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