset must be
529  * page aligned, whereas initial section which contains the
530  * vfio_device_migration_info structure, might not end at the offset, which is
531  * page aligned. The user is not required to access through mmap regardless
532  * of the capabilities of the region mmap.
533  * The vendor driver should determine whether and how to partition the data
534  * section. The vendor driver should return data_offset accordingly.
535  *
536  * The sequence to be followed while in pre-copy state and stop-and-copy state
537  * is as follows:
538  * a. Read pending_bytes, indicating the start of a new iteration to get device
539  *    data. Repeated read on pending_bytes at this stage should have no side
540  *    effects.
541  *    If pending_bytes == 0, the user application should not iterate to get data
542  *    for that device.
543  *    If pending_bytes > 0, perform the following steps.
544  * b. Read data_offset, indicating that the vendor driver should make data
545  *    available through the data section. The vendor driver should return this
546  *    read operation only after data is available from (region + data_offset)
547  *    to (region + data_offset + data_size).
548  * c. Read data_size, which is the amount of data in bytes available through
549  *    the migration region.
550  *    Read on data_offset and data_size should return the offset and size of
551  *    the current buffer if the user application reads data_offset and
552  *    data_size more than once here.
553  * d. Read data_size bytes of data from (region + data_offset) from the
554  *    migration region.
555  * e. Process the data.
556  * f. Read pending_bytes, which indicates that the data from the previous
557  *    iteration has been read. If pending_bytes > 0, go to step b.
558  *
559  * The user application can transition from the _SAVING|_RUNNING
560  * (pre-copy state) to the _SAVING (stop-and-copy) state regardless of the
561  * number of pending bytes. The user application should iterate in _SAVING
562  * (stop-and-copy) until pending_bytes is 0.
563  *
564  * The sequence to be followed while _RESUMING device state is as follows:
565  * While data for this device is available, repeat the following steps:
566  * a. Read data_offset from where the user application should write data.
567  * b. Write migration data starting at the migration region + data_offset for
568  *    the length determined by data_size from the migration source.
569  * c. Write data_size, which indicates to the vendor driver that data is
570  *    written in the migration region. Vendor driver must return this write
571  *    operations on consuming data. Vendor driver should apply the
572  *    user-provided migration region data to the device resume state.
573  *
574  * If an error occurs during the above sequences, the vendor driver can return
575  * an error code for next read() or write() operation, which will terminate the
576  * loop. The user application should then take the next necessary action, for
577  * example, failing migration or terminating the user application.
578  *
579  * For the user application, data is opaque. The user application should write
580  * data in the same order as the data is received and the data should be of
581  * same transaction size at the source.
582  */
583 
584 struct vfio_device_migration_info {
585 	__u32 device_state;         /* VFIO device state */
586 #define VFIO_DEVICE_STATE_STOP      (0)
587 #define VFIO_DEVICE_STATE_RUNNING   (1 << 0)
588 #define VFIO_DEVICE_STATE_SAVING    (1 << 1)
589 #define VFIO_DEVICE_STATE_RESUMING  (1 << 2)
590 #define VFIO_DEVICE_STATE_MASK      (VFIO_DEVICE_STATE_RUNNING | \
591 				     VFIO_DEVICE_STATE_SAVING |  \
592 				     VFIO_DEVICE_STATE_RESUMING)
593 
594 #define VFIO_DEVICE_STATE_VALID(state) \
595 	(state & VFIO_DEVICE_STATE_RESUMING ? \
596 	(state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_RESUMING : 1)
597 
598 #define VFIO_DEVICE_STATE_IS_ERROR(state) \
599 	((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_SAVING | \
600 					      VFIO_DEVICE_STATE_RESUMING))
601 
602 #define VFIO_DEVICE_STATE_SET_ERROR(state) \
603 	((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_SATE_SAVING | \
604 					     VFIO_DEVICE_STATE_RESUMING)
605 
606 	__u32 reserved;
607 	__u64 pending_bytes;
608 	__u64 data_offset;
609 	__u64 data_size;
610 };
611 
612 /*
613  * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped
614  * which allows direct access to non-MSIX registers which happened to be within
615  * the same system page.
616  *
617  * Even though the userspace gets direct access to the MSIX data, the existing
618  * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration.
619  */
620 #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE	3
621 
622 /*
623  * Capability with compressed real address (aka SSA - small system address)
624  * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing
625  * and by the userspace to associate a NVLink bridge with a GPU.
626  */
627 #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT	4
628 
629 struct vfio_region_info_cap_nvlink2_ssatgt {
630 	struct vfio_info_cap_header header;
631 	__u64 tgt;
632 };
633 
634 /*
635  * Capability with an NVLink link speed. The value is read by
636  * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed"
637  * property in the device tree. The value is fixed in the hardware
638  * and failing to provide the correct value results in the link
639  * not working with no indication from the driver why.
640  */
641 #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD	5
642 
643 struct vfio_region_info_cap_nvlink2_lnkspd {
644 	struct vfio_info_cap_header header;
645 	__u32 link_speed;
646 	__u32 __pad;
647 };
648 
649 /**
650  * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9,
651  *				    struct vfio_irq_info)
652  *
653  * Retrieve information about a device IRQ.  Caller provides
654  * struct vfio_irq_info with index value set.  Caller sets argsz.
655  * Implementation of IRQ mapping is bus driver specific.  Indexes
656  * using multiple IRQs are primarily intended to support MSI-like
657  * interrupt blocks.  Zero count irq blocks may be used to describe
658  * unimplemented interrupt types.
659  *
660  * The EVENTFD flag indicates the interrupt index supports eventfd based
661  * signaling.
662  *
663  * The MASKABLE flags indicates the index supports MASK and UNMASK
664  * actions described below.
665  *
666  * AUTOMASKED indicates that after signaling, the interrupt line is
667  * automatically masked by VFIO and the user needs to unmask the line
668  * to receive new interrupts.  This is primarily intended to distinguish
669  * level triggered interrupts.
670  *
671  * The NORESIZE flag indicates that the interrupt lines within the index
672  * are setup as a set and new subindexes cannot be enabled without first
673  * disabling the entire index.  This is used for interrupts like PCI MSI
674  * and MSI-X where the driver may only use a subset of the available
675  * indexes, but VFIO needs to enable a specific number of vectors
676  * upfront.  In the case of MSI-X, where the user can enable MSI-X and
677  * then add and unmask vectors, it's up to userspace to make the decision
678  * whether to allocate the maximum supported number of vectors or tear
679  * down setup and incrementally increase the vectors as each is enabled.
680  */
681 struct vfio_irq_info {
682 	__u32	argsz;
683 	__u32	flags;
684 #define VFIO_IRQ_INFO_EVENTFD		(1 << 0)
685 #define VFIO_IRQ_INFO_MASKABLE		(1 << 1)
686 #define VFIO_IRQ_INFO_AUTOMASKED	(1 << 2)
687 #define VFIO_IRQ_INFO_NORESIZE		(1 << 3)
688 	__u32	index;		/* IRQ index */
689 	__u32	count;		/* Number of IRQs within this index */
690 };
691 #define VFIO_DEVICE_GET_IRQ_INFO	_IO(VFIO_TYPE, VFIO_BASE + 9)
692 
693 /**
694  * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set)
695  *
696  * Set signaling, masking, and unmasking of interrupts.  Caller provides
697  * struct vfio_irq_set with all fields set.  'start' and 'count' indicate
698  * the range of subindexes being specified.
699  *
700  * The DATA flags specify the type of data provided.  If DATA_NONE, the
701  * operation performs the specified action immediately on the specified
702  * interrupt(s).  For example, to unmask AUTOMASKED interrupt [0,0]:
703  * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1.
704  *
705  * DATA_BOOL allows sparse support for the same on arrays of interrupts.
706  * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]):
707  * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3,
708  * data = {1,0,1}
709  *
710  * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd.
711  * A value of -1 can be used to either de-assign interrupts if already
712  * assigned or skip un-assigned interrupts.  For example, to set an eventfd
713  * to be trigger for interrupts [0,0] and [0,2]:
714  * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3,
715  * data = {fd1, -1, fd2}
716  * If index [0,1] is previously set, two count = 1 ioctls calls would be
717  * required to set [0,0] and [0,2] without changing [0,1].
718  *
719  * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used
720  * with ACTION_TRIGGER to perform kernel level interrupt loopback testing
721  * from userspace (ie. simulate hardware triggering).
722  *
723  * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER
724  * enables the interrupt index for the device.  Individual subindex interrupts
725  * can be disabled using the -1 value for DATA_EVENTFD or the index can be
726  * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0.
727  *
728  * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while
729  * ACTION_TRIGGER specifies kernel->user signaling.
730  */
731 struct vfio_irq_set {
732 	__u32	argsz;
733 	__u32	flags;
734 #define VFIO_IRQ_SET_DATA_NONE		(1 << 0) /* Data not present */
735 #define VFIO_IRQ_SET_DATA_BOOL		(1 << 1) /* Data is bool (u8) */
736 #define VFIO_IRQ_SET_DATA_EVENTFD	(1 << 2) /* Data is eventfd (s32) */
737 #define VFIO_IRQ_SET_ACTION_MASK	(1 << 3) /* Mask interrupt */
738 #define VFIO_IRQ_SET_ACTION_UNMASK	(1 << 4) /* Unmask interrupt */
739 #define VFIO_IRQ_SET_ACTION_TRIGGER	(1 << 5) /* Trigger interrupt */
740 	__u32	index;
741 	__u32	start;
742 	__u32	count;
743 	__u8	data[];
744 };
745 #define VFIO_DEVICE_SET_IRQS		_IO(VFIO_TYPE, VFIO_BASE + 10)
746 
747 #define VFIO_IRQ_SET_DATA_TYPE_MASK	(VFIO_IRQ_SET_DATA_NONE | \
748 					 VFIO_IRQ_SET_DATA_BOOL | \
749 					 VFIO_IRQ_SET_DATA_EVENTFD)
750 #define VFIO_IRQ_SET_ACTION_TYPE_MASK	(VFIO_IRQ_SET_ACTION_MASK | \
751 					 VFIO_IRQ_SET_ACTION_UNMASK | \
752 					 VFIO_IRQ_SET_ACTION_TRIGGER)
753 /**
754  * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11)
755  *
756  * Reset a device.
757  */
758 #define VFIO_DEVICE_RESET		_IO(VFIO_TYPE, VFIO_BASE + 11)
759 
760 /*
761  * The VFIO-PCI bus driver makes use of the following fixed region and
762  * IRQ index mapping.  Unimplemented regions return a size of zero.
763  * Unimplemented IRQ types return a count of zero.
764  */
765 
766 enum {
767 	VFIO_PCI_BAR0_REGION_INDEX,
768 	VFIO_PCI_BAR1_REGION_INDEX,
769 	VFIO_PCI_BAR2_REGION_INDEX,
770 	VFIO_PCI_BAR3_REGION_INDEX,
771 	VFIO_PCI_BAR4_REGION_INDEX,
772 	VFIO_PCI_BAR5_REGION_INDEX,
773 	VFIO_PCI_ROM_REGION_INDEX,
774 	VFIO_PCI_CONFIG_REGION_INDEX,
775 	/*
776 	 * Expose VGA regions defined for PCI base class 03, subclass 00.
777 	 * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df
778 	 * as well as the MMIO range 0xa0000 to 0xbffff.  Each implemented
779 	 * range is found at it's identity mapped offset from the region
780 	 * offset, for example 0x3b0 is region_info.offset + 0x3b0.  Areas
781 	 * between described ranges are unimplemented.
782 	 */
783 	VFIO_PCI_VGA_REGION_INDEX,
784 	VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */
785 				 /* device specific cap to define content. */
786 };
787 
788 enum {
789 	VFIO_PCI_INTX_IRQ_INDEX,
790 	VFIO_PCI_MSI_IRQ_INDEX,
791 	VFIO_PCI_MSIX_IRQ_INDEX,
792 	VFIO_PCI_ERR_IRQ_INDEX,
793 	VFIO_PCI_REQ_IRQ_INDEX,
794 	VFIO_PCI_NUM_IRQS
795 };
796 
797 /*
798  * The vfio-ccw bus driver makes use of the following fixed region and
799  * IRQ index mapping. Unimplemented regions return a size of zero.
800  * Unimplemented IRQ types return a count of zero.
801  */
802 
803 enum {
804 	VFIO_CCW_CONFIG_REGION_INDEX,
805 	VFIO_CCW_NUM_REGIONS
806 };
807 
808 enum {
809 	VFIO_CCW_IO_IRQ_INDEX,
810 	VFIO_CCW_CRW_IRQ_INDEX,
811 	VFIO_CCW_NUM_IRQS
812 };
813 
814 /**
815  * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IORW(VFIO_TYPE, VFIO_BASE + 12,
816  *					      struct vfio_pci_hot_reset_info)
817  *
818  * Return: 0 on success, -errno on failure:
819  *	-enospc = insufficient buffer, -enodev = unsupported for device.
820  */
821 struct vfio_pci_dependent_device {
822 	__u32	group_id;
823 	__u16	segment;
824 	__u8	bus;
825 	__u8	devfn; /* Use PCI_SLOT/PCI_FUNC */
826 };
827 
828 struct vfio_pci_hot_reset_info {
829 	__u32	argsz;
830 	__u32	flags;
831 	__u32	count;
832 	struct vfio_pci_dependent_device	devices[];
833 };
834 
835 #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
836 
837 /**
838  * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13,
839  *				    struct vfio_pci_hot_reset)
840  *
841  * Return: 0 on success, -errno on failure.
842  */
843 struct vfio_pci_hot_reset {
844 	__u32	argsz;
845 	__u32	flags;
846 	__u32	count;
847 	__s32	group_fds[];
848 };
849 
850 #define VFIO_DEVICE_PCI_HOT_RESET	_IO(VFIO_TYPE, VFIO_BASE + 13)
851 
852 /**
853  * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14,
854  *                                    struct vfio_device_query_gfx_plane)
855  *
856  * Set the drm_plane_type and flags, then retrieve the gfx plane info.
857  *
858  * flags supported:
859  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set
860  *   to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no
861  *   support for dma-buf.
862  * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set
863  *   to ask if the mdev supports region. 0 on support, -EINVAL on no
864  *   support for region.
865  * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set
866  *   with each call to query the plane info.
867  * - Others are invalid and return -EINVAL.
868  *
869  * Note:
870  * 1. Plane could be disabled by guest. In that case, success will be
871  *    returned with zero-initialized drm_format, size, width and height
872  *    fields.
873  * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available
874  *
875  * Return: 0 on success, -errno on other failure.
876  */
877 struct vfio_device_gfx_plane_info {
878 	__u32 argsz;
879 	__u32 flags;
880 #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0)
881 #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1)
882 #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2)
883 	/* in */
884 	__u32 drm_plane_type;	/* type of plane: DRM_PLANE_TYPE_* */
885 	/* out */
886 	__u32 drm_format;	/* drm format of plane */
887 	__u64 drm_format_mod;   /* tiled mode */
888 	__u32 width;	/* width of plane */
889 	__u32 height;	/* height of plane */
890 	__u32 stride;	/* stride of plane */
891 	__u32 size;	/* size of plane in bytes, align on page*/
892 	__u32 x_pos;	/* horizontal position of cursor plane */
893 	__u32 y_pos;	/* vertical position of cursor plane*/
894 	__u32 x_hot;    /* horizontal position of cursor hotspot */
895 	__u32 y_hot;    /* vertical position of cursor hotspot */
896 	union {
897 		__u32 region_index;	/* region index */
898 		__u32 dmabuf_id;	/* dma-buf id */
899 	};
900 };
901 
902 #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14)
903 
904 /**
905  * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32)
906  *
907  * Return a new dma-buf file descriptor for an exposed guest framebuffer
908  * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_
909  * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer.
910  */
911 
912 #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15)
913 
914 /**
915  * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16,
916  *                              struct vfio_device_ioeventfd)
917  *
918  * Perform a write to the device at the specified device fd offset, with
919  * the specified data and width when the provided eventfd is triggered.
920  * vfio bus drivers may not support this for all regions, for all widths,
921  * or at all.  vfio-pci currently only enables support for BAR regions,
922  * excluding the MSI-X vector table.
923  *
924  * Return: 0 on success, -errno on failure.
925  */
926 struct vfio_device_ioeventfd {
927 	__u32	argsz;
928 	__u32	flags;
929 #define VFIO_DEVICE_IOEVENTFD_8		(1 << 0) /* 1-byte write */
930 #define VFIO_DEVICE_IOEVENTFD_16	(1 << 1) /* 2-byte write */
931 #define VFIO_DEVICE_IOEVENTFD_32	(1 << 2) /* 4-byte write */
932 #define VFIO_DEVICE_IOEVENTFD_64	(1 << 3) /* 8-byte write */
933 #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK	(0xf)
934 	__u64	offset;			/* device fd offset of write */
935 	__u64	data;			/* data to be written */
936 	__s32	fd;			/* -1 for de-assignment */
937 };
938 
939 #define VFIO_DEVICE_IOEVENTFD		_IO(VFIO_TYPE, VFIO_BASE + 16)
940 
941 /**
942  * VFIO_DEVICE_FEATURE - _IORW(VFIO_TYPE, VFIO_BASE + 17,
943  *			       struct vfio_device_feature)
944  *
945  * Get, set, or probe feature data of the device.  The feature is selected
946  * using the FEATURE_MASK portion of the flags field.  Support for a feature
947  * can be probed by setting both the FEATURE_MASK and PROBE bits.  A probe
948  * may optionally include the GET and/or SET bits to determine read vs write
949  * access of the feature respectively.  Probing a feature will return success
950  * if the feature is supported and all of the optionally indicated GET/SET
951  * methods are supported.  The format of the data portion of the structure is
952  * specific to the given feature.  The data portion is not required for
953  * probing.  GET and SET are mutually exclusive, except for use with PROBE.
954  *
955  * Return 0 on success, -errno on failure.
956  */
957 struct vfio_device_feature {
958 	__u32	argsz;
959 	__u32	flags;
960 #define VFIO_DEVICE_FEATURE_MASK	(0xffff) /* 16-bit feature index */
961 #define VFIO_DEVICE_FEATURE_GET		(1 << 16) /* Get feature into data[] */
962 #define VFIO_DEVICE_FEATURE_SET		(1 << 17) /* Set feature from data[] */
963 #define VFIO_DEVICE_FEATURE_PROBE	(1 << 18) /* Probe feature support */
964 	__u8	data[];
965 };
966 
967 #define VFIO_DEVICE_FEATURE		_IO(VFIO_TYPE, VFIO_BASE + 17)
968 
969 /*
970  * Provide support for setting a PCI VF Token, which is used as a shared
971  * secret between PF and VF drivers.  This feature may only be set on a
972  * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing
973  * open VFs.  Data provided when setting this feature is a 16-byte array
974  * (__u8 b[16]), representing a UUID.
975  */
976 #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN	(0)
977 
978 /* -------- API for Type1 VFIO IOMMU -------- */
979 
980 /**
981  * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info)
982  *
983  * Retrieve information about the IOMMU object. Fills in provided
984  * struct vfio_iommu_info. Caller sets argsz.
985  *
986  * XXX Should we do these by CHECK_EXTENSION too?
987  */
988 struct vfio_iommu_type1_info {
989 	__u32	argsz;
990 	__u32	flags;
991 #define VFIO_IOMMU_INFO_PGSIZES (1 << 0)	/* supported page sizes info */
992 #define VFIO_IOMMU_INFO_CAPS	(1 << 1)	/* Info supports caps */
993 	__u64	iova_pgsizes;	/* Bitmap of supported page sizes */
994 	__u32   cap_offset;	/* Offset within info struct of first cap */
995 };
996 
997 /*
998  * The IOVA capability allows to report the valid IOVA range(s)
999  * excluding any non-relaxable reserved regions exposed by
1000  * devices attached to the container. Any DMA map attempt
1001  * outside the valid iova range will return error.
1002  *
1003  * The structures below define version 1 of this capability.
1004  */
1005 #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE  1
1006 
1007 struct vfio_iova_range {
1008 	__u64	start;
1009 	__u64	end;
1010 };
1011 
1012 struct vfio_iommu_type1_info_cap_iova_range {
1013 	struct	vfio_info_cap_header header;
1014 	__u32	nr_iovas;
1015 	__u32	reserved;
1016 	struct	vfio_iova_range iova_ranges[];
1017 };
1018 
1019 /*
1020  * The migration capability allows to report supported features for migration.
1021  *
1022  * The structures below define version 1 of this capability.
1023  *
1024  * The existence of this capability indicates that IOMMU kernel driver supports
1025  * dirty page logging.
1026  *
1027  * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty
1028  * page logging.
1029  * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap
1030  * size in bytes that can be used by user applications when getting the dirty
1031  * bitmap.
1032  */
1033 #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION  2
1034 
1035 struct vfio_iommu_type1_info_cap_migration {
1036 	struct	vfio_info_cap_header header;
1037 	__u32	flags;
1038 	__u64	pgsize_bitmap;
1039 	__u64	max_dirty_bitmap_size;		/* in bytes */
1040 };
1041 
1042 #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12)
1043 
1044 /**
1045  * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map)
1046  *
1047  * Map process virtual addresses to IO virtual addresses using the
1048  * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required.
1049  */
1050 struct vfio_iommu_type1_dma_map {
1051 	__u32	argsz;
1052 	__u32	flags;
1053 #define VFIO_DMA_MAP_FLAG_READ (1 << 0)		/* readable from device */
1054 #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1)	/* writable from device */
1055 	__u64	vaddr;				/* Process virtual address */
1056 	__u64	iova;				/* IO virtual address */
1057 	__u64	size;				/* Size of mapping (bytes) */
1058 };
1059 
1060 #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13)
1061 
1062 struct vfio_bitmap {
1063 	__u64        pgsize;	/* page size for bitmap in bytes */
1064 	__u64        size;	/* in bytes */
1065 	__u64 *data;	/* one bit per page */
1066 };
1067 
1068 /**
1069  * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14,
1070  *							struct vfio_dma_unmap)
1071  *
1072  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
1073  * Caller sets argsz.  The actual unmapped size is returned in the size
1074  * field.  No guarantee is made to the user that arbitrary unmaps of iova
1075  * or size different from those used in the original mapping call will
1076  * succeed.
1077  * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap
1078  * before unmapping IO virtual addresses. When this flag is set, the user must
1079  * provide a struct vfio_bitmap in data[]. User must provide zero-allocated
1080  * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field.
1081  * A bit in the bitmap represents one page, of user provided page size in
1082  * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set
1083  * indicates that the page at that offset from iova is dirty. A Bitmap of the
1084  * pages in the range of unmapped size is returned in the user-provided
1085  * vfio_bitmap.data.
1086  */
1087 struct vfio_iommu_type1_dma_unmap {
1088 	__u32	argsz;
1089 	__u32	flags;
1090 #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0)
1091 	__u64	iova;				/* IO virtual address */
1092 	__u64	size;				/* Size of mapping (bytes) */
1093 	__u8    data[];
1094 };
1095 
1096 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
1097 
1098 /*
1099  * IOCTLs to enable/disable IOMMU container usage.
1100  * No parameters are supported.
1101  */
1102 #define VFIO_IOMMU_ENABLE	_IO(VFIO_TYPE, VFIO_BASE + 15)
1103 #define VFIO_IOMMU_DISABLE	_IO(VFIO_TYPE, VFIO_BASE + 16)
1104 
1105 /**
1106  * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17,
1107  *                                     struct vfio_iommu_type1_dirty_bitmap)
1108  * IOCTL is used for dirty pages logging.
1109  * Caller should set flag depending on which operation to perform, details as
1110  * below:
1111  *
1112  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs
1113  * the IOMMU driver to log pages that are dirtied or potentially dirtied by
1114  * the device; designed to be used when a migration is in progress. Dirty pages
1115  * are logged until logging is disabled by user application by calling the IOCTL
1116  * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag.
1117  *
1118  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs
1119  * the IOMMU driver to stop logging dirtied pages.
1120  *
1121  * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set
1122  * returns the dirty pages bitmap for IOMMU container for a given IOVA range.
1123  * The user must specify the IOVA range and the pgsize through the structure
1124  * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface
1125  * supports getting a bitmap of the smallest supported pgsize only and can be
1126  * modified in future to get a bitmap of any specified supported pgsize. The
1127  * user must provide a zeroed memory area for the bitmap memory and specify its
1128  * size in bitmap.size. One bit is used to represent one page consecutively
1129  * starting from iova offset. The user should provide page size in bitmap.pgsize
1130  * field. A bit set in the bitmap indicates that the page at that offset from
1131  * iova is dirty. The caller must set argsz to a value including the size of
1132  * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the
1133  * actual bitmap. If dirty pages logging is not enabled, an error will be
1134  * returned.
1135  *
1136  * Only one of the flags _START, _STOP and _GET may be specified at a time.
1137  *
1138  */
1139 struct vfio_iommu_type1_dirty_bitmap {
1140 	__u32        argsz;
1141 	__u32        flags;
1142 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START	(1 << 0)
1143 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP	(1 << 1)
1144 #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP	(1 << 2)
1145 	__u8         data[];
1146 };
1147 
1148 struct vfio_iommu_type1_dirty_bitmap_get {
1149 	__u64              iova;	/* IO virtual address */
1150 	__u64              size;	/* Size of iova range */
1151 	struct vfio_bitmap bitmap;
1152 };
1153 
1154 #define VFIO_IOMMU_DIRTY_PAGES             _IO(VFIO_TYPE, VFIO_BASE + 17)
1155 
1156 /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
1157 
1158 /*
1159  * The SPAPR TCE DDW info struct provides the information about
1160  * the details of Dynamic DMA window capability.
1161  *
1162  * @pgsizes contains a page size bitmask, 4K/64K/16M are supported.
1163  * @max_dynamic_windows_supported tells the maximum number of windows
1164  * which the platform can create.
1165  * @levels tells the maximum number of levels in multi-level IOMMU tables;
1166  * this allows splitting a table into smaller chunks which reduces
1167  * the amount of physically contiguous memory required for the table.
1168  */
1169 struct vfio_iommu_spapr_tce_ddw_info {
1170 	__u64 pgsizes;			/* Bitmap of supported page sizes */
1171 	__u32 max_dynamic_windows_supported;
1172 	__u32 levels;
1173 };
1174 
1175 /*
1176  * The SPAPR TCE info struct provides the information about the PCI bus
1177  * address ranges available for DMA, these values are programmed into
1178  * the hardware so the guest has to know that information.
1179  *
1180  * The DMA 32 bit window start is an absolute PCI bus address.
1181  * The IOVA address passed via map/unmap ioctls are absolute PCI bus
1182  * addresses too so the window works as a filter rather than an offset
1183  * for IOVA addresses.
1184  *
1185  * Flags supported:
1186  * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows
1187  *   (DDW) support is present. @ddw is only supported when DDW is present.
1188  */
1189 struct vfio_iommu_spapr_tce_info {
1190 	__u32 argsz;
1191 	__u32 flags;
1192 #define VFIO_IOMMU_SPAPR_INFO_DDW	(1 << 0)	/* DDW supported */
1193 	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
1194 	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
1195 	struct vfio_iommu_spapr_tce_ddw_info ddw;
1196 };
1197 
1198 #define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
1199 
1200 /*
1201  * EEH PE operation struct provides ways to:
1202  * - enable/disable EEH functionality;
1203  * - unfreeze IO/DMA for frozen PE;
1204  * - read PE state;
1205  * - reset PE;
1206  * - configure PE;
1207  * - inject EEH error.
1208  */
1209 struct vfio_eeh_pe_err {
1210 	__u32 type;
1211 	__u32 func;
1212 	__u64 addr;
1213 	__u64 mask;
1214 };
1215 
1216 struct vfio_eeh_pe_op {
1217 	__u32 argsz;
1218 	__u32 flags;
1219 	__u32 op;
1220 	union {
1221 		struct vfio_eeh_pe_err err;
1222 	};
1223 };
1224 
1225 #define VFIO_EEH_PE_DISABLE		0	/* Disable EEH functionality */
1226 #define VFIO_EEH_PE_ENABLE		1	/* Enable EEH functionality  */
1227 #define VFIO_EEH_PE_UNFREEZE_IO		2	/* Enable IO for frozen PE   */
1228 #define VFIO_EEH_PE_UNFREEZE_DMA	3	/* Enable DMA for frozen PE  */
1229 #define VFIO_EEH_PE_GET_STATE		4	/* PE state retrieval        */
1230 #define  VFIO_EEH_PE_STATE_NORMAL	0	/* PE in functional state    */
1231 #define  VFIO_EEH_PE_STATE_RESET	1	/* PE reset in progress      */
1232 #define  VFIO_EEH_PE_STATE_STOPPED	2	/* Stopped DMA and IO        */
1233 #define  VFIO_EEH_PE_STATE_STOPPED_DMA	4	/* Stopped DMA only          */
1234 #define  VFIO_EEH_PE_STATE_UNAVAIL	5	/* State unavailable         */
1235 #define VFIO_EEH_PE_RESET_DEACTIVATE	5	/* Deassert PE reset         */
1236 #define VFIO_EEH_PE_RESET_HOT		6	/* Assert hot reset          */
1237 #define VFIO_EEH_PE_RESET_FUNDAMENTAL	7	/* Assert fundamental reset  */
1238 #define VFIO_EEH_PE_CONFIGURE		8	/* PE configuration          */
1239 #define VFIO_EEH_PE_INJECT_ERR		9	/* Inject EEH error          */
1240 
1241 #define VFIO_EEH_PE_OP			_IO(VFIO_TYPE, VFIO_BASE + 21)
1242 
1243 /**
1244  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory)
1245  *
1246  * Registers user space memory where DMA is allowed. It pins
1247  * user pages and does the locked memory accounting so
1248  * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls
1249  * get faster.
1250  */
1251 struct vfio_iommu_spapr_register_memory {
1252 	__u32	argsz;
1253 	__u32	flags;
1254 	__u64	vaddr;				/* Process virtual address */
1255 	__u64	size;				/* Size of mapping (bytes) */
1256 };
1257 #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 17)
1258 
1259 /**
1260  * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory)
1261  *
1262  * Unregisters user space memory registered with
1263  * VFIO_IOMMU_SPAPR_REGISTER_MEMORY.
1264  * Uses vfio_iommu_spapr_register_memory for parameters.
1265  */
1266 #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY	_IO(VFIO_TYPE, VFIO_BASE + 18)
1267 
1268 /**
1269  * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create)
1270  *
1271  * Creates an additional TCE table and programs it (sets a new DMA window)
1272  * to every IOMMU group in the container. It receives page shift, window
1273  * size and number of levels in the TCE table being created.
1274  *
1275  * It allocates and returns an offset on a PCI bus of the new DMA window.
1276  */
1277 struct vfio_iommu_spapr_tce_create {
1278 	__u32 argsz;
1279 	__u32 flags;
1280 	/* in */
1281 	__u32 page_shift;
1282 	__u32 __resv1;
1283 	__u64 window_size;
1284 	__u32 levels;
1285 	__u32 __resv2;
1286 	/* out */
1287 	__u64 start_addr;
1288 };
1289 #define VFIO_IOMMU_SPAPR_TCE_CREATE	_IO(VFIO_TYPE, VFIO_BASE + 19)
1290 
1291 /**
1292  * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove)
1293  *
1294  * Unprograms a TCE table from all groups in the container and destroys it.
1295  * It receives a PCI bus offset as a window id.
1296  */
1297 struct vfio_iommu_spapr_tce_remove {
1298 	__u32 argsz;
1299 	__u32 flags;
1300 	/* in */
1301 	__u64 start_addr;
1302 };
1303 #define VFIO_IOMMU_SPAPR_TCE_REMOVE	_IO(VFIO_TYPE, VFIO_BASE + 20)
1304 
1305 /* ***************************************************************** */
1306 
1307 #endif /* VFIO_H */
1308