1 /* 2 * Physical memory management API 3 * 4 * Copyright 2011 Red Hat, Inc. and/or its affiliates 5 * 6 * Authors: 7 * Avi Kivity <avi@redhat.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 */ 13 14 #ifndef MEMORY_H 15 #define MEMORY_H 16 17 #ifndef CONFIG_USER_ONLY 18 19 #include "exec/cpu-common.h" 20 #include "exec/hwaddr.h" 21 #include "exec/memattrs.h" 22 #include "exec/memop.h" 23 #include "exec/ramlist.h" 24 #include "qemu/bswap.h" 25 #include "qemu/queue.h" 26 #include "qemu/int128.h" 27 #include "qemu/notify.h" 28 #include "qom/object.h" 29 #include "qemu/rcu.h" 30 31 #define RAM_ADDR_INVALID (~(ram_addr_t)0) 32 33 #define MAX_PHYS_ADDR_SPACE_BITS 62 34 #define MAX_PHYS_ADDR (((hwaddr)1 << MAX_PHYS_ADDR_SPACE_BITS) - 1) 35 36 #define TYPE_MEMORY_REGION "memory-region" 37 DECLARE_INSTANCE_CHECKER(MemoryRegion, MEMORY_REGION, 38 TYPE_MEMORY_REGION) 39 40 #define TYPE_IOMMU_MEMORY_REGION "iommu-memory-region" 41 typedef struct IOMMUMemoryRegionClass IOMMUMemoryRegionClass; 42 DECLARE_OBJ_CHECKERS(IOMMUMemoryRegion, IOMMUMemoryRegionClass, 43 IOMMU_MEMORY_REGION, TYPE_IOMMU_MEMORY_REGION) 44 45 #define TYPE_RAM_DISCARD_MANAGER "qemu:ram-discard-manager" 46 typedef struct RamDiscardManagerClass RamDiscardManagerClass; 47 typedef struct RamDiscardManager RamDiscardManager; 48 DECLARE_OBJ_CHECKERS(RamDiscardManager, RamDiscardManagerClass, 49 RAM_DISCARD_MANAGER, TYPE_RAM_DISCARD_MANAGER); 50 51 #ifdef CONFIG_FUZZ 52 void fuzz_dma_read_cb(size_t addr, 53 size_t len, 54 MemoryRegion *mr); 55 #else 56 static inline void fuzz_dma_read_cb(size_t addr, 57 size_t len, 58 MemoryRegion *mr) 59 { 60 /* Do Nothing */ 61 } 62 #endif 63 64 /* Possible bits for global_dirty_log_{start|stop} */ 65 66 /* Dirty tracking enabled because migration is running */ 67 #define GLOBAL_DIRTY_MIGRATION (1U << 0) 68 69 /* Dirty tracking enabled because measuring dirty rate */ 70 #define GLOBAL_DIRTY_DIRTY_RATE (1U << 1) 71 72 /* Dirty tracking enabled because dirty limit */ 73 #define GLOBAL_DIRTY_LIMIT (1U << 2) 74 75 #define GLOBAL_DIRTY_MASK (0x7) 76 77 extern unsigned int global_dirty_tracking; 78 79 typedef struct MemoryRegionOps MemoryRegionOps; 80 81 struct ReservedRegion { 82 hwaddr low; 83 hwaddr high; 84 unsigned type; 85 }; 86 87 /** 88 * struct MemoryRegionSection: describes a fragment of a #MemoryRegion 89 * 90 * @mr: the region, or %NULL if empty 91 * @fv: the flat view of the address space the region is mapped in 92 * @offset_within_region: the beginning of the section, relative to @mr's start 93 * @size: the size of the section; will not exceed @mr's boundaries 94 * @offset_within_address_space: the address of the first byte of the section 95 * relative to the region's address space 96 * @readonly: writes to this section are ignored 97 * @nonvolatile: this section is non-volatile 98 */ 99 struct MemoryRegionSection { 100 Int128 size; 101 MemoryRegion *mr; 102 FlatView *fv; 103 hwaddr offset_within_region; 104 hwaddr offset_within_address_space; 105 bool readonly; 106 bool nonvolatile; 107 }; 108 109 typedef struct IOMMUTLBEntry IOMMUTLBEntry; 110 111 /* See address_space_translate: bit 0 is read, bit 1 is write. */ 112 typedef enum { 113 IOMMU_NONE = 0, 114 IOMMU_RO = 1, 115 IOMMU_WO = 2, 116 IOMMU_RW = 3, 117 } IOMMUAccessFlags; 118 119 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0)) 120 121 struct IOMMUTLBEntry { 122 AddressSpace *target_as; 123 hwaddr iova; 124 hwaddr translated_addr; 125 hwaddr addr_mask; /* 0xfff = 4k translation */ 126 IOMMUAccessFlags perm; 127 }; 128 129 /* 130 * Bitmap for different IOMMUNotifier capabilities. Each notifier can 131 * register with one or multiple IOMMU Notifier capability bit(s). 132 * 133 * Normally there're two use cases for the notifiers: 134 * 135 * (1) When the device needs accurate synchronizations of the vIOMMU page 136 * tables, it needs to register with both MAP|UNMAP notifies (which 137 * is defined as IOMMU_NOTIFIER_IOTLB_EVENTS below). 138 * 139 * Regarding to accurate synchronization, it's when the notified 140 * device maintains a shadow page table and must be notified on each 141 * guest MAP (page table entry creation) and UNMAP (invalidation) 142 * events (e.g. VFIO). Both notifications must be accurate so that 143 * the shadow page table is fully in sync with the guest view. 144 * 145 * (2) When the device doesn't need accurate synchronizations of the 146 * vIOMMU page tables, it needs to register only with UNMAP or 147 * DEVIOTLB_UNMAP notifies. 148 * 149 * It's when the device maintains a cache of IOMMU translations 150 * (IOTLB) and is able to fill that cache by requesting translations 151 * from the vIOMMU through a protocol similar to ATS (Address 152 * Translation Service). 153 * 154 * Note that in this mode the vIOMMU will not maintain a shadowed 155 * page table for the address space, and the UNMAP messages can cover 156 * more than the pages that used to get mapped. The IOMMU notifiee 157 * should be able to take care of over-sized invalidations. 158 */ 159 typedef enum { 160 IOMMU_NOTIFIER_NONE = 0, 161 /* Notify cache invalidations */ 162 IOMMU_NOTIFIER_UNMAP = 0x1, 163 /* Notify entry changes (newly created entries) */ 164 IOMMU_NOTIFIER_MAP = 0x2, 165 /* Notify changes on device IOTLB entries */ 166 IOMMU_NOTIFIER_DEVIOTLB_UNMAP = 0x04, 167 } IOMMUNotifierFlag; 168 169 #define IOMMU_NOTIFIER_IOTLB_EVENTS (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP) 170 #define IOMMU_NOTIFIER_DEVIOTLB_EVENTS IOMMU_NOTIFIER_DEVIOTLB_UNMAP 171 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_IOTLB_EVENTS | \ 172 IOMMU_NOTIFIER_DEVIOTLB_EVENTS) 173 174 struct IOMMUNotifier; 175 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier, 176 IOMMUTLBEntry *data); 177 178 struct IOMMUNotifier { 179 IOMMUNotify notify; 180 IOMMUNotifierFlag notifier_flags; 181 /* Notify for address space range start <= addr <= end */ 182 hwaddr start; 183 hwaddr end; 184 int iommu_idx; 185 QLIST_ENTRY(IOMMUNotifier) node; 186 }; 187 typedef struct IOMMUNotifier IOMMUNotifier; 188 189 typedef struct IOMMUTLBEvent { 190 IOMMUNotifierFlag type; 191 IOMMUTLBEntry entry; 192 } IOMMUTLBEvent; 193 194 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */ 195 #define RAM_PREALLOC (1 << 0) 196 197 /* RAM is mmap-ed with MAP_SHARED */ 198 #define RAM_SHARED (1 << 1) 199 200 /* Only a portion of RAM (used_length) is actually used, and migrated. 201 * Resizing RAM while migrating can result in the migration being canceled. 202 */ 203 #define RAM_RESIZEABLE (1 << 2) 204 205 /* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically 206 * zero the page and wake waiting processes. 207 * (Set during postcopy) 208 */ 209 #define RAM_UF_ZEROPAGE (1 << 3) 210 211 /* RAM can be migrated */ 212 #define RAM_MIGRATABLE (1 << 4) 213 214 /* RAM is a persistent kind memory */ 215 #define RAM_PMEM (1 << 5) 216 217 218 /* 219 * UFFDIO_WRITEPROTECT is used on this RAMBlock to 220 * support 'write-tracking' migration type. 221 * Implies ram_state->ram_wt_enabled. 222 */ 223 #define RAM_UF_WRITEPROTECT (1 << 6) 224 225 /* 226 * RAM is mmap-ed with MAP_NORESERVE. When set, reserving swap space (or huge 227 * pages if applicable) is skipped: will bail out if not supported. When not 228 * set, the OS will do the reservation, if supported for the memory type. 229 */ 230 #define RAM_NORESERVE (1 << 7) 231 232 /* RAM that isn't accessible through normal means. */ 233 #define RAM_PROTECTED (1 << 8) 234 235 /* RAM is an mmap-ed named file */ 236 #define RAM_NAMED_FILE (1 << 9) 237 238 /* RAM is mmap-ed read-only */ 239 #define RAM_READONLY (1 << 10) 240 241 /* RAM FD is opened read-only */ 242 #define RAM_READONLY_FD (1 << 11) 243 244 static inline void iommu_notifier_init(IOMMUNotifier *n, IOMMUNotify fn, 245 IOMMUNotifierFlag flags, 246 hwaddr start, hwaddr end, 247 int iommu_idx) 248 { 249 n->notify = fn; 250 n->notifier_flags = flags; 251 n->start = start; 252 n->end = end; 253 n->iommu_idx = iommu_idx; 254 } 255 256 /* 257 * Memory region callbacks 258 */ 259 struct MemoryRegionOps { 260 /* Read from the memory region. @addr is relative to @mr; @size is 261 * in bytes. */ 262 uint64_t (*read)(void *opaque, 263 hwaddr addr, 264 unsigned size); 265 /* Write to the memory region. @addr is relative to @mr; @size is 266 * in bytes. */ 267 void (*write)(void *opaque, 268 hwaddr addr, 269 uint64_t data, 270 unsigned size); 271 272 MemTxResult (*read_with_attrs)(void *opaque, 273 hwaddr addr, 274 uint64_t *data, 275 unsigned size, 276 MemTxAttrs attrs); 277 MemTxResult (*write_with_attrs)(void *opaque, 278 hwaddr addr, 279 uint64_t data, 280 unsigned size, 281 MemTxAttrs attrs); 282 283 enum device_endian endianness; 284 /* Guest-visible constraints: */ 285 struct { 286 /* If nonzero, specify bounds on access sizes beyond which a machine 287 * check is thrown. 288 */ 289 unsigned min_access_size; 290 unsigned max_access_size; 291 /* If true, unaligned accesses are supported. Otherwise unaligned 292 * accesses throw machine checks. 293 */ 294 bool unaligned; 295 /* 296 * If present, and returns #false, the transaction is not accepted 297 * by the device (and results in machine dependent behaviour such 298 * as a machine check exception). 299 */ 300 bool (*accepts)(void *opaque, hwaddr addr, 301 unsigned size, bool is_write, 302 MemTxAttrs attrs); 303 } valid; 304 /* Internal implementation constraints: */ 305 struct { 306 /* If nonzero, specifies the minimum size implemented. Smaller sizes 307 * will be rounded upwards and a partial result will be returned. 308 */ 309 unsigned min_access_size; 310 /* If nonzero, specifies the maximum size implemented. Larger sizes 311 * will be done as a series of accesses with smaller sizes. 312 */ 313 unsigned max_access_size; 314 /* If true, unaligned accesses are supported. Otherwise all accesses 315 * are converted to (possibly multiple) naturally aligned accesses. 316 */ 317 bool unaligned; 318 } impl; 319 }; 320 321 typedef struct MemoryRegionClass { 322 /* private */ 323 ObjectClass parent_class; 324 } MemoryRegionClass; 325 326 327 enum IOMMUMemoryRegionAttr { 328 IOMMU_ATTR_SPAPR_TCE_FD 329 }; 330 331 /* 332 * IOMMUMemoryRegionClass: 333 * 334 * All IOMMU implementations need to subclass TYPE_IOMMU_MEMORY_REGION 335 * and provide an implementation of at least the @translate method here 336 * to handle requests to the memory region. Other methods are optional. 337 * 338 * The IOMMU implementation must use the IOMMU notifier infrastructure 339 * to report whenever mappings are changed, by calling 340 * memory_region_notify_iommu() (or, if necessary, by calling 341 * memory_region_notify_iommu_one() for each registered notifier). 342 * 343 * Conceptually an IOMMU provides a mapping from input address 344 * to an output TLB entry. If the IOMMU is aware of memory transaction 345 * attributes and the output TLB entry depends on the transaction 346 * attributes, we represent this using IOMMU indexes. Each index 347 * selects a particular translation table that the IOMMU has: 348 * 349 * @attrs_to_index returns the IOMMU index for a set of transaction attributes 350 * 351 * @translate takes an input address and an IOMMU index 352 * 353 * and the mapping returned can only depend on the input address and the 354 * IOMMU index. 355 * 356 * Most IOMMUs don't care about the transaction attributes and support 357 * only a single IOMMU index. A more complex IOMMU might have one index 358 * for secure transactions and one for non-secure transactions. 359 */ 360 struct IOMMUMemoryRegionClass { 361 /* private: */ 362 MemoryRegionClass parent_class; 363 364 /* public: */ 365 /** 366 * @translate: 367 * 368 * Return a TLB entry that contains a given address. 369 * 370 * The IOMMUAccessFlags indicated via @flag are optional and may 371 * be specified as IOMMU_NONE to indicate that the caller needs 372 * the full translation information for both reads and writes. If 373 * the access flags are specified then the IOMMU implementation 374 * may use this as an optimization, to stop doing a page table 375 * walk as soon as it knows that the requested permissions are not 376 * allowed. If IOMMU_NONE is passed then the IOMMU must do the 377 * full page table walk and report the permissions in the returned 378 * IOMMUTLBEntry. (Note that this implies that an IOMMU may not 379 * return different mappings for reads and writes.) 380 * 381 * The returned information remains valid while the caller is 382 * holding the big QEMU lock or is inside an RCU critical section; 383 * if the caller wishes to cache the mapping beyond that it must 384 * register an IOMMU notifier so it can invalidate its cached 385 * information when the IOMMU mapping changes. 386 * 387 * @iommu: the IOMMUMemoryRegion 388 * 389 * @hwaddr: address to be translated within the memory region 390 * 391 * @flag: requested access permission 392 * 393 * @iommu_idx: IOMMU index for the translation 394 */ 395 IOMMUTLBEntry (*translate)(IOMMUMemoryRegion *iommu, hwaddr addr, 396 IOMMUAccessFlags flag, int iommu_idx); 397 /** 398 * @get_min_page_size: 399 * 400 * Returns minimum supported page size in bytes. 401 * 402 * If this method is not provided then the minimum is assumed to 403 * be TARGET_PAGE_SIZE. 404 * 405 * @iommu: the IOMMUMemoryRegion 406 */ 407 uint64_t (*get_min_page_size)(IOMMUMemoryRegion *iommu); 408 /** 409 * @notify_flag_changed: 410 * 411 * Called when IOMMU Notifier flag changes (ie when the set of 412 * events which IOMMU users are requesting notification for changes). 413 * Optional method -- need not be provided if the IOMMU does not 414 * need to know exactly which events must be notified. 415 * 416 * @iommu: the IOMMUMemoryRegion 417 * 418 * @old_flags: events which previously needed to be notified 419 * 420 * @new_flags: events which now need to be notified 421 * 422 * Returns 0 on success, or a negative errno; in particular 423 * returns -EINVAL if the new flag bitmap is not supported by the 424 * IOMMU memory region. In case of failure, the error object 425 * must be created 426 */ 427 int (*notify_flag_changed)(IOMMUMemoryRegion *iommu, 428 IOMMUNotifierFlag old_flags, 429 IOMMUNotifierFlag new_flags, 430 Error **errp); 431 /** 432 * @replay: 433 * 434 * Called to handle memory_region_iommu_replay(). 435 * 436 * The default implementation of memory_region_iommu_replay() is to 437 * call the IOMMU translate method for every page in the address space 438 * with flag == IOMMU_NONE and then call the notifier if translate 439 * returns a valid mapping. If this method is implemented then it 440 * overrides the default behaviour, and must provide the full semantics 441 * of memory_region_iommu_replay(), by calling @notifier for every 442 * translation present in the IOMMU. 443 * 444 * Optional method -- an IOMMU only needs to provide this method 445 * if the default is inefficient or produces undesirable side effects. 446 * 447 * Note: this is not related to record-and-replay functionality. 448 */ 449 void (*replay)(IOMMUMemoryRegion *iommu, IOMMUNotifier *notifier); 450 451 /** 452 * @get_attr: 453 * 454 * Get IOMMU misc attributes. This is an optional method that 455 * can be used to allow users of the IOMMU to get implementation-specific 456 * information. The IOMMU implements this method to handle calls 457 * by IOMMU users to memory_region_iommu_get_attr() by filling in 458 * the arbitrary data pointer for any IOMMUMemoryRegionAttr values that 459 * the IOMMU supports. If the method is unimplemented then 460 * memory_region_iommu_get_attr() will always return -EINVAL. 461 * 462 * @iommu: the IOMMUMemoryRegion 463 * 464 * @attr: attribute being queried 465 * 466 * @data: memory to fill in with the attribute data 467 * 468 * Returns 0 on success, or a negative errno; in particular 469 * returns -EINVAL for unrecognized or unimplemented attribute types. 470 */ 471 int (*get_attr)(IOMMUMemoryRegion *iommu, enum IOMMUMemoryRegionAttr attr, 472 void *data); 473 474 /** 475 * @attrs_to_index: 476 * 477 * Return the IOMMU index to use for a given set of transaction attributes. 478 * 479 * Optional method: if an IOMMU only supports a single IOMMU index then 480 * the default implementation of memory_region_iommu_attrs_to_index() 481 * will return 0. 482 * 483 * The indexes supported by an IOMMU must be contiguous, starting at 0. 484 * 485 * @iommu: the IOMMUMemoryRegion 486 * @attrs: memory transaction attributes 487 */ 488 int (*attrs_to_index)(IOMMUMemoryRegion *iommu, MemTxAttrs attrs); 489 490 /** 491 * @num_indexes: 492 * 493 * Return the number of IOMMU indexes this IOMMU supports. 494 * 495 * Optional method: if this method is not provided, then 496 * memory_region_iommu_num_indexes() will return 1, indicating that 497 * only a single IOMMU index is supported. 498 * 499 * @iommu: the IOMMUMemoryRegion 500 */ 501 int (*num_indexes)(IOMMUMemoryRegion *iommu); 502 503 /** 504 * @iommu_set_page_size_mask: 505 * 506 * Restrict the page size mask that can be supported with a given IOMMU 507 * memory region. Used for example to propagate host physical IOMMU page 508 * size mask limitations to the virtual IOMMU. 509 * 510 * Optional method: if this method is not provided, then the default global 511 * page mask is used. 512 * 513 * @iommu: the IOMMUMemoryRegion 514 * 515 * @page_size_mask: a bitmask of supported page sizes. At least one bit, 516 * representing the smallest page size, must be set. Additional set bits 517 * represent supported block sizes. For example a host physical IOMMU that 518 * uses page tables with a page size of 4kB, and supports 2MB and 4GB 519 * blocks, will set mask 0x40201000. A granule of 4kB with indiscriminate 520 * block sizes is specified with mask 0xfffffffffffff000. 521 * 522 * Returns 0 on success, or a negative error. In case of failure, the error 523 * object must be created. 524 */ 525 int (*iommu_set_page_size_mask)(IOMMUMemoryRegion *iommu, 526 uint64_t page_size_mask, 527 Error **errp); 528 }; 529 530 typedef struct RamDiscardListener RamDiscardListener; 531 typedef int (*NotifyRamPopulate)(RamDiscardListener *rdl, 532 MemoryRegionSection *section); 533 typedef void (*NotifyRamDiscard)(RamDiscardListener *rdl, 534 MemoryRegionSection *section); 535 536 struct RamDiscardListener { 537 /* 538 * @notify_populate: 539 * 540 * Notification that previously discarded memory is about to get populated. 541 * Listeners are able to object. If any listener objects, already 542 * successfully notified listeners are notified about a discard again. 543 * 544 * @rdl: the #RamDiscardListener getting notified 545 * @section: the #MemoryRegionSection to get populated. The section 546 * is aligned within the memory region to the minimum granularity 547 * unless it would exceed the registered section. 548 * 549 * Returns 0 on success. If the notification is rejected by the listener, 550 * an error is returned. 551 */ 552 NotifyRamPopulate notify_populate; 553 554 /* 555 * @notify_discard: 556 * 557 * Notification that previously populated memory was discarded successfully 558 * and listeners should drop all references to such memory and prevent 559 * new population (e.g., unmap). 560 * 561 * @rdl: the #RamDiscardListener getting notified 562 * @section: the #MemoryRegionSection to get populated. The section 563 * is aligned within the memory region to the minimum granularity 564 * unless it would exceed the registered section. 565 */ 566 NotifyRamDiscard notify_discard; 567 568 /* 569 * @double_discard_supported: 570 * 571 * The listener suppors getting @notify_discard notifications that span 572 * already discarded parts. 573 */ 574 bool double_discard_supported; 575 576 MemoryRegionSection *section; 577 QLIST_ENTRY(RamDiscardListener) next; 578 }; 579 580 static inline void ram_discard_listener_init(RamDiscardListener *rdl, 581 NotifyRamPopulate populate_fn, 582 NotifyRamDiscard discard_fn, 583 bool double_discard_supported) 584 { 585 rdl->notify_populate = populate_fn; 586 rdl->notify_discard = discard_fn; 587 rdl->double_discard_supported = double_discard_supported; 588 } 589 590 typedef int (*ReplayRamPopulate)(MemoryRegionSection *section, void *opaque); 591 typedef void (*ReplayRamDiscard)(MemoryRegionSection *section, void *opaque); 592 593 /* 594 * RamDiscardManagerClass: 595 * 596 * A #RamDiscardManager coordinates which parts of specific RAM #MemoryRegion 597 * regions are currently populated to be used/accessed by the VM, notifying 598 * after parts were discarded (freeing up memory) and before parts will be 599 * populated (consuming memory), to be used/accessed by the VM. 600 * 601 * A #RamDiscardManager can only be set for a RAM #MemoryRegion while the 602 * #MemoryRegion isn't mapped yet; it cannot change while the #MemoryRegion is 603 * mapped. 604 * 605 * The #RamDiscardManager is intended to be used by technologies that are 606 * incompatible with discarding of RAM (e.g., VFIO, which may pin all 607 * memory inside a #MemoryRegion), and require proper coordination to only 608 * map the currently populated parts, to hinder parts that are expected to 609 * remain discarded from silently getting populated and consuming memory. 610 * Technologies that support discarding of RAM don't have to bother and can 611 * simply map the whole #MemoryRegion. 612 * 613 * An example #RamDiscardManager is virtio-mem, which logically (un)plugs 614 * memory within an assigned RAM #MemoryRegion, coordinated with the VM. 615 * Logically unplugging memory consists of discarding RAM. The VM agreed to not 616 * access unplugged (discarded) memory - especially via DMA. virtio-mem will 617 * properly coordinate with listeners before memory is plugged (populated), 618 * and after memory is unplugged (discarded). 619 * 620 * Listeners are called in multiples of the minimum granularity (unless it 621 * would exceed the registered range) and changes are aligned to the minimum 622 * granularity within the #MemoryRegion. Listeners have to prepare for memory 623 * becoming discarded in a different granularity than it was populated and the 624 * other way around. 625 */ 626 struct RamDiscardManagerClass { 627 /* private */ 628 InterfaceClass parent_class; 629 630 /* public */ 631 632 /** 633 * @get_min_granularity: 634 * 635 * Get the minimum granularity in which listeners will get notified 636 * about changes within the #MemoryRegion via the #RamDiscardManager. 637 * 638 * @rdm: the #RamDiscardManager 639 * @mr: the #MemoryRegion 640 * 641 * Returns the minimum granularity. 642 */ 643 uint64_t (*get_min_granularity)(const RamDiscardManager *rdm, 644 const MemoryRegion *mr); 645 646 /** 647 * @is_populated: 648 * 649 * Check whether the given #MemoryRegionSection is completely populated 650 * (i.e., no parts are currently discarded) via the #RamDiscardManager. 651 * There are no alignment requirements. 652 * 653 * @rdm: the #RamDiscardManager 654 * @section: the #MemoryRegionSection 655 * 656 * Returns whether the given range is completely populated. 657 */ 658 bool (*is_populated)(const RamDiscardManager *rdm, 659 const MemoryRegionSection *section); 660 661 /** 662 * @replay_populated: 663 * 664 * Call the #ReplayRamPopulate callback for all populated parts within the 665 * #MemoryRegionSection via the #RamDiscardManager. 666 * 667 * In case any call fails, no further calls are made. 668 * 669 * @rdm: the #RamDiscardManager 670 * @section: the #MemoryRegionSection 671 * @replay_fn: the #ReplayRamPopulate callback 672 * @opaque: pointer to forward to the callback 673 * 674 * Returns 0 on success, or a negative error if any notification failed. 675 */ 676 int (*replay_populated)(const RamDiscardManager *rdm, 677 MemoryRegionSection *section, 678 ReplayRamPopulate replay_fn, void *opaque); 679 680 /** 681 * @replay_discarded: 682 * 683 * Call the #ReplayRamDiscard callback for all discarded parts within the 684 * #MemoryRegionSection via the #RamDiscardManager. 685 * 686 * @rdm: the #RamDiscardManager 687 * @section: the #MemoryRegionSection 688 * @replay_fn: the #ReplayRamDiscard callback 689 * @opaque: pointer to forward to the callback 690 */ 691 void (*replay_discarded)(const RamDiscardManager *rdm, 692 MemoryRegionSection *section, 693 ReplayRamDiscard replay_fn, void *opaque); 694 695 /** 696 * @register_listener: 697 * 698 * Register a #RamDiscardListener for the given #MemoryRegionSection and 699 * immediately notify the #RamDiscardListener about all populated parts 700 * within the #MemoryRegionSection via the #RamDiscardManager. 701 * 702 * In case any notification fails, no further notifications are triggered 703 * and an error is logged. 704 * 705 * @rdm: the #RamDiscardManager 706 * @rdl: the #RamDiscardListener 707 * @section: the #MemoryRegionSection 708 */ 709 void (*register_listener)(RamDiscardManager *rdm, 710 RamDiscardListener *rdl, 711 MemoryRegionSection *section); 712 713 /** 714 * @unregister_listener: 715 * 716 * Unregister a previously registered #RamDiscardListener via the 717 * #RamDiscardManager after notifying the #RamDiscardListener about all 718 * populated parts becoming unpopulated within the registered 719 * #MemoryRegionSection. 720 * 721 * @rdm: the #RamDiscardManager 722 * @rdl: the #RamDiscardListener 723 */ 724 void (*unregister_listener)(RamDiscardManager *rdm, 725 RamDiscardListener *rdl); 726 }; 727 728 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm, 729 const MemoryRegion *mr); 730 731 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm, 732 const MemoryRegionSection *section); 733 734 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm, 735 MemoryRegionSection *section, 736 ReplayRamPopulate replay_fn, 737 void *opaque); 738 739 void ram_discard_manager_replay_discarded(const RamDiscardManager *rdm, 740 MemoryRegionSection *section, 741 ReplayRamDiscard replay_fn, 742 void *opaque); 743 744 void ram_discard_manager_register_listener(RamDiscardManager *rdm, 745 RamDiscardListener *rdl, 746 MemoryRegionSection *section); 747 748 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm, 749 RamDiscardListener *rdl); 750 751 bool memory_get_xlat_addr(IOMMUTLBEntry *iotlb, void **vaddr, 752 ram_addr_t *ram_addr, bool *read_only, 753 bool *mr_has_discard_manager); 754 755 typedef struct CoalescedMemoryRange CoalescedMemoryRange; 756 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd; 757 758 /** MemoryRegion: 759 * 760 * A struct representing a memory region. 761 */ 762 struct MemoryRegion { 763 Object parent_obj; 764 765 /* private: */ 766 767 /* The following fields should fit in a cache line */ 768 bool romd_mode; 769 bool ram; 770 bool subpage; 771 bool readonly; /* For RAM regions */ 772 bool nonvolatile; 773 bool rom_device; 774 bool flush_coalesced_mmio; 775 uint8_t dirty_log_mask; 776 bool is_iommu; 777 RAMBlock *ram_block; 778 Object *owner; 779 /* owner as TYPE_DEVICE. Used for re-entrancy checks in MR access hotpath */ 780 DeviceState *dev; 781 782 const MemoryRegionOps *ops; 783 void *opaque; 784 MemoryRegion *container; 785 int mapped_via_alias; /* Mapped via an alias, container might be NULL */ 786 Int128 size; 787 hwaddr addr; 788 void (*destructor)(MemoryRegion *mr); 789 uint64_t align; 790 bool terminates; 791 bool ram_device; 792 bool enabled; 793 bool warning_printed; /* For reservations */ 794 uint8_t vga_logging_count; 795 MemoryRegion *alias; 796 hwaddr alias_offset; 797 int32_t priority; 798 QTAILQ_HEAD(, MemoryRegion) subregions; 799 QTAILQ_ENTRY(MemoryRegion) subregions_link; 800 QTAILQ_HEAD(, CoalescedMemoryRange) coalesced; 801 const char *name; 802 unsigned ioeventfd_nb; 803 MemoryRegionIoeventfd *ioeventfds; 804 RamDiscardManager *rdm; /* Only for RAM */ 805 806 /* For devices designed to perform re-entrant IO into their own IO MRs */ 807 bool disable_reentrancy_guard; 808 }; 809 810 struct IOMMUMemoryRegion { 811 MemoryRegion parent_obj; 812 813 QLIST_HEAD(, IOMMUNotifier) iommu_notify; 814 IOMMUNotifierFlag iommu_notify_flags; 815 }; 816 817 #define IOMMU_NOTIFIER_FOREACH(n, mr) \ 818 QLIST_FOREACH((n), &(mr)->iommu_notify, node) 819 820 #define MEMORY_LISTENER_PRIORITY_MIN 0 821 #define MEMORY_LISTENER_PRIORITY_ACCEL 10 822 #define MEMORY_LISTENER_PRIORITY_DEV_BACKEND 10 823 824 /** 825 * struct MemoryListener: callbacks structure for updates to the physical memory map 826 * 827 * Allows a component to adjust to changes in the guest-visible memory map. 828 * Use with memory_listener_register() and memory_listener_unregister(). 829 */ 830 struct MemoryListener { 831 /** 832 * @begin: 833 * 834 * Called at the beginning of an address space update transaction. 835 * Followed by calls to #MemoryListener.region_add(), 836 * #MemoryListener.region_del(), #MemoryListener.region_nop(), 837 * #MemoryListener.log_start() and #MemoryListener.log_stop() in 838 * increasing address order. 839 * 840 * @listener: The #MemoryListener. 841 */ 842 void (*begin)(MemoryListener *listener); 843 844 /** 845 * @commit: 846 * 847 * Called at the end of an address space update transaction, 848 * after the last call to #MemoryListener.region_add(), 849 * #MemoryListener.region_del() or #MemoryListener.region_nop(), 850 * #MemoryListener.log_start() and #MemoryListener.log_stop(). 851 * 852 * @listener: The #MemoryListener. 853 */ 854 void (*commit)(MemoryListener *listener); 855 856 /** 857 * @region_add: 858 * 859 * Called during an address space update transaction, 860 * for a section of the address space that is new in this address space 861 * space since the last transaction. 862 * 863 * @listener: The #MemoryListener. 864 * @section: The new #MemoryRegionSection. 865 */ 866 void (*region_add)(MemoryListener *listener, MemoryRegionSection *section); 867 868 /** 869 * @region_del: 870 * 871 * Called during an address space update transaction, 872 * for a section of the address space that has disappeared in the address 873 * space since the last transaction. 874 * 875 * @listener: The #MemoryListener. 876 * @section: The old #MemoryRegionSection. 877 */ 878 void (*region_del)(MemoryListener *listener, MemoryRegionSection *section); 879 880 /** 881 * @region_nop: 882 * 883 * Called during an address space update transaction, 884 * for a section of the address space that is in the same place in the address 885 * space as in the last transaction. 886 * 887 * @listener: The #MemoryListener. 888 * @section: The #MemoryRegionSection. 889 */ 890 void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section); 891 892 /** 893 * @log_start: 894 * 895 * Called during an address space update transaction, after 896 * one of #MemoryListener.region_add(), #MemoryListener.region_del() or 897 * #MemoryListener.region_nop(), if dirty memory logging clients have 898 * become active since the last transaction. 899 * 900 * @listener: The #MemoryListener. 901 * @section: The #MemoryRegionSection. 902 * @old: A bitmap of dirty memory logging clients that were active in 903 * the previous transaction. 904 * @new: A bitmap of dirty memory logging clients that are active in 905 * the current transaction. 906 */ 907 void (*log_start)(MemoryListener *listener, MemoryRegionSection *section, 908 int old, int new); 909 910 /** 911 * @log_stop: 912 * 913 * Called during an address space update transaction, after 914 * one of #MemoryListener.region_add(), #MemoryListener.region_del() or 915 * #MemoryListener.region_nop() and possibly after 916 * #MemoryListener.log_start(), if dirty memory logging clients have 917 * become inactive since the last transaction. 918 * 919 * @listener: The #MemoryListener. 920 * @section: The #MemoryRegionSection. 921 * @old: A bitmap of dirty memory logging clients that were active in 922 * the previous transaction. 923 * @new: A bitmap of dirty memory logging clients that are active in 924 * the current transaction. 925 */ 926 void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section, 927 int old, int new); 928 929 /** 930 * @log_sync: 931 * 932 * Called by memory_region_snapshot_and_clear_dirty() and 933 * memory_global_dirty_log_sync(), before accessing QEMU's "official" 934 * copy of the dirty memory bitmap for a #MemoryRegionSection. 935 * 936 * @listener: The #MemoryListener. 937 * @section: The #MemoryRegionSection. 938 */ 939 void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section); 940 941 /** 942 * @log_sync_global: 943 * 944 * This is the global version of @log_sync when the listener does 945 * not have a way to synchronize the log with finer granularity. 946 * When the listener registers with @log_sync_global defined, then 947 * its @log_sync must be NULL. Vice versa. 948 * 949 * @listener: The #MemoryListener. 950 * @last_stage: The last stage to synchronize the log during migration. 951 * The caller should guarantee that the synchronization with true for 952 * @last_stage is triggered for once after all VCPUs have been stopped. 953 */ 954 void (*log_sync_global)(MemoryListener *listener, bool last_stage); 955 956 /** 957 * @log_clear: 958 * 959 * Called before reading the dirty memory bitmap for a 960 * #MemoryRegionSection. 961 * 962 * @listener: The #MemoryListener. 963 * @section: The #MemoryRegionSection. 964 */ 965 void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section); 966 967 /** 968 * @log_global_start: 969 * 970 * Called by memory_global_dirty_log_start(), which 971 * enables the %DIRTY_LOG_MIGRATION client on all memory regions in 972 * the address space. #MemoryListener.log_global_start() is also 973 * called when a #MemoryListener is added, if global dirty logging is 974 * active at that time. 975 * 976 * @listener: The #MemoryListener. 977 */ 978 void (*log_global_start)(MemoryListener *listener); 979 980 /** 981 * @log_global_stop: 982 * 983 * Called by memory_global_dirty_log_stop(), which 984 * disables the %DIRTY_LOG_MIGRATION client on all memory regions in 985 * the address space. 986 * 987 * @listener: The #MemoryListener. 988 */ 989 void (*log_global_stop)(MemoryListener *listener); 990 991 /** 992 * @log_global_after_sync: 993 * 994 * Called after reading the dirty memory bitmap 995 * for any #MemoryRegionSection. 996 * 997 * @listener: The #MemoryListener. 998 */ 999 void (*log_global_after_sync)(MemoryListener *listener); 1000 1001 /** 1002 * @eventfd_add: 1003 * 1004 * Called during an address space update transaction, 1005 * for a section of the address space that has had a new ioeventfd 1006 * registration since the last transaction. 1007 * 1008 * @listener: The #MemoryListener. 1009 * @section: The new #MemoryRegionSection. 1010 * @match_data: The @match_data parameter for the new ioeventfd. 1011 * @data: The @data parameter for the new ioeventfd. 1012 * @e: The #EventNotifier parameter for the new ioeventfd. 1013 */ 1014 void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section, 1015 bool match_data, uint64_t data, EventNotifier *e); 1016 1017 /** 1018 * @eventfd_del: 1019 * 1020 * Called during an address space update transaction, 1021 * for a section of the address space that has dropped an ioeventfd 1022 * registration since the last transaction. 1023 * 1024 * @listener: The #MemoryListener. 1025 * @section: The new #MemoryRegionSection. 1026 * @match_data: The @match_data parameter for the dropped ioeventfd. 1027 * @data: The @data parameter for the dropped ioeventfd. 1028 * @e: The #EventNotifier parameter for the dropped ioeventfd. 1029 */ 1030 void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section, 1031 bool match_data, uint64_t data, EventNotifier *e); 1032 1033 /** 1034 * @coalesced_io_add: 1035 * 1036 * Called during an address space update transaction, 1037 * for a section of the address space that has had a new coalesced 1038 * MMIO range registration since the last transaction. 1039 * 1040 * @listener: The #MemoryListener. 1041 * @section: The new #MemoryRegionSection. 1042 * @addr: The starting address for the coalesced MMIO range. 1043 * @len: The length of the coalesced MMIO range. 1044 */ 1045 void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section, 1046 hwaddr addr, hwaddr len); 1047 1048 /** 1049 * @coalesced_io_del: 1050 * 1051 * Called during an address space update transaction, 1052 * for a section of the address space that has dropped a coalesced 1053 * MMIO range since the last transaction. 1054 * 1055 * @listener: The #MemoryListener. 1056 * @section: The new #MemoryRegionSection. 1057 * @addr: The starting address for the coalesced MMIO range. 1058 * @len: The length of the coalesced MMIO range. 1059 */ 1060 void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section, 1061 hwaddr addr, hwaddr len); 1062 /** 1063 * @priority: 1064 * 1065 * Govern the order in which memory listeners are invoked. Lower priorities 1066 * are invoked earlier for "add" or "start" callbacks, and later for "delete" 1067 * or "stop" callbacks. 1068 */ 1069 unsigned priority; 1070 1071 /** 1072 * @name: 1073 * 1074 * Name of the listener. It can be used in contexts where we'd like to 1075 * identify one memory listener with the rest. 1076 */ 1077 const char *name; 1078 1079 /* private: */ 1080 AddressSpace *address_space; 1081 QTAILQ_ENTRY(MemoryListener) link; 1082 QTAILQ_ENTRY(MemoryListener) link_as; 1083 }; 1084 1085 /** 1086 * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects 1087 */ 1088 struct AddressSpace { 1089 /* private: */ 1090 struct rcu_head rcu; 1091 char *name; 1092 MemoryRegion *root; 1093 1094 /* Accessed via RCU. */ 1095 struct FlatView *current_map; 1096 1097 int ioeventfd_nb; 1098 struct MemoryRegionIoeventfd *ioeventfds; 1099 QTAILQ_HEAD(, MemoryListener) listeners; 1100 QTAILQ_ENTRY(AddressSpace) address_spaces_link; 1101 }; 1102 1103 typedef struct AddressSpaceDispatch AddressSpaceDispatch; 1104 typedef struct FlatRange FlatRange; 1105 1106 /* Flattened global view of current active memory hierarchy. Kept in sorted 1107 * order. 1108 */ 1109 struct FlatView { 1110 struct rcu_head rcu; 1111 unsigned ref; 1112 FlatRange *ranges; 1113 unsigned nr; 1114 unsigned nr_allocated; 1115 struct AddressSpaceDispatch *dispatch; 1116 MemoryRegion *root; 1117 }; 1118 1119 static inline FlatView *address_space_to_flatview(AddressSpace *as) 1120 { 1121 return qatomic_rcu_read(&as->current_map); 1122 } 1123 1124 /** 1125 * typedef flatview_cb: callback for flatview_for_each_range() 1126 * 1127 * @start: start address of the range within the FlatView 1128 * @len: length of the range in bytes 1129 * @mr: MemoryRegion covering this range 1130 * @offset_in_region: offset of the first byte of the range within @mr 1131 * @opaque: data pointer passed to flatview_for_each_range() 1132 * 1133 * Returns: true to stop the iteration, false to keep going. 1134 */ 1135 typedef bool (*flatview_cb)(Int128 start, 1136 Int128 len, 1137 const MemoryRegion *mr, 1138 hwaddr offset_in_region, 1139 void *opaque); 1140 1141 /** 1142 * flatview_for_each_range: Iterate through a FlatView 1143 * @fv: the FlatView to iterate through 1144 * @cb: function to call for each range 1145 * @opaque: opaque data pointer to pass to @cb 1146 * 1147 * A FlatView is made up of a list of non-overlapping ranges, each of 1148 * which is a slice of a MemoryRegion. This function iterates through 1149 * each range in @fv, calling @cb. The callback function can terminate 1150 * iteration early by returning 'true'. 1151 */ 1152 void flatview_for_each_range(FlatView *fv, flatview_cb cb, void *opaque); 1153 1154 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a, 1155 MemoryRegionSection *b) 1156 { 1157 return a->mr == b->mr && 1158 a->fv == b->fv && 1159 a->offset_within_region == b->offset_within_region && 1160 a->offset_within_address_space == b->offset_within_address_space && 1161 int128_eq(a->size, b->size) && 1162 a->readonly == b->readonly && 1163 a->nonvolatile == b->nonvolatile; 1164 } 1165 1166 /** 1167 * memory_region_section_new_copy: Copy a memory region section 1168 * 1169 * Allocate memory for a new copy, copy the memory region section, and 1170 * properly take a reference on all relevant members. 1171 * 1172 * @s: the #MemoryRegionSection to copy 1173 */ 1174 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s); 1175 1176 /** 1177 * memory_region_section_new_copy: Free a copied memory region section 1178 * 1179 * Free a copy of a memory section created via memory_region_section_new_copy(). 1180 * properly dropping references on all relevant members. 1181 * 1182 * @s: the #MemoryRegionSection to copy 1183 */ 1184 void memory_region_section_free_copy(MemoryRegionSection *s); 1185 1186 /** 1187 * memory_region_init: Initialize a memory region 1188 * 1189 * The region typically acts as a container for other memory regions. Use 1190 * memory_region_add_subregion() to add subregions. 1191 * 1192 * @mr: the #MemoryRegion to be initialized 1193 * @owner: the object that tracks the region's reference count 1194 * @name: used for debugging; not visible to the user or ABI 1195 * @size: size of the region; any subregions beyond this size will be clipped 1196 */ 1197 void memory_region_init(MemoryRegion *mr, 1198 Object *owner, 1199 const char *name, 1200 uint64_t size); 1201 1202 /** 1203 * memory_region_ref: Add 1 to a memory region's reference count 1204 * 1205 * Whenever memory regions are accessed outside the BQL, they need to be 1206 * preserved against hot-unplug. MemoryRegions actually do not have their 1207 * own reference count; they piggyback on a QOM object, their "owner". 1208 * This function adds a reference to the owner. 1209 * 1210 * All MemoryRegions must have an owner if they can disappear, even if the 1211 * device they belong to operates exclusively under the BQL. This is because 1212 * the region could be returned at any time by memory_region_find, and this 1213 * is usually under guest control. 1214 * 1215 * @mr: the #MemoryRegion 1216 */ 1217 void memory_region_ref(MemoryRegion *mr); 1218 1219 /** 1220 * memory_region_unref: Remove 1 to a memory region's reference count 1221 * 1222 * Whenever memory regions are accessed outside the BQL, they need to be 1223 * preserved against hot-unplug. MemoryRegions actually do not have their 1224 * own reference count; they piggyback on a QOM object, their "owner". 1225 * This function removes a reference to the owner and possibly destroys it. 1226 * 1227 * @mr: the #MemoryRegion 1228 */ 1229 void memory_region_unref(MemoryRegion *mr); 1230 1231 /** 1232 * memory_region_init_io: Initialize an I/O memory region. 1233 * 1234 * Accesses into the region will cause the callbacks in @ops to be called. 1235 * if @size is nonzero, subregions will be clipped to @size. 1236 * 1237 * @mr: the #MemoryRegion to be initialized. 1238 * @owner: the object that tracks the region's reference count 1239 * @ops: a structure containing read and write callbacks to be used when 1240 * I/O is performed on the region. 1241 * @opaque: passed to the read and write callbacks of the @ops structure. 1242 * @name: used for debugging; not visible to the user or ABI 1243 * @size: size of the region. 1244 */ 1245 void memory_region_init_io(MemoryRegion *mr, 1246 Object *owner, 1247 const MemoryRegionOps *ops, 1248 void *opaque, 1249 const char *name, 1250 uint64_t size); 1251 1252 /** 1253 * memory_region_init_ram_nomigrate: Initialize RAM memory region. Accesses 1254 * into the region will modify memory 1255 * directly. 1256 * 1257 * @mr: the #MemoryRegion to be initialized. 1258 * @owner: the object that tracks the region's reference count 1259 * @name: Region name, becomes part of RAMBlock name used in migration stream 1260 * must be unique within any device 1261 * @size: size of the region. 1262 * @errp: pointer to Error*, to store an error if it happens. 1263 * 1264 * Note that this function does not do anything to cause the data in the 1265 * RAM memory region to be migrated; that is the responsibility of the caller. 1266 */ 1267 void memory_region_init_ram_nomigrate(MemoryRegion *mr, 1268 Object *owner, 1269 const char *name, 1270 uint64_t size, 1271 Error **errp); 1272 1273 /** 1274 * memory_region_init_ram_flags_nomigrate: Initialize RAM memory region. 1275 * Accesses into the region will 1276 * modify memory directly. 1277 * 1278 * @mr: the #MemoryRegion to be initialized. 1279 * @owner: the object that tracks the region's reference count 1280 * @name: Region name, becomes part of RAMBlock name used in migration stream 1281 * must be unique within any device 1282 * @size: size of the region. 1283 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_NORESERVE. 1284 * @errp: pointer to Error*, to store an error if it happens. 1285 * 1286 * Note that this function does not do anything to cause the data in the 1287 * RAM memory region to be migrated; that is the responsibility of the caller. 1288 */ 1289 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr, 1290 Object *owner, 1291 const char *name, 1292 uint64_t size, 1293 uint32_t ram_flags, 1294 Error **errp); 1295 1296 /** 1297 * memory_region_init_resizeable_ram: Initialize memory region with resizable 1298 * RAM. Accesses into the region will 1299 * modify memory directly. Only an initial 1300 * portion of this RAM is actually used. 1301 * Changing the size while migrating 1302 * can result in the migration being 1303 * canceled. 1304 * 1305 * @mr: the #MemoryRegion to be initialized. 1306 * @owner: the object that tracks the region's reference count 1307 * @name: Region name, becomes part of RAMBlock name used in migration stream 1308 * must be unique within any device 1309 * @size: used size of the region. 1310 * @max_size: max size of the region. 1311 * @resized: callback to notify owner about used size change. 1312 * @errp: pointer to Error*, to store an error if it happens. 1313 * 1314 * Note that this function does not do anything to cause the data in the 1315 * RAM memory region to be migrated; that is the responsibility of the caller. 1316 */ 1317 void memory_region_init_resizeable_ram(MemoryRegion *mr, 1318 Object *owner, 1319 const char *name, 1320 uint64_t size, 1321 uint64_t max_size, 1322 void (*resized)(const char*, 1323 uint64_t length, 1324 void *host), 1325 Error **errp); 1326 #ifdef CONFIG_POSIX 1327 1328 /** 1329 * memory_region_init_ram_from_file: Initialize RAM memory region with a 1330 * mmap-ed backend. 1331 * 1332 * @mr: the #MemoryRegion to be initialized. 1333 * @owner: the object that tracks the region's reference count 1334 * @name: Region name, becomes part of RAMBlock name used in migration stream 1335 * must be unique within any device 1336 * @size: size of the region. 1337 * @align: alignment of the region base address; if 0, the default alignment 1338 * (getpagesize()) will be used. 1339 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM, 1340 * RAM_NORESERVE, RAM_PROTECTED, RAM_NAMED_FILE, RAM_READONLY, 1341 * RAM_READONLY_FD 1342 * @path: the path in which to allocate the RAM. 1343 * @offset: offset within the file referenced by path 1344 * @errp: pointer to Error*, to store an error if it happens. 1345 * 1346 * Note that this function does not do anything to cause the data in the 1347 * RAM memory region to be migrated; that is the responsibility of the caller. 1348 */ 1349 void memory_region_init_ram_from_file(MemoryRegion *mr, 1350 Object *owner, 1351 const char *name, 1352 uint64_t size, 1353 uint64_t align, 1354 uint32_t ram_flags, 1355 const char *path, 1356 ram_addr_t offset, 1357 Error **errp); 1358 1359 /** 1360 * memory_region_init_ram_from_fd: Initialize RAM memory region with a 1361 * mmap-ed backend. 1362 * 1363 * @mr: the #MemoryRegion to be initialized. 1364 * @owner: the object that tracks the region's reference count 1365 * @name: the name of the region. 1366 * @size: size of the region. 1367 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM, 1368 * RAM_NORESERVE, RAM_PROTECTED, RAM_NAMED_FILE, RAM_READONLY, 1369 * RAM_READONLY_FD 1370 * @fd: the fd to mmap. 1371 * @offset: offset within the file referenced by fd 1372 * @errp: pointer to Error*, to store an error if it happens. 1373 * 1374 * Note that this function does not do anything to cause the data in the 1375 * RAM memory region to be migrated; that is the responsibility of the caller. 1376 */ 1377 void memory_region_init_ram_from_fd(MemoryRegion *mr, 1378 Object *owner, 1379 const char *name, 1380 uint64_t size, 1381 uint32_t ram_flags, 1382 int fd, 1383 ram_addr_t offset, 1384 Error **errp); 1385 #endif 1386 1387 /** 1388 * memory_region_init_ram_ptr: Initialize RAM memory region from a 1389 * user-provided pointer. Accesses into the 1390 * region will modify memory directly. 1391 * 1392 * @mr: the #MemoryRegion to be initialized. 1393 * @owner: the object that tracks the region's reference count 1394 * @name: Region name, becomes part of RAMBlock name used in migration stream 1395 * must be unique within any device 1396 * @size: size of the region. 1397 * @ptr: memory to be mapped; must contain at least @size bytes. 1398 * 1399 * Note that this function does not do anything to cause the data in the 1400 * RAM memory region to be migrated; that is the responsibility of the caller. 1401 */ 1402 void memory_region_init_ram_ptr(MemoryRegion *mr, 1403 Object *owner, 1404 const char *name, 1405 uint64_t size, 1406 void *ptr); 1407 1408 /** 1409 * memory_region_init_ram_device_ptr: Initialize RAM device memory region from 1410 * a user-provided pointer. 1411 * 1412 * A RAM device represents a mapping to a physical device, such as to a PCI 1413 * MMIO BAR of an vfio-pci assigned device. The memory region may be mapped 1414 * into the VM address space and access to the region will modify memory 1415 * directly. However, the memory region should not be included in a memory 1416 * dump (device may not be enabled/mapped at the time of the dump), and 1417 * operations incompatible with manipulating MMIO should be avoided. Replaces 1418 * skip_dump flag. 1419 * 1420 * @mr: the #MemoryRegion to be initialized. 1421 * @owner: the object that tracks the region's reference count 1422 * @name: the name of the region. 1423 * @size: size of the region. 1424 * @ptr: memory to be mapped; must contain at least @size bytes. 1425 * 1426 * Note that this function does not do anything to cause the data in the 1427 * RAM memory region to be migrated; that is the responsibility of the caller. 1428 * (For RAM device memory regions, migrating the contents rarely makes sense.) 1429 */ 1430 void memory_region_init_ram_device_ptr(MemoryRegion *mr, 1431 Object *owner, 1432 const char *name, 1433 uint64_t size, 1434 void *ptr); 1435 1436 /** 1437 * memory_region_init_alias: Initialize a memory region that aliases all or a 1438 * part of another memory region. 1439 * 1440 * @mr: the #MemoryRegion to be initialized. 1441 * @owner: the object that tracks the region's reference count 1442 * @name: used for debugging; not visible to the user or ABI 1443 * @orig: the region to be referenced; @mr will be equivalent to 1444 * @orig between @offset and @offset + @size - 1. 1445 * @offset: start of the section in @orig to be referenced. 1446 * @size: size of the region. 1447 */ 1448 void memory_region_init_alias(MemoryRegion *mr, 1449 Object *owner, 1450 const char *name, 1451 MemoryRegion *orig, 1452 hwaddr offset, 1453 uint64_t size); 1454 1455 /** 1456 * memory_region_init_rom_nomigrate: Initialize a ROM memory region. 1457 * 1458 * This has the same effect as calling memory_region_init_ram_nomigrate() 1459 * and then marking the resulting region read-only with 1460 * memory_region_set_readonly(). 1461 * 1462 * Note that this function does not do anything to cause the data in the 1463 * RAM side of the memory region to be migrated; that is the responsibility 1464 * of the caller. 1465 * 1466 * @mr: the #MemoryRegion to be initialized. 1467 * @owner: the object that tracks the region's reference count 1468 * @name: Region name, becomes part of RAMBlock name used in migration stream 1469 * must be unique within any device 1470 * @size: size of the region. 1471 * @errp: pointer to Error*, to store an error if it happens. 1472 */ 1473 void memory_region_init_rom_nomigrate(MemoryRegion *mr, 1474 Object *owner, 1475 const char *name, 1476 uint64_t size, 1477 Error **errp); 1478 1479 /** 1480 * memory_region_init_rom_device_nomigrate: Initialize a ROM memory region. 1481 * Writes are handled via callbacks. 1482 * 1483 * Note that this function does not do anything to cause the data in the 1484 * RAM side of the memory region to be migrated; that is the responsibility 1485 * of the caller. 1486 * 1487 * @mr: the #MemoryRegion to be initialized. 1488 * @owner: the object that tracks the region's reference count 1489 * @ops: callbacks for write access handling (must not be NULL). 1490 * @opaque: passed to the read and write callbacks of the @ops structure. 1491 * @name: Region name, becomes part of RAMBlock name used in migration stream 1492 * must be unique within any device 1493 * @size: size of the region. 1494 * @errp: pointer to Error*, to store an error if it happens. 1495 */ 1496 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr, 1497 Object *owner, 1498 const MemoryRegionOps *ops, 1499 void *opaque, 1500 const char *name, 1501 uint64_t size, 1502 Error **errp); 1503 1504 /** 1505 * memory_region_init_iommu: Initialize a memory region of a custom type 1506 * that translates addresses 1507 * 1508 * An IOMMU region translates addresses and forwards accesses to a target 1509 * memory region. 1510 * 1511 * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION. 1512 * @_iommu_mr should be a pointer to enough memory for an instance of 1513 * that subclass, @instance_size is the size of that subclass, and 1514 * @mrtypename is its name. This function will initialize @_iommu_mr as an 1515 * instance of the subclass, and its methods will then be called to handle 1516 * accesses to the memory region. See the documentation of 1517 * #IOMMUMemoryRegionClass for further details. 1518 * 1519 * @_iommu_mr: the #IOMMUMemoryRegion to be initialized 1520 * @instance_size: the IOMMUMemoryRegion subclass instance size 1521 * @mrtypename: the type name of the #IOMMUMemoryRegion 1522 * @owner: the object that tracks the region's reference count 1523 * @name: used for debugging; not visible to the user or ABI 1524 * @size: size of the region. 1525 */ 1526 void memory_region_init_iommu(void *_iommu_mr, 1527 size_t instance_size, 1528 const char *mrtypename, 1529 Object *owner, 1530 const char *name, 1531 uint64_t size); 1532 1533 /** 1534 * memory_region_init_ram - Initialize RAM memory region. Accesses into the 1535 * region will modify memory directly. 1536 * 1537 * @mr: the #MemoryRegion to be initialized 1538 * @owner: the object that tracks the region's reference count (must be 1539 * TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL) 1540 * @name: name of the memory region 1541 * @size: size of the region in bytes 1542 * @errp: pointer to Error*, to store an error if it happens. 1543 * 1544 * This function allocates RAM for a board model or device, and 1545 * arranges for it to be migrated (by calling vmstate_register_ram() 1546 * if @owner is a DeviceState, or vmstate_register_ram_global() if 1547 * @owner is NULL). 1548 * 1549 * TODO: Currently we restrict @owner to being either NULL (for 1550 * global RAM regions with no owner) or devices, so that we can 1551 * give the RAM block a unique name for migration purposes. 1552 * We should lift this restriction and allow arbitrary Objects. 1553 * If you pass a non-NULL non-device @owner then we will assert. 1554 */ 1555 void memory_region_init_ram(MemoryRegion *mr, 1556 Object *owner, 1557 const char *name, 1558 uint64_t size, 1559 Error **errp); 1560 1561 /** 1562 * memory_region_init_rom: Initialize a ROM memory region. 1563 * 1564 * This has the same effect as calling memory_region_init_ram() 1565 * and then marking the resulting region read-only with 1566 * memory_region_set_readonly(). This includes arranging for the 1567 * contents to be migrated. 1568 * 1569 * TODO: Currently we restrict @owner to being either NULL (for 1570 * global RAM regions with no owner) or devices, so that we can 1571 * give the RAM block a unique name for migration purposes. 1572 * We should lift this restriction and allow arbitrary Objects. 1573 * If you pass a non-NULL non-device @owner then we will assert. 1574 * 1575 * @mr: the #MemoryRegion to be initialized. 1576 * @owner: the object that tracks the region's reference count 1577 * @name: Region name, becomes part of RAMBlock name used in migration stream 1578 * must be unique within any device 1579 * @size: size of the region. 1580 * @errp: pointer to Error*, to store an error if it happens. 1581 */ 1582 void memory_region_init_rom(MemoryRegion *mr, 1583 Object *owner, 1584 const char *name, 1585 uint64_t size, 1586 Error **errp); 1587 1588 /** 1589 * memory_region_init_rom_device: Initialize a ROM memory region. 1590 * Writes are handled via callbacks. 1591 * 1592 * This function initializes a memory region backed by RAM for reads 1593 * and callbacks for writes, and arranges for the RAM backing to 1594 * be migrated (by calling vmstate_register_ram() 1595 * if @owner is a DeviceState, or vmstate_register_ram_global() if 1596 * @owner is NULL). 1597 * 1598 * TODO: Currently we restrict @owner to being either NULL (for 1599 * global RAM regions with no owner) or devices, so that we can 1600 * give the RAM block a unique name for migration purposes. 1601 * We should lift this restriction and allow arbitrary Objects. 1602 * If you pass a non-NULL non-device @owner then we will assert. 1603 * 1604 * @mr: the #MemoryRegion to be initialized. 1605 * @owner: the object that tracks the region's reference count 1606 * @ops: callbacks for write access handling (must not be NULL). 1607 * @opaque: passed to the read and write callbacks of the @ops structure. 1608 * @name: Region name, becomes part of RAMBlock name used in migration stream 1609 * must be unique within any device 1610 * @size: size of the region. 1611 * @errp: pointer to Error*, to store an error if it happens. 1612 */ 1613 void memory_region_init_rom_device(MemoryRegion *mr, 1614 Object *owner, 1615 const MemoryRegionOps *ops, 1616 void *opaque, 1617 const char *name, 1618 uint64_t size, 1619 Error **errp); 1620 1621 1622 /** 1623 * memory_region_owner: get a memory region's owner. 1624 * 1625 * @mr: the memory region being queried. 1626 */ 1627 Object *memory_region_owner(MemoryRegion *mr); 1628 1629 /** 1630 * memory_region_size: get a memory region's size. 1631 * 1632 * @mr: the memory region being queried. 1633 */ 1634 uint64_t memory_region_size(MemoryRegion *mr); 1635 1636 /** 1637 * memory_region_is_ram: check whether a memory region is random access 1638 * 1639 * Returns %true if a memory region is random access. 1640 * 1641 * @mr: the memory region being queried 1642 */ 1643 static inline bool memory_region_is_ram(MemoryRegion *mr) 1644 { 1645 return mr->ram; 1646 } 1647 1648 /** 1649 * memory_region_is_ram_device: check whether a memory region is a ram device 1650 * 1651 * Returns %true if a memory region is a device backed ram region 1652 * 1653 * @mr: the memory region being queried 1654 */ 1655 bool memory_region_is_ram_device(MemoryRegion *mr); 1656 1657 /** 1658 * memory_region_is_romd: check whether a memory region is in ROMD mode 1659 * 1660 * Returns %true if a memory region is a ROM device and currently set to allow 1661 * direct reads. 1662 * 1663 * @mr: the memory region being queried 1664 */ 1665 static inline bool memory_region_is_romd(MemoryRegion *mr) 1666 { 1667 return mr->rom_device && mr->romd_mode; 1668 } 1669 1670 /** 1671 * memory_region_is_protected: check whether a memory region is protected 1672 * 1673 * Returns %true if a memory region is protected RAM and cannot be accessed 1674 * via standard mechanisms, e.g. DMA. 1675 * 1676 * @mr: the memory region being queried 1677 */ 1678 bool memory_region_is_protected(MemoryRegion *mr); 1679 1680 /** 1681 * memory_region_get_iommu: check whether a memory region is an iommu 1682 * 1683 * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu, 1684 * otherwise NULL. 1685 * 1686 * @mr: the memory region being queried 1687 */ 1688 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr) 1689 { 1690 if (mr->alias) { 1691 return memory_region_get_iommu(mr->alias); 1692 } 1693 if (mr->is_iommu) { 1694 return (IOMMUMemoryRegion *) mr; 1695 } 1696 return NULL; 1697 } 1698 1699 /** 1700 * memory_region_get_iommu_class_nocheck: returns iommu memory region class 1701 * if an iommu or NULL if not 1702 * 1703 * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu, 1704 * otherwise NULL. This is fast path avoiding QOM checking, use with caution. 1705 * 1706 * @iommu_mr: the memory region being queried 1707 */ 1708 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck( 1709 IOMMUMemoryRegion *iommu_mr) 1710 { 1711 return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class); 1712 } 1713 1714 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL) 1715 1716 /** 1717 * memory_region_iommu_get_min_page_size: get minimum supported page size 1718 * for an iommu 1719 * 1720 * Returns minimum supported page size for an iommu. 1721 * 1722 * @iommu_mr: the memory region being queried 1723 */ 1724 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr); 1725 1726 /** 1727 * memory_region_notify_iommu: notify a change in an IOMMU translation entry. 1728 * 1729 * Note: for any IOMMU implementation, an in-place mapping change 1730 * should be notified with an UNMAP followed by a MAP. 1731 * 1732 * @iommu_mr: the memory region that was changed 1733 * @iommu_idx: the IOMMU index for the translation table which has changed 1734 * @event: TLB event with the new entry in the IOMMU translation table. 1735 * The entry replaces all old entries for the same virtual I/O address 1736 * range. 1737 */ 1738 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr, 1739 int iommu_idx, 1740 IOMMUTLBEvent event); 1741 1742 /** 1743 * memory_region_notify_iommu_one: notify a change in an IOMMU translation 1744 * entry to a single notifier 1745 * 1746 * This works just like memory_region_notify_iommu(), but it only 1747 * notifies a specific notifier, not all of them. 1748 * 1749 * @notifier: the notifier to be notified 1750 * @event: TLB event with the new entry in the IOMMU translation table. 1751 * The entry replaces all old entries for the same virtual I/O address 1752 * range. 1753 */ 1754 void memory_region_notify_iommu_one(IOMMUNotifier *notifier, 1755 IOMMUTLBEvent *event); 1756 1757 /** 1758 * memory_region_unmap_iommu_notifier_range: notify a unmap for an IOMMU 1759 * translation that covers the 1760 * range of a notifier 1761 * 1762 * @notifier: the notifier to be notified 1763 */ 1764 void memory_region_unmap_iommu_notifier_range(IOMMUNotifier *notifier); 1765 1766 1767 /** 1768 * memory_region_register_iommu_notifier: register a notifier for changes to 1769 * IOMMU translation entries. 1770 * 1771 * Returns 0 on success, or a negative errno otherwise. In particular, 1772 * -EINVAL indicates that at least one of the attributes of the notifier 1773 * is not supported (flag/range) by the IOMMU memory region. In case of error 1774 * the error object must be created. 1775 * 1776 * @mr: the memory region to observe 1777 * @n: the IOMMUNotifier to be added; the notify callback receives a 1778 * pointer to an #IOMMUTLBEntry as the opaque value; the pointer 1779 * ceases to be valid on exit from the notifier. 1780 * @errp: pointer to Error*, to store an error if it happens. 1781 */ 1782 int memory_region_register_iommu_notifier(MemoryRegion *mr, 1783 IOMMUNotifier *n, Error **errp); 1784 1785 /** 1786 * memory_region_iommu_replay: replay existing IOMMU translations to 1787 * a notifier with the minimum page granularity returned by 1788 * mr->iommu_ops->get_page_size(). 1789 * 1790 * Note: this is not related to record-and-replay functionality. 1791 * 1792 * @iommu_mr: the memory region to observe 1793 * @n: the notifier to which to replay iommu mappings 1794 */ 1795 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n); 1796 1797 /** 1798 * memory_region_unregister_iommu_notifier: unregister a notifier for 1799 * changes to IOMMU translation entries. 1800 * 1801 * @mr: the memory region which was observed and for which notity_stopped() 1802 * needs to be called 1803 * @n: the notifier to be removed. 1804 */ 1805 void memory_region_unregister_iommu_notifier(MemoryRegion *mr, 1806 IOMMUNotifier *n); 1807 1808 /** 1809 * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is 1810 * defined on the IOMMU. 1811 * 1812 * Returns 0 on success, or a negative errno otherwise. In particular, 1813 * -EINVAL indicates that the IOMMU does not support the requested 1814 * attribute. 1815 * 1816 * @iommu_mr: the memory region 1817 * @attr: the requested attribute 1818 * @data: a pointer to the requested attribute data 1819 */ 1820 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr, 1821 enum IOMMUMemoryRegionAttr attr, 1822 void *data); 1823 1824 /** 1825 * memory_region_iommu_attrs_to_index: return the IOMMU index to 1826 * use for translations with the given memory transaction attributes. 1827 * 1828 * @iommu_mr: the memory region 1829 * @attrs: the memory transaction attributes 1830 */ 1831 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr, 1832 MemTxAttrs attrs); 1833 1834 /** 1835 * memory_region_iommu_num_indexes: return the total number of IOMMU 1836 * indexes that this IOMMU supports. 1837 * 1838 * @iommu_mr: the memory region 1839 */ 1840 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr); 1841 1842 /** 1843 * memory_region_iommu_set_page_size_mask: set the supported page 1844 * sizes for a given IOMMU memory region 1845 * 1846 * @iommu_mr: IOMMU memory region 1847 * @page_size_mask: supported page size mask 1848 * @errp: pointer to Error*, to store an error if it happens. 1849 */ 1850 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr, 1851 uint64_t page_size_mask, 1852 Error **errp); 1853 1854 /** 1855 * memory_region_name: get a memory region's name 1856 * 1857 * Returns the string that was used to initialize the memory region. 1858 * 1859 * @mr: the memory region being queried 1860 */ 1861 const char *memory_region_name(const MemoryRegion *mr); 1862 1863 /** 1864 * memory_region_is_logging: return whether a memory region is logging writes 1865 * 1866 * Returns %true if the memory region is logging writes for the given client 1867 * 1868 * @mr: the memory region being queried 1869 * @client: the client being queried 1870 */ 1871 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client); 1872 1873 /** 1874 * memory_region_get_dirty_log_mask: return the clients for which a 1875 * memory region is logging writes. 1876 * 1877 * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants 1878 * are the bit indices. 1879 * 1880 * @mr: the memory region being queried 1881 */ 1882 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr); 1883 1884 /** 1885 * memory_region_is_rom: check whether a memory region is ROM 1886 * 1887 * Returns %true if a memory region is read-only memory. 1888 * 1889 * @mr: the memory region being queried 1890 */ 1891 static inline bool memory_region_is_rom(MemoryRegion *mr) 1892 { 1893 return mr->ram && mr->readonly; 1894 } 1895 1896 /** 1897 * memory_region_is_nonvolatile: check whether a memory region is non-volatile 1898 * 1899 * Returns %true is a memory region is non-volatile memory. 1900 * 1901 * @mr: the memory region being queried 1902 */ 1903 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr) 1904 { 1905 return mr->nonvolatile; 1906 } 1907 1908 /** 1909 * memory_region_get_fd: Get a file descriptor backing a RAM memory region. 1910 * 1911 * Returns a file descriptor backing a file-based RAM memory region, 1912 * or -1 if the region is not a file-based RAM memory region. 1913 * 1914 * @mr: the RAM or alias memory region being queried. 1915 */ 1916 int memory_region_get_fd(MemoryRegion *mr); 1917 1918 /** 1919 * memory_region_from_host: Convert a pointer into a RAM memory region 1920 * and an offset within it. 1921 * 1922 * Given a host pointer inside a RAM memory region (created with 1923 * memory_region_init_ram() or memory_region_init_ram_ptr()), return 1924 * the MemoryRegion and the offset within it. 1925 * 1926 * Use with care; by the time this function returns, the returned pointer is 1927 * not protected by RCU anymore. If the caller is not within an RCU critical 1928 * section and does not hold the iothread lock, it must have other means of 1929 * protecting the pointer, such as a reference to the region that includes 1930 * the incoming ram_addr_t. 1931 * 1932 * @ptr: the host pointer to be converted 1933 * @offset: the offset within memory region 1934 */ 1935 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset); 1936 1937 /** 1938 * memory_region_get_ram_ptr: Get a pointer into a RAM memory region. 1939 * 1940 * Returns a host pointer to a RAM memory region (created with 1941 * memory_region_init_ram() or memory_region_init_ram_ptr()). 1942 * 1943 * Use with care; by the time this function returns, the returned pointer is 1944 * not protected by RCU anymore. If the caller is not within an RCU critical 1945 * section and does not hold the iothread lock, it must have other means of 1946 * protecting the pointer, such as a reference to the region that includes 1947 * the incoming ram_addr_t. 1948 * 1949 * @mr: the memory region being queried. 1950 */ 1951 void *memory_region_get_ram_ptr(MemoryRegion *mr); 1952 1953 /* memory_region_ram_resize: Resize a RAM region. 1954 * 1955 * Resizing RAM while migrating can result in the migration being canceled. 1956 * Care has to be taken if the guest might have already detected the memory. 1957 * 1958 * @mr: a memory region created with @memory_region_init_resizeable_ram. 1959 * @newsize: the new size the region 1960 * @errp: pointer to Error*, to store an error if it happens. 1961 */ 1962 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, 1963 Error **errp); 1964 1965 /** 1966 * memory_region_msync: Synchronize selected address range of 1967 * a memory mapped region 1968 * 1969 * @mr: the memory region to be msync 1970 * @addr: the initial address of the range to be sync 1971 * @size: the size of the range to be sync 1972 */ 1973 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size); 1974 1975 /** 1976 * memory_region_writeback: Trigger cache writeback for 1977 * selected address range 1978 * 1979 * @mr: the memory region to be updated 1980 * @addr: the initial address of the range to be written back 1981 * @size: the size of the range to be written back 1982 */ 1983 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size); 1984 1985 /** 1986 * memory_region_set_log: Turn dirty logging on or off for a region. 1987 * 1988 * Turns dirty logging on or off for a specified client (display, migration). 1989 * Only meaningful for RAM regions. 1990 * 1991 * @mr: the memory region being updated. 1992 * @log: whether dirty logging is to be enabled or disabled. 1993 * @client: the user of the logging information; %DIRTY_MEMORY_VGA only. 1994 */ 1995 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client); 1996 1997 /** 1998 * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region. 1999 * 2000 * Marks a range of bytes as dirty, after it has been dirtied outside 2001 * guest code. 2002 * 2003 * @mr: the memory region being dirtied. 2004 * @addr: the address (relative to the start of the region) being dirtied. 2005 * @size: size of the range being dirtied. 2006 */ 2007 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr, 2008 hwaddr size); 2009 2010 /** 2011 * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range 2012 * 2013 * This function is called when the caller wants to clear the remote 2014 * dirty bitmap of a memory range within the memory region. This can 2015 * be used by e.g. KVM to manually clear dirty log when 2016 * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host 2017 * kernel. 2018 * 2019 * @mr: the memory region to clear the dirty log upon 2020 * @start: start address offset within the memory region 2021 * @len: length of the memory region to clear dirty bitmap 2022 */ 2023 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start, 2024 hwaddr len); 2025 2026 /** 2027 * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty 2028 * bitmap and clear it. 2029 * 2030 * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and 2031 * returns the snapshot. The snapshot can then be used to query dirty 2032 * status, using memory_region_snapshot_get_dirty. Snapshotting allows 2033 * querying the same page multiple times, which is especially useful for 2034 * display updates where the scanlines often are not page aligned. 2035 * 2036 * The dirty bitmap region which gets copied into the snapshot (and 2037 * cleared afterwards) can be larger than requested. The boundaries 2038 * are rounded up/down so complete bitmap longs (covering 64 pages on 2039 * 64bit hosts) can be copied over into the bitmap snapshot. Which 2040 * isn't a problem for display updates as the extra pages are outside 2041 * the visible area, and in case the visible area changes a full 2042 * display redraw is due anyway. Should other use cases for this 2043 * function emerge we might have to revisit this implementation 2044 * detail. 2045 * 2046 * Use g_free to release DirtyBitmapSnapshot. 2047 * 2048 * @mr: the memory region being queried. 2049 * @addr: the address (relative to the start of the region) being queried. 2050 * @size: the size of the range being queried. 2051 * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA. 2052 */ 2053 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr, 2054 hwaddr addr, 2055 hwaddr size, 2056 unsigned client); 2057 2058 /** 2059 * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty 2060 * in the specified dirty bitmap snapshot. 2061 * 2062 * @mr: the memory region being queried. 2063 * @snap: the dirty bitmap snapshot 2064 * @addr: the address (relative to the start of the region) being queried. 2065 * @size: the size of the range being queried. 2066 */ 2067 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, 2068 DirtyBitmapSnapshot *snap, 2069 hwaddr addr, hwaddr size); 2070 2071 /** 2072 * memory_region_reset_dirty: Mark a range of pages as clean, for a specified 2073 * client. 2074 * 2075 * Marks a range of pages as no longer dirty. 2076 * 2077 * @mr: the region being updated. 2078 * @addr: the start of the subrange being cleaned. 2079 * @size: the size of the subrange being cleaned. 2080 * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or 2081 * %DIRTY_MEMORY_VGA. 2082 */ 2083 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr, 2084 hwaddr size, unsigned client); 2085 2086 /** 2087 * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate 2088 * TBs (for self-modifying code). 2089 * 2090 * The MemoryRegionOps->write() callback of a ROM device must use this function 2091 * to mark byte ranges that have been modified internally, such as by directly 2092 * accessing the memory returned by memory_region_get_ram_ptr(). 2093 * 2094 * This function marks the range dirty and invalidates TBs so that TCG can 2095 * detect self-modifying code. 2096 * 2097 * @mr: the region being flushed. 2098 * @addr: the start, relative to the start of the region, of the range being 2099 * flushed. 2100 * @size: the size, in bytes, of the range being flushed. 2101 */ 2102 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size); 2103 2104 /** 2105 * memory_region_set_readonly: Turn a memory region read-only (or read-write) 2106 * 2107 * Allows a memory region to be marked as read-only (turning it into a ROM). 2108 * only useful on RAM regions. 2109 * 2110 * @mr: the region being updated. 2111 * @readonly: whether rhe region is to be ROM or RAM. 2112 */ 2113 void memory_region_set_readonly(MemoryRegion *mr, bool readonly); 2114 2115 /** 2116 * memory_region_set_nonvolatile: Turn a memory region non-volatile 2117 * 2118 * Allows a memory region to be marked as non-volatile. 2119 * only useful on RAM regions. 2120 * 2121 * @mr: the region being updated. 2122 * @nonvolatile: whether rhe region is to be non-volatile. 2123 */ 2124 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile); 2125 2126 /** 2127 * memory_region_rom_device_set_romd: enable/disable ROMD mode 2128 * 2129 * Allows a ROM device (initialized with memory_region_init_rom_device() to 2130 * set to ROMD mode (default) or MMIO mode. When it is in ROMD mode, the 2131 * device is mapped to guest memory and satisfies read access directly. 2132 * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function. 2133 * Writes are always handled by the #MemoryRegion.write function. 2134 * 2135 * @mr: the memory region to be updated 2136 * @romd_mode: %true to put the region into ROMD mode 2137 */ 2138 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode); 2139 2140 /** 2141 * memory_region_set_coalescing: Enable memory coalescing for the region. 2142 * 2143 * Enabled writes to a region to be queued for later processing. MMIO ->write 2144 * callbacks may be delayed until a non-coalesced MMIO is issued. 2145 * Only useful for IO regions. Roughly similar to write-combining hardware. 2146 * 2147 * @mr: the memory region to be write coalesced 2148 */ 2149 void memory_region_set_coalescing(MemoryRegion *mr); 2150 2151 /** 2152 * memory_region_add_coalescing: Enable memory coalescing for a sub-range of 2153 * a region. 2154 * 2155 * Like memory_region_set_coalescing(), but works on a sub-range of a region. 2156 * Multiple calls can be issued coalesced disjoint ranges. 2157 * 2158 * @mr: the memory region to be updated. 2159 * @offset: the start of the range within the region to be coalesced. 2160 * @size: the size of the subrange to be coalesced. 2161 */ 2162 void memory_region_add_coalescing(MemoryRegion *mr, 2163 hwaddr offset, 2164 uint64_t size); 2165 2166 /** 2167 * memory_region_clear_coalescing: Disable MMIO coalescing for the region. 2168 * 2169 * Disables any coalescing caused by memory_region_set_coalescing() or 2170 * memory_region_add_coalescing(). Roughly equivalent to uncacheble memory 2171 * hardware. 2172 * 2173 * @mr: the memory region to be updated. 2174 */ 2175 void memory_region_clear_coalescing(MemoryRegion *mr); 2176 2177 /** 2178 * memory_region_set_flush_coalesced: Enforce memory coalescing flush before 2179 * accesses. 2180 * 2181 * Ensure that pending coalesced MMIO request are flushed before the memory 2182 * region is accessed. This property is automatically enabled for all regions 2183 * passed to memory_region_set_coalescing() and memory_region_add_coalescing(). 2184 * 2185 * @mr: the memory region to be updated. 2186 */ 2187 void memory_region_set_flush_coalesced(MemoryRegion *mr); 2188 2189 /** 2190 * memory_region_clear_flush_coalesced: Disable memory coalescing flush before 2191 * accesses. 2192 * 2193 * Clear the automatic coalesced MMIO flushing enabled via 2194 * memory_region_set_flush_coalesced. Note that this service has no effect on 2195 * memory regions that have MMIO coalescing enabled for themselves. For them, 2196 * automatic flushing will stop once coalescing is disabled. 2197 * 2198 * @mr: the memory region to be updated. 2199 */ 2200 void memory_region_clear_flush_coalesced(MemoryRegion *mr); 2201 2202 /** 2203 * memory_region_add_eventfd: Request an eventfd to be triggered when a word 2204 * is written to a location. 2205 * 2206 * Marks a word in an IO region (initialized with memory_region_init_io()) 2207 * as a trigger for an eventfd event. The I/O callback will not be called. 2208 * The caller must be prepared to handle failure (that is, take the required 2209 * action if the callback _is_ called). 2210 * 2211 * @mr: the memory region being updated. 2212 * @addr: the address within @mr that is to be monitored 2213 * @size: the size of the access to trigger the eventfd 2214 * @match_data: whether to match against @data, instead of just @addr 2215 * @data: the data to match against the guest write 2216 * @e: event notifier to be triggered when @addr, @size, and @data all match. 2217 **/ 2218 void memory_region_add_eventfd(MemoryRegion *mr, 2219 hwaddr addr, 2220 unsigned size, 2221 bool match_data, 2222 uint64_t data, 2223 EventNotifier *e); 2224 2225 /** 2226 * memory_region_del_eventfd: Cancel an eventfd. 2227 * 2228 * Cancels an eventfd trigger requested by a previous 2229 * memory_region_add_eventfd() call. 2230 * 2231 * @mr: the memory region being updated. 2232 * @addr: the address within @mr that is to be monitored 2233 * @size: the size of the access to trigger the eventfd 2234 * @match_data: whether to match against @data, instead of just @addr 2235 * @data: the data to match against the guest write 2236 * @e: event notifier to be triggered when @addr, @size, and @data all match. 2237 */ 2238 void memory_region_del_eventfd(MemoryRegion *mr, 2239 hwaddr addr, 2240 unsigned size, 2241 bool match_data, 2242 uint64_t data, 2243 EventNotifier *e); 2244 2245 /** 2246 * memory_region_add_subregion: Add a subregion to a container. 2247 * 2248 * Adds a subregion at @offset. The subregion may not overlap with other 2249 * subregions (except for those explicitly marked as overlapping). A region 2250 * may only be added once as a subregion (unless removed with 2251 * memory_region_del_subregion()); use memory_region_init_alias() if you 2252 * want a region to be a subregion in multiple locations. 2253 * 2254 * @mr: the region to contain the new subregion; must be a container 2255 * initialized with memory_region_init(). 2256 * @offset: the offset relative to @mr where @subregion is added. 2257 * @subregion: the subregion to be added. 2258 */ 2259 void memory_region_add_subregion(MemoryRegion *mr, 2260 hwaddr offset, 2261 MemoryRegion *subregion); 2262 /** 2263 * memory_region_add_subregion_overlap: Add a subregion to a container 2264 * with overlap. 2265 * 2266 * Adds a subregion at @offset. The subregion may overlap with other 2267 * subregions. Conflicts are resolved by having a higher @priority hide a 2268 * lower @priority. Subregions without priority are taken as @priority 0. 2269 * A region may only be added once as a subregion (unless removed with 2270 * memory_region_del_subregion()); use memory_region_init_alias() if you 2271 * want a region to be a subregion in multiple locations. 2272 * 2273 * @mr: the region to contain the new subregion; must be a container 2274 * initialized with memory_region_init(). 2275 * @offset: the offset relative to @mr where @subregion is added. 2276 * @subregion: the subregion to be added. 2277 * @priority: used for resolving overlaps; highest priority wins. 2278 */ 2279 void memory_region_add_subregion_overlap(MemoryRegion *mr, 2280 hwaddr offset, 2281 MemoryRegion *subregion, 2282 int priority); 2283 2284 /** 2285 * memory_region_get_ram_addr: Get the ram address associated with a memory 2286 * region 2287 * 2288 * @mr: the region to be queried 2289 */ 2290 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr); 2291 2292 uint64_t memory_region_get_alignment(const MemoryRegion *mr); 2293 /** 2294 * memory_region_del_subregion: Remove a subregion. 2295 * 2296 * Removes a subregion from its container. 2297 * 2298 * @mr: the container to be updated. 2299 * @subregion: the region being removed; must be a current subregion of @mr. 2300 */ 2301 void memory_region_del_subregion(MemoryRegion *mr, 2302 MemoryRegion *subregion); 2303 2304 /* 2305 * memory_region_set_enabled: dynamically enable or disable a region 2306 * 2307 * Enables or disables a memory region. A disabled memory region 2308 * ignores all accesses to itself and its subregions. It does not 2309 * obscure sibling subregions with lower priority - it simply behaves as 2310 * if it was removed from the hierarchy. 2311 * 2312 * Regions default to being enabled. 2313 * 2314 * @mr: the region to be updated 2315 * @enabled: whether to enable or disable the region 2316 */ 2317 void memory_region_set_enabled(MemoryRegion *mr, bool enabled); 2318 2319 /* 2320 * memory_region_set_address: dynamically update the address of a region 2321 * 2322 * Dynamically updates the address of a region, relative to its container. 2323 * May be used on regions are currently part of a memory hierarchy. 2324 * 2325 * @mr: the region to be updated 2326 * @addr: new address, relative to container region 2327 */ 2328 void memory_region_set_address(MemoryRegion *mr, hwaddr addr); 2329 2330 /* 2331 * memory_region_set_size: dynamically update the size of a region. 2332 * 2333 * Dynamically updates the size of a region. 2334 * 2335 * @mr: the region to be updated 2336 * @size: used size of the region. 2337 */ 2338 void memory_region_set_size(MemoryRegion *mr, uint64_t size); 2339 2340 /* 2341 * memory_region_set_alias_offset: dynamically update a memory alias's offset 2342 * 2343 * Dynamically updates the offset into the target region that an alias points 2344 * to, as if the fourth argument to memory_region_init_alias() has changed. 2345 * 2346 * @mr: the #MemoryRegion to be updated; should be an alias. 2347 * @offset: the new offset into the target memory region 2348 */ 2349 void memory_region_set_alias_offset(MemoryRegion *mr, 2350 hwaddr offset); 2351 2352 /** 2353 * memory_region_present: checks if an address relative to a @container 2354 * translates into #MemoryRegion within @container 2355 * 2356 * Answer whether a #MemoryRegion within @container covers the address 2357 * @addr. 2358 * 2359 * @container: a #MemoryRegion within which @addr is a relative address 2360 * @addr: the area within @container to be searched 2361 */ 2362 bool memory_region_present(MemoryRegion *container, hwaddr addr); 2363 2364 /** 2365 * memory_region_is_mapped: returns true if #MemoryRegion is mapped 2366 * into another memory region, which does not necessarily imply that it is 2367 * mapped into an address space. 2368 * 2369 * @mr: a #MemoryRegion which should be checked if it's mapped 2370 */ 2371 bool memory_region_is_mapped(MemoryRegion *mr); 2372 2373 /** 2374 * memory_region_get_ram_discard_manager: get the #RamDiscardManager for a 2375 * #MemoryRegion 2376 * 2377 * The #RamDiscardManager cannot change while a memory region is mapped. 2378 * 2379 * @mr: the #MemoryRegion 2380 */ 2381 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr); 2382 2383 /** 2384 * memory_region_has_ram_discard_manager: check whether a #MemoryRegion has a 2385 * #RamDiscardManager assigned 2386 * 2387 * @mr: the #MemoryRegion 2388 */ 2389 static inline bool memory_region_has_ram_discard_manager(MemoryRegion *mr) 2390 { 2391 return !!memory_region_get_ram_discard_manager(mr); 2392 } 2393 2394 /** 2395 * memory_region_set_ram_discard_manager: set the #RamDiscardManager for a 2396 * #MemoryRegion 2397 * 2398 * This function must not be called for a mapped #MemoryRegion, a #MemoryRegion 2399 * that does not cover RAM, or a #MemoryRegion that already has a 2400 * #RamDiscardManager assigned. 2401 * 2402 * @mr: the #MemoryRegion 2403 * @rdm: #RamDiscardManager to set 2404 */ 2405 void memory_region_set_ram_discard_manager(MemoryRegion *mr, 2406 RamDiscardManager *rdm); 2407 2408 /** 2409 * memory_region_find: translate an address/size relative to a 2410 * MemoryRegion into a #MemoryRegionSection. 2411 * 2412 * Locates the first #MemoryRegion within @mr that overlaps the range 2413 * given by @addr and @size. 2414 * 2415 * Returns a #MemoryRegionSection that describes a contiguous overlap. 2416 * It will have the following characteristics: 2417 * - @size = 0 iff no overlap was found 2418 * - @mr is non-%NULL iff an overlap was found 2419 * 2420 * Remember that in the return value the @offset_within_region is 2421 * relative to the returned region (in the .@mr field), not to the 2422 * @mr argument. 2423 * 2424 * Similarly, the .@offset_within_address_space is relative to the 2425 * address space that contains both regions, the passed and the 2426 * returned one. However, in the special case where the @mr argument 2427 * has no container (and thus is the root of the address space), the 2428 * following will hold: 2429 * - @offset_within_address_space >= @addr 2430 * - @offset_within_address_space + .@size <= @addr + @size 2431 * 2432 * @mr: a MemoryRegion within which @addr is a relative address 2433 * @addr: start of the area within @as to be searched 2434 * @size: size of the area to be searched 2435 */ 2436 MemoryRegionSection memory_region_find(MemoryRegion *mr, 2437 hwaddr addr, uint64_t size); 2438 2439 /** 2440 * memory_global_dirty_log_sync: synchronize the dirty log for all memory 2441 * 2442 * Synchronizes the dirty page log for all address spaces. 2443 * 2444 * @last_stage: whether this is the last stage of live migration 2445 */ 2446 void memory_global_dirty_log_sync(bool last_stage); 2447 2448 /** 2449 * memory_global_dirty_log_sync: synchronize the dirty log for all memory 2450 * 2451 * Synchronizes the vCPUs with a thread that is reading the dirty bitmap. 2452 * This function must be called after the dirty log bitmap is cleared, and 2453 * before dirty guest memory pages are read. If you are using 2454 * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes 2455 * care of doing this. 2456 */ 2457 void memory_global_after_dirty_log_sync(void); 2458 2459 /** 2460 * memory_region_transaction_begin: Start a transaction. 2461 * 2462 * During a transaction, changes will be accumulated and made visible 2463 * only when the transaction ends (is committed). 2464 */ 2465 void memory_region_transaction_begin(void); 2466 2467 /** 2468 * memory_region_transaction_commit: Commit a transaction and make changes 2469 * visible to the guest. 2470 */ 2471 void memory_region_transaction_commit(void); 2472 2473 /** 2474 * memory_listener_register: register callbacks to be called when memory 2475 * sections are mapped or unmapped into an address 2476 * space 2477 * 2478 * @listener: an object containing the callbacks to be called 2479 * @filter: if non-%NULL, only regions in this address space will be observed 2480 */ 2481 void memory_listener_register(MemoryListener *listener, AddressSpace *filter); 2482 2483 /** 2484 * memory_listener_unregister: undo the effect of memory_listener_register() 2485 * 2486 * @listener: an object containing the callbacks to be removed 2487 */ 2488 void memory_listener_unregister(MemoryListener *listener); 2489 2490 /** 2491 * memory_global_dirty_log_start: begin dirty logging for all regions 2492 * 2493 * @flags: purpose of starting dirty log, migration or dirty rate 2494 */ 2495 void memory_global_dirty_log_start(unsigned int flags); 2496 2497 /** 2498 * memory_global_dirty_log_stop: end dirty logging for all regions 2499 * 2500 * @flags: purpose of stopping dirty log, migration or dirty rate 2501 */ 2502 void memory_global_dirty_log_stop(unsigned int flags); 2503 2504 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled); 2505 2506 bool memory_region_access_valid(MemoryRegion *mr, hwaddr addr, 2507 unsigned size, bool is_write, 2508 MemTxAttrs attrs); 2509 2510 /** 2511 * memory_region_dispatch_read: perform a read directly to the specified 2512 * MemoryRegion. 2513 * 2514 * @mr: #MemoryRegion to access 2515 * @addr: address within that region 2516 * @pval: pointer to uint64_t which the data is written to 2517 * @op: size, sign, and endianness of the memory operation 2518 * @attrs: memory transaction attributes to use for the access 2519 */ 2520 MemTxResult memory_region_dispatch_read(MemoryRegion *mr, 2521 hwaddr addr, 2522 uint64_t *pval, 2523 MemOp op, 2524 MemTxAttrs attrs); 2525 /** 2526 * memory_region_dispatch_write: perform a write directly to the specified 2527 * MemoryRegion. 2528 * 2529 * @mr: #MemoryRegion to access 2530 * @addr: address within that region 2531 * @data: data to write 2532 * @op: size, sign, and endianness of the memory operation 2533 * @attrs: memory transaction attributes to use for the access 2534 */ 2535 MemTxResult memory_region_dispatch_write(MemoryRegion *mr, 2536 hwaddr addr, 2537 uint64_t data, 2538 MemOp op, 2539 MemTxAttrs attrs); 2540 2541 /** 2542 * address_space_init: initializes an address space 2543 * 2544 * @as: an uninitialized #AddressSpace 2545 * @root: a #MemoryRegion that routes addresses for the address space 2546 * @name: an address space name. The name is only used for debugging 2547 * output. 2548 */ 2549 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name); 2550 2551 /** 2552 * address_space_destroy: destroy an address space 2553 * 2554 * Releases all resources associated with an address space. After an address space 2555 * is destroyed, its root memory region (given by address_space_init()) may be destroyed 2556 * as well. 2557 * 2558 * @as: address space to be destroyed 2559 */ 2560 void address_space_destroy(AddressSpace *as); 2561 2562 /** 2563 * address_space_remove_listeners: unregister all listeners of an address space 2564 * 2565 * Removes all callbacks previously registered with memory_listener_register() 2566 * for @as. 2567 * 2568 * @as: an initialized #AddressSpace 2569 */ 2570 void address_space_remove_listeners(AddressSpace *as); 2571 2572 /** 2573 * address_space_rw: read from or write to an address space. 2574 * 2575 * Return a MemTxResult indicating whether the operation succeeded 2576 * or failed (eg unassigned memory, device rejected the transaction, 2577 * IOMMU fault). 2578 * 2579 * @as: #AddressSpace to be accessed 2580 * @addr: address within that address space 2581 * @attrs: memory transaction attributes 2582 * @buf: buffer with the data transferred 2583 * @len: the number of bytes to read or write 2584 * @is_write: indicates the transfer direction 2585 */ 2586 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, 2587 MemTxAttrs attrs, void *buf, 2588 hwaddr len, bool is_write); 2589 2590 /** 2591 * address_space_write: write to address space. 2592 * 2593 * Return a MemTxResult indicating whether the operation succeeded 2594 * or failed (eg unassigned memory, device rejected the transaction, 2595 * IOMMU fault). 2596 * 2597 * @as: #AddressSpace to be accessed 2598 * @addr: address within that address space 2599 * @attrs: memory transaction attributes 2600 * @buf: buffer with the data transferred 2601 * @len: the number of bytes to write 2602 */ 2603 MemTxResult address_space_write(AddressSpace *as, hwaddr addr, 2604 MemTxAttrs attrs, 2605 const void *buf, hwaddr len); 2606 2607 /** 2608 * address_space_write_rom: write to address space, including ROM. 2609 * 2610 * This function writes to the specified address space, but will 2611 * write data to both ROM and RAM. This is used for non-guest 2612 * writes like writes from the gdb debug stub or initial loading 2613 * of ROM contents. 2614 * 2615 * Note that portions of the write which attempt to write data to 2616 * a device will be silently ignored -- only real RAM and ROM will 2617 * be written to. 2618 * 2619 * Return a MemTxResult indicating whether the operation succeeded 2620 * or failed (eg unassigned memory, device rejected the transaction, 2621 * IOMMU fault). 2622 * 2623 * @as: #AddressSpace to be accessed 2624 * @addr: address within that address space 2625 * @attrs: memory transaction attributes 2626 * @buf: buffer with the data transferred 2627 * @len: the number of bytes to write 2628 */ 2629 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr, 2630 MemTxAttrs attrs, 2631 const void *buf, hwaddr len); 2632 2633 /* address_space_ld*: load from an address space 2634 * address_space_st*: store to an address space 2635 * 2636 * These functions perform a load or store of the byte, word, 2637 * longword or quad to the specified address within the AddressSpace. 2638 * The _le suffixed functions treat the data as little endian; 2639 * _be indicates big endian; no suffix indicates "same endianness 2640 * as guest CPU". 2641 * 2642 * The "guest CPU endianness" accessors are deprecated for use outside 2643 * target-* code; devices should be CPU-agnostic and use either the LE 2644 * or the BE accessors. 2645 * 2646 * @as #AddressSpace to be accessed 2647 * @addr: address within that address space 2648 * @val: data value, for stores 2649 * @attrs: memory transaction attributes 2650 * @result: location to write the success/failure of the transaction; 2651 * if NULL, this information is discarded 2652 */ 2653 2654 #define SUFFIX 2655 #define ARG1 as 2656 #define ARG1_DECL AddressSpace *as 2657 #include "exec/memory_ldst.h.inc" 2658 2659 #define SUFFIX 2660 #define ARG1 as 2661 #define ARG1_DECL AddressSpace *as 2662 #include "exec/memory_ldst_phys.h.inc" 2663 2664 struct MemoryRegionCache { 2665 void *ptr; 2666 hwaddr xlat; 2667 hwaddr len; 2668 FlatView *fv; 2669 MemoryRegionSection mrs; 2670 bool is_write; 2671 }; 2672 2673 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL }) 2674 2675 2676 /* address_space_ld*_cached: load from a cached #MemoryRegion 2677 * address_space_st*_cached: store into a cached #MemoryRegion 2678 * 2679 * These functions perform a load or store of the byte, word, 2680 * longword or quad to the specified address. The address is 2681 * a physical address in the AddressSpace, but it must lie within 2682 * a #MemoryRegion that was mapped with address_space_cache_init. 2683 * 2684 * The _le suffixed functions treat the data as little endian; 2685 * _be indicates big endian; no suffix indicates "same endianness 2686 * as guest CPU". 2687 * 2688 * The "guest CPU endianness" accessors are deprecated for use outside 2689 * target-* code; devices should be CPU-agnostic and use either the LE 2690 * or the BE accessors. 2691 * 2692 * @cache: previously initialized #MemoryRegionCache to be accessed 2693 * @addr: address within the address space 2694 * @val: data value, for stores 2695 * @attrs: memory transaction attributes 2696 * @result: location to write the success/failure of the transaction; 2697 * if NULL, this information is discarded 2698 */ 2699 2700 #define SUFFIX _cached_slow 2701 #define ARG1 cache 2702 #define ARG1_DECL MemoryRegionCache *cache 2703 #include "exec/memory_ldst.h.inc" 2704 2705 /* Inline fast path for direct RAM access. */ 2706 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache, 2707 hwaddr addr, MemTxAttrs attrs, MemTxResult *result) 2708 { 2709 assert(addr < cache->len); 2710 if (likely(cache->ptr)) { 2711 return ldub_p(cache->ptr + addr); 2712 } else { 2713 return address_space_ldub_cached_slow(cache, addr, attrs, result); 2714 } 2715 } 2716 2717 static inline void address_space_stb_cached(MemoryRegionCache *cache, 2718 hwaddr addr, uint8_t val, MemTxAttrs attrs, MemTxResult *result) 2719 { 2720 assert(addr < cache->len); 2721 if (likely(cache->ptr)) { 2722 stb_p(cache->ptr + addr, val); 2723 } else { 2724 address_space_stb_cached_slow(cache, addr, val, attrs, result); 2725 } 2726 } 2727 2728 #define ENDIANNESS _le 2729 #include "exec/memory_ldst_cached.h.inc" 2730 2731 #define ENDIANNESS _be 2732 #include "exec/memory_ldst_cached.h.inc" 2733 2734 #define SUFFIX _cached 2735 #define ARG1 cache 2736 #define ARG1_DECL MemoryRegionCache *cache 2737 #include "exec/memory_ldst_phys.h.inc" 2738 2739 /* address_space_cache_init: prepare for repeated access to a physical 2740 * memory region 2741 * 2742 * @cache: #MemoryRegionCache to be filled 2743 * @as: #AddressSpace to be accessed 2744 * @addr: address within that address space 2745 * @len: length of buffer 2746 * @is_write: indicates the transfer direction 2747 * 2748 * Will only work with RAM, and may map a subset of the requested range by 2749 * returning a value that is less than @len. On failure, return a negative 2750 * errno value. 2751 * 2752 * Because it only works with RAM, this function can be used for 2753 * read-modify-write operations. In this case, is_write should be %true. 2754 * 2755 * Note that addresses passed to the address_space_*_cached functions 2756 * are relative to @addr. 2757 */ 2758 int64_t address_space_cache_init(MemoryRegionCache *cache, 2759 AddressSpace *as, 2760 hwaddr addr, 2761 hwaddr len, 2762 bool is_write); 2763 2764 /** 2765 * address_space_cache_invalidate: complete a write to a #MemoryRegionCache 2766 * 2767 * @cache: The #MemoryRegionCache to operate on. 2768 * @addr: The first physical address that was written, relative to the 2769 * address that was passed to @address_space_cache_init. 2770 * @access_len: The number of bytes that were written starting at @addr. 2771 */ 2772 void address_space_cache_invalidate(MemoryRegionCache *cache, 2773 hwaddr addr, 2774 hwaddr access_len); 2775 2776 /** 2777 * address_space_cache_destroy: free a #MemoryRegionCache 2778 * 2779 * @cache: The #MemoryRegionCache whose memory should be released. 2780 */ 2781 void address_space_cache_destroy(MemoryRegionCache *cache); 2782 2783 /* address_space_get_iotlb_entry: translate an address into an IOTLB 2784 * entry. Should be called from an RCU critical section. 2785 */ 2786 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr, 2787 bool is_write, MemTxAttrs attrs); 2788 2789 /* address_space_translate: translate an address range into an address space 2790 * into a MemoryRegion and an address range into that section. Should be 2791 * called from an RCU critical section, to avoid that the last reference 2792 * to the returned region disappears after address_space_translate returns. 2793 * 2794 * @fv: #FlatView to be accessed 2795 * @addr: address within that address space 2796 * @xlat: pointer to address within the returned memory region section's 2797 * #MemoryRegion. 2798 * @len: pointer to length 2799 * @is_write: indicates the transfer direction 2800 * @attrs: memory attributes 2801 */ 2802 MemoryRegion *flatview_translate(FlatView *fv, 2803 hwaddr addr, hwaddr *xlat, 2804 hwaddr *len, bool is_write, 2805 MemTxAttrs attrs); 2806 2807 static inline MemoryRegion *address_space_translate(AddressSpace *as, 2808 hwaddr addr, hwaddr *xlat, 2809 hwaddr *len, bool is_write, 2810 MemTxAttrs attrs) 2811 { 2812 return flatview_translate(address_space_to_flatview(as), 2813 addr, xlat, len, is_write, attrs); 2814 } 2815 2816 /* address_space_access_valid: check for validity of accessing an address 2817 * space range 2818 * 2819 * Check whether memory is assigned to the given address space range, and 2820 * access is permitted by any IOMMU regions that are active for the address 2821 * space. 2822 * 2823 * For now, addr and len should be aligned to a page size. This limitation 2824 * will be lifted in the future. 2825 * 2826 * @as: #AddressSpace to be accessed 2827 * @addr: address within that address space 2828 * @len: length of the area to be checked 2829 * @is_write: indicates the transfer direction 2830 * @attrs: memory attributes 2831 */ 2832 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len, 2833 bool is_write, MemTxAttrs attrs); 2834 2835 /* address_space_map: map a physical memory region into a host virtual address 2836 * 2837 * May map a subset of the requested range, given by and returned in @plen. 2838 * May return %NULL and set *@plen to zero(0), if resources needed to perform 2839 * the mapping are exhausted. 2840 * Use only for reads OR writes - not for read-modify-write operations. 2841 * Use cpu_register_map_client() to know when retrying the map operation is 2842 * likely to succeed. 2843 * 2844 * @as: #AddressSpace to be accessed 2845 * @addr: address within that address space 2846 * @plen: pointer to length of buffer; updated on return 2847 * @is_write: indicates the transfer direction 2848 * @attrs: memory attributes 2849 */ 2850 void *address_space_map(AddressSpace *as, hwaddr addr, 2851 hwaddr *plen, bool is_write, MemTxAttrs attrs); 2852 2853 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map() 2854 * 2855 * Will also mark the memory as dirty if @is_write == %true. @access_len gives 2856 * the amount of memory that was actually read or written by the caller. 2857 * 2858 * @as: #AddressSpace used 2859 * @buffer: host pointer as returned by address_space_map() 2860 * @len: buffer length as returned by address_space_map() 2861 * @access_len: amount of data actually transferred 2862 * @is_write: indicates the transfer direction 2863 */ 2864 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len, 2865 bool is_write, hwaddr access_len); 2866 2867 2868 /* Internal functions, part of the implementation of address_space_read. */ 2869 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr, 2870 MemTxAttrs attrs, void *buf, hwaddr len); 2871 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr, 2872 MemTxAttrs attrs, void *buf, 2873 hwaddr len, hwaddr addr1, hwaddr l, 2874 MemoryRegion *mr); 2875 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr); 2876 2877 /* Internal functions, part of the implementation of address_space_read_cached 2878 * and address_space_write_cached. */ 2879 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache, 2880 hwaddr addr, void *buf, hwaddr len); 2881 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache, 2882 hwaddr addr, const void *buf, 2883 hwaddr len); 2884 2885 int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr); 2886 bool prepare_mmio_access(MemoryRegion *mr); 2887 2888 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write) 2889 { 2890 if (is_write) { 2891 return memory_region_is_ram(mr) && !mr->readonly && 2892 !mr->rom_device && !memory_region_is_ram_device(mr); 2893 } else { 2894 return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) || 2895 memory_region_is_romd(mr); 2896 } 2897 } 2898 2899 /** 2900 * address_space_read: read from an address space. 2901 * 2902 * Return a MemTxResult indicating whether the operation succeeded 2903 * or failed (eg unassigned memory, device rejected the transaction, 2904 * IOMMU fault). Called within RCU critical section. 2905 * 2906 * @as: #AddressSpace to be accessed 2907 * @addr: address within that address space 2908 * @attrs: memory transaction attributes 2909 * @buf: buffer with the data transferred 2910 * @len: length of the data transferred 2911 */ 2912 static inline __attribute__((__always_inline__)) 2913 MemTxResult address_space_read(AddressSpace *as, hwaddr addr, 2914 MemTxAttrs attrs, void *buf, 2915 hwaddr len) 2916 { 2917 MemTxResult result = MEMTX_OK; 2918 hwaddr l, addr1; 2919 void *ptr; 2920 MemoryRegion *mr; 2921 FlatView *fv; 2922 2923 if (__builtin_constant_p(len)) { 2924 if (len) { 2925 RCU_READ_LOCK_GUARD(); 2926 fv = address_space_to_flatview(as); 2927 l = len; 2928 mr = flatview_translate(fv, addr, &addr1, &l, false, attrs); 2929 if (len == l && memory_access_is_direct(mr, false)) { 2930 ptr = qemu_map_ram_ptr(mr->ram_block, addr1); 2931 memcpy(buf, ptr, len); 2932 } else { 2933 result = flatview_read_continue(fv, addr, attrs, buf, len, 2934 addr1, l, mr); 2935 } 2936 } 2937 } else { 2938 result = address_space_read_full(as, addr, attrs, buf, len); 2939 } 2940 return result; 2941 } 2942 2943 /** 2944 * address_space_read_cached: read from a cached RAM region 2945 * 2946 * @cache: Cached region to be addressed 2947 * @addr: address relative to the base of the RAM region 2948 * @buf: buffer with the data transferred 2949 * @len: length of the data transferred 2950 */ 2951 static inline MemTxResult 2952 address_space_read_cached(MemoryRegionCache *cache, hwaddr addr, 2953 void *buf, hwaddr len) 2954 { 2955 assert(addr < cache->len && len <= cache->len - addr); 2956 fuzz_dma_read_cb(cache->xlat + addr, len, cache->mrs.mr); 2957 if (likely(cache->ptr)) { 2958 memcpy(buf, cache->ptr + addr, len); 2959 return MEMTX_OK; 2960 } else { 2961 return address_space_read_cached_slow(cache, addr, buf, len); 2962 } 2963 } 2964 2965 /** 2966 * address_space_write_cached: write to a cached RAM region 2967 * 2968 * @cache: Cached region to be addressed 2969 * @addr: address relative to the base of the RAM region 2970 * @buf: buffer with the data transferred 2971 * @len: length of the data transferred 2972 */ 2973 static inline MemTxResult 2974 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr, 2975 const void *buf, hwaddr len) 2976 { 2977 assert(addr < cache->len && len <= cache->len - addr); 2978 if (likely(cache->ptr)) { 2979 memcpy(cache->ptr + addr, buf, len); 2980 return MEMTX_OK; 2981 } else { 2982 return address_space_write_cached_slow(cache, addr, buf, len); 2983 } 2984 } 2985 2986 /** 2987 * address_space_set: Fill address space with a constant byte. 2988 * 2989 * Return a MemTxResult indicating whether the operation succeeded 2990 * or failed (eg unassigned memory, device rejected the transaction, 2991 * IOMMU fault). 2992 * 2993 * @as: #AddressSpace to be accessed 2994 * @addr: address within that address space 2995 * @c: constant byte to fill the memory 2996 * @len: the number of bytes to fill with the constant byte 2997 * @attrs: memory transaction attributes 2998 */ 2999 MemTxResult address_space_set(AddressSpace *as, hwaddr addr, 3000 uint8_t c, hwaddr len, MemTxAttrs attrs); 3001 3002 #ifdef NEED_CPU_H 3003 /* enum device_endian to MemOp. */ 3004 static inline MemOp devend_memop(enum device_endian end) 3005 { 3006 QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN && 3007 DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN); 3008 3009 #if HOST_BIG_ENDIAN != TARGET_BIG_ENDIAN 3010 /* Swap if non-host endianness or native (target) endianness */ 3011 return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP; 3012 #else 3013 const int non_host_endianness = 3014 DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN; 3015 3016 /* In this case, native (target) endianness needs no swap. */ 3017 return (end == non_host_endianness) ? MO_BSWAP : 0; 3018 #endif 3019 } 3020 #endif 3021 3022 /* 3023 * Inhibit technologies that require discarding of pages in RAM blocks, e.g., 3024 * to manage the actual amount of memory consumed by the VM (then, the memory 3025 * provided by RAM blocks might be bigger than the desired memory consumption). 3026 * This *must* be set if: 3027 * - Discarding parts of a RAM blocks does not result in the change being 3028 * reflected in the VM and the pages getting freed. 3029 * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous 3030 * discards blindly. 3031 * - Discarding parts of a RAM blocks will result in integrity issues (e.g., 3032 * encrypted VMs). 3033 * Technologies that only temporarily pin the current working set of a 3034 * driver are fine, because we don't expect such pages to be discarded 3035 * (esp. based on guest action like balloon inflation). 3036 * 3037 * This is *not* to be used to protect from concurrent discards (esp., 3038 * postcopy). 3039 * 3040 * Returns 0 if successful. Returns -EBUSY if a technology that relies on 3041 * discards to work reliably is active. 3042 */ 3043 int ram_block_discard_disable(bool state); 3044 3045 /* 3046 * See ram_block_discard_disable(): only disable uncoordinated discards, 3047 * keeping coordinated discards (via the RamDiscardManager) enabled. 3048 */ 3049 int ram_block_uncoordinated_discard_disable(bool state); 3050 3051 /* 3052 * Inhibit technologies that disable discarding of pages in RAM blocks. 3053 * 3054 * Returns 0 if successful. Returns -EBUSY if discards are already set to 3055 * broken. 3056 */ 3057 int ram_block_discard_require(bool state); 3058 3059 /* 3060 * See ram_block_discard_require(): only inhibit technologies that disable 3061 * uncoordinated discarding of pages in RAM blocks, allowing co-existance with 3062 * technologies that only inhibit uncoordinated discards (via the 3063 * RamDiscardManager). 3064 */ 3065 int ram_block_coordinated_discard_require(bool state); 3066 3067 /* 3068 * Test if any discarding of memory in ram blocks is disabled. 3069 */ 3070 bool ram_block_discard_is_disabled(void); 3071 3072 /* 3073 * Test if any discarding of memory in ram blocks is required to work reliably. 3074 */ 3075 bool ram_block_discard_is_required(void); 3076 3077 #endif 3078 3079 #endif 3080