1.. SPDX-License-Identifier: GPL-2.0+ 2 3======= 4IOMMUFD 5======= 6 7:Author: Jason Gunthorpe 8:Author: Kevin Tian 9 10Overview 11======== 12 13IOMMUFD is the user API to control the IOMMU subsystem as it relates to managing 14IO page tables from userspace using file descriptors. It intends to be general 15and consumable by any driver that wants to expose DMA to userspace. These 16drivers are eventually expected to deprecate any internal IOMMU logic 17they may already/historically implement (e.g. vfio_iommu_type1.c). 18 19At minimum iommufd provides universal support of managing I/O address spaces and 20I/O page tables for all IOMMUs, with room in the design to add non-generic 21features to cater to specific hardware functionality. 22 23In this context the capital letter (IOMMUFD) refers to the subsystem while the 24small letter (iommufd) refers to the file descriptors created via /dev/iommu for 25use by userspace. 26 27Key Concepts 28============ 29 30User Visible Objects 31-------------------- 32 33Following IOMMUFD objects are exposed to userspace: 34 35- IOMMUFD_OBJ_IOAS, representing an I/O address space (IOAS), allowing map/unmap 36 of user space memory into ranges of I/O Virtual Address (IOVA). 37 38 The IOAS is a functional replacement for the VFIO container, and like the VFIO 39 container it copies an IOVA map to a list of iommu_domains held within it. 40 41- IOMMUFD_OBJ_DEVICE, representing a device that is bound to iommufd by an 42 external driver. 43 44- IOMMUFD_OBJ_HWPT_PAGING, representing an actual hardware I/O page table 45 (i.e. a single struct iommu_domain) managed by the iommu driver. "PAGING" 46 primarly indicates this type of HWPT should be linked to an IOAS. It also 47 indicates that it is backed by an iommu_domain with __IOMMU_DOMAIN_PAGING 48 feature flag. This can be either an UNMANAGED stage-1 domain for a device 49 running in the user space, or a nesting parent stage-2 domain for mappings 50 from guest-level physical addresses to host-level physical addresses. 51 52 The IOAS has a list of HWPT_PAGINGs that share the same IOVA mapping and 53 it will synchronize its mapping with each member HWPT_PAGING. 54 55- IOMMUFD_OBJ_HWPT_NESTED, representing an actual hardware I/O page table 56 (i.e. a single struct iommu_domain) managed by user space (e.g. guest OS). 57 "NESTED" indicates that this type of HWPT should be linked to an HWPT_PAGING. 58 It also indicates that it is backed by an iommu_domain that has a type of 59 IOMMU_DOMAIN_NESTED. This must be a stage-1 domain for a device running in 60 the user space (e.g. in a guest VM enabling the IOMMU nested translation 61 feature.) As such, it must be created with a given nesting parent stage-2 62 domain to associate to. This nested stage-1 page table managed by the user 63 space usually has mappings from guest-level I/O virtual addresses to guest- 64 level physical addresses. 65 66- IOMMUFD_OBJ_VIOMMU, representing a slice of the physical IOMMU instance, 67 passed to or shared with a VM. It may be some HW-accelerated virtualization 68 features and some SW resources used by the VM. For examples: 69 70 * Security namespace for guest owned ID, e.g. guest-controlled cache tags 71 * Non-device-affiliated event reporting, e.g. invalidation queue errors 72 * Access to a sharable nesting parent pagetable across physical IOMMUs 73 * Virtualization of various platforms IDs, e.g. RIDs and others 74 * Delivery of paravirtualized invalidation 75 * Direct assigned invalidation queues 76 * Direct assigned interrupts 77 78 Such a vIOMMU object generally has the access to a nesting parent pagetable 79 to support some HW-accelerated virtualization features. So, a vIOMMU object 80 must be created given a nesting parent HWPT_PAGING object, and then it would 81 encapsulate that HWPT_PAGING object. Therefore, a vIOMMU object can be used 82 to allocate an HWPT_NESTED object in place of the encapsulated HWPT_PAGING. 83 84 .. note:: 85 86 The name "vIOMMU" isn't necessarily identical to a virtualized IOMMU in a 87 VM. A VM can have one giant virtualized IOMMU running on a machine having 88 multiple physical IOMMUs, in which case the VMM will dispatch the requests 89 or configurations from this single virtualized IOMMU instance to multiple 90 vIOMMU objects created for individual slices of different physical IOMMUs. 91 In other words, a vIOMMU object is always a representation of one physical 92 IOMMU, not necessarily of a virtualized IOMMU. For VMMs that want the full 93 virtualization features from physical IOMMUs, it is suggested to build the 94 same number of virtualized IOMMUs as the number of physical IOMMUs, so the 95 passed-through devices would be connected to their own virtualized IOMMUs 96 backed by corresponding vIOMMU objects, in which case a guest OS would do 97 the "dispatch" naturally instead of VMM trappings. 98 99- IOMMUFD_OBJ_VDEVICE, representing a virtual device for an IOMMUFD_OBJ_DEVICE 100 against an IOMMUFD_OBJ_VIOMMU. This virtual device holds the device's virtual 101 information or attributes (related to the vIOMMU) in a VM. An immediate vDATA 102 example can be the virtual ID of the device on a vIOMMU, which is a unique ID 103 that VMM assigns to the device for a translation channel/port of the vIOMMU, 104 e.g. vSID of ARM SMMUv3, vDeviceID of AMD IOMMU, and vRID of Intel VT-d to a 105 Context Table. Potential use cases of some advanced security information can 106 be forwarded via this object too, such as security level or realm information 107 in a Confidential Compute Architecture. A VMM should create a vDEVICE object 108 to forward all the device information in a VM, when it connects a device to a 109 vIOMMU, which is a separate ioctl call from attaching the same device to an 110 HWPT_PAGING that the vIOMMU holds. 111 112All user-visible objects are destroyed via the IOMMU_DESTROY uAPI. 113 114The diagrams below show relationships between user-visible objects and kernel 115datastructures (external to iommufd), with numbers referred to operations 116creating the objects and links:: 117 118 _______________________________________________________________________ 119 | iommufd (HWPT_PAGING only) | 120 | | 121 | [1] [3] [2] | 122 | ________________ _____________ ________ | 123 | | | | | | | | 124 | | IOAS |<---| HWPT_PAGING |<---------------------| DEVICE | | 125 | |________________| |_____________| |________| | 126 | | | | | 127 |_________|____________________|__________________________________|_____| 128 | | | 129 | ______v_____ ___v__ 130 | PFN storage | (paging) | |struct| 131 |------------>|iommu_domain|<-----------------------|device| 132 |____________| |______| 133 134 _______________________________________________________________________ 135 | iommufd (with HWPT_NESTED) | 136 | | 137 | [1] [3] [4] [2] | 138 | ________________ _____________ _____________ ________ | 139 | | | | | | | | | | 140 | | IOAS |<---| HWPT_PAGING |<---| HWPT_NESTED |<--| DEVICE | | 141 | |________________| |_____________| |_____________| |________| | 142 | | | | | | 143 |_________|____________________|__________________|_______________|_____| 144 | | | | 145 | ______v_____ ______v_____ ___v__ 146 | PFN storage | (paging) | | (nested) | |struct| 147 |------------>|iommu_domain|<----|iommu_domain|<----|device| 148 |____________| |____________| |______| 149 150 _______________________________________________________________________ 151 | iommufd (with vIOMMU/vDEVICE) | 152 | | 153 | [5] [6] | 154 | _____________ _____________ | 155 | | | | | | 156 | |----------------| vIOMMU |<---| vDEVICE |<----| | 157 | | | | |_____________| | | 158 | | | | | | 159 | | [1] | | [4] | [2] | 160 | | ______ | | _____________ _|______ | 161 | | | | | [3] | | | | | | 162 | | | IOAS |<---|(HWPT_PAGING)|<---| HWPT_NESTED |<--| DEVICE | | 163 | | |______| |_____________| |_____________| |________| | 164 | | | | | | | 165 |______|________|______________|__________________|_______________|_____| 166 | | | | | 167 ______v_____ | ______v_____ ______v_____ ___v__ 168 | struct | | PFN | (paging) | | (nested) | |struct| 169 |iommu_device| |------>|iommu_domain|<----|iommu_domain|<----|device| 170 |____________| storage|____________| |____________| |______| 171 1721. IOMMUFD_OBJ_IOAS is created via the IOMMU_IOAS_ALLOC uAPI. An iommufd can 173 hold multiple IOAS objects. IOAS is the most generic object and does not 174 expose interfaces that are specific to single IOMMU drivers. All operations 175 on the IOAS must operate equally on each of the iommu_domains inside of it. 176 1772. IOMMUFD_OBJ_DEVICE is created when an external driver calls the IOMMUFD kAPI 178 to bind a device to an iommufd. The driver is expected to implement a set of 179 ioctls to allow userspace to initiate the binding operation. Successful 180 completion of this operation establishes the desired DMA ownership over the 181 device. The driver must also set the driver_managed_dma flag and must not 182 touch the device until this operation succeeds. 183 1843. IOMMUFD_OBJ_HWPT_PAGING can be created in two ways: 185 186 * IOMMUFD_OBJ_HWPT_PAGING is automatically created when an external driver 187 calls the IOMMUFD kAPI to attach a bound device to an IOAS. Similarly the 188 external driver uAPI allows userspace to initiate the attaching operation. 189 If a compatible member HWPT_PAGING object exists in the IOAS's HWPT_PAGING 190 list, then it will be reused. Otherwise a new HWPT_PAGING that represents 191 an iommu_domain to userspace will be created, and then added to the list. 192 Successful completion of this operation sets up the linkages among IOAS, 193 device and iommu_domain. Once this completes the device could do DMA. 194 195 * IOMMUFD_OBJ_HWPT_PAGING can be manually created via the IOMMU_HWPT_ALLOC 196 uAPI, provided an ioas_id via @pt_id to associate the new HWPT_PAGING to 197 the corresponding IOAS object. The benefit of this manual allocation is to 198 allow allocation flags (defined in enum iommufd_hwpt_alloc_flags), e.g. it 199 allocates a nesting parent HWPT_PAGING if the IOMMU_HWPT_ALLOC_NEST_PARENT 200 flag is set. 201 2024. IOMMUFD_OBJ_HWPT_NESTED can be only manually created via the IOMMU_HWPT_ALLOC 203 uAPI, provided an hwpt_id or a viommu_id of a vIOMMU object encapsulating a 204 nesting parent HWPT_PAGING via @pt_id to associate the new HWPT_NESTED object 205 to the corresponding HWPT_PAGING object. The associating HWPT_PAGING object 206 must be a nesting parent manually allocated via the same uAPI previously with 207 an IOMMU_HWPT_ALLOC_NEST_PARENT flag, otherwise the allocation will fail. The 208 allocation will be further validated by the IOMMU driver to ensure that the 209 nesting parent domain and the nested domain being allocated are compatible. 210 Successful completion of this operation sets up linkages among IOAS, device, 211 and iommu_domains. Once this completes the device could do DMA via a 2-stage 212 translation, a.k.a nested translation. Note that multiple HWPT_NESTED objects 213 can be allocated by (and then associated to) the same nesting parent. 214 215 .. note:: 216 217 Either a manual IOMMUFD_OBJ_HWPT_PAGING or an IOMMUFD_OBJ_HWPT_NESTED is 218 created via the same IOMMU_HWPT_ALLOC uAPI. The difference is at the type 219 of the object passed in via the @pt_id field of struct iommufd_hwpt_alloc. 220 2215. IOMMUFD_OBJ_VIOMMU can be only manually created via the IOMMU_VIOMMU_ALLOC 222 uAPI, provided a dev_id (for the device's physical IOMMU to back the vIOMMU) 223 and an hwpt_id (to associate the vIOMMU to a nesting parent HWPT_PAGING). The 224 iommufd core will link the vIOMMU object to the struct iommu_device that the 225 struct device is behind. And an IOMMU driver can implement a viommu_alloc op 226 to allocate its own vIOMMU data structure embedding the core-level structure 227 iommufd_viommu and some driver-specific data. If necessary, the driver can 228 also configure its HW virtualization feature for that vIOMMU (and thus for 229 the VM). Successful completion of this operation sets up the linkages between 230 the vIOMMU object and the HWPT_PAGING, then this vIOMMU object can be used 231 as a nesting parent object to allocate an HWPT_NESTED object described above. 232 2336. IOMMUFD_OBJ_VDEVICE can be only manually created via the IOMMU_VDEVICE_ALLOC 234 uAPI, provided a viommu_id for an iommufd_viommu object and a dev_id for an 235 iommufd_device object. The vDEVICE object will be the binding between these 236 two parent objects. Another @virt_id will be also set via the uAPI providing 237 the iommufd core an index to store the vDEVICE object to a vDEVICE array per 238 vIOMMU. If necessary, the IOMMU driver may choose to implement a vdevce_alloc 239 op to init its HW for virtualization feature related to a vDEVICE. Successful 240 completion of this operation sets up the linkages between vIOMMU and device. 241 242A device can only bind to an iommufd due to DMA ownership claim and attach to at 243most one IOAS object (no support of PASID yet). 244 245Kernel Datastructure 246-------------------- 247 248User visible objects are backed by following datastructures: 249 250- iommufd_ioas for IOMMUFD_OBJ_IOAS. 251- iommufd_device for IOMMUFD_OBJ_DEVICE. 252- iommufd_hwpt_paging for IOMMUFD_OBJ_HWPT_PAGING. 253- iommufd_hwpt_nested for IOMMUFD_OBJ_HWPT_NESTED. 254- iommufd_viommu for IOMMUFD_OBJ_VIOMMU. 255- iommufd_vdevice for IOMMUFD_OBJ_VDEVICE. 256 257Several terminologies when looking at these datastructures: 258 259- Automatic domain - refers to an iommu domain created automatically when 260 attaching a device to an IOAS object. This is compatible to the semantics of 261 VFIO type1. 262 263- Manual domain - refers to an iommu domain designated by the user as the 264 target pagetable to be attached to by a device. Though currently there are 265 no uAPIs to directly create such domain, the datastructure and algorithms 266 are ready for handling that use case. 267 268- In-kernel user - refers to something like a VFIO mdev that is using the 269 IOMMUFD access interface to access the IOAS. This starts by creating an 270 iommufd_access object that is similar to the domain binding a physical device 271 would do. The access object will then allow converting IOVA ranges into struct 272 page * lists, or doing direct read/write to an IOVA. 273 274iommufd_ioas serves as the metadata datastructure to manage how IOVA ranges are 275mapped to memory pages, composed of: 276 277- struct io_pagetable holding the IOVA map 278- struct iopt_area's representing populated portions of IOVA 279- struct iopt_pages representing the storage of PFNs 280- struct iommu_domain representing the IO page table in the IOMMU 281- struct iopt_pages_access representing in-kernel users of PFNs 282- struct xarray pinned_pfns holding a list of pages pinned by in-kernel users 283 284Each iopt_pages represents a logical linear array of full PFNs. The PFNs are 285ultimately derived from userspace VAs via an mm_struct. Once they have been 286pinned the PFNs are stored in IOPTEs of an iommu_domain or inside the pinned_pfns 287xarray if they have been pinned through an iommufd_access. 288 289PFN have to be copied between all combinations of storage locations, depending 290on what domains are present and what kinds of in-kernel "software access" users 291exist. The mechanism ensures that a page is pinned only once. 292 293An io_pagetable is composed of iopt_areas pointing at iopt_pages, along with a 294list of iommu_domains that mirror the IOVA to PFN map. 295 296Multiple io_pagetable-s, through their iopt_area-s, can share a single 297iopt_pages which avoids multi-pinning and double accounting of page 298consumption. 299 300iommufd_ioas is shareable between subsystems, e.g. VFIO and VDPA, as long as 301devices managed by different subsystems are bound to a same iommufd. 302 303IOMMUFD User API 304================ 305 306.. kernel-doc:: include/uapi/linux/iommufd.h 307 308IOMMUFD Kernel API 309================== 310 311The IOMMUFD kAPI is device-centric with group-related tricks managed behind the 312scene. This allows the external drivers calling such kAPI to implement a simple 313device-centric uAPI for connecting its device to an iommufd, instead of 314explicitly imposing the group semantics in its uAPI as VFIO does. 315 316.. kernel-doc:: drivers/iommu/iommufd/device.c 317 :export: 318 319.. kernel-doc:: drivers/iommu/iommufd/main.c 320 :export: 321 322VFIO and IOMMUFD 323---------------- 324 325Connecting a VFIO device to iommufd can be done in two ways. 326 327First is a VFIO compatible way by directly implementing the /dev/vfio/vfio 328container IOCTLs by mapping them into io_pagetable operations. Doing so allows 329the use of iommufd in legacy VFIO applications by symlinking /dev/vfio/vfio to 330/dev/iommufd or extending VFIO to SET_CONTAINER using an iommufd instead of a 331container fd. 332 333The second approach directly extends VFIO to support a new set of device-centric 334user API based on aforementioned IOMMUFD kernel API. It requires userspace 335change but better matches the IOMMUFD API semantics and easier to support new 336iommufd features when comparing it to the first approach. 337 338Currently both approaches are still work-in-progress. 339 340There are still a few gaps to be resolved to catch up with VFIO type1, as 341documented in iommufd_vfio_check_extension(). 342 343Future TODOs 344============ 345 346Currently IOMMUFD supports only kernel-managed I/O page table, similar to VFIO 347type1. New features on the radar include: 348 349 - Binding iommu_domain's to PASID/SSID 350 - Userspace page tables, for ARM, x86 and S390 351 - Kernel bypass'd invalidation of user page tables 352 - Re-use of the KVM page table in the IOMMU 353 - Dirty page tracking in the IOMMU 354 - Runtime Increase/Decrease of IOPTE size 355 - PRI support with faults resolved in userspace 356