1dd873966SEric Auger /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ 2883f0b85SAlex Williamson /* 3883f0b85SAlex Williamson * VFIO API definition 4883f0b85SAlex Williamson * 5883f0b85SAlex Williamson * Copyright (C) 2012 Red Hat, Inc. All rights reserved. 6883f0b85SAlex Williamson * Author: Alex Williamson <alex.williamson@redhat.com> 7883f0b85SAlex Williamson * 8883f0b85SAlex Williamson * This program is free software; you can redistribute it and/or modify 9883f0b85SAlex Williamson * it under the terms of the GNU General Public License version 2 as 10883f0b85SAlex Williamson * published by the Free Software Foundation. 11883f0b85SAlex Williamson */ 12d4834ff9SAlexander Graf #ifndef VFIO_H 13d4834ff9SAlexander Graf #define VFIO_H 14883f0b85SAlex Williamson 15883f0b85SAlex Williamson #include <linux/types.h> 16883f0b85SAlex Williamson #include <linux/ioctl.h> 17883f0b85SAlex Williamson 18883f0b85SAlex Williamson #define VFIO_API_VERSION 0 19883f0b85SAlex Williamson 20883f0b85SAlex Williamson 21883f0b85SAlex Williamson /* Kernel & User level defines for VFIO IOCTLs. */ 22883f0b85SAlex Williamson 23883f0b85SAlex Williamson /* Extensions */ 24883f0b85SAlex Williamson 25883f0b85SAlex Williamson #define VFIO_TYPE1_IOMMU 1 26c5daeae1SAlexey Kardashevskiy #define VFIO_SPAPR_TCE_IOMMU 2 279ffd2685SCornelia Huck #define VFIO_TYPE1v2_IOMMU 3 289ffd2685SCornelia Huck /* 299ffd2685SCornelia Huck * IOMMU enforces DMA cache coherence (ex. PCIe NoSnoop stripping). This 309ffd2685SCornelia Huck * capability is subject to change as groups are added or removed. 319ffd2685SCornelia Huck */ 329ffd2685SCornelia Huck #define VFIO_DMA_CC_IOMMU 4 33883f0b85SAlex Williamson 34a9fd1654SJens Freimann /* Check if EEH is supported */ 35a9fd1654SJens Freimann #define VFIO_EEH 5 36a9fd1654SJens Freimann 37444b1996SArd Biesheuvel /* Two-stage IOMMU */ 3844fe383cSHendrik Brueckner #define __VFIO_RESERVED_TYPE1_NESTING_IOMMU 6 /* Implies v2 */ 39444b1996SArd Biesheuvel 4025b8b39bSAlexey Kardashevskiy #define VFIO_SPAPR_TCE_v2_IOMMU 7 4125b8b39bSAlexey Kardashevskiy 42883f0b85SAlex Williamson /* 4366fb2d54SCornelia Huck * The No-IOMMU IOMMU offers no translation or isolation for devices and 4466fb2d54SCornelia Huck * supports no ioctls outside of VFIO_CHECK_EXTENSION. Use of VFIO's No-IOMMU 4566fb2d54SCornelia Huck * code will taint the host kernel and should be used with extreme caution. 4666fb2d54SCornelia Huck */ 4766fb2d54SCornelia Huck #define VFIO_NOIOMMU_IOMMU 8 4866fb2d54SCornelia Huck 49278f064eSEduardo Habkost /* Supports VFIO_DMA_UNMAP_FLAG_ALL */ 50278f064eSEduardo Habkost #define VFIO_UNMAP_ALL 9 51278f064eSEduardo Habkost 52c5c0fdbeSDavid 'Digit' Turner /* 53c5c0fdbeSDavid 'Digit' Turner * Supports the vaddr flag for DMA map and unmap. Not supported for mediated 54c5c0fdbeSDavid 'Digit' Turner * devices, so this capability is subject to change as groups are added or 55c5c0fdbeSDavid 'Digit' Turner * removed. 56c5c0fdbeSDavid 'Digit' Turner */ 57278f064eSEduardo Habkost #define VFIO_UPDATE_VADDR 10 58278f064eSEduardo Habkost 5966fb2d54SCornelia Huck /* 60883f0b85SAlex Williamson * The IOCTL interface is designed for extensibility by embedding the 61883f0b85SAlex Williamson * structure length (argsz) and flags into structures passed between 62883f0b85SAlex Williamson * kernel and userspace. We therefore use the _IO() macro for these 63883f0b85SAlex Williamson * defines to avoid implicitly embedding a size into the ioctl request. 64883f0b85SAlex Williamson * As structure fields are added, argsz will increase to match and flag 65883f0b85SAlex Williamson * bits will be defined to indicate additional fields with valid data. 66883f0b85SAlex Williamson * It's *always* the caller's responsibility to indicate the size of 67883f0b85SAlex Williamson * the structure passed by setting argsz appropriately. 68883f0b85SAlex Williamson */ 69883f0b85SAlex Williamson 70883f0b85SAlex Williamson #define VFIO_TYPE (';') 71883f0b85SAlex Williamson #define VFIO_BASE 100 72883f0b85SAlex Williamson 73b89485a5SPaolo Bonzini /* 74b89485a5SPaolo Bonzini * For extension of INFO ioctls, VFIO makes use of a capability chain 75b89485a5SPaolo Bonzini * designed after PCI/e capabilities. A flag bit indicates whether 76b89485a5SPaolo Bonzini * this capability chain is supported and a field defined in the fixed 77b89485a5SPaolo Bonzini * structure defines the offset of the first capability in the chain. 78b89485a5SPaolo Bonzini * This field is only valid when the corresponding bit in the flags 79b89485a5SPaolo Bonzini * bitmap is set. This offset field is relative to the start of the 80b89485a5SPaolo Bonzini * INFO buffer, as is the next field within each capability header. 81b89485a5SPaolo Bonzini * The id within the header is a shared address space per INFO ioctl, 82b89485a5SPaolo Bonzini * while the version field is specific to the capability id. The 83b89485a5SPaolo Bonzini * contents following the header are specific to the capability id. 84b89485a5SPaolo Bonzini */ 85b89485a5SPaolo Bonzini struct vfio_info_cap_header { 86b89485a5SPaolo Bonzini __u16 id; /* Identifies capability */ 87b89485a5SPaolo Bonzini __u16 version; /* Version specific to the capability ID */ 88b89485a5SPaolo Bonzini __u32 next; /* Offset of next capability */ 89b89485a5SPaolo Bonzini }; 90b89485a5SPaolo Bonzini 91b89485a5SPaolo Bonzini /* 92b89485a5SPaolo Bonzini * Callers of INFO ioctls passing insufficiently sized buffers will see 93b89485a5SPaolo Bonzini * the capability chain flag bit set, a zero value for the first capability 94b89485a5SPaolo Bonzini * offset (if available within the provided argsz), and argsz will be 95b89485a5SPaolo Bonzini * updated to report the necessary buffer size. For compatibility, the 96b89485a5SPaolo Bonzini * INFO ioctl will not report error in this case, but the capability chain 97b89485a5SPaolo Bonzini * will not be available. 98b89485a5SPaolo Bonzini */ 99b89485a5SPaolo Bonzini 100883f0b85SAlex Williamson /* -------- IOCTLs for VFIO file descriptor (/dev/vfio/vfio) -------- */ 101883f0b85SAlex Williamson 102883f0b85SAlex Williamson /** 103883f0b85SAlex Williamson * VFIO_GET_API_VERSION - _IO(VFIO_TYPE, VFIO_BASE + 0) 104883f0b85SAlex Williamson * 105883f0b85SAlex Williamson * Report the version of the VFIO API. This allows us to bump the entire 106883f0b85SAlex Williamson * API version should we later need to add or change features in incompatible 107883f0b85SAlex Williamson * ways. 108883f0b85SAlex Williamson * Return: VFIO_API_VERSION 109883f0b85SAlex Williamson * Availability: Always 110883f0b85SAlex Williamson */ 111883f0b85SAlex Williamson #define VFIO_GET_API_VERSION _IO(VFIO_TYPE, VFIO_BASE + 0) 112883f0b85SAlex Williamson 113883f0b85SAlex Williamson /** 114883f0b85SAlex Williamson * VFIO_CHECK_EXTENSION - _IOW(VFIO_TYPE, VFIO_BASE + 1, __u32) 115883f0b85SAlex Williamson * 116883f0b85SAlex Williamson * Check whether an extension is supported. 117883f0b85SAlex Williamson * Return: 0 if not supported, 1 (or some other positive integer) if supported. 118883f0b85SAlex Williamson * Availability: Always 119883f0b85SAlex Williamson */ 120883f0b85SAlex Williamson #define VFIO_CHECK_EXTENSION _IO(VFIO_TYPE, VFIO_BASE + 1) 121883f0b85SAlex Williamson 122883f0b85SAlex Williamson /** 123883f0b85SAlex Williamson * VFIO_SET_IOMMU - _IOW(VFIO_TYPE, VFIO_BASE + 2, __s32) 124883f0b85SAlex Williamson * 125883f0b85SAlex Williamson * Set the iommu to the given type. The type must be supported by an 126883f0b85SAlex Williamson * iommu driver as verified by calling CHECK_EXTENSION using the same 127883f0b85SAlex Williamson * type. A group must be set to this file descriptor before this 128883f0b85SAlex Williamson * ioctl is available. The IOMMU interfaces enabled by this call are 129883f0b85SAlex Williamson * specific to the value set. 130883f0b85SAlex Williamson * Return: 0 on success, -errno on failure 131883f0b85SAlex Williamson * Availability: When VFIO group attached 132883f0b85SAlex Williamson */ 133883f0b85SAlex Williamson #define VFIO_SET_IOMMU _IO(VFIO_TYPE, VFIO_BASE + 2) 134883f0b85SAlex Williamson 135883f0b85SAlex Williamson /* -------- IOCTLs for GROUP file descriptors (/dev/vfio/$GROUP) -------- */ 136883f0b85SAlex Williamson 137883f0b85SAlex Williamson /** 138883f0b85SAlex Williamson * VFIO_GROUP_GET_STATUS - _IOR(VFIO_TYPE, VFIO_BASE + 3, 139883f0b85SAlex Williamson * struct vfio_group_status) 140883f0b85SAlex Williamson * 141883f0b85SAlex Williamson * Retrieve information about the group. Fills in provided 142883f0b85SAlex Williamson * struct vfio_group_info. Caller sets argsz. 143883f0b85SAlex Williamson * Return: 0 on succes, -errno on failure. 144883f0b85SAlex Williamson * Availability: Always 145883f0b85SAlex Williamson */ 146883f0b85SAlex Williamson struct vfio_group_status { 147883f0b85SAlex Williamson __u32 argsz; 148883f0b85SAlex Williamson __u32 flags; 149883f0b85SAlex Williamson #define VFIO_GROUP_FLAGS_VIABLE (1 << 0) 150883f0b85SAlex Williamson #define VFIO_GROUP_FLAGS_CONTAINER_SET (1 << 1) 151883f0b85SAlex Williamson }; 152883f0b85SAlex Williamson #define VFIO_GROUP_GET_STATUS _IO(VFIO_TYPE, VFIO_BASE + 3) 153883f0b85SAlex Williamson 154883f0b85SAlex Williamson /** 155883f0b85SAlex Williamson * VFIO_GROUP_SET_CONTAINER - _IOW(VFIO_TYPE, VFIO_BASE + 4, __s32) 156883f0b85SAlex Williamson * 157883f0b85SAlex Williamson * Set the container for the VFIO group to the open VFIO file 158883f0b85SAlex Williamson * descriptor provided. Groups may only belong to a single 159883f0b85SAlex Williamson * container. Containers may, at their discretion, support multiple 160883f0b85SAlex Williamson * groups. Only when a container is set are all of the interfaces 161883f0b85SAlex Williamson * of the VFIO file descriptor and the VFIO group file descriptor 162883f0b85SAlex Williamson * available to the user. 163883f0b85SAlex Williamson * Return: 0 on success, -errno on failure. 164883f0b85SAlex Williamson * Availability: Always 165883f0b85SAlex Williamson */ 166883f0b85SAlex Williamson #define VFIO_GROUP_SET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 4) 167883f0b85SAlex Williamson 168883f0b85SAlex Williamson /** 169883f0b85SAlex Williamson * VFIO_GROUP_UNSET_CONTAINER - _IO(VFIO_TYPE, VFIO_BASE + 5) 170883f0b85SAlex Williamson * 171883f0b85SAlex Williamson * Remove the group from the attached container. This is the 172883f0b85SAlex Williamson * opposite of the SET_CONTAINER call and returns the group to 173883f0b85SAlex Williamson * an initial state. All device file descriptors must be released 174883f0b85SAlex Williamson * prior to calling this interface. When removing the last group 175883f0b85SAlex Williamson * from a container, the IOMMU will be disabled and all state lost, 176883f0b85SAlex Williamson * effectively also returning the VFIO file descriptor to an initial 177883f0b85SAlex Williamson * state. 178883f0b85SAlex Williamson * Return: 0 on success, -errno on failure. 179883f0b85SAlex Williamson * Availability: When attached to container 180883f0b85SAlex Williamson */ 181883f0b85SAlex Williamson #define VFIO_GROUP_UNSET_CONTAINER _IO(VFIO_TYPE, VFIO_BASE + 5) 182883f0b85SAlex Williamson 183883f0b85SAlex Williamson /** 184883f0b85SAlex Williamson * VFIO_GROUP_GET_DEVICE_FD - _IOW(VFIO_TYPE, VFIO_BASE + 6, char) 185883f0b85SAlex Williamson * 186883f0b85SAlex Williamson * Return a new file descriptor for the device object described by 187883f0b85SAlex Williamson * the provided string. The string should match a device listed in 188883f0b85SAlex Williamson * the devices subdirectory of the IOMMU group sysfs entry. The 189883f0b85SAlex Williamson * group containing the device must already be added to this context. 190883f0b85SAlex Williamson * Return: new file descriptor on success, -errno on failure. 191883f0b85SAlex Williamson * Availability: When attached to container 192883f0b85SAlex Williamson */ 193883f0b85SAlex Williamson #define VFIO_GROUP_GET_DEVICE_FD _IO(VFIO_TYPE, VFIO_BASE + 6) 194883f0b85SAlex Williamson 195883f0b85SAlex Williamson /* --------------- IOCTLs for DEVICE file descriptors --------------- */ 196883f0b85SAlex Williamson 197883f0b85SAlex Williamson /** 198883f0b85SAlex Williamson * VFIO_DEVICE_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 7, 199883f0b85SAlex Williamson * struct vfio_device_info) 200883f0b85SAlex Williamson * 201883f0b85SAlex Williamson * Retrieve information about the device. Fills in provided 202883f0b85SAlex Williamson * struct vfio_device_info. Caller sets argsz. 203883f0b85SAlex Williamson * Return: 0 on success, -errno on failure. 204883f0b85SAlex Williamson */ 205883f0b85SAlex Williamson struct vfio_device_info { 206883f0b85SAlex Williamson __u32 argsz; 207883f0b85SAlex Williamson __u32 flags; 208883f0b85SAlex Williamson #define VFIO_DEVICE_FLAGS_RESET (1 << 0) /* Device supports reset */ 209883f0b85SAlex Williamson #define VFIO_DEVICE_FLAGS_PCI (1 << 1) /* vfio-pci device */ 2107a52ce8aSCornelia Huck #define VFIO_DEVICE_FLAGS_PLATFORM (1 << 2) /* vfio-platform device */ 2117a52ce8aSCornelia Huck #define VFIO_DEVICE_FLAGS_AMBA (1 << 3) /* vfio-amba device */ 21274c98e20SCornelia Huck #define VFIO_DEVICE_FLAGS_CCW (1 << 4) /* vfio-ccw device */ 2138f3cd250SCornelia Huck #define VFIO_DEVICE_FLAGS_AP (1 << 5) /* vfio-ap device */ 21453ba2eeeSMatthew Rosato #define VFIO_DEVICE_FLAGS_FSL_MC (1 << 6) /* vfio-fsl-mc device */ 21553ba2eeeSMatthew Rosato #define VFIO_DEVICE_FLAGS_CAPS (1 << 7) /* Info supports caps */ 216d0bf492fSCédric Le Goater #define VFIO_DEVICE_FLAGS_CDX (1 << 8) /* vfio-cdx device */ 217883f0b85SAlex Williamson __u32 num_regions; /* Max region index + 1 */ 218883f0b85SAlex Williamson __u32 num_irqs; /* Max IRQ index + 1 */ 21953ba2eeeSMatthew Rosato __u32 cap_offset; /* Offset within info struct of first cap */ 220da3c22c7SThomas Huth __u32 pad; 221883f0b85SAlex Williamson }; 222883f0b85SAlex Williamson #define VFIO_DEVICE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 7) 223883f0b85SAlex Williamson 2243a5eb5b4SPaolo Bonzini /* 2253a5eb5b4SPaolo Bonzini * Vendor driver using Mediated device framework should provide device_api 2263a5eb5b4SPaolo Bonzini * attribute in supported type attribute groups. Device API string should be one 2273a5eb5b4SPaolo Bonzini * of the following corresponding to device flags in vfio_device_info structure. 2283a5eb5b4SPaolo Bonzini */ 2293a5eb5b4SPaolo Bonzini 2303a5eb5b4SPaolo Bonzini #define VFIO_DEVICE_API_PCI_STRING "vfio-pci" 2313a5eb5b4SPaolo Bonzini #define VFIO_DEVICE_API_PLATFORM_STRING "vfio-platform" 2323a5eb5b4SPaolo Bonzini #define VFIO_DEVICE_API_AMBA_STRING "vfio-amba" 23374c98e20SCornelia Huck #define VFIO_DEVICE_API_CCW_STRING "vfio-ccw" 2348f3cd250SCornelia Huck #define VFIO_DEVICE_API_AP_STRING "vfio-ap" 2353a5eb5b4SPaolo Bonzini 23653ba2eeeSMatthew Rosato /* 23753ba2eeeSMatthew Rosato * The following capabilities are unique to s390 zPCI devices. Their contents 23853ba2eeeSMatthew Rosato * are further-defined in vfio_zdev.h 23953ba2eeeSMatthew Rosato */ 24053ba2eeeSMatthew Rosato #define VFIO_DEVICE_INFO_CAP_ZPCI_BASE 1 24153ba2eeeSMatthew Rosato #define VFIO_DEVICE_INFO_CAP_ZPCI_GROUP 2 24253ba2eeeSMatthew Rosato #define VFIO_DEVICE_INFO_CAP_ZPCI_UTIL 3 24353ba2eeeSMatthew Rosato #define VFIO_DEVICE_INFO_CAP_ZPCI_PFIP 4 24453ba2eeeSMatthew Rosato 245d0bf492fSCédric Le Goater /* 246d0bf492fSCédric Le Goater * The following VFIO_DEVICE_INFO capability reports support for PCIe AtomicOp 247d0bf492fSCédric Le Goater * completion to the root bus with supported widths provided via flags. 248d0bf492fSCédric Le Goater */ 249d0bf492fSCédric Le Goater #define VFIO_DEVICE_INFO_CAP_PCI_ATOMIC_COMP 5 250d0bf492fSCédric Le Goater struct vfio_device_info_cap_pci_atomic_comp { 251d0bf492fSCédric Le Goater struct vfio_info_cap_header header; 252d0bf492fSCédric Le Goater __u32 flags; 253d0bf492fSCédric Le Goater #define VFIO_PCI_ATOMIC_COMP32 (1 << 0) 254d0bf492fSCédric Le Goater #define VFIO_PCI_ATOMIC_COMP64 (1 << 1) 255d0bf492fSCédric Le Goater #define VFIO_PCI_ATOMIC_COMP128 (1 << 2) 256d0bf492fSCédric Le Goater __u32 reserved; 257d0bf492fSCédric Le Goater }; 258d0bf492fSCédric Le Goater 259883f0b85SAlex Williamson /** 260883f0b85SAlex Williamson * VFIO_DEVICE_GET_REGION_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 8, 261883f0b85SAlex Williamson * struct vfio_region_info) 262883f0b85SAlex Williamson * 263883f0b85SAlex Williamson * Retrieve information about a device region. Caller provides 264883f0b85SAlex Williamson * struct vfio_region_info with index value set. Caller sets argsz. 265883f0b85SAlex Williamson * Implementation of region mapping is bus driver specific. This is 266883f0b85SAlex Williamson * intended to describe MMIO, I/O port, as well as bus specific 267883f0b85SAlex Williamson * regions (ex. PCI config space). Zero sized regions may be used 268883f0b85SAlex Williamson * to describe unimplemented regions (ex. unimplemented PCI BARs). 269883f0b85SAlex Williamson * Return: 0 on success, -errno on failure. 270883f0b85SAlex Williamson */ 271883f0b85SAlex Williamson struct vfio_region_info { 272883f0b85SAlex Williamson __u32 argsz; 273883f0b85SAlex Williamson __u32 flags; 274883f0b85SAlex Williamson #define VFIO_REGION_INFO_FLAG_READ (1 << 0) /* Region supports read */ 275883f0b85SAlex Williamson #define VFIO_REGION_INFO_FLAG_WRITE (1 << 1) /* Region supports write */ 276883f0b85SAlex Williamson #define VFIO_REGION_INFO_FLAG_MMAP (1 << 2) /* Region supports mmap */ 277b89485a5SPaolo Bonzini #define VFIO_REGION_INFO_FLAG_CAPS (1 << 3) /* Info supports caps */ 278883f0b85SAlex Williamson __u32 index; /* Region index */ 279b89485a5SPaolo Bonzini __u32 cap_offset; /* Offset within info struct of first cap */ 280efb91426SDaniel Henrique Barboza __aligned_u64 size; /* Region size (bytes) */ 281efb91426SDaniel Henrique Barboza __aligned_u64 offset; /* Region offset from start of device fd */ 282883f0b85SAlex Williamson }; 283883f0b85SAlex Williamson #define VFIO_DEVICE_GET_REGION_INFO _IO(VFIO_TYPE, VFIO_BASE + 8) 284883f0b85SAlex Williamson 285b89485a5SPaolo Bonzini /* 286b89485a5SPaolo Bonzini * The sparse mmap capability allows finer granularity of specifying areas 287b89485a5SPaolo Bonzini * within a region with mmap support. When specified, the user should only 288b89485a5SPaolo Bonzini * mmap the offset ranges specified by the areas array. mmaps outside of the 289b89485a5SPaolo Bonzini * areas specified may fail (such as the range covering a PCI MSI-X table) or 290b89485a5SPaolo Bonzini * may result in improper device behavior. 291b89485a5SPaolo Bonzini * 292b89485a5SPaolo Bonzini * The structures below define version 1 of this capability. 293b89485a5SPaolo Bonzini */ 294b89485a5SPaolo Bonzini #define VFIO_REGION_INFO_CAP_SPARSE_MMAP 1 295b89485a5SPaolo Bonzini 296b89485a5SPaolo Bonzini struct vfio_region_sparse_mmap_area { 297efb91426SDaniel Henrique Barboza __aligned_u64 offset; /* Offset of mmap'able area within region */ 298efb91426SDaniel Henrique Barboza __aligned_u64 size; /* Size of mmap'able area */ 299b89485a5SPaolo Bonzini }; 300b89485a5SPaolo Bonzini 301b89485a5SPaolo Bonzini struct vfio_region_info_cap_sparse_mmap { 302b89485a5SPaolo Bonzini struct vfio_info_cap_header header; 303b89485a5SPaolo Bonzini __u32 nr_areas; 304b89485a5SPaolo Bonzini __u32 reserved; 305b89485a5SPaolo Bonzini struct vfio_region_sparse_mmap_area areas[]; 306b89485a5SPaolo Bonzini }; 307b89485a5SPaolo Bonzini 308b89485a5SPaolo Bonzini /* 309b89485a5SPaolo Bonzini * The device specific type capability allows regions unique to a specific 310b89485a5SPaolo Bonzini * device or class of devices to be exposed. This helps solve the problem for 311b89485a5SPaolo Bonzini * vfio bus drivers of defining which region indexes correspond to which region 312b89485a5SPaolo Bonzini * on the device, without needing to resort to static indexes, as done by 313b89485a5SPaolo Bonzini * vfio-pci. For instance, if we were to go back in time, we might remove 314b89485a5SPaolo Bonzini * VFIO_PCI_VGA_REGION_INDEX and let vfio-pci simply define that all indexes 315b89485a5SPaolo Bonzini * greater than or equal to VFIO_PCI_NUM_REGIONS are device specific and we'd 316b89485a5SPaolo Bonzini * make a "VGA" device specific type to describe the VGA access space. This 317b89485a5SPaolo Bonzini * means that non-VGA devices wouldn't need to waste this index, and thus the 318b89485a5SPaolo Bonzini * address space associated with it due to implementation of device file 319b89485a5SPaolo Bonzini * descriptor offsets in vfio-pci. 320b89485a5SPaolo Bonzini * 321b89485a5SPaolo Bonzini * The current implementation is now part of the user ABI, so we can't use this 322b89485a5SPaolo Bonzini * for VGA, but there are other upcoming use cases, such as opregions for Intel 323b89485a5SPaolo Bonzini * IGD devices and framebuffers for vGPU devices. We missed VGA, but we'll 324b89485a5SPaolo Bonzini * use this for future additions. 325b89485a5SPaolo Bonzini * 326b89485a5SPaolo Bonzini * The structure below defines version 1 of this capability. 327b89485a5SPaolo Bonzini */ 328b89485a5SPaolo Bonzini #define VFIO_REGION_INFO_CAP_TYPE 2 329b89485a5SPaolo Bonzini 330b89485a5SPaolo Bonzini struct vfio_region_info_cap_type { 331b89485a5SPaolo Bonzini struct vfio_info_cap_header header; 332b89485a5SPaolo Bonzini __u32 type; /* global per bus driver */ 333b89485a5SPaolo Bonzini __u32 subtype; /* type specific */ 334b89485a5SPaolo Bonzini }; 335b89485a5SPaolo Bonzini 336f363d039SEric Auger /* 337f363d039SEric Auger * List of region types, global per bus driver. 338f363d039SEric Auger * If you introduce a new type, please add it here. 339f363d039SEric Auger */ 340f363d039SEric Auger 341f363d039SEric Auger /* PCI region type containing a PCI vendor part */ 342b89485a5SPaolo Bonzini #define VFIO_REGION_TYPE_PCI_VENDOR_TYPE (1 << 31) 343b89485a5SPaolo Bonzini #define VFIO_REGION_TYPE_PCI_VENDOR_MASK (0xffff) 344f363d039SEric Auger #define VFIO_REGION_TYPE_GFX (1) 345f363d039SEric Auger #define VFIO_REGION_TYPE_CCW (2) 346e4082063SAlex Williamson #define VFIO_REGION_TYPE_MIGRATION_DEPRECATED (3) 347b89485a5SPaolo Bonzini 348f363d039SEric Auger /* sub-types for VFIO_REGION_TYPE_PCI_* */ 349f363d039SEric Auger 350f363d039SEric Auger /* 8086 vendor PCI sub-types */ 351b89485a5SPaolo Bonzini #define VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION (1) 352b89485a5SPaolo Bonzini #define VFIO_REGION_SUBTYPE_INTEL_IGD_HOST_CFG (2) 353b89485a5SPaolo Bonzini #define VFIO_REGION_SUBTYPE_INTEL_IGD_LPC_CFG (3) 354b89485a5SPaolo Bonzini 355f363d039SEric Auger /* 10de vendor PCI sub-types */ 356f363d039SEric Auger /* 357f363d039SEric Auger * NVIDIA GPU NVlink2 RAM is coherent RAM mapped onto the host address space. 358278f064eSEduardo Habkost * 359278f064eSEduardo Habkost * Deprecated, region no longer provided 360f363d039SEric Auger */ 361f363d039SEric Auger #define VFIO_REGION_SUBTYPE_NVIDIA_NVLINK2_RAM (1) 362f363d039SEric Auger 363f363d039SEric Auger /* 1014 vendor PCI sub-types */ 364f363d039SEric Auger /* 365f363d039SEric Auger * IBM NPU NVlink2 ATSD (Address Translation Shootdown) register of NPU 366f363d039SEric Auger * to do TLB invalidation on a GPU. 367278f064eSEduardo Habkost * 368278f064eSEduardo Habkost * Deprecated, region no longer provided 369f363d039SEric Auger */ 370f363d039SEric Auger #define VFIO_REGION_SUBTYPE_IBM_NVLINK2_ATSD (1) 371f363d039SEric Auger 372f363d039SEric Auger /* sub-types for VFIO_REGION_TYPE_GFX */ 373da054c64SPaolo Bonzini #define VFIO_REGION_SUBTYPE_GFX_EDID (1) 374da054c64SPaolo Bonzini 375da054c64SPaolo Bonzini /** 376da054c64SPaolo Bonzini * struct vfio_region_gfx_edid - EDID region layout. 377da054c64SPaolo Bonzini * 378da054c64SPaolo Bonzini * Set display link state and EDID blob. 379da054c64SPaolo Bonzini * 380da054c64SPaolo Bonzini * The EDID blob has monitor information such as brand, name, serial 381da054c64SPaolo Bonzini * number, physical size, supported video modes and more. 382da054c64SPaolo Bonzini * 383da054c64SPaolo Bonzini * This special region allows userspace (typically qemu) set a virtual 384da054c64SPaolo Bonzini * EDID for the virtual monitor, which allows a flexible display 385da054c64SPaolo Bonzini * configuration. 386da054c64SPaolo Bonzini * 387da054c64SPaolo Bonzini * For the edid blob spec look here: 388da054c64SPaolo Bonzini * https://en.wikipedia.org/wiki/Extended_Display_Identification_Data 389da054c64SPaolo Bonzini * 390da054c64SPaolo Bonzini * On linux systems you can find the EDID blob in sysfs: 391da054c64SPaolo Bonzini * /sys/class/drm/${card}/${connector}/edid 392da054c64SPaolo Bonzini * 393da054c64SPaolo Bonzini * You can use the edid-decode ulility (comes with xorg-x11-utils) to 394da054c64SPaolo Bonzini * decode the EDID blob. 395da054c64SPaolo Bonzini * 396da054c64SPaolo Bonzini * @edid_offset: location of the edid blob, relative to the 397da054c64SPaolo Bonzini * start of the region (readonly). 398da054c64SPaolo Bonzini * @edid_max_size: max size of the edid blob (readonly). 399da054c64SPaolo Bonzini * @edid_size: actual edid size (read/write). 400da054c64SPaolo Bonzini * @link_state: display link state (read/write). 401da054c64SPaolo Bonzini * VFIO_DEVICE_GFX_LINK_STATE_UP: Monitor is turned on. 402da054c64SPaolo Bonzini * VFIO_DEVICE_GFX_LINK_STATE_DOWN: Monitor is turned off. 403da054c64SPaolo Bonzini * @max_xres: max display width (0 == no limitation, readonly). 404da054c64SPaolo Bonzini * @max_yres: max display height (0 == no limitation, readonly). 405da054c64SPaolo Bonzini * 406da054c64SPaolo Bonzini * EDID update protocol: 407da054c64SPaolo Bonzini * (1) set link-state to down. 408da054c64SPaolo Bonzini * (2) update edid blob and size. 409da054c64SPaolo Bonzini * (3) set link-state to up. 410da054c64SPaolo Bonzini */ 411da054c64SPaolo Bonzini struct vfio_region_gfx_edid { 412da054c64SPaolo Bonzini __u32 edid_offset; 413da054c64SPaolo Bonzini __u32 edid_max_size; 414da054c64SPaolo Bonzini __u32 edid_size; 415da054c64SPaolo Bonzini __u32 max_xres; 416da054c64SPaolo Bonzini __u32 max_yres; 417da054c64SPaolo Bonzini __u32 link_state; 418da054c64SPaolo Bonzini #define VFIO_DEVICE_GFX_LINK_STATE_UP 1 419da054c64SPaolo Bonzini #define VFIO_DEVICE_GFX_LINK_STATE_DOWN 2 420da054c64SPaolo Bonzini }; 421da054c64SPaolo Bonzini 422f363d039SEric Auger /* sub-types for VFIO_REGION_TYPE_CCW */ 423d9cb4336SCornelia Huck #define VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD (1) 424f76b348eSCornelia Huck #define VFIO_REGION_SUBTYPE_CCW_SCHIB (2) 425f76b348eSCornelia Huck #define VFIO_REGION_SUBTYPE_CCW_CRW (3) 426f76b348eSCornelia Huck 427f76b348eSCornelia Huck /* sub-types for VFIO_REGION_TYPE_MIGRATION */ 428e4082063SAlex Williamson #define VFIO_REGION_SUBTYPE_MIGRATION_DEPRECATED (1) 429f76b348eSCornelia Huck 430f76b348eSCornelia Huck struct vfio_device_migration_info { 431f76b348eSCornelia Huck __u32 device_state; /* VFIO device state */ 432e4082063SAlex Williamson #define VFIO_DEVICE_STATE_V1_STOP (0) 433e4082063SAlex Williamson #define VFIO_DEVICE_STATE_V1_RUNNING (1 << 0) 434e4082063SAlex Williamson #define VFIO_DEVICE_STATE_V1_SAVING (1 << 1) 435e4082063SAlex Williamson #define VFIO_DEVICE_STATE_V1_RESUMING (1 << 2) 436e4082063SAlex Williamson #define VFIO_DEVICE_STATE_MASK (VFIO_DEVICE_STATE_V1_RUNNING | \ 437e4082063SAlex Williamson VFIO_DEVICE_STATE_V1_SAVING | \ 438e4082063SAlex Williamson VFIO_DEVICE_STATE_V1_RESUMING) 439f76b348eSCornelia Huck 440f76b348eSCornelia Huck #define VFIO_DEVICE_STATE_VALID(state) \ 441e4082063SAlex Williamson (state & VFIO_DEVICE_STATE_V1_RESUMING ? \ 442e4082063SAlex Williamson (state & VFIO_DEVICE_STATE_MASK) == VFIO_DEVICE_STATE_V1_RESUMING : 1) 443f76b348eSCornelia Huck 444f76b348eSCornelia Huck #define VFIO_DEVICE_STATE_IS_ERROR(state) \ 445e4082063SAlex Williamson ((state & VFIO_DEVICE_STATE_MASK) == (VFIO_DEVICE_STATE_V1_SAVING | \ 446e4082063SAlex Williamson VFIO_DEVICE_STATE_V1_RESUMING)) 447f76b348eSCornelia Huck 448f76b348eSCornelia Huck #define VFIO_DEVICE_STATE_SET_ERROR(state) \ 449e4082063SAlex Williamson ((state & ~VFIO_DEVICE_STATE_MASK) | VFIO_DEVICE_STATE_V1_SAVING | \ 450e4082063SAlex Williamson VFIO_DEVICE_STATE_V1_RESUMING) 451f76b348eSCornelia Huck 452f76b348eSCornelia Huck __u32 reserved; 453efb91426SDaniel Henrique Barboza __aligned_u64 pending_bytes; 454efb91426SDaniel Henrique Barboza __aligned_u64 data_offset; 455efb91426SDaniel Henrique Barboza __aligned_u64 data_size; 456f76b348eSCornelia Huck }; 457d9cb4336SCornelia Huck 458da054c64SPaolo Bonzini /* 4599f2d175dSPaolo Bonzini * The MSIX mappable capability informs that MSIX data of a BAR can be mmapped 4609f2d175dSPaolo Bonzini * which allows direct access to non-MSIX registers which happened to be within 4619f2d175dSPaolo Bonzini * the same system page. 4629f2d175dSPaolo Bonzini * 4639f2d175dSPaolo Bonzini * Even though the userspace gets direct access to the MSIX data, the existing 4649f2d175dSPaolo Bonzini * VFIO_DEVICE_SET_IRQS interface must still be used for MSIX configuration. 4659f2d175dSPaolo Bonzini */ 4669f2d175dSPaolo Bonzini #define VFIO_REGION_INFO_CAP_MSIX_MAPPABLE 3 4679f2d175dSPaolo Bonzini 468da054c64SPaolo Bonzini /* 469da054c64SPaolo Bonzini * Capability with compressed real address (aka SSA - small system address) 470da054c64SPaolo Bonzini * where GPU RAM is mapped on a system bus. Used by a GPU for DMA routing 471da054c64SPaolo Bonzini * and by the userspace to associate a NVLink bridge with a GPU. 472278f064eSEduardo Habkost * 473278f064eSEduardo Habkost * Deprecated, capability no longer provided 474da054c64SPaolo Bonzini */ 475da054c64SPaolo Bonzini #define VFIO_REGION_INFO_CAP_NVLINK2_SSATGT 4 476da054c64SPaolo Bonzini 477da054c64SPaolo Bonzini struct vfio_region_info_cap_nvlink2_ssatgt { 478da054c64SPaolo Bonzini struct vfio_info_cap_header header; 479efb91426SDaniel Henrique Barboza __aligned_u64 tgt; 480da054c64SPaolo Bonzini }; 481da054c64SPaolo Bonzini 482da054c64SPaolo Bonzini /* 483da054c64SPaolo Bonzini * Capability with an NVLink link speed. The value is read by 484da054c64SPaolo Bonzini * the NVlink2 bridge driver from the bridge's "ibm,nvlink-speed" 485da054c64SPaolo Bonzini * property in the device tree. The value is fixed in the hardware 486da054c64SPaolo Bonzini * and failing to provide the correct value results in the link 487da054c64SPaolo Bonzini * not working with no indication from the driver why. 488278f064eSEduardo Habkost * 489278f064eSEduardo Habkost * Deprecated, capability no longer provided 490da054c64SPaolo Bonzini */ 491da054c64SPaolo Bonzini #define VFIO_REGION_INFO_CAP_NVLINK2_LNKSPD 5 492da054c64SPaolo Bonzini 493da054c64SPaolo Bonzini struct vfio_region_info_cap_nvlink2_lnkspd { 494da054c64SPaolo Bonzini struct vfio_info_cap_header header; 495da054c64SPaolo Bonzini __u32 link_speed; 496da054c64SPaolo Bonzini __u32 __pad; 497da054c64SPaolo Bonzini }; 498da054c64SPaolo Bonzini 499883f0b85SAlex Williamson /** 500883f0b85SAlex Williamson * VFIO_DEVICE_GET_IRQ_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 9, 501883f0b85SAlex Williamson * struct vfio_irq_info) 502883f0b85SAlex Williamson * 503883f0b85SAlex Williamson * Retrieve information about a device IRQ. Caller provides 504883f0b85SAlex Williamson * struct vfio_irq_info with index value set. Caller sets argsz. 505883f0b85SAlex Williamson * Implementation of IRQ mapping is bus driver specific. Indexes 506883f0b85SAlex Williamson * using multiple IRQs are primarily intended to support MSI-like 507883f0b85SAlex Williamson * interrupt blocks. Zero count irq blocks may be used to describe 508883f0b85SAlex Williamson * unimplemented interrupt types. 509883f0b85SAlex Williamson * 510883f0b85SAlex Williamson * The EVENTFD flag indicates the interrupt index supports eventfd based 511883f0b85SAlex Williamson * signaling. 512883f0b85SAlex Williamson * 513883f0b85SAlex Williamson * The MASKABLE flags indicates the index supports MASK and UNMASK 514883f0b85SAlex Williamson * actions described below. 515883f0b85SAlex Williamson * 516883f0b85SAlex Williamson * AUTOMASKED indicates that after signaling, the interrupt line is 517883f0b85SAlex Williamson * automatically masked by VFIO and the user needs to unmask the line 518883f0b85SAlex Williamson * to receive new interrupts. This is primarily intended to distinguish 519883f0b85SAlex Williamson * level triggered interrupts. 520883f0b85SAlex Williamson * 521883f0b85SAlex Williamson * The NORESIZE flag indicates that the interrupt lines within the index 522883f0b85SAlex Williamson * are setup as a set and new subindexes cannot be enabled without first 523883f0b85SAlex Williamson * disabling the entire index. This is used for interrupts like PCI MSI 524883f0b85SAlex Williamson * and MSI-X where the driver may only use a subset of the available 525883f0b85SAlex Williamson * indexes, but VFIO needs to enable a specific number of vectors 526883f0b85SAlex Williamson * upfront. In the case of MSI-X, where the user can enable MSI-X and 527883f0b85SAlex Williamson * then add and unmask vectors, it's up to userspace to make the decision 528883f0b85SAlex Williamson * whether to allocate the maximum supported number of vectors or tear 529883f0b85SAlex Williamson * down setup and incrementally increase the vectors as each is enabled. 530d0bf492fSCédric Le Goater * Absence of the NORESIZE flag indicates that vectors can be enabled 531d0bf492fSCédric Le Goater * and disabled dynamically without impacting other vectors within the 532d0bf492fSCédric Le Goater * index. 533883f0b85SAlex Williamson */ 534883f0b85SAlex Williamson struct vfio_irq_info { 535883f0b85SAlex Williamson __u32 argsz; 536883f0b85SAlex Williamson __u32 flags; 537883f0b85SAlex Williamson #define VFIO_IRQ_INFO_EVENTFD (1 << 0) 538883f0b85SAlex Williamson #define VFIO_IRQ_INFO_MASKABLE (1 << 1) 539883f0b85SAlex Williamson #define VFIO_IRQ_INFO_AUTOMASKED (1 << 2) 540883f0b85SAlex Williamson #define VFIO_IRQ_INFO_NORESIZE (1 << 3) 541883f0b85SAlex Williamson __u32 index; /* IRQ index */ 542883f0b85SAlex Williamson __u32 count; /* Number of IRQs within this index */ 543883f0b85SAlex Williamson }; 544883f0b85SAlex Williamson #define VFIO_DEVICE_GET_IRQ_INFO _IO(VFIO_TYPE, VFIO_BASE + 9) 545883f0b85SAlex Williamson 546883f0b85SAlex Williamson /** 547883f0b85SAlex Williamson * VFIO_DEVICE_SET_IRQS - _IOW(VFIO_TYPE, VFIO_BASE + 10, struct vfio_irq_set) 548883f0b85SAlex Williamson * 549883f0b85SAlex Williamson * Set signaling, masking, and unmasking of interrupts. Caller provides 550883f0b85SAlex Williamson * struct vfio_irq_set with all fields set. 'start' and 'count' indicate 551883f0b85SAlex Williamson * the range of subindexes being specified. 552883f0b85SAlex Williamson * 553883f0b85SAlex Williamson * The DATA flags specify the type of data provided. If DATA_NONE, the 554883f0b85SAlex Williamson * operation performs the specified action immediately on the specified 555883f0b85SAlex Williamson * interrupt(s). For example, to unmask AUTOMASKED interrupt [0,0]: 556883f0b85SAlex Williamson * flags = (DATA_NONE|ACTION_UNMASK), index = 0, start = 0, count = 1. 557883f0b85SAlex Williamson * 558883f0b85SAlex Williamson * DATA_BOOL allows sparse support for the same on arrays of interrupts. 559883f0b85SAlex Williamson * For example, to mask interrupts [0,1] and [0,3] (but not [0,2]): 560883f0b85SAlex Williamson * flags = (DATA_BOOL|ACTION_MASK), index = 0, start = 1, count = 3, 561883f0b85SAlex Williamson * data = {1,0,1} 562883f0b85SAlex Williamson * 563883f0b85SAlex Williamson * DATA_EVENTFD binds the specified ACTION to the provided __s32 eventfd. 564883f0b85SAlex Williamson * A value of -1 can be used to either de-assign interrupts if already 565883f0b85SAlex Williamson * assigned or skip un-assigned interrupts. For example, to set an eventfd 566883f0b85SAlex Williamson * to be trigger for interrupts [0,0] and [0,2]: 567883f0b85SAlex Williamson * flags = (DATA_EVENTFD|ACTION_TRIGGER), index = 0, start = 0, count = 3, 568883f0b85SAlex Williamson * data = {fd1, -1, fd2} 569883f0b85SAlex Williamson * If index [0,1] is previously set, two count = 1 ioctls calls would be 570883f0b85SAlex Williamson * required to set [0,0] and [0,2] without changing [0,1]. 571883f0b85SAlex Williamson * 572883f0b85SAlex Williamson * Once a signaling mechanism is set, DATA_BOOL or DATA_NONE can be used 573883f0b85SAlex Williamson * with ACTION_TRIGGER to perform kernel level interrupt loopback testing 574883f0b85SAlex Williamson * from userspace (ie. simulate hardware triggering). 575883f0b85SAlex Williamson * 576883f0b85SAlex Williamson * Setting of an event triggering mechanism to userspace for ACTION_TRIGGER 577883f0b85SAlex Williamson * enables the interrupt index for the device. Individual subindex interrupts 578883f0b85SAlex Williamson * can be disabled using the -1 value for DATA_EVENTFD or the index can be 579883f0b85SAlex Williamson * disabled as a whole with: flags = (DATA_NONE|ACTION_TRIGGER), count = 0. 580883f0b85SAlex Williamson * 581883f0b85SAlex Williamson * Note that ACTION_[UN]MASK specify user->kernel signaling (irqfds) while 582883f0b85SAlex Williamson * ACTION_TRIGGER specifies kernel->user signaling. 583883f0b85SAlex Williamson */ 584883f0b85SAlex Williamson struct vfio_irq_set { 585883f0b85SAlex Williamson __u32 argsz; 586883f0b85SAlex Williamson __u32 flags; 587883f0b85SAlex Williamson #define VFIO_IRQ_SET_DATA_NONE (1 << 0) /* Data not present */ 588883f0b85SAlex Williamson #define VFIO_IRQ_SET_DATA_BOOL (1 << 1) /* Data is bool (u8) */ 589883f0b85SAlex Williamson #define VFIO_IRQ_SET_DATA_EVENTFD (1 << 2) /* Data is eventfd (s32) */ 590883f0b85SAlex Williamson #define VFIO_IRQ_SET_ACTION_MASK (1 << 3) /* Mask interrupt */ 591883f0b85SAlex Williamson #define VFIO_IRQ_SET_ACTION_UNMASK (1 << 4) /* Unmask interrupt */ 592883f0b85SAlex Williamson #define VFIO_IRQ_SET_ACTION_TRIGGER (1 << 5) /* Trigger interrupt */ 593883f0b85SAlex Williamson __u32 index; 594883f0b85SAlex Williamson __u32 start; 595883f0b85SAlex Williamson __u32 count; 596883f0b85SAlex Williamson __u8 data[]; 597883f0b85SAlex Williamson }; 598883f0b85SAlex Williamson #define VFIO_DEVICE_SET_IRQS _IO(VFIO_TYPE, VFIO_BASE + 10) 599883f0b85SAlex Williamson 600883f0b85SAlex Williamson #define VFIO_IRQ_SET_DATA_TYPE_MASK (VFIO_IRQ_SET_DATA_NONE | \ 601883f0b85SAlex Williamson VFIO_IRQ_SET_DATA_BOOL | \ 602883f0b85SAlex Williamson VFIO_IRQ_SET_DATA_EVENTFD) 603883f0b85SAlex Williamson #define VFIO_IRQ_SET_ACTION_TYPE_MASK (VFIO_IRQ_SET_ACTION_MASK | \ 604883f0b85SAlex Williamson VFIO_IRQ_SET_ACTION_UNMASK | \ 605883f0b85SAlex Williamson VFIO_IRQ_SET_ACTION_TRIGGER) 606883f0b85SAlex Williamson /** 607883f0b85SAlex Williamson * VFIO_DEVICE_RESET - _IO(VFIO_TYPE, VFIO_BASE + 11) 608883f0b85SAlex Williamson * 609883f0b85SAlex Williamson * Reset a device. 610883f0b85SAlex Williamson */ 611883f0b85SAlex Williamson #define VFIO_DEVICE_RESET _IO(VFIO_TYPE, VFIO_BASE + 11) 612883f0b85SAlex Williamson 613883f0b85SAlex Williamson /* 614883f0b85SAlex Williamson * The VFIO-PCI bus driver makes use of the following fixed region and 615883f0b85SAlex Williamson * IRQ index mapping. Unimplemented regions return a size of zero. 616883f0b85SAlex Williamson * Unimplemented IRQ types return a count of zero. 617883f0b85SAlex Williamson */ 618883f0b85SAlex Williamson 619883f0b85SAlex Williamson enum { 620883f0b85SAlex Williamson VFIO_PCI_BAR0_REGION_INDEX, 621883f0b85SAlex Williamson VFIO_PCI_BAR1_REGION_INDEX, 622883f0b85SAlex Williamson VFIO_PCI_BAR2_REGION_INDEX, 623883f0b85SAlex Williamson VFIO_PCI_BAR3_REGION_INDEX, 624883f0b85SAlex Williamson VFIO_PCI_BAR4_REGION_INDEX, 625883f0b85SAlex Williamson VFIO_PCI_BAR5_REGION_INDEX, 626883f0b85SAlex Williamson VFIO_PCI_ROM_REGION_INDEX, 627883f0b85SAlex Williamson VFIO_PCI_CONFIG_REGION_INDEX, 628010ca0b3SAlex Williamson /* 629010ca0b3SAlex Williamson * Expose VGA regions defined for PCI base class 03, subclass 00. 630010ca0b3SAlex Williamson * This includes I/O port ranges 0x3b0 to 0x3bb and 0x3c0 to 0x3df 631010ca0b3SAlex Williamson * as well as the MMIO range 0xa0000 to 0xbffff. Each implemented 632010ca0b3SAlex Williamson * range is found at it's identity mapped offset from the region 633010ca0b3SAlex Williamson * offset, for example 0x3b0 is region_info.offset + 0x3b0. Areas 634010ca0b3SAlex Williamson * between described ranges are unimplemented. 635010ca0b3SAlex Williamson */ 636010ca0b3SAlex Williamson VFIO_PCI_VGA_REGION_INDEX, 637b89485a5SPaolo Bonzini VFIO_PCI_NUM_REGIONS = 9 /* Fixed user ABI, region indexes >=9 use */ 638b89485a5SPaolo Bonzini /* device specific cap to define content. */ 639883f0b85SAlex Williamson }; 640883f0b85SAlex Williamson 641883f0b85SAlex Williamson enum { 642883f0b85SAlex Williamson VFIO_PCI_INTX_IRQ_INDEX, 643883f0b85SAlex Williamson VFIO_PCI_MSI_IRQ_INDEX, 644883f0b85SAlex Williamson VFIO_PCI_MSIX_IRQ_INDEX, 645e098b453SAlexey Kardashevskiy VFIO_PCI_ERR_IRQ_INDEX, 64647cbe50cSAlex Williamson VFIO_PCI_REQ_IRQ_INDEX, 647883f0b85SAlex Williamson VFIO_PCI_NUM_IRQS 648883f0b85SAlex Williamson }; 649883f0b85SAlex Williamson 65074c98e20SCornelia Huck /* 65174c98e20SCornelia Huck * The vfio-ccw bus driver makes use of the following fixed region and 65274c98e20SCornelia Huck * IRQ index mapping. Unimplemented regions return a size of zero. 65374c98e20SCornelia Huck * Unimplemented IRQ types return a count of zero. 65474c98e20SCornelia Huck */ 65574c98e20SCornelia Huck 65674c98e20SCornelia Huck enum { 65774c98e20SCornelia Huck VFIO_CCW_CONFIG_REGION_INDEX, 65874c98e20SCornelia Huck VFIO_CCW_NUM_REGIONS 65974c98e20SCornelia Huck }; 66074c98e20SCornelia Huck 66174c98e20SCornelia Huck enum { 66274c98e20SCornelia Huck VFIO_CCW_IO_IRQ_INDEX, 663f76b348eSCornelia Huck VFIO_CCW_CRW_IRQ_INDEX, 664b3c818a4SEric Farman VFIO_CCW_REQ_IRQ_INDEX, 66574c98e20SCornelia Huck VFIO_CCW_NUM_IRQS 66674c98e20SCornelia Huck }; 66774c98e20SCornelia Huck 668d0bf492fSCédric Le Goater /* 669d0bf492fSCédric Le Goater * The vfio-ap bus driver makes use of the following IRQ index mapping. 670d0bf492fSCédric Le Goater * Unimplemented IRQ types return a count of zero. 671d0bf492fSCédric Le Goater */ 672d0bf492fSCédric Le Goater enum { 673d0bf492fSCédric Le Goater VFIO_AP_REQ_IRQ_INDEX, 674*1cab5a02SRorie Reyes VFIO_AP_CFG_CHG_IRQ_INDEX, 675d0bf492fSCédric Le Goater VFIO_AP_NUM_IRQS 676d0bf492fSCédric Le Goater }; 677d0bf492fSCédric Le Goater 6784f265607SAndrew Jones /** 679d525f73fSChenyi Qiang * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO - _IOWR(VFIO_TYPE, VFIO_BASE + 12, 6804f265607SAndrew Jones * struct vfio_pci_hot_reset_info) 6814f265607SAndrew Jones * 682da3c22c7SThomas Huth * This command is used to query the affected devices in the hot reset for 683da3c22c7SThomas Huth * a given device. 684da3c22c7SThomas Huth * 685da3c22c7SThomas Huth * This command always reports the segment, bus, and devfn information for 686da3c22c7SThomas Huth * each affected device, and selectively reports the group_id or devid per 687da3c22c7SThomas Huth * the way how the calling device is opened. 688da3c22c7SThomas Huth * 689da3c22c7SThomas Huth * - If the calling device is opened via the traditional group/container 690da3c22c7SThomas Huth * API, group_id is reported. User should check if it has owned all 691da3c22c7SThomas Huth * the affected devices and provides a set of group fds to prove the 692da3c22c7SThomas Huth * ownership in VFIO_DEVICE_PCI_HOT_RESET ioctl. 693da3c22c7SThomas Huth * 694da3c22c7SThomas Huth * - If the calling device is opened as a cdev, devid is reported. 695da3c22c7SThomas Huth * Flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set to indicate this 696da3c22c7SThomas Huth * data type. All the affected devices should be represented in 697da3c22c7SThomas Huth * the dev_set, ex. bound to a vfio driver, and also be owned by 698da3c22c7SThomas Huth * this interface which is determined by the following conditions: 699da3c22c7SThomas Huth * 1) Has a valid devid within the iommufd_ctx of the calling device. 700da3c22c7SThomas Huth * Ownership cannot be determined across separate iommufd_ctx and 701da3c22c7SThomas Huth * the cdev calling conventions do not support a proof-of-ownership 702da3c22c7SThomas Huth * model as provided in the legacy group interface. In this case 703da3c22c7SThomas Huth * valid devid with value greater than zero is provided in the return 704da3c22c7SThomas Huth * structure. 705da3c22c7SThomas Huth * 2) Does not have a valid devid within the iommufd_ctx of the calling 706da3c22c7SThomas Huth * device, but belongs to the same IOMMU group as the calling device 707da3c22c7SThomas Huth * or another opened device that has a valid devid within the 708da3c22c7SThomas Huth * iommufd_ctx of the calling device. This provides implicit ownership 709da3c22c7SThomas Huth * for devices within the same DMA isolation context. In this case 710da3c22c7SThomas Huth * the devid value of VFIO_PCI_DEVID_OWNED is provided in the return 711da3c22c7SThomas Huth * structure. 712da3c22c7SThomas Huth * 713da3c22c7SThomas Huth * A devid value of VFIO_PCI_DEVID_NOT_OWNED is provided in the return 714da3c22c7SThomas Huth * structure for affected devices where device is NOT represented in the 715da3c22c7SThomas Huth * dev_set or ownership is not available. Such devices prevent the use 716da3c22c7SThomas Huth * of VFIO_DEVICE_PCI_HOT_RESET ioctl outside of the proof-of-ownership 717da3c22c7SThomas Huth * calling conventions (ie. via legacy group accessed devices). Flag 718da3c22c7SThomas Huth * VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED would be set when all the 719da3c22c7SThomas Huth * affected devices are represented in the dev_set and also owned by 720da3c22c7SThomas Huth * the user. This flag is available only when 721da3c22c7SThomas Huth * flag VFIO_PCI_HOT_RESET_FLAG_DEV_ID is set, otherwise reserved. 722da3c22c7SThomas Huth * When set, user could invoke VFIO_DEVICE_PCI_HOT_RESET with a zero 723da3c22c7SThomas Huth * length fd array on the calling device as the ownership is validated 724da3c22c7SThomas Huth * by iommufd_ctx. 725da3c22c7SThomas Huth * 7264f265607SAndrew Jones * Return: 0 on success, -errno on failure: 7274f265607SAndrew Jones * -enospc = insufficient buffer, -enodev = unsupported for device. 7284f265607SAndrew Jones */ 7294f265607SAndrew Jones struct vfio_pci_dependent_device { 730da3c22c7SThomas Huth union { 7314f265607SAndrew Jones __u32 group_id; 732da3c22c7SThomas Huth __u32 devid; 733da3c22c7SThomas Huth #define VFIO_PCI_DEVID_OWNED 0 734da3c22c7SThomas Huth #define VFIO_PCI_DEVID_NOT_OWNED -1 735da3c22c7SThomas Huth }; 7364f265607SAndrew Jones __u16 segment; 7374f265607SAndrew Jones __u8 bus; 7384f265607SAndrew Jones __u8 devfn; /* Use PCI_SLOT/PCI_FUNC */ 7394f265607SAndrew Jones }; 7404f265607SAndrew Jones 7414f265607SAndrew Jones struct vfio_pci_hot_reset_info { 7424f265607SAndrew Jones __u32 argsz; 7434f265607SAndrew Jones __u32 flags; 744da3c22c7SThomas Huth #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID (1 << 0) 745da3c22c7SThomas Huth #define VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED (1 << 1) 7464f265607SAndrew Jones __u32 count; 7474f265607SAndrew Jones struct vfio_pci_dependent_device devices[]; 7484f265607SAndrew Jones }; 7494f265607SAndrew Jones 7504f265607SAndrew Jones #define VFIO_DEVICE_GET_PCI_HOT_RESET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 7514f265607SAndrew Jones 7524f265607SAndrew Jones /** 7534f265607SAndrew Jones * VFIO_DEVICE_PCI_HOT_RESET - _IOW(VFIO_TYPE, VFIO_BASE + 13, 7544f265607SAndrew Jones * struct vfio_pci_hot_reset) 7554f265607SAndrew Jones * 756da3c22c7SThomas Huth * A PCI hot reset results in either a bus or slot reset which may affect 757da3c22c7SThomas Huth * other devices sharing the bus/slot. The calling user must have 758da3c22c7SThomas Huth * ownership of the full set of affected devices as determined by the 759da3c22c7SThomas Huth * VFIO_DEVICE_GET_PCI_HOT_RESET_INFO ioctl. 760da3c22c7SThomas Huth * 761da3c22c7SThomas Huth * When called on a device file descriptor acquired through the vfio 762da3c22c7SThomas Huth * group interface, the user is required to provide proof of ownership 763da3c22c7SThomas Huth * of those affected devices via the group_fds array in struct 764da3c22c7SThomas Huth * vfio_pci_hot_reset. 765da3c22c7SThomas Huth * 766da3c22c7SThomas Huth * When called on a direct cdev opened vfio device, the flags field of 767da3c22c7SThomas Huth * struct vfio_pci_hot_reset_info reports the ownership status of the 768da3c22c7SThomas Huth * affected devices and this ioctl must be called with an empty group_fds 769da3c22c7SThomas Huth * array. See above INFO ioctl definition for ownership requirements. 770da3c22c7SThomas Huth * 771da3c22c7SThomas Huth * Mixed usage of legacy groups and cdevs across the set of affected 772da3c22c7SThomas Huth * devices is not supported. 773da3c22c7SThomas Huth * 7744f265607SAndrew Jones * Return: 0 on success, -errno on failure. 7754f265607SAndrew Jones */ 7764f265607SAndrew Jones struct vfio_pci_hot_reset { 7774f265607SAndrew Jones __u32 argsz; 7784f265607SAndrew Jones __u32 flags; 7794f265607SAndrew Jones __u32 count; 7804f265607SAndrew Jones __s32 group_fds[]; 7814f265607SAndrew Jones }; 7824f265607SAndrew Jones 7834f265607SAndrew Jones #define VFIO_DEVICE_PCI_HOT_RESET _IO(VFIO_TYPE, VFIO_BASE + 13) 7844f265607SAndrew Jones 7859f2d175dSPaolo Bonzini /** 7869f2d175dSPaolo Bonzini * VFIO_DEVICE_QUERY_GFX_PLANE - _IOW(VFIO_TYPE, VFIO_BASE + 14, 7879f2d175dSPaolo Bonzini * struct vfio_device_query_gfx_plane) 7889f2d175dSPaolo Bonzini * 7899f2d175dSPaolo Bonzini * Set the drm_plane_type and flags, then retrieve the gfx plane info. 7909f2d175dSPaolo Bonzini * 7919f2d175dSPaolo Bonzini * flags supported: 7929f2d175dSPaolo Bonzini * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_DMABUF are set 7939f2d175dSPaolo Bonzini * to ask if the mdev supports dma-buf. 0 on support, -EINVAL on no 7949f2d175dSPaolo Bonzini * support for dma-buf. 7959f2d175dSPaolo Bonzini * - VFIO_GFX_PLANE_TYPE_PROBE and VFIO_GFX_PLANE_TYPE_REGION are set 7969f2d175dSPaolo Bonzini * to ask if the mdev supports region. 0 on support, -EINVAL on no 7979f2d175dSPaolo Bonzini * support for region. 7989f2d175dSPaolo Bonzini * - VFIO_GFX_PLANE_TYPE_DMABUF or VFIO_GFX_PLANE_TYPE_REGION is set 7999f2d175dSPaolo Bonzini * with each call to query the plane info. 8009f2d175dSPaolo Bonzini * - Others are invalid and return -EINVAL. 8019f2d175dSPaolo Bonzini * 8029f2d175dSPaolo Bonzini * Note: 8039f2d175dSPaolo Bonzini * 1. Plane could be disabled by guest. In that case, success will be 8049f2d175dSPaolo Bonzini * returned with zero-initialized drm_format, size, width and height 8059f2d175dSPaolo Bonzini * fields. 8069f2d175dSPaolo Bonzini * 2. x_hot/y_hot is set to 0xFFFFFFFF if no hotspot information available 8079f2d175dSPaolo Bonzini * 8089f2d175dSPaolo Bonzini * Return: 0 on success, -errno on other failure. 8099f2d175dSPaolo Bonzini */ 8109f2d175dSPaolo Bonzini struct vfio_device_gfx_plane_info { 8119f2d175dSPaolo Bonzini __u32 argsz; 8129f2d175dSPaolo Bonzini __u32 flags; 8139f2d175dSPaolo Bonzini #define VFIO_GFX_PLANE_TYPE_PROBE (1 << 0) 8149f2d175dSPaolo Bonzini #define VFIO_GFX_PLANE_TYPE_DMABUF (1 << 1) 8159f2d175dSPaolo Bonzini #define VFIO_GFX_PLANE_TYPE_REGION (1 << 2) 8169f2d175dSPaolo Bonzini /* in */ 8179f2d175dSPaolo Bonzini __u32 drm_plane_type; /* type of plane: DRM_PLANE_TYPE_* */ 8189f2d175dSPaolo Bonzini /* out */ 8199f2d175dSPaolo Bonzini __u32 drm_format; /* drm format of plane */ 820efb91426SDaniel Henrique Barboza __aligned_u64 drm_format_mod; /* tiled mode */ 8219f2d175dSPaolo Bonzini __u32 width; /* width of plane */ 8229f2d175dSPaolo Bonzini __u32 height; /* height of plane */ 8239f2d175dSPaolo Bonzini __u32 stride; /* stride of plane */ 8249f2d175dSPaolo Bonzini __u32 size; /* size of plane in bytes, align on page*/ 8259f2d175dSPaolo Bonzini __u32 x_pos; /* horizontal position of cursor plane */ 8269f2d175dSPaolo Bonzini __u32 y_pos; /* vertical position of cursor plane*/ 8279f2d175dSPaolo Bonzini __u32 x_hot; /* horizontal position of cursor hotspot */ 8289f2d175dSPaolo Bonzini __u32 y_hot; /* vertical position of cursor hotspot */ 8299f2d175dSPaolo Bonzini union { 8309f2d175dSPaolo Bonzini __u32 region_index; /* region index */ 8319f2d175dSPaolo Bonzini __u32 dmabuf_id; /* dma-buf id */ 8329f2d175dSPaolo Bonzini }; 833efb91426SDaniel Henrique Barboza __u32 reserved; 8349f2d175dSPaolo Bonzini }; 8359f2d175dSPaolo Bonzini 8369f2d175dSPaolo Bonzini #define VFIO_DEVICE_QUERY_GFX_PLANE _IO(VFIO_TYPE, VFIO_BASE + 14) 8379f2d175dSPaolo Bonzini 8389f2d175dSPaolo Bonzini /** 8399f2d175dSPaolo Bonzini * VFIO_DEVICE_GET_GFX_DMABUF - _IOW(VFIO_TYPE, VFIO_BASE + 15, __u32) 8409f2d175dSPaolo Bonzini * 8419f2d175dSPaolo Bonzini * Return a new dma-buf file descriptor for an exposed guest framebuffer 8429f2d175dSPaolo Bonzini * described by the provided dmabuf_id. The dmabuf_id is returned from VFIO_ 8439f2d175dSPaolo Bonzini * DEVICE_QUERY_GFX_PLANE as a token of the exposed guest framebuffer. 8449f2d175dSPaolo Bonzini */ 8459f2d175dSPaolo Bonzini 8469f2d175dSPaolo Bonzini #define VFIO_DEVICE_GET_GFX_DMABUF _IO(VFIO_TYPE, VFIO_BASE + 15) 8479f2d175dSPaolo Bonzini 84865a6d8ddSPeter Maydell /** 84965a6d8ddSPeter Maydell * VFIO_DEVICE_IOEVENTFD - _IOW(VFIO_TYPE, VFIO_BASE + 16, 85065a6d8ddSPeter Maydell * struct vfio_device_ioeventfd) 85165a6d8ddSPeter Maydell * 85265a6d8ddSPeter Maydell * Perform a write to the device at the specified device fd offset, with 85365a6d8ddSPeter Maydell * the specified data and width when the provided eventfd is triggered. 85465a6d8ddSPeter Maydell * vfio bus drivers may not support this for all regions, for all widths, 85565a6d8ddSPeter Maydell * or at all. vfio-pci currently only enables support for BAR regions, 85665a6d8ddSPeter Maydell * excluding the MSI-X vector table. 85765a6d8ddSPeter Maydell * 85865a6d8ddSPeter Maydell * Return: 0 on success, -errno on failure. 85965a6d8ddSPeter Maydell */ 86065a6d8ddSPeter Maydell struct vfio_device_ioeventfd { 86165a6d8ddSPeter Maydell __u32 argsz; 86265a6d8ddSPeter Maydell __u32 flags; 86365a6d8ddSPeter Maydell #define VFIO_DEVICE_IOEVENTFD_8 (1 << 0) /* 1-byte write */ 86465a6d8ddSPeter Maydell #define VFIO_DEVICE_IOEVENTFD_16 (1 << 1) /* 2-byte write */ 86565a6d8ddSPeter Maydell #define VFIO_DEVICE_IOEVENTFD_32 (1 << 2) /* 4-byte write */ 86665a6d8ddSPeter Maydell #define VFIO_DEVICE_IOEVENTFD_64 (1 << 3) /* 8-byte write */ 86765a6d8ddSPeter Maydell #define VFIO_DEVICE_IOEVENTFD_SIZE_MASK (0xf) 868efb91426SDaniel Henrique Barboza __aligned_u64 offset; /* device fd offset of write */ 869efb91426SDaniel Henrique Barboza __aligned_u64 data; /* data to be written */ 87065a6d8ddSPeter Maydell __s32 fd; /* -1 for de-assignment */ 871efb91426SDaniel Henrique Barboza __u32 reserved; 87265a6d8ddSPeter Maydell }; 87365a6d8ddSPeter Maydell 87465a6d8ddSPeter Maydell #define VFIO_DEVICE_IOEVENTFD _IO(VFIO_TYPE, VFIO_BASE + 16) 87565a6d8ddSPeter Maydell 876dc6f8d45SCornelia Huck /** 877d525f73fSChenyi Qiang * VFIO_DEVICE_FEATURE - _IOWR(VFIO_TYPE, VFIO_BASE + 17, 878dc6f8d45SCornelia Huck * struct vfio_device_feature) 879dc6f8d45SCornelia Huck * 880dc6f8d45SCornelia Huck * Get, set, or probe feature data of the device. The feature is selected 881dc6f8d45SCornelia Huck * using the FEATURE_MASK portion of the flags field. Support for a feature 882dc6f8d45SCornelia Huck * can be probed by setting both the FEATURE_MASK and PROBE bits. A probe 883dc6f8d45SCornelia Huck * may optionally include the GET and/or SET bits to determine read vs write 884dc6f8d45SCornelia Huck * access of the feature respectively. Probing a feature will return success 885dc6f8d45SCornelia Huck * if the feature is supported and all of the optionally indicated GET/SET 886dc6f8d45SCornelia Huck * methods are supported. The format of the data portion of the structure is 887dc6f8d45SCornelia Huck * specific to the given feature. The data portion is not required for 888dc6f8d45SCornelia Huck * probing. GET and SET are mutually exclusive, except for use with PROBE. 889dc6f8d45SCornelia Huck * 890dc6f8d45SCornelia Huck * Return 0 on success, -errno on failure. 891dc6f8d45SCornelia Huck */ 892dc6f8d45SCornelia Huck struct vfio_device_feature { 893dc6f8d45SCornelia Huck __u32 argsz; 894dc6f8d45SCornelia Huck __u32 flags; 895dc6f8d45SCornelia Huck #define VFIO_DEVICE_FEATURE_MASK (0xffff) /* 16-bit feature index */ 896dc6f8d45SCornelia Huck #define VFIO_DEVICE_FEATURE_GET (1 << 16) /* Get feature into data[] */ 897dc6f8d45SCornelia Huck #define VFIO_DEVICE_FEATURE_SET (1 << 17) /* Set feature from data[] */ 898dc6f8d45SCornelia Huck #define VFIO_DEVICE_FEATURE_PROBE (1 << 18) /* Probe feature support */ 899dc6f8d45SCornelia Huck __u8 data[]; 900dc6f8d45SCornelia Huck }; 901dc6f8d45SCornelia Huck 902dc6f8d45SCornelia Huck #define VFIO_DEVICE_FEATURE _IO(VFIO_TYPE, VFIO_BASE + 17) 903dc6f8d45SCornelia Huck 904dc6f8d45SCornelia Huck /* 905da3c22c7SThomas Huth * VFIO_DEVICE_BIND_IOMMUFD - _IOR(VFIO_TYPE, VFIO_BASE + 18, 906da3c22c7SThomas Huth * struct vfio_device_bind_iommufd) 907da3c22c7SThomas Huth * @argsz: User filled size of this data. 908da3c22c7SThomas Huth * @flags: Must be 0. 909da3c22c7SThomas Huth * @iommufd: iommufd to bind. 910da3c22c7SThomas Huth * @out_devid: The device id generated by this bind. devid is a handle for 911da3c22c7SThomas Huth * this device/iommufd bond and can be used in IOMMUFD commands. 912da3c22c7SThomas Huth * 913da3c22c7SThomas Huth * Bind a vfio_device to the specified iommufd. 914da3c22c7SThomas Huth * 915da3c22c7SThomas Huth * User is restricted from accessing the device before the binding operation 916da3c22c7SThomas Huth * is completed. Only allowed on cdev fds. 917da3c22c7SThomas Huth * 918da3c22c7SThomas Huth * Unbind is automatically conducted when device fd is closed. 919da3c22c7SThomas Huth * 920da3c22c7SThomas Huth * Return: 0 on success, -errno on failure. 921da3c22c7SThomas Huth */ 922da3c22c7SThomas Huth struct vfio_device_bind_iommufd { 923da3c22c7SThomas Huth __u32 argsz; 924da3c22c7SThomas Huth __u32 flags; 925da3c22c7SThomas Huth __s32 iommufd; 926da3c22c7SThomas Huth __u32 out_devid; 927da3c22c7SThomas Huth }; 928da3c22c7SThomas Huth 929da3c22c7SThomas Huth #define VFIO_DEVICE_BIND_IOMMUFD _IO(VFIO_TYPE, VFIO_BASE + 18) 930da3c22c7SThomas Huth 931da3c22c7SThomas Huth /* 932da3c22c7SThomas Huth * VFIO_DEVICE_ATTACH_IOMMUFD_PT - _IOW(VFIO_TYPE, VFIO_BASE + 19, 933da3c22c7SThomas Huth * struct vfio_device_attach_iommufd_pt) 934da3c22c7SThomas Huth * @argsz: User filled size of this data. 935*1cab5a02SRorie Reyes * @flags: Flags for attach. 936da3c22c7SThomas Huth * @pt_id: Input the target id which can represent an ioas or a hwpt 937da3c22c7SThomas Huth * allocated via iommufd subsystem. 938da3c22c7SThomas Huth * Output the input ioas id or the attached hwpt id which could 939da3c22c7SThomas Huth * be the specified hwpt itself or a hwpt automatically created 940da3c22c7SThomas Huth * for the specified ioas by kernel during the attachment. 941*1cab5a02SRorie Reyes * @pasid: The pasid to be attached, only meaningful when 942*1cab5a02SRorie Reyes * VFIO_DEVICE_ATTACH_PASID is set in @flags 943da3c22c7SThomas Huth * 944da3c22c7SThomas Huth * Associate the device with an address space within the bound iommufd. 945da3c22c7SThomas Huth * Undo by VFIO_DEVICE_DETACH_IOMMUFD_PT or device fd close. This is only 946da3c22c7SThomas Huth * allowed on cdev fds. 947da3c22c7SThomas Huth * 948*1cab5a02SRorie Reyes * If a vfio device or a pasid of this device is currently attached to a valid 949*1cab5a02SRorie Reyes * hw_pagetable (hwpt), without doing a VFIO_DEVICE_DETACH_IOMMUFD_PT, a second 950*1cab5a02SRorie Reyes * VFIO_DEVICE_ATTACH_IOMMUFD_PT ioctl passing in another hwpt id is allowed. 951*1cab5a02SRorie Reyes * This action, also known as a hw_pagetable replacement, will replace the 952*1cab5a02SRorie Reyes * currently attached hwpt of the device or the pasid of this device with a new 953*1cab5a02SRorie Reyes * hwpt corresponding to the given pt_id. 954da3c22c7SThomas Huth * 955da3c22c7SThomas Huth * Return: 0 on success, -errno on failure. 956da3c22c7SThomas Huth */ 957da3c22c7SThomas Huth struct vfio_device_attach_iommufd_pt { 958da3c22c7SThomas Huth __u32 argsz; 959da3c22c7SThomas Huth __u32 flags; 960*1cab5a02SRorie Reyes #define VFIO_DEVICE_ATTACH_PASID (1 << 0) 961da3c22c7SThomas Huth __u32 pt_id; 962*1cab5a02SRorie Reyes __u32 pasid; 963da3c22c7SThomas Huth }; 964da3c22c7SThomas Huth 965da3c22c7SThomas Huth #define VFIO_DEVICE_ATTACH_IOMMUFD_PT _IO(VFIO_TYPE, VFIO_BASE + 19) 966da3c22c7SThomas Huth 967da3c22c7SThomas Huth /* 968da3c22c7SThomas Huth * VFIO_DEVICE_DETACH_IOMMUFD_PT - _IOW(VFIO_TYPE, VFIO_BASE + 20, 969da3c22c7SThomas Huth * struct vfio_device_detach_iommufd_pt) 970da3c22c7SThomas Huth * @argsz: User filled size of this data. 971*1cab5a02SRorie Reyes * @flags: Flags for detach. 972*1cab5a02SRorie Reyes * @pasid: The pasid to be detached, only meaningful when 973*1cab5a02SRorie Reyes * VFIO_DEVICE_DETACH_PASID is set in @flags 974da3c22c7SThomas Huth * 975*1cab5a02SRorie Reyes * Remove the association of the device or a pasid of the device and its current 976*1cab5a02SRorie Reyes * associated address space. After it, the device or the pasid should be in a 977*1cab5a02SRorie Reyes * blocking DMA state. This is only allowed on cdev fds. 978da3c22c7SThomas Huth * 979da3c22c7SThomas Huth * Return: 0 on success, -errno on failure. 980da3c22c7SThomas Huth */ 981da3c22c7SThomas Huth struct vfio_device_detach_iommufd_pt { 982da3c22c7SThomas Huth __u32 argsz; 983da3c22c7SThomas Huth __u32 flags; 984*1cab5a02SRorie Reyes #define VFIO_DEVICE_DETACH_PASID (1 << 0) 985*1cab5a02SRorie Reyes __u32 pasid; 986da3c22c7SThomas Huth }; 987da3c22c7SThomas Huth 988da3c22c7SThomas Huth #define VFIO_DEVICE_DETACH_IOMMUFD_PT _IO(VFIO_TYPE, VFIO_BASE + 20) 989da3c22c7SThomas Huth 990da3c22c7SThomas Huth /* 991dc6f8d45SCornelia Huck * Provide support for setting a PCI VF Token, which is used as a shared 992dc6f8d45SCornelia Huck * secret between PF and VF drivers. This feature may only be set on a 993dc6f8d45SCornelia Huck * PCI SR-IOV PF when SR-IOV is enabled on the PF and there are no existing 994dc6f8d45SCornelia Huck * open VFs. Data provided when setting this feature is a 16-byte array 995dc6f8d45SCornelia Huck * (__u8 b[16]), representing a UUID. 996dc6f8d45SCornelia Huck */ 997dc6f8d45SCornelia Huck #define VFIO_DEVICE_FEATURE_PCI_VF_TOKEN (0) 998dc6f8d45SCornelia Huck 999e4082063SAlex Williamson /* 1000e4082063SAlex Williamson * Indicates the device can support the migration API through 1001e4082063SAlex Williamson * VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. If this GET succeeds, the RUNNING and 1002e4082063SAlex Williamson * ERROR states are always supported. Support for additional states is 1003e4082063SAlex Williamson * indicated via the flags field; at least VFIO_MIGRATION_STOP_COPY must be 1004e4082063SAlex Williamson * set. 1005e4082063SAlex Williamson * 1006e4082063SAlex Williamson * VFIO_MIGRATION_STOP_COPY means that STOP, STOP_COPY and 1007e4082063SAlex Williamson * RESUMING are supported. 1008e4082063SAlex Williamson * 1009e4082063SAlex Williamson * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P means that RUNNING_P2P 1010e4082063SAlex Williamson * is supported in addition to the STOP_COPY states. 1011e4082063SAlex Williamson * 101293d7620cSAvihai Horon * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY means that 101393d7620cSAvihai Horon * PRE_COPY is supported in addition to the STOP_COPY states. 101493d7620cSAvihai Horon * 101593d7620cSAvihai Horon * VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY 101693d7620cSAvihai Horon * means that RUNNING_P2P, PRE_COPY and PRE_COPY_P2P are supported 101793d7620cSAvihai Horon * in addition to the STOP_COPY states. 101893d7620cSAvihai Horon * 1019e4082063SAlex Williamson * Other combinations of flags have behavior to be defined in the future. 1020e4082063SAlex Williamson */ 1021e4082063SAlex Williamson struct vfio_device_feature_migration { 1022e4082063SAlex Williamson __aligned_u64 flags; 1023e4082063SAlex Williamson #define VFIO_MIGRATION_STOP_COPY (1 << 0) 1024e4082063SAlex Williamson #define VFIO_MIGRATION_P2P (1 << 1) 102593d7620cSAvihai Horon #define VFIO_MIGRATION_PRE_COPY (1 << 2) 1026e4082063SAlex Williamson }; 1027e4082063SAlex Williamson #define VFIO_DEVICE_FEATURE_MIGRATION 1 1028e4082063SAlex Williamson 1029e4082063SAlex Williamson /* 1030e4082063SAlex Williamson * Upon VFIO_DEVICE_FEATURE_SET, execute a migration state change on the VFIO 1031e4082063SAlex Williamson * device. The new state is supplied in device_state, see enum 1032e4082063SAlex Williamson * vfio_device_mig_state for details 1033e4082063SAlex Williamson * 1034e4082063SAlex Williamson * The kernel migration driver must fully transition the device to the new state 1035e4082063SAlex Williamson * value before the operation returns to the user. 1036e4082063SAlex Williamson * 1037e4082063SAlex Williamson * The kernel migration driver must not generate asynchronous device state 1038e4082063SAlex Williamson * transitions outside of manipulation by the user or the VFIO_DEVICE_RESET 1039e4082063SAlex Williamson * ioctl as described above. 1040e4082063SAlex Williamson * 1041e4082063SAlex Williamson * If this function fails then current device_state may be the original 1042e4082063SAlex Williamson * operating state or some other state along the combination transition path. 1043e4082063SAlex Williamson * The user can then decide if it should execute a VFIO_DEVICE_RESET, attempt 1044e4082063SAlex Williamson * to return to the original state, or attempt to return to some other state 1045e4082063SAlex Williamson * such as RUNNING or STOP. 1046e4082063SAlex Williamson * 1047e4082063SAlex Williamson * If the new_state starts a new data transfer session then the FD associated 1048e4082063SAlex Williamson * with that session is returned in data_fd. The user is responsible to close 1049e4082063SAlex Williamson * this FD when it is finished. The user must consider the migration data stream 1050e4082063SAlex Williamson * carried over the FD to be opaque and must preserve the byte order of the 1051e4082063SAlex Williamson * stream. The user is not required to preserve buffer segmentation when writing 1052e4082063SAlex Williamson * the data stream during the RESUMING operation. 1053e4082063SAlex Williamson * 1054e4082063SAlex Williamson * Upon VFIO_DEVICE_FEATURE_GET, get the current migration state of the VFIO 1055e4082063SAlex Williamson * device, data_fd will be -1. 1056e4082063SAlex Williamson */ 1057e4082063SAlex Williamson struct vfio_device_feature_mig_state { 1058e4082063SAlex Williamson __u32 device_state; /* From enum vfio_device_mig_state */ 1059e4082063SAlex Williamson __s32 data_fd; 1060e4082063SAlex Williamson }; 1061e4082063SAlex Williamson #define VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE 2 1062e4082063SAlex Williamson 1063e4082063SAlex Williamson /* 1064e4082063SAlex Williamson * The device migration Finite State Machine is described by the enum 1065e4082063SAlex Williamson * vfio_device_mig_state. Some of the FSM arcs will create a migration data 1066e4082063SAlex Williamson * transfer session by returning a FD, in this case the migration data will 1067e4082063SAlex Williamson * flow over the FD using read() and write() as discussed below. 1068e4082063SAlex Williamson * 1069e4082063SAlex Williamson * There are 5 states to support VFIO_MIGRATION_STOP_COPY: 1070e4082063SAlex Williamson * RUNNING - The device is running normally 1071e4082063SAlex Williamson * STOP - The device does not change the internal or external state 1072e4082063SAlex Williamson * STOP_COPY - The device internal state can be read out 1073e4082063SAlex Williamson * RESUMING - The device is stopped and is loading a new internal state 1074e4082063SAlex Williamson * ERROR - The device has failed and must be reset 1075e4082063SAlex Williamson * 107693d7620cSAvihai Horon * And optional states to support VFIO_MIGRATION_P2P: 1077e4082063SAlex Williamson * RUNNING_P2P - RUNNING, except the device cannot do peer to peer DMA 107893d7620cSAvihai Horon * And VFIO_MIGRATION_PRE_COPY: 107993d7620cSAvihai Horon * PRE_COPY - The device is running normally but tracking internal state 108093d7620cSAvihai Horon * changes 108193d7620cSAvihai Horon * And VFIO_MIGRATION_P2P | VFIO_MIGRATION_PRE_COPY: 108293d7620cSAvihai Horon * PRE_COPY_P2P - PRE_COPY, except the device cannot do peer to peer DMA 1083e4082063SAlex Williamson * 1084e4082063SAlex Williamson * The FSM takes actions on the arcs between FSM states. The driver implements 1085e4082063SAlex Williamson * the following behavior for the FSM arcs: 1086e4082063SAlex Williamson * 1087e4082063SAlex Williamson * RUNNING_P2P -> STOP 1088e4082063SAlex Williamson * STOP_COPY -> STOP 1089e4082063SAlex Williamson * While in STOP the device must stop the operation of the device. The device 1090e4082063SAlex Williamson * must not generate interrupts, DMA, or any other change to external state. 1091e4082063SAlex Williamson * It must not change its internal state. When stopped the device and kernel 1092e4082063SAlex Williamson * migration driver must accept and respond to interaction to support external 1093e4082063SAlex Williamson * subsystems in the STOP state, for example PCI MSI-X and PCI config space. 1094e4082063SAlex Williamson * Failure by the user to restrict device access while in STOP must not result 1095e4082063SAlex Williamson * in error conditions outside the user context (ex. host system faults). 1096e4082063SAlex Williamson * 1097e4082063SAlex Williamson * The STOP_COPY arc will terminate a data transfer session. 1098e4082063SAlex Williamson * 1099e4082063SAlex Williamson * RESUMING -> STOP 1100e4082063SAlex Williamson * Leaving RESUMING terminates a data transfer session and indicates the 1101e4082063SAlex Williamson * device should complete processing of the data delivered by write(). The 1102e4082063SAlex Williamson * kernel migration driver should complete the incorporation of data written 1103e4082063SAlex Williamson * to the data transfer FD into the device internal state and perform 1104e4082063SAlex Williamson * final validity and consistency checking of the new device state. If the 1105e4082063SAlex Williamson * user provided data is found to be incomplete, inconsistent, or otherwise 1106e4082063SAlex Williamson * invalid, the migration driver must fail the SET_STATE ioctl and 1107e4082063SAlex Williamson * optionally go to the ERROR state as described below. 1108e4082063SAlex Williamson * 1109e4082063SAlex Williamson * While in STOP the device has the same behavior as other STOP states 1110e4082063SAlex Williamson * described above. 1111e4082063SAlex Williamson * 1112e4082063SAlex Williamson * To abort a RESUMING session the device must be reset. 1113e4082063SAlex Williamson * 111493d7620cSAvihai Horon * PRE_COPY -> RUNNING 1115e4082063SAlex Williamson * RUNNING_P2P -> RUNNING 1116e4082063SAlex Williamson * While in RUNNING the device is fully operational, the device may generate 1117e4082063SAlex Williamson * interrupts, DMA, respond to MMIO, all vfio device regions are functional, 1118e4082063SAlex Williamson * and the device may advance its internal state. 1119e4082063SAlex Williamson * 112093d7620cSAvihai Horon * The PRE_COPY arc will terminate a data transfer session. 112193d7620cSAvihai Horon * 112293d7620cSAvihai Horon * PRE_COPY_P2P -> RUNNING_P2P 1123e4082063SAlex Williamson * RUNNING -> RUNNING_P2P 1124e4082063SAlex Williamson * STOP -> RUNNING_P2P 1125e4082063SAlex Williamson * While in RUNNING_P2P the device is partially running in the P2P quiescent 1126e4082063SAlex Williamson * state defined below. 1127e4082063SAlex Williamson * 112893d7620cSAvihai Horon * The PRE_COPY_P2P arc will terminate a data transfer session. 1129e4082063SAlex Williamson * 113093d7620cSAvihai Horon * RUNNING -> PRE_COPY 113193d7620cSAvihai Horon * RUNNING_P2P -> PRE_COPY_P2P 113293d7620cSAvihai Horon * STOP -> STOP_COPY 113393d7620cSAvihai Horon * PRE_COPY, PRE_COPY_P2P and STOP_COPY form the "saving group" of states 113493d7620cSAvihai Horon * which share a data transfer session. Moving between these states alters 113593d7620cSAvihai Horon * what is streamed in session, but does not terminate or otherwise affect 113693d7620cSAvihai Horon * the associated fd. 113793d7620cSAvihai Horon * 113893d7620cSAvihai Horon * These arcs begin the process of saving the device state and will return a 113993d7620cSAvihai Horon * new data_fd. The migration driver may perform actions such as enabling 114093d7620cSAvihai Horon * dirty logging of device state when entering PRE_COPY or PER_COPY_P2P. 114193d7620cSAvihai Horon * 114293d7620cSAvihai Horon * Each arc does not change the device operation, the device remains 114393d7620cSAvihai Horon * RUNNING, P2P quiesced or in STOP. The STOP_COPY state is described below 114493d7620cSAvihai Horon * in PRE_COPY_P2P -> STOP_COPY. 114593d7620cSAvihai Horon * 114693d7620cSAvihai Horon * PRE_COPY -> PRE_COPY_P2P 114793d7620cSAvihai Horon * Entering PRE_COPY_P2P continues all the behaviors of PRE_COPY above. 114893d7620cSAvihai Horon * However, while in the PRE_COPY_P2P state, the device is partially running 114993d7620cSAvihai Horon * in the P2P quiescent state defined below, like RUNNING_P2P. 115093d7620cSAvihai Horon * 115193d7620cSAvihai Horon * PRE_COPY_P2P -> PRE_COPY 115293d7620cSAvihai Horon * This arc allows returning the device to a full RUNNING behavior while 115393d7620cSAvihai Horon * continuing all the behaviors of PRE_COPY. 115493d7620cSAvihai Horon * 115593d7620cSAvihai Horon * PRE_COPY_P2P -> STOP_COPY 1156e4082063SAlex Williamson * While in the STOP_COPY state the device has the same behavior as STOP 1157e4082063SAlex Williamson * with the addition that the data transfers session continues to stream the 1158e4082063SAlex Williamson * migration state. End of stream on the FD indicates the entire device 1159e4082063SAlex Williamson * state has been transferred. 1160e4082063SAlex Williamson * 1161e4082063SAlex Williamson * The user should take steps to restrict access to vfio device regions while 1162e4082063SAlex Williamson * the device is in STOP_COPY or risk corruption of the device migration data 1163e4082063SAlex Williamson * stream. 1164e4082063SAlex Williamson * 1165e4082063SAlex Williamson * STOP -> RESUMING 1166e4082063SAlex Williamson * Entering the RESUMING state starts a process of restoring the device state 1167e4082063SAlex Williamson * and will return a new data_fd. The data stream fed into the data_fd should 1168e4082063SAlex Williamson * be taken from the data transfer output of a single FD during saving from 1169e4082063SAlex Williamson * a compatible device. The migration driver may alter/reset the internal 1170e4082063SAlex Williamson * device state for this arc if required to prepare the device to receive the 1171e4082063SAlex Williamson * migration data. 1172e4082063SAlex Williamson * 117393d7620cSAvihai Horon * STOP_COPY -> PRE_COPY 117493d7620cSAvihai Horon * STOP_COPY -> PRE_COPY_P2P 117593d7620cSAvihai Horon * These arcs are not permitted and return error if requested. Future 117693d7620cSAvihai Horon * revisions of this API may define behaviors for these arcs, in this case 117793d7620cSAvihai Horon * support will be discoverable by a new flag in 117893d7620cSAvihai Horon * VFIO_DEVICE_FEATURE_MIGRATION. 117993d7620cSAvihai Horon * 1180e4082063SAlex Williamson * any -> ERROR 1181e4082063SAlex Williamson * ERROR cannot be specified as a device state, however any transition request 1182e4082063SAlex Williamson * can be failed with an errno return and may then move the device_state into 1183e4082063SAlex Williamson * ERROR. In this case the device was unable to execute the requested arc and 1184e4082063SAlex Williamson * was also unable to restore the device to any valid device_state. 1185e4082063SAlex Williamson * To recover from ERROR VFIO_DEVICE_RESET must be used to return the 1186e4082063SAlex Williamson * device_state back to RUNNING. 1187e4082063SAlex Williamson * 1188e4082063SAlex Williamson * The optional peer to peer (P2P) quiescent state is intended to be a quiescent 1189e4082063SAlex Williamson * state for the device for the purposes of managing multiple devices within a 1190e4082063SAlex Williamson * user context where peer-to-peer DMA between devices may be active. The 119193d7620cSAvihai Horon * RUNNING_P2P and PRE_COPY_P2P states must prevent the device from initiating 1192e4082063SAlex Williamson * any new P2P DMA transactions. If the device can identify P2P transactions 1193e4082063SAlex Williamson * then it can stop only P2P DMA, otherwise it must stop all DMA. The migration 1194e4082063SAlex Williamson * driver must complete any such outstanding operations prior to completing the 1195e4082063SAlex Williamson * FSM arc into a P2P state. For the purpose of specification the states 1196e4082063SAlex Williamson * behave as though the device was fully running if not supported. Like while in 1197e4082063SAlex Williamson * STOP or STOP_COPY the user must not touch the device, otherwise the state 1198e4082063SAlex Williamson * can be exited. 1199e4082063SAlex Williamson * 1200e4082063SAlex Williamson * The remaining possible transitions are interpreted as combinations of the 1201e4082063SAlex Williamson * above FSM arcs. As there are multiple paths through the FSM arcs the path 1202e4082063SAlex Williamson * should be selected based on the following rules: 1203e4082063SAlex Williamson * - Select the shortest path. 120493d7620cSAvihai Horon * - The path cannot have saving group states as interior arcs, only 120593d7620cSAvihai Horon * starting/end states. 1206e4082063SAlex Williamson * Refer to vfio_mig_get_next_state() for the result of the algorithm. 1207e4082063SAlex Williamson * 1208e4082063SAlex Williamson * The automatic transit through the FSM arcs that make up the combination 1209e4082063SAlex Williamson * transition is invisible to the user. When working with combination arcs the 1210e4082063SAlex Williamson * user may see any step along the path in the device_state if SET_STATE 1211e4082063SAlex Williamson * fails. When handling these types of errors users should anticipate future 1212e4082063SAlex Williamson * revisions of this protocol using new states and those states becoming 1213e4082063SAlex Williamson * visible in this case. 1214e4082063SAlex Williamson * 1215e4082063SAlex Williamson * The optional states cannot be used with SET_STATE if the device does not 1216e4082063SAlex Williamson * support them. The user can discover if these states are supported by using 1217e4082063SAlex Williamson * VFIO_DEVICE_FEATURE_MIGRATION. By using combination transitions the user can 1218e4082063SAlex Williamson * avoid knowing about these optional states if the kernel driver supports them. 121993d7620cSAvihai Horon * 122093d7620cSAvihai Horon * Arcs touching PRE_COPY and PRE_COPY_P2P are removed if support for PRE_COPY 122193d7620cSAvihai Horon * is not present. 1222e4082063SAlex Williamson */ 1223e4082063SAlex Williamson enum vfio_device_mig_state { 1224e4082063SAlex Williamson VFIO_DEVICE_STATE_ERROR = 0, 1225e4082063SAlex Williamson VFIO_DEVICE_STATE_STOP = 1, 1226e4082063SAlex Williamson VFIO_DEVICE_STATE_RUNNING = 2, 1227e4082063SAlex Williamson VFIO_DEVICE_STATE_STOP_COPY = 3, 1228e4082063SAlex Williamson VFIO_DEVICE_STATE_RESUMING = 4, 1229e4082063SAlex Williamson VFIO_DEVICE_STATE_RUNNING_P2P = 5, 123093d7620cSAvihai Horon VFIO_DEVICE_STATE_PRE_COPY = 6, 123193d7620cSAvihai Horon VFIO_DEVICE_STATE_PRE_COPY_P2P = 7, 12326a02465fSDaniel Henrique Barboza VFIO_DEVICE_STATE_NR, 1233e4082063SAlex Williamson }; 1234e4082063SAlex Williamson 123593d7620cSAvihai Horon /** 123693d7620cSAvihai Horon * VFIO_MIG_GET_PRECOPY_INFO - _IO(VFIO_TYPE, VFIO_BASE + 21) 123793d7620cSAvihai Horon * 123893d7620cSAvihai Horon * This ioctl is used on the migration data FD in the precopy phase of the 123993d7620cSAvihai Horon * migration data transfer. It returns an estimate of the current data sizes 124093d7620cSAvihai Horon * remaining to be transferred. It allows the user to judge when it is 124193d7620cSAvihai Horon * appropriate to leave PRE_COPY for STOP_COPY. 124293d7620cSAvihai Horon * 124393d7620cSAvihai Horon * This ioctl is valid only in PRE_COPY states and kernel driver should 124493d7620cSAvihai Horon * return -EINVAL from any other migration state. 124593d7620cSAvihai Horon * 124693d7620cSAvihai Horon * The vfio_precopy_info data structure returned by this ioctl provides 124793d7620cSAvihai Horon * estimates of data available from the device during the PRE_COPY states. 124893d7620cSAvihai Horon * This estimate is split into two categories, initial_bytes and 124993d7620cSAvihai Horon * dirty_bytes. 125093d7620cSAvihai Horon * 125193d7620cSAvihai Horon * The initial_bytes field indicates the amount of initial precopy 125293d7620cSAvihai Horon * data available from the device. This field should have a non-zero initial 125393d7620cSAvihai Horon * value and decrease as migration data is read from the device. 125493d7620cSAvihai Horon * It is recommended to leave PRE_COPY for STOP_COPY only after this field 125593d7620cSAvihai Horon * reaches zero. Leaving PRE_COPY earlier might make things slower. 125693d7620cSAvihai Horon * 125793d7620cSAvihai Horon * The dirty_bytes field tracks device state changes relative to data 125893d7620cSAvihai Horon * previously retrieved. This field starts at zero and may increase as 125993d7620cSAvihai Horon * the internal device state is modified or decrease as that modified 126093d7620cSAvihai Horon * state is read from the device. 126193d7620cSAvihai Horon * 126293d7620cSAvihai Horon * Userspace may use the combination of these fields to estimate the 126393d7620cSAvihai Horon * potential data size available during the PRE_COPY phases, as well as 126493d7620cSAvihai Horon * trends relative to the rate the device is dirtying its internal 126593d7620cSAvihai Horon * state, but these fields are not required to have any bearing relative 126693d7620cSAvihai Horon * to the data size available during the STOP_COPY phase. 126793d7620cSAvihai Horon * 126893d7620cSAvihai Horon * Drivers have a lot of flexibility in when and what they transfer during the 126993d7620cSAvihai Horon * PRE_COPY phase, and how they report this from VFIO_MIG_GET_PRECOPY_INFO. 127093d7620cSAvihai Horon * 127193d7620cSAvihai Horon * During pre-copy the migration data FD has a temporary "end of stream" that is 127293d7620cSAvihai Horon * reached when both initial_bytes and dirty_byte are zero. For instance, this 127393d7620cSAvihai Horon * may indicate that the device is idle and not currently dirtying any internal 127493d7620cSAvihai Horon * state. When read() is done on this temporary end of stream the kernel driver 127593d7620cSAvihai Horon * should return ENOMSG from read(). Userspace can wait for more data (which may 127693d7620cSAvihai Horon * never come) by using poll. 127793d7620cSAvihai Horon * 127893d7620cSAvihai Horon * Once in STOP_COPY the migration data FD has a permanent end of stream 127993d7620cSAvihai Horon * signaled in the usual way by read() always returning 0 and poll always 128093d7620cSAvihai Horon * returning readable. ENOMSG may not be returned in STOP_COPY. 128193d7620cSAvihai Horon * Support for this ioctl is mandatory if a driver claims to support 128293d7620cSAvihai Horon * VFIO_MIGRATION_PRE_COPY. 128393d7620cSAvihai Horon * 128493d7620cSAvihai Horon * Return: 0 on success, -1 and errno set on failure. 128593d7620cSAvihai Horon */ 128693d7620cSAvihai Horon struct vfio_precopy_info { 128793d7620cSAvihai Horon __u32 argsz; 128893d7620cSAvihai Horon __u32 flags; 128993d7620cSAvihai Horon __aligned_u64 initial_bytes; 129093d7620cSAvihai Horon __aligned_u64 dirty_bytes; 129193d7620cSAvihai Horon }; 129293d7620cSAvihai Horon 129393d7620cSAvihai Horon #define VFIO_MIG_GET_PRECOPY_INFO _IO(VFIO_TYPE, VFIO_BASE + 21) 129493d7620cSAvihai Horon 129593e0932bSPeter Xu /* 129693e0932bSPeter Xu * Upon VFIO_DEVICE_FEATURE_SET, allow the device to be moved into a low power 129793e0932bSPeter Xu * state with the platform-based power management. Device use of lower power 129893e0932bSPeter Xu * states depends on factors managed by the runtime power management core, 129993e0932bSPeter Xu * including system level support and coordinating support among dependent 130093e0932bSPeter Xu * devices. Enabling device low power entry does not guarantee lower power 130193e0932bSPeter Xu * usage by the device, nor is a mechanism provided through this feature to 130293e0932bSPeter Xu * know the current power state of the device. If any device access happens 130393e0932bSPeter Xu * (either from the host or through the vfio uAPI) when the device is in the 130493e0932bSPeter Xu * low power state, then the host will move the device out of the low power 130593e0932bSPeter Xu * state as necessary prior to the access. Once the access is completed, the 130693e0932bSPeter Xu * device may re-enter the low power state. For single shot low power support 130793e0932bSPeter Xu * with wake-up notification, see 130893e0932bSPeter Xu * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP below. Access to mmap'd 130993e0932bSPeter Xu * device regions is disabled on LOW_POWER_ENTRY and may only be resumed after 131093e0932bSPeter Xu * calling LOW_POWER_EXIT. 131193e0932bSPeter Xu */ 131293e0932bSPeter Xu #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY 3 131393e0932bSPeter Xu 131493e0932bSPeter Xu /* 131593e0932bSPeter Xu * This device feature has the same behavior as 131693e0932bSPeter Xu * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY with the exception that the user 131793e0932bSPeter Xu * provides an eventfd for wake-up notification. When the device moves out of 131893e0932bSPeter Xu * the low power state for the wake-up, the host will not allow the device to 131993e0932bSPeter Xu * re-enter a low power state without a subsequent user call to one of the low 132093e0932bSPeter Xu * power entry device feature IOCTLs. Access to mmap'd device regions is 132193e0932bSPeter Xu * disabled on LOW_POWER_ENTRY_WITH_WAKEUP and may only be resumed after the 132293e0932bSPeter Xu * low power exit. The low power exit can happen either through LOW_POWER_EXIT 132393e0932bSPeter Xu * or through any other access (where the wake-up notification has been 132493e0932bSPeter Xu * generated). The access to mmap'd device regions will not trigger low power 132593e0932bSPeter Xu * exit. 132693e0932bSPeter Xu * 132793e0932bSPeter Xu * The notification through the provided eventfd will be generated only when 132893e0932bSPeter Xu * the device has entered and is resumed from a low power state after 132993e0932bSPeter Xu * calling this device feature IOCTL. A device that has not entered low power 133093e0932bSPeter Xu * state, as managed through the runtime power management core, will not 133193e0932bSPeter Xu * generate a notification through the provided eventfd on access. Calling the 133293e0932bSPeter Xu * LOW_POWER_EXIT feature is optional in the case where notification has been 133393e0932bSPeter Xu * signaled on the provided eventfd that a resume from low power has occurred. 133493e0932bSPeter Xu */ 133593e0932bSPeter Xu struct vfio_device_low_power_entry_with_wakeup { 133693e0932bSPeter Xu __s32 wakeup_eventfd; 133793e0932bSPeter Xu __u32 reserved; 133893e0932bSPeter Xu }; 133993e0932bSPeter Xu 134093e0932bSPeter Xu #define VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP 4 134193e0932bSPeter Xu 134293e0932bSPeter Xu /* 134393e0932bSPeter Xu * Upon VFIO_DEVICE_FEATURE_SET, disallow use of device low power states as 134493e0932bSPeter Xu * previously enabled via VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY or 134593e0932bSPeter Xu * VFIO_DEVICE_FEATURE_LOW_POWER_ENTRY_WITH_WAKEUP device features. 134693e0932bSPeter Xu * This device feature IOCTL may itself generate a wakeup eventfd notification 134793e0932bSPeter Xu * in the latter case if the device had previously entered a low power state. 134893e0932bSPeter Xu */ 134993e0932bSPeter Xu #define VFIO_DEVICE_FEATURE_LOW_POWER_EXIT 5 135093e0932bSPeter Xu 135193e0932bSPeter Xu /* 135293e0932bSPeter Xu * Upon VFIO_DEVICE_FEATURE_SET start/stop device DMA logging. 135393e0932bSPeter Xu * VFIO_DEVICE_FEATURE_PROBE can be used to detect if the device supports 135493e0932bSPeter Xu * DMA logging. 135593e0932bSPeter Xu * 135693e0932bSPeter Xu * DMA logging allows a device to internally record what DMAs the device is 135793e0932bSPeter Xu * initiating and report them back to userspace. It is part of the VFIO 135893e0932bSPeter Xu * migration infrastructure that allows implementing dirty page tracking 135993e0932bSPeter Xu * during the pre copy phase of live migration. Only DMA WRITEs are logged, 136093e0932bSPeter Xu * and this API is not connected to VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE. 136193e0932bSPeter Xu * 136293e0932bSPeter Xu * When DMA logging is started a range of IOVAs to monitor is provided and the 136393e0932bSPeter Xu * device can optimize its logging to cover only the IOVA range given. Each 136493e0932bSPeter Xu * DMA that the device initiates inside the range will be logged by the device 136593e0932bSPeter Xu * for later retrieval. 136693e0932bSPeter Xu * 136793e0932bSPeter Xu * page_size is an input that hints what tracking granularity the device 136893e0932bSPeter Xu * should try to achieve. If the device cannot do the hinted page size then 136993e0932bSPeter Xu * it's the driver choice which page size to pick based on its support. 137093e0932bSPeter Xu * On output the device will return the page size it selected. 137193e0932bSPeter Xu * 137293e0932bSPeter Xu * ranges is a pointer to an array of 137393e0932bSPeter Xu * struct vfio_device_feature_dma_logging_range. 137493e0932bSPeter Xu * 137593e0932bSPeter Xu * The core kernel code guarantees to support by minimum num_ranges that fit 137693e0932bSPeter Xu * into a single kernel page. User space can try higher values but should give 137793e0932bSPeter Xu * up if the above can't be achieved as of some driver limitations. 137893e0932bSPeter Xu * 137993e0932bSPeter Xu * A single call to start device DMA logging can be issued and a matching stop 138093e0932bSPeter Xu * should follow at the end. Another start is not allowed in the meantime. 138193e0932bSPeter Xu */ 138293e0932bSPeter Xu struct vfio_device_feature_dma_logging_control { 138393e0932bSPeter Xu __aligned_u64 page_size; 138493e0932bSPeter Xu __u32 num_ranges; 138593e0932bSPeter Xu __u32 __reserved; 138693e0932bSPeter Xu __aligned_u64 ranges; 138793e0932bSPeter Xu }; 138893e0932bSPeter Xu 138993e0932bSPeter Xu struct vfio_device_feature_dma_logging_range { 139093e0932bSPeter Xu __aligned_u64 iova; 139193e0932bSPeter Xu __aligned_u64 length; 139293e0932bSPeter Xu }; 139393e0932bSPeter Xu 139493e0932bSPeter Xu #define VFIO_DEVICE_FEATURE_DMA_LOGGING_START 6 139593e0932bSPeter Xu 139693e0932bSPeter Xu /* 139793e0932bSPeter Xu * Upon VFIO_DEVICE_FEATURE_SET stop device DMA logging that was started 139893e0932bSPeter Xu * by VFIO_DEVICE_FEATURE_DMA_LOGGING_START 139993e0932bSPeter Xu */ 140093e0932bSPeter Xu #define VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP 7 140193e0932bSPeter Xu 140293e0932bSPeter Xu /* 140393e0932bSPeter Xu * Upon VFIO_DEVICE_FEATURE_GET read back and clear the device DMA log 140493e0932bSPeter Xu * 140593e0932bSPeter Xu * Query the device's DMA log for written pages within the given IOVA range. 140693e0932bSPeter Xu * During querying the log is cleared for the IOVA range. 140793e0932bSPeter Xu * 140893e0932bSPeter Xu * bitmap is a pointer to an array of u64s that will hold the output bitmap 140993e0932bSPeter Xu * with 1 bit reporting a page_size unit of IOVA. The mapping of IOVA to bits 141093e0932bSPeter Xu * is given by: 141193e0932bSPeter Xu * bitmap[(addr - iova)/page_size] & (1ULL << (addr % 64)) 141293e0932bSPeter Xu * 141393e0932bSPeter Xu * The input page_size can be any power of two value and does not have to 141493e0932bSPeter Xu * match the value given to VFIO_DEVICE_FEATURE_DMA_LOGGING_START. The driver 141593e0932bSPeter Xu * will format its internal logging to match the reporting page size, possibly 141693e0932bSPeter Xu * by replicating bits if the internal page size is lower than requested. 141793e0932bSPeter Xu * 141893e0932bSPeter Xu * The LOGGING_REPORT will only set bits in the bitmap and never clear or 141993e0932bSPeter Xu * perform any initialization of the user provided bitmap. 142093e0932bSPeter Xu * 142193e0932bSPeter Xu * If any error is returned userspace should assume that the dirty log is 142293e0932bSPeter Xu * corrupted. Error recovery is to consider all memory dirty and try to 142393e0932bSPeter Xu * restart the dirty tracking, or to abort/restart the whole migration. 142493e0932bSPeter Xu * 142593e0932bSPeter Xu * If DMA logging is not enabled, an error will be returned. 142693e0932bSPeter Xu * 142793e0932bSPeter Xu */ 142893e0932bSPeter Xu struct vfio_device_feature_dma_logging_report { 142993e0932bSPeter Xu __aligned_u64 iova; 143093e0932bSPeter Xu __aligned_u64 length; 143193e0932bSPeter Xu __aligned_u64 page_size; 143293e0932bSPeter Xu __aligned_u64 bitmap; 143393e0932bSPeter Xu }; 143493e0932bSPeter Xu 143593e0932bSPeter Xu #define VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT 8 143693e0932bSPeter Xu 143793d7620cSAvihai Horon /* 143893d7620cSAvihai Horon * Upon VFIO_DEVICE_FEATURE_GET read back the estimated data length that will 143993d7620cSAvihai Horon * be required to complete stop copy. 144093d7620cSAvihai Horon * 144193d7620cSAvihai Horon * Note: Can be called on each device state. 144293d7620cSAvihai Horon */ 144393d7620cSAvihai Horon 144493d7620cSAvihai Horon struct vfio_device_feature_mig_data_size { 144593d7620cSAvihai Horon __aligned_u64 stop_copy_length; 144693d7620cSAvihai Horon }; 144793d7620cSAvihai Horon 144893d7620cSAvihai Horon #define VFIO_DEVICE_FEATURE_MIG_DATA_SIZE 9 144993d7620cSAvihai Horon 1450efb91426SDaniel Henrique Barboza /** 1451efb91426SDaniel Henrique Barboza * Upon VFIO_DEVICE_FEATURE_SET, set or clear the BUS mastering for the device 1452efb91426SDaniel Henrique Barboza * based on the operation specified in op flag. 1453efb91426SDaniel Henrique Barboza * 1454efb91426SDaniel Henrique Barboza * The functionality is incorporated for devices that needs bus master control, 1455efb91426SDaniel Henrique Barboza * but the in-band device interface lacks the support. Consequently, it is not 1456efb91426SDaniel Henrique Barboza * applicable to PCI devices, as bus master control for PCI devices is managed 1457efb91426SDaniel Henrique Barboza * in-band through the configuration space. At present, this feature is supported 1458efb91426SDaniel Henrique Barboza * only for CDX devices. 1459efb91426SDaniel Henrique Barboza * When the device's BUS MASTER setting is configured as CLEAR, it will result in 1460efb91426SDaniel Henrique Barboza * blocking all incoming DMA requests from the device. On the other hand, configuring 1461efb91426SDaniel Henrique Barboza * the device's BUS MASTER setting as SET (enable) will grant the device the 1462efb91426SDaniel Henrique Barboza * capability to perform DMA to the host memory. 1463efb91426SDaniel Henrique Barboza */ 1464efb91426SDaniel Henrique Barboza struct vfio_device_feature_bus_master { 1465efb91426SDaniel Henrique Barboza __u32 op; 1466efb91426SDaniel Henrique Barboza #define VFIO_DEVICE_FEATURE_CLEAR_MASTER 0 /* Clear Bus Master */ 1467efb91426SDaniel Henrique Barboza #define VFIO_DEVICE_FEATURE_SET_MASTER 1 /* Set Bus Master */ 1468efb91426SDaniel Henrique Barboza }; 1469efb91426SDaniel Henrique Barboza #define VFIO_DEVICE_FEATURE_BUS_MASTER 10 1470efb91426SDaniel Henrique Barboza 1471883f0b85SAlex Williamson /* -------- API for Type1 VFIO IOMMU -------- */ 1472883f0b85SAlex Williamson 1473883f0b85SAlex Williamson /** 1474883f0b85SAlex Williamson * VFIO_IOMMU_GET_INFO - _IOR(VFIO_TYPE, VFIO_BASE + 12, struct vfio_iommu_info) 1475883f0b85SAlex Williamson * 1476883f0b85SAlex Williamson * Retrieve information about the IOMMU object. Fills in provided 1477883f0b85SAlex Williamson * struct vfio_iommu_info. Caller sets argsz. 1478883f0b85SAlex Williamson * 1479883f0b85SAlex Williamson * XXX Should we do these by CHECK_EXTENSION too? 1480883f0b85SAlex Williamson */ 1481883f0b85SAlex Williamson struct vfio_iommu_type1_info { 1482883f0b85SAlex Williamson __u32 argsz; 1483883f0b85SAlex Williamson __u32 flags; 1484883f0b85SAlex Williamson #define VFIO_IOMMU_INFO_PGSIZES (1 << 0) /* supported page sizes info */ 1485f363d039SEric Auger #define VFIO_IOMMU_INFO_CAPS (1 << 1) /* Info supports caps */ 1486efb91426SDaniel Henrique Barboza __aligned_u64 iova_pgsizes; /* Bitmap of supported page sizes */ 1487f363d039SEric Auger __u32 cap_offset; /* Offset within info struct of first cap */ 1488da3c22c7SThomas Huth __u32 pad; 1489f363d039SEric Auger }; 1490f363d039SEric Auger 1491f363d039SEric Auger /* 1492f363d039SEric Auger * The IOVA capability allows to report the valid IOVA range(s) 1493f363d039SEric Auger * excluding any non-relaxable reserved regions exposed by 1494f363d039SEric Auger * devices attached to the container. Any DMA map attempt 1495f363d039SEric Auger * outside the valid iova range will return error. 1496f363d039SEric Auger * 1497f363d039SEric Auger * The structures below define version 1 of this capability. 1498f363d039SEric Auger */ 1499f363d039SEric Auger #define VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE 1 1500f363d039SEric Auger 1501f363d039SEric Auger struct vfio_iova_range { 1502f363d039SEric Auger __u64 start; 1503f363d039SEric Auger __u64 end; 1504f363d039SEric Auger }; 1505f363d039SEric Auger 1506f363d039SEric Auger struct vfio_iommu_type1_info_cap_iova_range { 1507f363d039SEric Auger struct vfio_info_cap_header header; 1508f363d039SEric Auger __u32 nr_iovas; 1509f363d039SEric Auger __u32 reserved; 1510f363d039SEric Auger struct vfio_iova_range iova_ranges[]; 1511883f0b85SAlex Williamson }; 1512883f0b85SAlex Williamson 1513f76b348eSCornelia Huck /* 1514f76b348eSCornelia Huck * The migration capability allows to report supported features for migration. 1515f76b348eSCornelia Huck * 1516f76b348eSCornelia Huck * The structures below define version 1 of this capability. 1517f76b348eSCornelia Huck * 1518f76b348eSCornelia Huck * The existence of this capability indicates that IOMMU kernel driver supports 1519f76b348eSCornelia Huck * dirty page logging. 1520f76b348eSCornelia Huck * 1521f76b348eSCornelia Huck * pgsize_bitmap: Kernel driver returns bitmap of supported page sizes for dirty 1522f76b348eSCornelia Huck * page logging. 1523f76b348eSCornelia Huck * max_dirty_bitmap_size: Kernel driver returns maximum supported dirty bitmap 1524f76b348eSCornelia Huck * size in bytes that can be used by user applications when getting the dirty 1525f76b348eSCornelia Huck * bitmap. 1526f76b348eSCornelia Huck */ 1527e6546342SJason Wang #define VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION 2 1528f76b348eSCornelia Huck 1529f76b348eSCornelia Huck struct vfio_iommu_type1_info_cap_migration { 1530f76b348eSCornelia Huck struct vfio_info_cap_header header; 1531f76b348eSCornelia Huck __u32 flags; 1532f76b348eSCornelia Huck __u64 pgsize_bitmap; 1533f76b348eSCornelia Huck __u64 max_dirty_bitmap_size; /* in bytes */ 1534f76b348eSCornelia Huck }; 1535f76b348eSCornelia Huck 153653ba2eeeSMatthew Rosato /* 153753ba2eeeSMatthew Rosato * The DMA available capability allows to report the current number of 153853ba2eeeSMatthew Rosato * simultaneously outstanding DMA mappings that are allowed. 153953ba2eeeSMatthew Rosato * 154053ba2eeeSMatthew Rosato * The structure below defines version 1 of this capability. 154153ba2eeeSMatthew Rosato * 154253ba2eeeSMatthew Rosato * avail: specifies the current number of outstanding DMA mappings allowed. 154353ba2eeeSMatthew Rosato */ 154453ba2eeeSMatthew Rosato #define VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL 3 154553ba2eeeSMatthew Rosato 154653ba2eeeSMatthew Rosato struct vfio_iommu_type1_info_dma_avail { 154753ba2eeeSMatthew Rosato struct vfio_info_cap_header header; 154853ba2eeeSMatthew Rosato __u32 avail; 154953ba2eeeSMatthew Rosato }; 155053ba2eeeSMatthew Rosato 1551883f0b85SAlex Williamson #define VFIO_IOMMU_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 1552883f0b85SAlex Williamson 1553883f0b85SAlex Williamson /** 1554883f0b85SAlex Williamson * VFIO_IOMMU_MAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 13, struct vfio_dma_map) 1555883f0b85SAlex Williamson * 1556883f0b85SAlex Williamson * Map process virtual addresses to IO virtual addresses using the 1557883f0b85SAlex Williamson * provided struct vfio_dma_map. Caller sets argsz. READ &/ WRITE required. 1558278f064eSEduardo Habkost * 1559c5c0fdbeSDavid 'Digit' Turner * If flags & VFIO_DMA_MAP_FLAG_VADDR, update the base vaddr for iova. The vaddr 1560278f064eSEduardo Habkost * must have previously been invalidated with VFIO_DMA_UNMAP_FLAG_VADDR. To 1561278f064eSEduardo Habkost * maintain memory consistency within the user application, the updated vaddr 1562278f064eSEduardo Habkost * must address the same memory object as originally mapped. Failure to do so 1563278f064eSEduardo Habkost * will result in user memory corruption and/or device misbehavior. iova and 1564278f064eSEduardo Habkost * size must match those in the original MAP_DMA call. Protection is not 1565278f064eSEduardo Habkost * changed, and the READ & WRITE flags must be 0. 1566883f0b85SAlex Williamson */ 1567883f0b85SAlex Williamson struct vfio_iommu_type1_dma_map { 1568883f0b85SAlex Williamson __u32 argsz; 1569883f0b85SAlex Williamson __u32 flags; 1570883f0b85SAlex Williamson #define VFIO_DMA_MAP_FLAG_READ (1 << 0) /* readable from device */ 1571883f0b85SAlex Williamson #define VFIO_DMA_MAP_FLAG_WRITE (1 << 1) /* writable from device */ 1572278f064eSEduardo Habkost #define VFIO_DMA_MAP_FLAG_VADDR (1 << 2) 1573883f0b85SAlex Williamson __u64 vaddr; /* Process virtual address */ 1574883f0b85SAlex Williamson __u64 iova; /* IO virtual address */ 1575883f0b85SAlex Williamson __u64 size; /* Size of mapping (bytes) */ 1576883f0b85SAlex Williamson }; 1577883f0b85SAlex Williamson 1578883f0b85SAlex Williamson #define VFIO_IOMMU_MAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 13) 1579883f0b85SAlex Williamson 1580f76b348eSCornelia Huck struct vfio_bitmap { 1581f76b348eSCornelia Huck __u64 pgsize; /* page size for bitmap in bytes */ 1582f76b348eSCornelia Huck __u64 size; /* in bytes */ 1583f76b348eSCornelia Huck __u64 *data; /* one bit per page */ 1584f76b348eSCornelia Huck }; 1585f76b348eSCornelia Huck 1586883f0b85SAlex Williamson /** 1587c5daeae1SAlexey Kardashevskiy * VFIO_IOMMU_UNMAP_DMA - _IOWR(VFIO_TYPE, VFIO_BASE + 14, 1588c5daeae1SAlexey Kardashevskiy * struct vfio_dma_unmap) 1589883f0b85SAlex Williamson * 1590883f0b85SAlex Williamson * Unmap IO virtual addresses using the provided struct vfio_dma_unmap. 1591c5daeae1SAlexey Kardashevskiy * Caller sets argsz. The actual unmapped size is returned in the size 1592c5daeae1SAlexey Kardashevskiy * field. No guarantee is made to the user that arbitrary unmaps of iova 1593c5daeae1SAlexey Kardashevskiy * or size different from those used in the original mapping call will 1594c5daeae1SAlexey Kardashevskiy * succeed. 1595278f064eSEduardo Habkost * 1596f76b348eSCornelia Huck * VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP should be set to get the dirty bitmap 1597f76b348eSCornelia Huck * before unmapping IO virtual addresses. When this flag is set, the user must 1598f76b348eSCornelia Huck * provide a struct vfio_bitmap in data[]. User must provide zero-allocated 1599f76b348eSCornelia Huck * memory via vfio_bitmap.data and its size in the vfio_bitmap.size field. 1600f76b348eSCornelia Huck * A bit in the bitmap represents one page, of user provided page size in 1601f76b348eSCornelia Huck * vfio_bitmap.pgsize field, consecutively starting from iova offset. Bit set 1602f76b348eSCornelia Huck * indicates that the page at that offset from iova is dirty. A Bitmap of the 1603f76b348eSCornelia Huck * pages in the range of unmapped size is returned in the user-provided 1604f76b348eSCornelia Huck * vfio_bitmap.data. 1605278f064eSEduardo Habkost * 1606278f064eSEduardo Habkost * If flags & VFIO_DMA_UNMAP_FLAG_ALL, unmap all addresses. iova and size 1607278f064eSEduardo Habkost * must be 0. This cannot be combined with the get-dirty-bitmap flag. 1608278f064eSEduardo Habkost * 1609278f064eSEduardo Habkost * If flags & VFIO_DMA_UNMAP_FLAG_VADDR, do not unmap, but invalidate host 1610c5c0fdbeSDavid 'Digit' Turner * virtual addresses in the iova range. DMA to already-mapped pages continues. 1611c5c0fdbeSDavid 'Digit' Turner * Groups may not be added to the container while any addresses are invalid. 1612c5c0fdbeSDavid 'Digit' Turner * This cannot be combined with the get-dirty-bitmap flag. 1613883f0b85SAlex Williamson */ 1614883f0b85SAlex Williamson struct vfio_iommu_type1_dma_unmap { 1615883f0b85SAlex Williamson __u32 argsz; 1616883f0b85SAlex Williamson __u32 flags; 1617f76b348eSCornelia Huck #define VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP (1 << 0) 1618278f064eSEduardo Habkost #define VFIO_DMA_UNMAP_FLAG_ALL (1 << 1) 1619278f064eSEduardo Habkost #define VFIO_DMA_UNMAP_FLAG_VADDR (1 << 2) 1620883f0b85SAlex Williamson __u64 iova; /* IO virtual address */ 1621883f0b85SAlex Williamson __u64 size; /* Size of mapping (bytes) */ 1622f76b348eSCornelia Huck __u8 data[]; 1623883f0b85SAlex Williamson }; 1624883f0b85SAlex Williamson 1625883f0b85SAlex Williamson #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14) 1626883f0b85SAlex Williamson 1627c5daeae1SAlexey Kardashevskiy /* 1628c5daeae1SAlexey Kardashevskiy * IOCTLs to enable/disable IOMMU container usage. 1629c5daeae1SAlexey Kardashevskiy * No parameters are supported. 1630c5daeae1SAlexey Kardashevskiy */ 1631c5daeae1SAlexey Kardashevskiy #define VFIO_IOMMU_ENABLE _IO(VFIO_TYPE, VFIO_BASE + 15) 1632c5daeae1SAlexey Kardashevskiy #define VFIO_IOMMU_DISABLE _IO(VFIO_TYPE, VFIO_BASE + 16) 1633c5daeae1SAlexey Kardashevskiy 1634f76b348eSCornelia Huck /** 1635f76b348eSCornelia Huck * VFIO_IOMMU_DIRTY_PAGES - _IOWR(VFIO_TYPE, VFIO_BASE + 17, 1636f76b348eSCornelia Huck * struct vfio_iommu_type1_dirty_bitmap) 1637f76b348eSCornelia Huck * IOCTL is used for dirty pages logging. 1638f76b348eSCornelia Huck * Caller should set flag depending on which operation to perform, details as 1639f76b348eSCornelia Huck * below: 1640f76b348eSCornelia Huck * 1641f76b348eSCornelia Huck * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_START flag set, instructs 1642f76b348eSCornelia Huck * the IOMMU driver to log pages that are dirtied or potentially dirtied by 1643f76b348eSCornelia Huck * the device; designed to be used when a migration is in progress. Dirty pages 1644f76b348eSCornelia Huck * are logged until logging is disabled by user application by calling the IOCTL 1645f76b348eSCornelia Huck * with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag. 1646f76b348eSCornelia Huck * 1647f76b348eSCornelia Huck * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP flag set, instructs 1648f76b348eSCornelia Huck * the IOMMU driver to stop logging dirtied pages. 1649f76b348eSCornelia Huck * 1650f76b348eSCornelia Huck * Calling the IOCTL with VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP flag set 1651f76b348eSCornelia Huck * returns the dirty pages bitmap for IOMMU container for a given IOVA range. 1652f76b348eSCornelia Huck * The user must specify the IOVA range and the pgsize through the structure 1653f76b348eSCornelia Huck * vfio_iommu_type1_dirty_bitmap_get in the data[] portion. This interface 1654f76b348eSCornelia Huck * supports getting a bitmap of the smallest supported pgsize only and can be 1655f76b348eSCornelia Huck * modified in future to get a bitmap of any specified supported pgsize. The 1656f76b348eSCornelia Huck * user must provide a zeroed memory area for the bitmap memory and specify its 1657f76b348eSCornelia Huck * size in bitmap.size. One bit is used to represent one page consecutively 1658f76b348eSCornelia Huck * starting from iova offset. The user should provide page size in bitmap.pgsize 1659f76b348eSCornelia Huck * field. A bit set in the bitmap indicates that the page at that offset from 1660f76b348eSCornelia Huck * iova is dirty. The caller must set argsz to a value including the size of 1661f76b348eSCornelia Huck * structure vfio_iommu_type1_dirty_bitmap_get, but excluding the size of the 1662f76b348eSCornelia Huck * actual bitmap. If dirty pages logging is not enabled, an error will be 1663f76b348eSCornelia Huck * returned. 1664f76b348eSCornelia Huck * 1665f76b348eSCornelia Huck * Only one of the flags _START, _STOP and _GET may be specified at a time. 1666f76b348eSCornelia Huck * 1667f76b348eSCornelia Huck */ 1668f76b348eSCornelia Huck struct vfio_iommu_type1_dirty_bitmap { 1669f76b348eSCornelia Huck __u32 argsz; 1670f76b348eSCornelia Huck __u32 flags; 1671f76b348eSCornelia Huck #define VFIO_IOMMU_DIRTY_PAGES_FLAG_START (1 << 0) 1672f76b348eSCornelia Huck #define VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP (1 << 1) 1673f76b348eSCornelia Huck #define VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP (1 << 2) 1674f76b348eSCornelia Huck __u8 data[]; 1675f76b348eSCornelia Huck }; 1676f76b348eSCornelia Huck 1677f76b348eSCornelia Huck struct vfio_iommu_type1_dirty_bitmap_get { 1678f76b348eSCornelia Huck __u64 iova; /* IO virtual address */ 1679f76b348eSCornelia Huck __u64 size; /* Size of iova range */ 1680f76b348eSCornelia Huck struct vfio_bitmap bitmap; 1681f76b348eSCornelia Huck }; 1682f76b348eSCornelia Huck 1683f76b348eSCornelia Huck #define VFIO_IOMMU_DIRTY_PAGES _IO(VFIO_TYPE, VFIO_BASE + 17) 1684f76b348eSCornelia Huck 1685c5daeae1SAlexey Kardashevskiy /* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */ 1686c5daeae1SAlexey Kardashevskiy 1687c5daeae1SAlexey Kardashevskiy /* 168825b8b39bSAlexey Kardashevskiy * The SPAPR TCE DDW info struct provides the information about 168925b8b39bSAlexey Kardashevskiy * the details of Dynamic DMA window capability. 169025b8b39bSAlexey Kardashevskiy * 169125b8b39bSAlexey Kardashevskiy * @pgsizes contains a page size bitmask, 4K/64K/16M are supported. 169225b8b39bSAlexey Kardashevskiy * @max_dynamic_windows_supported tells the maximum number of windows 169325b8b39bSAlexey Kardashevskiy * which the platform can create. 169425b8b39bSAlexey Kardashevskiy * @levels tells the maximum number of levels in multi-level IOMMU tables; 169525b8b39bSAlexey Kardashevskiy * this allows splitting a table into smaller chunks which reduces 169625b8b39bSAlexey Kardashevskiy * the amount of physically contiguous memory required for the table. 169725b8b39bSAlexey Kardashevskiy */ 169825b8b39bSAlexey Kardashevskiy struct vfio_iommu_spapr_tce_ddw_info { 169925b8b39bSAlexey Kardashevskiy __u64 pgsizes; /* Bitmap of supported page sizes */ 170025b8b39bSAlexey Kardashevskiy __u32 max_dynamic_windows_supported; 170125b8b39bSAlexey Kardashevskiy __u32 levels; 170225b8b39bSAlexey Kardashevskiy }; 170325b8b39bSAlexey Kardashevskiy 170425b8b39bSAlexey Kardashevskiy /* 1705c5daeae1SAlexey Kardashevskiy * The SPAPR TCE info struct provides the information about the PCI bus 1706c5daeae1SAlexey Kardashevskiy * address ranges available for DMA, these values are programmed into 1707c5daeae1SAlexey Kardashevskiy * the hardware so the guest has to know that information. 1708c5daeae1SAlexey Kardashevskiy * 1709c5daeae1SAlexey Kardashevskiy * The DMA 32 bit window start is an absolute PCI bus address. 1710c5daeae1SAlexey Kardashevskiy * The IOVA address passed via map/unmap ioctls are absolute PCI bus 1711c5daeae1SAlexey Kardashevskiy * addresses too so the window works as a filter rather than an offset 1712c5daeae1SAlexey Kardashevskiy * for IOVA addresses. 1713c5daeae1SAlexey Kardashevskiy * 171425b8b39bSAlexey Kardashevskiy * Flags supported: 171525b8b39bSAlexey Kardashevskiy * - VFIO_IOMMU_SPAPR_INFO_DDW: informs the userspace that dynamic DMA windows 171625b8b39bSAlexey Kardashevskiy * (DDW) support is present. @ddw is only supported when DDW is present. 1717c5daeae1SAlexey Kardashevskiy */ 1718c5daeae1SAlexey Kardashevskiy struct vfio_iommu_spapr_tce_info { 1719c5daeae1SAlexey Kardashevskiy __u32 argsz; 172025b8b39bSAlexey Kardashevskiy __u32 flags; 172125b8b39bSAlexey Kardashevskiy #define VFIO_IOMMU_SPAPR_INFO_DDW (1 << 0) /* DDW supported */ 1722c5daeae1SAlexey Kardashevskiy __u32 dma32_window_start; /* 32 bit window start (bytes) */ 1723c5daeae1SAlexey Kardashevskiy __u32 dma32_window_size; /* 32 bit window size (bytes) */ 172425b8b39bSAlexey Kardashevskiy struct vfio_iommu_spapr_tce_ddw_info ddw; 1725c5daeae1SAlexey Kardashevskiy }; 1726c5daeae1SAlexey Kardashevskiy 1727c5daeae1SAlexey Kardashevskiy #define VFIO_IOMMU_SPAPR_TCE_GET_INFO _IO(VFIO_TYPE, VFIO_BASE + 12) 1728c5daeae1SAlexey Kardashevskiy 1729a9fd1654SJens Freimann /* 1730a9fd1654SJens Freimann * EEH PE operation struct provides ways to: 1731a9fd1654SJens Freimann * - enable/disable EEH functionality; 1732a9fd1654SJens Freimann * - unfreeze IO/DMA for frozen PE; 1733a9fd1654SJens Freimann * - read PE state; 1734a9fd1654SJens Freimann * - reset PE; 173525b8b39bSAlexey Kardashevskiy * - configure PE; 173625b8b39bSAlexey Kardashevskiy * - inject EEH error. 1737a9fd1654SJens Freimann */ 173825b8b39bSAlexey Kardashevskiy struct vfio_eeh_pe_err { 173925b8b39bSAlexey Kardashevskiy __u32 type; 174025b8b39bSAlexey Kardashevskiy __u32 func; 174125b8b39bSAlexey Kardashevskiy __u64 addr; 174225b8b39bSAlexey Kardashevskiy __u64 mask; 174325b8b39bSAlexey Kardashevskiy }; 174425b8b39bSAlexey Kardashevskiy 1745a9fd1654SJens Freimann struct vfio_eeh_pe_op { 1746a9fd1654SJens Freimann __u32 argsz; 1747a9fd1654SJens Freimann __u32 flags; 1748a9fd1654SJens Freimann __u32 op; 174925b8b39bSAlexey Kardashevskiy union { 175025b8b39bSAlexey Kardashevskiy struct vfio_eeh_pe_err err; 175125b8b39bSAlexey Kardashevskiy }; 1752a9fd1654SJens Freimann }; 1753a9fd1654SJens Freimann 1754a9fd1654SJens Freimann #define VFIO_EEH_PE_DISABLE 0 /* Disable EEH functionality */ 1755a9fd1654SJens Freimann #define VFIO_EEH_PE_ENABLE 1 /* Enable EEH functionality */ 1756a9fd1654SJens Freimann #define VFIO_EEH_PE_UNFREEZE_IO 2 /* Enable IO for frozen PE */ 1757a9fd1654SJens Freimann #define VFIO_EEH_PE_UNFREEZE_DMA 3 /* Enable DMA for frozen PE */ 1758a9fd1654SJens Freimann #define VFIO_EEH_PE_GET_STATE 4 /* PE state retrieval */ 1759a9fd1654SJens Freimann #define VFIO_EEH_PE_STATE_NORMAL 0 /* PE in functional state */ 1760a9fd1654SJens Freimann #define VFIO_EEH_PE_STATE_RESET 1 /* PE reset in progress */ 1761a9fd1654SJens Freimann #define VFIO_EEH_PE_STATE_STOPPED 2 /* Stopped DMA and IO */ 1762a9fd1654SJens Freimann #define VFIO_EEH_PE_STATE_STOPPED_DMA 4 /* Stopped DMA only */ 1763a9fd1654SJens Freimann #define VFIO_EEH_PE_STATE_UNAVAIL 5 /* State unavailable */ 1764a9fd1654SJens Freimann #define VFIO_EEH_PE_RESET_DEACTIVATE 5 /* Deassert PE reset */ 1765a9fd1654SJens Freimann #define VFIO_EEH_PE_RESET_HOT 6 /* Assert hot reset */ 1766a9fd1654SJens Freimann #define VFIO_EEH_PE_RESET_FUNDAMENTAL 7 /* Assert fundamental reset */ 1767a9fd1654SJens Freimann #define VFIO_EEH_PE_CONFIGURE 8 /* PE configuration */ 176825b8b39bSAlexey Kardashevskiy #define VFIO_EEH_PE_INJECT_ERR 9 /* Inject EEH error */ 1769a9fd1654SJens Freimann 1770a9fd1654SJens Freimann #define VFIO_EEH_PE_OP _IO(VFIO_TYPE, VFIO_BASE + 21) 1771a9fd1654SJens Freimann 177225b8b39bSAlexey Kardashevskiy /** 177325b8b39bSAlexey Kardashevskiy * VFIO_IOMMU_SPAPR_REGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 17, struct vfio_iommu_spapr_register_memory) 177425b8b39bSAlexey Kardashevskiy * 177525b8b39bSAlexey Kardashevskiy * Registers user space memory where DMA is allowed. It pins 177625b8b39bSAlexey Kardashevskiy * user pages and does the locked memory accounting so 177725b8b39bSAlexey Kardashevskiy * subsequent VFIO_IOMMU_MAP_DMA/VFIO_IOMMU_UNMAP_DMA calls 177825b8b39bSAlexey Kardashevskiy * get faster. 177925b8b39bSAlexey Kardashevskiy */ 178025b8b39bSAlexey Kardashevskiy struct vfio_iommu_spapr_register_memory { 178125b8b39bSAlexey Kardashevskiy __u32 argsz; 178225b8b39bSAlexey Kardashevskiy __u32 flags; 178325b8b39bSAlexey Kardashevskiy __u64 vaddr; /* Process virtual address */ 178425b8b39bSAlexey Kardashevskiy __u64 size; /* Size of mapping (bytes) */ 178525b8b39bSAlexey Kardashevskiy }; 178625b8b39bSAlexey Kardashevskiy #define VFIO_IOMMU_SPAPR_REGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 17) 178725b8b39bSAlexey Kardashevskiy 178825b8b39bSAlexey Kardashevskiy /** 178925b8b39bSAlexey Kardashevskiy * VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY - _IOW(VFIO_TYPE, VFIO_BASE + 18, struct vfio_iommu_spapr_register_memory) 179025b8b39bSAlexey Kardashevskiy * 179125b8b39bSAlexey Kardashevskiy * Unregisters user space memory registered with 179225b8b39bSAlexey Kardashevskiy * VFIO_IOMMU_SPAPR_REGISTER_MEMORY. 179325b8b39bSAlexey Kardashevskiy * Uses vfio_iommu_spapr_register_memory for parameters. 179425b8b39bSAlexey Kardashevskiy */ 179525b8b39bSAlexey Kardashevskiy #define VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY _IO(VFIO_TYPE, VFIO_BASE + 18) 179625b8b39bSAlexey Kardashevskiy 179725b8b39bSAlexey Kardashevskiy /** 179825b8b39bSAlexey Kardashevskiy * VFIO_IOMMU_SPAPR_TCE_CREATE - _IOWR(VFIO_TYPE, VFIO_BASE + 19, struct vfio_iommu_spapr_tce_create) 179925b8b39bSAlexey Kardashevskiy * 180025b8b39bSAlexey Kardashevskiy * Creates an additional TCE table and programs it (sets a new DMA window) 180125b8b39bSAlexey Kardashevskiy * to every IOMMU group in the container. It receives page shift, window 180225b8b39bSAlexey Kardashevskiy * size and number of levels in the TCE table being created. 180325b8b39bSAlexey Kardashevskiy * 180425b8b39bSAlexey Kardashevskiy * It allocates and returns an offset on a PCI bus of the new DMA window. 180525b8b39bSAlexey Kardashevskiy */ 180625b8b39bSAlexey Kardashevskiy struct vfio_iommu_spapr_tce_create { 180725b8b39bSAlexey Kardashevskiy __u32 argsz; 180825b8b39bSAlexey Kardashevskiy __u32 flags; 180925b8b39bSAlexey Kardashevskiy /* in */ 181025b8b39bSAlexey Kardashevskiy __u32 page_shift; 181166fb2d54SCornelia Huck __u32 __resv1; 181225b8b39bSAlexey Kardashevskiy __u64 window_size; 181325b8b39bSAlexey Kardashevskiy __u32 levels; 181466fb2d54SCornelia Huck __u32 __resv2; 181525b8b39bSAlexey Kardashevskiy /* out */ 181625b8b39bSAlexey Kardashevskiy __u64 start_addr; 181725b8b39bSAlexey Kardashevskiy }; 181825b8b39bSAlexey Kardashevskiy #define VFIO_IOMMU_SPAPR_TCE_CREATE _IO(VFIO_TYPE, VFIO_BASE + 19) 181925b8b39bSAlexey Kardashevskiy 182025b8b39bSAlexey Kardashevskiy /** 182125b8b39bSAlexey Kardashevskiy * VFIO_IOMMU_SPAPR_TCE_REMOVE - _IOW(VFIO_TYPE, VFIO_BASE + 20, struct vfio_iommu_spapr_tce_remove) 182225b8b39bSAlexey Kardashevskiy * 182325b8b39bSAlexey Kardashevskiy * Unprograms a TCE table from all groups in the container and destroys it. 182425b8b39bSAlexey Kardashevskiy * It receives a PCI bus offset as a window id. 182525b8b39bSAlexey Kardashevskiy */ 182625b8b39bSAlexey Kardashevskiy struct vfio_iommu_spapr_tce_remove { 182725b8b39bSAlexey Kardashevskiy __u32 argsz; 182825b8b39bSAlexey Kardashevskiy __u32 flags; 182925b8b39bSAlexey Kardashevskiy /* in */ 183025b8b39bSAlexey Kardashevskiy __u64 start_addr; 183125b8b39bSAlexey Kardashevskiy }; 183225b8b39bSAlexey Kardashevskiy #define VFIO_IOMMU_SPAPR_TCE_REMOVE _IO(VFIO_TYPE, VFIO_BASE + 20) 183325b8b39bSAlexey Kardashevskiy 1834c5daeae1SAlexey Kardashevskiy /* ***************************************************************** */ 1835c5daeae1SAlexey Kardashevskiy 1836d4834ff9SAlexander Graf #endif /* VFIO_H */ 1837