1 /* 2 * Physical memory management API 3 * 4 * Copyright 2011 Red Hat, Inc. and/or its affiliates 5 * 6 * Authors: 7 * Avi Kivity <avi@redhat.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 */ 13 14 #ifndef MEMORY_H 15 #define MEMORY_H 16 17 #ifndef CONFIG_USER_ONLY 18 19 #include "exec/cpu-common.h" 20 #include "exec/hwaddr.h" 21 #include "exec/memattrs.h" 22 #include "exec/memop.h" 23 #include "exec/ramlist.h" 24 #include "qemu/bswap.h" 25 #include "qemu/queue.h" 26 #include "qemu/int128.h" 27 #include "qemu/notify.h" 28 #include "qom/object.h" 29 #include "qemu/rcu.h" 30 31 #define RAM_ADDR_INVALID (~(ram_addr_t)0) 32 33 #define MAX_PHYS_ADDR_SPACE_BITS 62 34 #define MAX_PHYS_ADDR (((hwaddr)1 << MAX_PHYS_ADDR_SPACE_BITS) - 1) 35 36 #define TYPE_MEMORY_REGION "memory-region" 37 DECLARE_INSTANCE_CHECKER(MemoryRegion, MEMORY_REGION, 38 TYPE_MEMORY_REGION) 39 40 #define TYPE_IOMMU_MEMORY_REGION "iommu-memory-region" 41 typedef struct IOMMUMemoryRegionClass IOMMUMemoryRegionClass; 42 DECLARE_OBJ_CHECKERS(IOMMUMemoryRegion, IOMMUMemoryRegionClass, 43 IOMMU_MEMORY_REGION, TYPE_IOMMU_MEMORY_REGION) 44 45 #define TYPE_RAM_DISCARD_MANAGER "qemu:ram-discard-manager" 46 typedef struct RamDiscardManagerClass RamDiscardManagerClass; 47 typedef struct RamDiscardManager RamDiscardManager; 48 DECLARE_OBJ_CHECKERS(RamDiscardManager, RamDiscardManagerClass, 49 RAM_DISCARD_MANAGER, TYPE_RAM_DISCARD_MANAGER); 50 51 #ifdef CONFIG_FUZZ 52 void fuzz_dma_read_cb(size_t addr, 53 size_t len, 54 MemoryRegion *mr); 55 #else 56 static inline void fuzz_dma_read_cb(size_t addr, 57 size_t len, 58 MemoryRegion *mr) 59 { 60 /* Do Nothing */ 61 } 62 #endif 63 64 /* Possible bits for global_dirty_log_{start|stop} */ 65 66 /* Dirty tracking enabled because migration is running */ 67 #define GLOBAL_DIRTY_MIGRATION (1U << 0) 68 69 /* Dirty tracking enabled because measuring dirty rate */ 70 #define GLOBAL_DIRTY_DIRTY_RATE (1U << 1) 71 72 #define GLOBAL_DIRTY_MASK (0x3) 73 74 extern unsigned int global_dirty_tracking; 75 76 typedef struct MemoryRegionOps MemoryRegionOps; 77 78 struct ReservedRegion { 79 hwaddr low; 80 hwaddr high; 81 unsigned type; 82 }; 83 84 /** 85 * struct MemoryRegionSection: describes a fragment of a #MemoryRegion 86 * 87 * @mr: the region, or %NULL if empty 88 * @fv: the flat view of the address space the region is mapped in 89 * @offset_within_region: the beginning of the section, relative to @mr's start 90 * @size: the size of the section; will not exceed @mr's boundaries 91 * @offset_within_address_space: the address of the first byte of the section 92 * relative to the region's address space 93 * @readonly: writes to this section are ignored 94 * @nonvolatile: this section is non-volatile 95 */ 96 struct MemoryRegionSection { 97 Int128 size; 98 MemoryRegion *mr; 99 FlatView *fv; 100 hwaddr offset_within_region; 101 hwaddr offset_within_address_space; 102 bool readonly; 103 bool nonvolatile; 104 }; 105 106 typedef struct IOMMUTLBEntry IOMMUTLBEntry; 107 108 /* See address_space_translate: bit 0 is read, bit 1 is write. */ 109 typedef enum { 110 IOMMU_NONE = 0, 111 IOMMU_RO = 1, 112 IOMMU_WO = 2, 113 IOMMU_RW = 3, 114 } IOMMUAccessFlags; 115 116 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0)) 117 118 struct IOMMUTLBEntry { 119 AddressSpace *target_as; 120 hwaddr iova; 121 hwaddr translated_addr; 122 hwaddr addr_mask; /* 0xfff = 4k translation */ 123 IOMMUAccessFlags perm; 124 }; 125 126 /* 127 * Bitmap for different IOMMUNotifier capabilities. Each notifier can 128 * register with one or multiple IOMMU Notifier capability bit(s). 129 */ 130 typedef enum { 131 IOMMU_NOTIFIER_NONE = 0, 132 /* Notify cache invalidations */ 133 IOMMU_NOTIFIER_UNMAP = 0x1, 134 /* Notify entry changes (newly created entries) */ 135 IOMMU_NOTIFIER_MAP = 0x2, 136 /* Notify changes on device IOTLB entries */ 137 IOMMU_NOTIFIER_DEVIOTLB_UNMAP = 0x04, 138 } IOMMUNotifierFlag; 139 140 #define IOMMU_NOTIFIER_IOTLB_EVENTS (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP) 141 #define IOMMU_NOTIFIER_DEVIOTLB_EVENTS IOMMU_NOTIFIER_DEVIOTLB_UNMAP 142 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_IOTLB_EVENTS | \ 143 IOMMU_NOTIFIER_DEVIOTLB_EVENTS) 144 145 struct IOMMUNotifier; 146 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier, 147 IOMMUTLBEntry *data); 148 149 struct IOMMUNotifier { 150 IOMMUNotify notify; 151 IOMMUNotifierFlag notifier_flags; 152 /* Notify for address space range start <= addr <= end */ 153 hwaddr start; 154 hwaddr end; 155 int iommu_idx; 156 QLIST_ENTRY(IOMMUNotifier) node; 157 }; 158 typedef struct IOMMUNotifier IOMMUNotifier; 159 160 typedef struct IOMMUTLBEvent { 161 IOMMUNotifierFlag type; 162 IOMMUTLBEntry entry; 163 } IOMMUTLBEvent; 164 165 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */ 166 #define RAM_PREALLOC (1 << 0) 167 168 /* RAM is mmap-ed with MAP_SHARED */ 169 #define RAM_SHARED (1 << 1) 170 171 /* Only a portion of RAM (used_length) is actually used, and migrated. 172 * Resizing RAM while migrating can result in the migration being canceled. 173 */ 174 #define RAM_RESIZEABLE (1 << 2) 175 176 /* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically 177 * zero the page and wake waiting processes. 178 * (Set during postcopy) 179 */ 180 #define RAM_UF_ZEROPAGE (1 << 3) 181 182 /* RAM can be migrated */ 183 #define RAM_MIGRATABLE (1 << 4) 184 185 /* RAM is a persistent kind memory */ 186 #define RAM_PMEM (1 << 5) 187 188 189 /* 190 * UFFDIO_WRITEPROTECT is used on this RAMBlock to 191 * support 'write-tracking' migration type. 192 * Implies ram_state->ram_wt_enabled. 193 */ 194 #define RAM_UF_WRITEPROTECT (1 << 6) 195 196 /* 197 * RAM is mmap-ed with MAP_NORESERVE. When set, reserving swap space (or huge 198 * pages if applicable) is skipped: will bail out if not supported. When not 199 * set, the OS will do the reservation, if supported for the memory type. 200 */ 201 #define RAM_NORESERVE (1 << 7) 202 203 /* RAM that isn't accessible through normal means. */ 204 #define RAM_PROTECTED (1 << 8) 205 206 static inline void iommu_notifier_init(IOMMUNotifier *n, IOMMUNotify fn, 207 IOMMUNotifierFlag flags, 208 hwaddr start, hwaddr end, 209 int iommu_idx) 210 { 211 n->notify = fn; 212 n->notifier_flags = flags; 213 n->start = start; 214 n->end = end; 215 n->iommu_idx = iommu_idx; 216 } 217 218 /* 219 * Memory region callbacks 220 */ 221 struct MemoryRegionOps { 222 /* Read from the memory region. @addr is relative to @mr; @size is 223 * in bytes. */ 224 uint64_t (*read)(void *opaque, 225 hwaddr addr, 226 unsigned size); 227 /* Write to the memory region. @addr is relative to @mr; @size is 228 * in bytes. */ 229 void (*write)(void *opaque, 230 hwaddr addr, 231 uint64_t data, 232 unsigned size); 233 234 MemTxResult (*read_with_attrs)(void *opaque, 235 hwaddr addr, 236 uint64_t *data, 237 unsigned size, 238 MemTxAttrs attrs); 239 MemTxResult (*write_with_attrs)(void *opaque, 240 hwaddr addr, 241 uint64_t data, 242 unsigned size, 243 MemTxAttrs attrs); 244 245 enum device_endian endianness; 246 /* Guest-visible constraints: */ 247 struct { 248 /* If nonzero, specify bounds on access sizes beyond which a machine 249 * check is thrown. 250 */ 251 unsigned min_access_size; 252 unsigned max_access_size; 253 /* If true, unaligned accesses are supported. Otherwise unaligned 254 * accesses throw machine checks. 255 */ 256 bool unaligned; 257 /* 258 * If present, and returns #false, the transaction is not accepted 259 * by the device (and results in machine dependent behaviour such 260 * as a machine check exception). 261 */ 262 bool (*accepts)(void *opaque, hwaddr addr, 263 unsigned size, bool is_write, 264 MemTxAttrs attrs); 265 } valid; 266 /* Internal implementation constraints: */ 267 struct { 268 /* If nonzero, specifies the minimum size implemented. Smaller sizes 269 * will be rounded upwards and a partial result will be returned. 270 */ 271 unsigned min_access_size; 272 /* If nonzero, specifies the maximum size implemented. Larger sizes 273 * will be done as a series of accesses with smaller sizes. 274 */ 275 unsigned max_access_size; 276 /* If true, unaligned accesses are supported. Otherwise all accesses 277 * are converted to (possibly multiple) naturally aligned accesses. 278 */ 279 bool unaligned; 280 } impl; 281 }; 282 283 typedef struct MemoryRegionClass { 284 /* private */ 285 ObjectClass parent_class; 286 } MemoryRegionClass; 287 288 289 enum IOMMUMemoryRegionAttr { 290 IOMMU_ATTR_SPAPR_TCE_FD 291 }; 292 293 /* 294 * IOMMUMemoryRegionClass: 295 * 296 * All IOMMU implementations need to subclass TYPE_IOMMU_MEMORY_REGION 297 * and provide an implementation of at least the @translate method here 298 * to handle requests to the memory region. Other methods are optional. 299 * 300 * The IOMMU implementation must use the IOMMU notifier infrastructure 301 * to report whenever mappings are changed, by calling 302 * memory_region_notify_iommu() (or, if necessary, by calling 303 * memory_region_notify_iommu_one() for each registered notifier). 304 * 305 * Conceptually an IOMMU provides a mapping from input address 306 * to an output TLB entry. If the IOMMU is aware of memory transaction 307 * attributes and the output TLB entry depends on the transaction 308 * attributes, we represent this using IOMMU indexes. Each index 309 * selects a particular translation table that the IOMMU has: 310 * 311 * @attrs_to_index returns the IOMMU index for a set of transaction attributes 312 * 313 * @translate takes an input address and an IOMMU index 314 * 315 * and the mapping returned can only depend on the input address and the 316 * IOMMU index. 317 * 318 * Most IOMMUs don't care about the transaction attributes and support 319 * only a single IOMMU index. A more complex IOMMU might have one index 320 * for secure transactions and one for non-secure transactions. 321 */ 322 struct IOMMUMemoryRegionClass { 323 /* private: */ 324 MemoryRegionClass parent_class; 325 326 /* public: */ 327 /** 328 * @translate: 329 * 330 * Return a TLB entry that contains a given address. 331 * 332 * The IOMMUAccessFlags indicated via @flag are optional and may 333 * be specified as IOMMU_NONE to indicate that the caller needs 334 * the full translation information for both reads and writes. If 335 * the access flags are specified then the IOMMU implementation 336 * may use this as an optimization, to stop doing a page table 337 * walk as soon as it knows that the requested permissions are not 338 * allowed. If IOMMU_NONE is passed then the IOMMU must do the 339 * full page table walk and report the permissions in the returned 340 * IOMMUTLBEntry. (Note that this implies that an IOMMU may not 341 * return different mappings for reads and writes.) 342 * 343 * The returned information remains valid while the caller is 344 * holding the big QEMU lock or is inside an RCU critical section; 345 * if the caller wishes to cache the mapping beyond that it must 346 * register an IOMMU notifier so it can invalidate its cached 347 * information when the IOMMU mapping changes. 348 * 349 * @iommu: the IOMMUMemoryRegion 350 * 351 * @hwaddr: address to be translated within the memory region 352 * 353 * @flag: requested access permission 354 * 355 * @iommu_idx: IOMMU index for the translation 356 */ 357 IOMMUTLBEntry (*translate)(IOMMUMemoryRegion *iommu, hwaddr addr, 358 IOMMUAccessFlags flag, int iommu_idx); 359 /** 360 * @get_min_page_size: 361 * 362 * Returns minimum supported page size in bytes. 363 * 364 * If this method is not provided then the minimum is assumed to 365 * be TARGET_PAGE_SIZE. 366 * 367 * @iommu: the IOMMUMemoryRegion 368 */ 369 uint64_t (*get_min_page_size)(IOMMUMemoryRegion *iommu); 370 /** 371 * @notify_flag_changed: 372 * 373 * Called when IOMMU Notifier flag changes (ie when the set of 374 * events which IOMMU users are requesting notification for changes). 375 * Optional method -- need not be provided if the IOMMU does not 376 * need to know exactly which events must be notified. 377 * 378 * @iommu: the IOMMUMemoryRegion 379 * 380 * @old_flags: events which previously needed to be notified 381 * 382 * @new_flags: events which now need to be notified 383 * 384 * Returns 0 on success, or a negative errno; in particular 385 * returns -EINVAL if the new flag bitmap is not supported by the 386 * IOMMU memory region. In case of failure, the error object 387 * must be created 388 */ 389 int (*notify_flag_changed)(IOMMUMemoryRegion *iommu, 390 IOMMUNotifierFlag old_flags, 391 IOMMUNotifierFlag new_flags, 392 Error **errp); 393 /** 394 * @replay: 395 * 396 * Called to handle memory_region_iommu_replay(). 397 * 398 * The default implementation of memory_region_iommu_replay() is to 399 * call the IOMMU translate method for every page in the address space 400 * with flag == IOMMU_NONE and then call the notifier if translate 401 * returns a valid mapping. If this method is implemented then it 402 * overrides the default behaviour, and must provide the full semantics 403 * of memory_region_iommu_replay(), by calling @notifier for every 404 * translation present in the IOMMU. 405 * 406 * Optional method -- an IOMMU only needs to provide this method 407 * if the default is inefficient or produces undesirable side effects. 408 * 409 * Note: this is not related to record-and-replay functionality. 410 */ 411 void (*replay)(IOMMUMemoryRegion *iommu, IOMMUNotifier *notifier); 412 413 /** 414 * @get_attr: 415 * 416 * Get IOMMU misc attributes. This is an optional method that 417 * can be used to allow users of the IOMMU to get implementation-specific 418 * information. The IOMMU implements this method to handle calls 419 * by IOMMU users to memory_region_iommu_get_attr() by filling in 420 * the arbitrary data pointer for any IOMMUMemoryRegionAttr values that 421 * the IOMMU supports. If the method is unimplemented then 422 * memory_region_iommu_get_attr() will always return -EINVAL. 423 * 424 * @iommu: the IOMMUMemoryRegion 425 * 426 * @attr: attribute being queried 427 * 428 * @data: memory to fill in with the attribute data 429 * 430 * Returns 0 on success, or a negative errno; in particular 431 * returns -EINVAL for unrecognized or unimplemented attribute types. 432 */ 433 int (*get_attr)(IOMMUMemoryRegion *iommu, enum IOMMUMemoryRegionAttr attr, 434 void *data); 435 436 /** 437 * @attrs_to_index: 438 * 439 * Return the IOMMU index to use for a given set of transaction attributes. 440 * 441 * Optional method: if an IOMMU only supports a single IOMMU index then 442 * the default implementation of memory_region_iommu_attrs_to_index() 443 * will return 0. 444 * 445 * The indexes supported by an IOMMU must be contiguous, starting at 0. 446 * 447 * @iommu: the IOMMUMemoryRegion 448 * @attrs: memory transaction attributes 449 */ 450 int (*attrs_to_index)(IOMMUMemoryRegion *iommu, MemTxAttrs attrs); 451 452 /** 453 * @num_indexes: 454 * 455 * Return the number of IOMMU indexes this IOMMU supports. 456 * 457 * Optional method: if this method is not provided, then 458 * memory_region_iommu_num_indexes() will return 1, indicating that 459 * only a single IOMMU index is supported. 460 * 461 * @iommu: the IOMMUMemoryRegion 462 */ 463 int (*num_indexes)(IOMMUMemoryRegion *iommu); 464 465 /** 466 * @iommu_set_page_size_mask: 467 * 468 * Restrict the page size mask that can be supported with a given IOMMU 469 * memory region. Used for example to propagate host physical IOMMU page 470 * size mask limitations to the virtual IOMMU. 471 * 472 * Optional method: if this method is not provided, then the default global 473 * page mask is used. 474 * 475 * @iommu: the IOMMUMemoryRegion 476 * 477 * @page_size_mask: a bitmask of supported page sizes. At least one bit, 478 * representing the smallest page size, must be set. Additional set bits 479 * represent supported block sizes. For example a host physical IOMMU that 480 * uses page tables with a page size of 4kB, and supports 2MB and 4GB 481 * blocks, will set mask 0x40201000. A granule of 4kB with indiscriminate 482 * block sizes is specified with mask 0xfffffffffffff000. 483 * 484 * Returns 0 on success, or a negative error. In case of failure, the error 485 * object must be created. 486 */ 487 int (*iommu_set_page_size_mask)(IOMMUMemoryRegion *iommu, 488 uint64_t page_size_mask, 489 Error **errp); 490 }; 491 492 typedef struct RamDiscardListener RamDiscardListener; 493 typedef int (*NotifyRamPopulate)(RamDiscardListener *rdl, 494 MemoryRegionSection *section); 495 typedef void (*NotifyRamDiscard)(RamDiscardListener *rdl, 496 MemoryRegionSection *section); 497 498 struct RamDiscardListener { 499 /* 500 * @notify_populate: 501 * 502 * Notification that previously discarded memory is about to get populated. 503 * Listeners are able to object. If any listener objects, already 504 * successfully notified listeners are notified about a discard again. 505 * 506 * @rdl: the #RamDiscardListener getting notified 507 * @section: the #MemoryRegionSection to get populated. The section 508 * is aligned within the memory region to the minimum granularity 509 * unless it would exceed the registered section. 510 * 511 * Returns 0 on success. If the notification is rejected by the listener, 512 * an error is returned. 513 */ 514 NotifyRamPopulate notify_populate; 515 516 /* 517 * @notify_discard: 518 * 519 * Notification that previously populated memory was discarded successfully 520 * and listeners should drop all references to such memory and prevent 521 * new population (e.g., unmap). 522 * 523 * @rdl: the #RamDiscardListener getting notified 524 * @section: the #MemoryRegionSection to get populated. The section 525 * is aligned within the memory region to the minimum granularity 526 * unless it would exceed the registered section. 527 */ 528 NotifyRamDiscard notify_discard; 529 530 /* 531 * @double_discard_supported: 532 * 533 * The listener suppors getting @notify_discard notifications that span 534 * already discarded parts. 535 */ 536 bool double_discard_supported; 537 538 MemoryRegionSection *section; 539 QLIST_ENTRY(RamDiscardListener) next; 540 }; 541 542 static inline void ram_discard_listener_init(RamDiscardListener *rdl, 543 NotifyRamPopulate populate_fn, 544 NotifyRamDiscard discard_fn, 545 bool double_discard_supported) 546 { 547 rdl->notify_populate = populate_fn; 548 rdl->notify_discard = discard_fn; 549 rdl->double_discard_supported = double_discard_supported; 550 } 551 552 typedef int (*ReplayRamPopulate)(MemoryRegionSection *section, void *opaque); 553 554 /* 555 * RamDiscardManagerClass: 556 * 557 * A #RamDiscardManager coordinates which parts of specific RAM #MemoryRegion 558 * regions are currently populated to be used/accessed by the VM, notifying 559 * after parts were discarded (freeing up memory) and before parts will be 560 * populated (consuming memory), to be used/acessed by the VM. 561 * 562 * A #RamDiscardManager can only be set for a RAM #MemoryRegion while the 563 * #MemoryRegion isn't mapped yet; it cannot change while the #MemoryRegion is 564 * mapped. 565 * 566 * The #RamDiscardManager is intended to be used by technologies that are 567 * incompatible with discarding of RAM (e.g., VFIO, which may pin all 568 * memory inside a #MemoryRegion), and require proper coordination to only 569 * map the currently populated parts, to hinder parts that are expected to 570 * remain discarded from silently getting populated and consuming memory. 571 * Technologies that support discarding of RAM don't have to bother and can 572 * simply map the whole #MemoryRegion. 573 * 574 * An example #RamDiscardManager is virtio-mem, which logically (un)plugs 575 * memory within an assigned RAM #MemoryRegion, coordinated with the VM. 576 * Logically unplugging memory consists of discarding RAM. The VM agreed to not 577 * access unplugged (discarded) memory - especially via DMA. virtio-mem will 578 * properly coordinate with listeners before memory is plugged (populated), 579 * and after memory is unplugged (discarded). 580 * 581 * Listeners are called in multiples of the minimum granularity (unless it 582 * would exceed the registered range) and changes are aligned to the minimum 583 * granularity within the #MemoryRegion. Listeners have to prepare for memory 584 * becomming discarded in a different granularity than it was populated and the 585 * other way around. 586 */ 587 struct RamDiscardManagerClass { 588 /* private */ 589 InterfaceClass parent_class; 590 591 /* public */ 592 593 /** 594 * @get_min_granularity: 595 * 596 * Get the minimum granularity in which listeners will get notified 597 * about changes within the #MemoryRegion via the #RamDiscardManager. 598 * 599 * @rdm: the #RamDiscardManager 600 * @mr: the #MemoryRegion 601 * 602 * Returns the minimum granularity. 603 */ 604 uint64_t (*get_min_granularity)(const RamDiscardManager *rdm, 605 const MemoryRegion *mr); 606 607 /** 608 * @is_populated: 609 * 610 * Check whether the given #MemoryRegionSection is completely populated 611 * (i.e., no parts are currently discarded) via the #RamDiscardManager. 612 * There are no alignment requirements. 613 * 614 * @rdm: the #RamDiscardManager 615 * @section: the #MemoryRegionSection 616 * 617 * Returns whether the given range is completely populated. 618 */ 619 bool (*is_populated)(const RamDiscardManager *rdm, 620 const MemoryRegionSection *section); 621 622 /** 623 * @replay_populated: 624 * 625 * Call the #ReplayRamPopulate callback for all populated parts within the 626 * #MemoryRegionSection via the #RamDiscardManager. 627 * 628 * In case any call fails, no further calls are made. 629 * 630 * @rdm: the #RamDiscardManager 631 * @section: the #MemoryRegionSection 632 * @replay_fn: the #ReplayRamPopulate callback 633 * @opaque: pointer to forward to the callback 634 * 635 * Returns 0 on success, or a negative error if any notification failed. 636 */ 637 int (*replay_populated)(const RamDiscardManager *rdm, 638 MemoryRegionSection *section, 639 ReplayRamPopulate replay_fn, void *opaque); 640 641 /** 642 * @register_listener: 643 * 644 * Register a #RamDiscardListener for the given #MemoryRegionSection and 645 * immediately notify the #RamDiscardListener about all populated parts 646 * within the #MemoryRegionSection via the #RamDiscardManager. 647 * 648 * In case any notification fails, no further notifications are triggered 649 * and an error is logged. 650 * 651 * @rdm: the #RamDiscardManager 652 * @rdl: the #RamDiscardListener 653 * @section: the #MemoryRegionSection 654 */ 655 void (*register_listener)(RamDiscardManager *rdm, 656 RamDiscardListener *rdl, 657 MemoryRegionSection *section); 658 659 /** 660 * @unregister_listener: 661 * 662 * Unregister a previously registered #RamDiscardListener via the 663 * #RamDiscardManager after notifying the #RamDiscardListener about all 664 * populated parts becoming unpopulated within the registered 665 * #MemoryRegionSection. 666 * 667 * @rdm: the #RamDiscardManager 668 * @rdl: the #RamDiscardListener 669 */ 670 void (*unregister_listener)(RamDiscardManager *rdm, 671 RamDiscardListener *rdl); 672 }; 673 674 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm, 675 const MemoryRegion *mr); 676 677 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm, 678 const MemoryRegionSection *section); 679 680 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm, 681 MemoryRegionSection *section, 682 ReplayRamPopulate replay_fn, 683 void *opaque); 684 685 void ram_discard_manager_register_listener(RamDiscardManager *rdm, 686 RamDiscardListener *rdl, 687 MemoryRegionSection *section); 688 689 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm, 690 RamDiscardListener *rdl); 691 692 typedef struct CoalescedMemoryRange CoalescedMemoryRange; 693 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd; 694 695 /** MemoryRegion: 696 * 697 * A struct representing a memory region. 698 */ 699 struct MemoryRegion { 700 Object parent_obj; 701 702 /* private: */ 703 704 /* The following fields should fit in a cache line */ 705 bool romd_mode; 706 bool ram; 707 bool subpage; 708 bool readonly; /* For RAM regions */ 709 bool nonvolatile; 710 bool rom_device; 711 bool flush_coalesced_mmio; 712 uint8_t dirty_log_mask; 713 bool is_iommu; 714 RAMBlock *ram_block; 715 Object *owner; 716 717 const MemoryRegionOps *ops; 718 void *opaque; 719 MemoryRegion *container; 720 Int128 size; 721 hwaddr addr; 722 void (*destructor)(MemoryRegion *mr); 723 uint64_t align; 724 bool terminates; 725 bool ram_device; 726 bool enabled; 727 bool warning_printed; /* For reservations */ 728 uint8_t vga_logging_count; 729 MemoryRegion *alias; 730 hwaddr alias_offset; 731 int32_t priority; 732 QTAILQ_HEAD(, MemoryRegion) subregions; 733 QTAILQ_ENTRY(MemoryRegion) subregions_link; 734 QTAILQ_HEAD(, CoalescedMemoryRange) coalesced; 735 const char *name; 736 unsigned ioeventfd_nb; 737 MemoryRegionIoeventfd *ioeventfds; 738 RamDiscardManager *rdm; /* Only for RAM */ 739 }; 740 741 struct IOMMUMemoryRegion { 742 MemoryRegion parent_obj; 743 744 QLIST_HEAD(, IOMMUNotifier) iommu_notify; 745 IOMMUNotifierFlag iommu_notify_flags; 746 }; 747 748 #define IOMMU_NOTIFIER_FOREACH(n, mr) \ 749 QLIST_FOREACH((n), &(mr)->iommu_notify, node) 750 751 /** 752 * struct MemoryListener: callbacks structure for updates to the physical memory map 753 * 754 * Allows a component to adjust to changes in the guest-visible memory map. 755 * Use with memory_listener_register() and memory_listener_unregister(). 756 */ 757 struct MemoryListener { 758 /** 759 * @begin: 760 * 761 * Called at the beginning of an address space update transaction. 762 * Followed by calls to #MemoryListener.region_add(), 763 * #MemoryListener.region_del(), #MemoryListener.region_nop(), 764 * #MemoryListener.log_start() and #MemoryListener.log_stop() in 765 * increasing address order. 766 * 767 * @listener: The #MemoryListener. 768 */ 769 void (*begin)(MemoryListener *listener); 770 771 /** 772 * @commit: 773 * 774 * Called at the end of an address space update transaction, 775 * after the last call to #MemoryListener.region_add(), 776 * #MemoryListener.region_del() or #MemoryListener.region_nop(), 777 * #MemoryListener.log_start() and #MemoryListener.log_stop(). 778 * 779 * @listener: The #MemoryListener. 780 */ 781 void (*commit)(MemoryListener *listener); 782 783 /** 784 * @region_add: 785 * 786 * Called during an address space update transaction, 787 * for a section of the address space that is new in this address space 788 * space since the last transaction. 789 * 790 * @listener: The #MemoryListener. 791 * @section: The new #MemoryRegionSection. 792 */ 793 void (*region_add)(MemoryListener *listener, MemoryRegionSection *section); 794 795 /** 796 * @region_del: 797 * 798 * Called during an address space update transaction, 799 * for a section of the address space that has disappeared in the address 800 * space since the last transaction. 801 * 802 * @listener: The #MemoryListener. 803 * @section: The old #MemoryRegionSection. 804 */ 805 void (*region_del)(MemoryListener *listener, MemoryRegionSection *section); 806 807 /** 808 * @region_nop: 809 * 810 * Called during an address space update transaction, 811 * for a section of the address space that is in the same place in the address 812 * space as in the last transaction. 813 * 814 * @listener: The #MemoryListener. 815 * @section: The #MemoryRegionSection. 816 */ 817 void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section); 818 819 /** 820 * @log_start: 821 * 822 * Called during an address space update transaction, after 823 * one of #MemoryListener.region_add(), #MemoryListener.region_del() or 824 * #MemoryListener.region_nop(), if dirty memory logging clients have 825 * become active since the last transaction. 826 * 827 * @listener: The #MemoryListener. 828 * @section: The #MemoryRegionSection. 829 * @old: A bitmap of dirty memory logging clients that were active in 830 * the previous transaction. 831 * @new: A bitmap of dirty memory logging clients that are active in 832 * the current transaction. 833 */ 834 void (*log_start)(MemoryListener *listener, MemoryRegionSection *section, 835 int old, int new); 836 837 /** 838 * @log_stop: 839 * 840 * Called during an address space update transaction, after 841 * one of #MemoryListener.region_add(), #MemoryListener.region_del() or 842 * #MemoryListener.region_nop() and possibly after 843 * #MemoryListener.log_start(), if dirty memory logging clients have 844 * become inactive since the last transaction. 845 * 846 * @listener: The #MemoryListener. 847 * @section: The #MemoryRegionSection. 848 * @old: A bitmap of dirty memory logging clients that were active in 849 * the previous transaction. 850 * @new: A bitmap of dirty memory logging clients that are active in 851 * the current transaction. 852 */ 853 void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section, 854 int old, int new); 855 856 /** 857 * @log_sync: 858 * 859 * Called by memory_region_snapshot_and_clear_dirty() and 860 * memory_global_dirty_log_sync(), before accessing QEMU's "official" 861 * copy of the dirty memory bitmap for a #MemoryRegionSection. 862 * 863 * @listener: The #MemoryListener. 864 * @section: The #MemoryRegionSection. 865 */ 866 void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section); 867 868 /** 869 * @log_sync_global: 870 * 871 * This is the global version of @log_sync when the listener does 872 * not have a way to synchronize the log with finer granularity. 873 * When the listener registers with @log_sync_global defined, then 874 * its @log_sync must be NULL. Vice versa. 875 * 876 * @listener: The #MemoryListener. 877 */ 878 void (*log_sync_global)(MemoryListener *listener); 879 880 /** 881 * @log_clear: 882 * 883 * Called before reading the dirty memory bitmap for a 884 * #MemoryRegionSection. 885 * 886 * @listener: The #MemoryListener. 887 * @section: The #MemoryRegionSection. 888 */ 889 void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section); 890 891 /** 892 * @log_global_start: 893 * 894 * Called by memory_global_dirty_log_start(), which 895 * enables the %DIRTY_LOG_MIGRATION client on all memory regions in 896 * the address space. #MemoryListener.log_global_start() is also 897 * called when a #MemoryListener is added, if global dirty logging is 898 * active at that time. 899 * 900 * @listener: The #MemoryListener. 901 */ 902 void (*log_global_start)(MemoryListener *listener); 903 904 /** 905 * @log_global_stop: 906 * 907 * Called by memory_global_dirty_log_stop(), which 908 * disables the %DIRTY_LOG_MIGRATION client on all memory regions in 909 * the address space. 910 * 911 * @listener: The #MemoryListener. 912 */ 913 void (*log_global_stop)(MemoryListener *listener); 914 915 /** 916 * @log_global_after_sync: 917 * 918 * Called after reading the dirty memory bitmap 919 * for any #MemoryRegionSection. 920 * 921 * @listener: The #MemoryListener. 922 */ 923 void (*log_global_after_sync)(MemoryListener *listener); 924 925 /** 926 * @eventfd_add: 927 * 928 * Called during an address space update transaction, 929 * for a section of the address space that has had a new ioeventfd 930 * registration since the last transaction. 931 * 932 * @listener: The #MemoryListener. 933 * @section: The new #MemoryRegionSection. 934 * @match_data: The @match_data parameter for the new ioeventfd. 935 * @data: The @data parameter for the new ioeventfd. 936 * @e: The #EventNotifier parameter for the new ioeventfd. 937 */ 938 void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section, 939 bool match_data, uint64_t data, EventNotifier *e); 940 941 /** 942 * @eventfd_del: 943 * 944 * Called during an address space update transaction, 945 * for a section of the address space that has dropped an ioeventfd 946 * registration since the last transaction. 947 * 948 * @listener: The #MemoryListener. 949 * @section: The new #MemoryRegionSection. 950 * @match_data: The @match_data parameter for the dropped ioeventfd. 951 * @data: The @data parameter for the dropped ioeventfd. 952 * @e: The #EventNotifier parameter for the dropped ioeventfd. 953 */ 954 void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section, 955 bool match_data, uint64_t data, EventNotifier *e); 956 957 /** 958 * @coalesced_io_add: 959 * 960 * Called during an address space update transaction, 961 * for a section of the address space that has had a new coalesced 962 * MMIO range registration since the last transaction. 963 * 964 * @listener: The #MemoryListener. 965 * @section: The new #MemoryRegionSection. 966 * @addr: The starting address for the coalesced MMIO range. 967 * @len: The length of the coalesced MMIO range. 968 */ 969 void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section, 970 hwaddr addr, hwaddr len); 971 972 /** 973 * @coalesced_io_del: 974 * 975 * Called during an address space update transaction, 976 * for a section of the address space that has dropped a coalesced 977 * MMIO range since the last transaction. 978 * 979 * @listener: The #MemoryListener. 980 * @section: The new #MemoryRegionSection. 981 * @addr: The starting address for the coalesced MMIO range. 982 * @len: The length of the coalesced MMIO range. 983 */ 984 void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section, 985 hwaddr addr, hwaddr len); 986 /** 987 * @priority: 988 * 989 * Govern the order in which memory listeners are invoked. Lower priorities 990 * are invoked earlier for "add" or "start" callbacks, and later for "delete" 991 * or "stop" callbacks. 992 */ 993 unsigned priority; 994 995 /** 996 * @name: 997 * 998 * Name of the listener. It can be used in contexts where we'd like to 999 * identify one memory listener with the rest. 1000 */ 1001 const char *name; 1002 1003 /* private: */ 1004 AddressSpace *address_space; 1005 QTAILQ_ENTRY(MemoryListener) link; 1006 QTAILQ_ENTRY(MemoryListener) link_as; 1007 }; 1008 1009 /** 1010 * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects 1011 */ 1012 struct AddressSpace { 1013 /* private: */ 1014 struct rcu_head rcu; 1015 char *name; 1016 MemoryRegion *root; 1017 1018 /* Accessed via RCU. */ 1019 struct FlatView *current_map; 1020 1021 int ioeventfd_nb; 1022 struct MemoryRegionIoeventfd *ioeventfds; 1023 QTAILQ_HEAD(, MemoryListener) listeners; 1024 QTAILQ_ENTRY(AddressSpace) address_spaces_link; 1025 }; 1026 1027 typedef struct AddressSpaceDispatch AddressSpaceDispatch; 1028 typedef struct FlatRange FlatRange; 1029 1030 /* Flattened global view of current active memory hierarchy. Kept in sorted 1031 * order. 1032 */ 1033 struct FlatView { 1034 struct rcu_head rcu; 1035 unsigned ref; 1036 FlatRange *ranges; 1037 unsigned nr; 1038 unsigned nr_allocated; 1039 struct AddressSpaceDispatch *dispatch; 1040 MemoryRegion *root; 1041 }; 1042 1043 static inline FlatView *address_space_to_flatview(AddressSpace *as) 1044 { 1045 return qatomic_rcu_read(&as->current_map); 1046 } 1047 1048 /** 1049 * typedef flatview_cb: callback for flatview_for_each_range() 1050 * 1051 * @start: start address of the range within the FlatView 1052 * @len: length of the range in bytes 1053 * @mr: MemoryRegion covering this range 1054 * @offset_in_region: offset of the first byte of the range within @mr 1055 * @opaque: data pointer passed to flatview_for_each_range() 1056 * 1057 * Returns: true to stop the iteration, false to keep going. 1058 */ 1059 typedef bool (*flatview_cb)(Int128 start, 1060 Int128 len, 1061 const MemoryRegion *mr, 1062 hwaddr offset_in_region, 1063 void *opaque); 1064 1065 /** 1066 * flatview_for_each_range: Iterate through a FlatView 1067 * @fv: the FlatView to iterate through 1068 * @cb: function to call for each range 1069 * @opaque: opaque data pointer to pass to @cb 1070 * 1071 * A FlatView is made up of a list of non-overlapping ranges, each of 1072 * which is a slice of a MemoryRegion. This function iterates through 1073 * each range in @fv, calling @cb. The callback function can terminate 1074 * iteration early by returning 'true'. 1075 */ 1076 void flatview_for_each_range(FlatView *fv, flatview_cb cb, void *opaque); 1077 1078 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a, 1079 MemoryRegionSection *b) 1080 { 1081 return a->mr == b->mr && 1082 a->fv == b->fv && 1083 a->offset_within_region == b->offset_within_region && 1084 a->offset_within_address_space == b->offset_within_address_space && 1085 int128_eq(a->size, b->size) && 1086 a->readonly == b->readonly && 1087 a->nonvolatile == b->nonvolatile; 1088 } 1089 1090 /** 1091 * memory_region_section_new_copy: Copy a memory region section 1092 * 1093 * Allocate memory for a new copy, copy the memory region section, and 1094 * properly take a reference on all relevant members. 1095 * 1096 * @s: the #MemoryRegionSection to copy 1097 */ 1098 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s); 1099 1100 /** 1101 * memory_region_section_new_copy: Free a copied memory region section 1102 * 1103 * Free a copy of a memory section created via memory_region_section_new_copy(). 1104 * properly dropping references on all relevant members. 1105 * 1106 * @s: the #MemoryRegionSection to copy 1107 */ 1108 void memory_region_section_free_copy(MemoryRegionSection *s); 1109 1110 /** 1111 * memory_region_init: Initialize a memory region 1112 * 1113 * The region typically acts as a container for other memory regions. Use 1114 * memory_region_add_subregion() to add subregions. 1115 * 1116 * @mr: the #MemoryRegion to be initialized 1117 * @owner: the object that tracks the region's reference count 1118 * @name: used for debugging; not visible to the user or ABI 1119 * @size: size of the region; any subregions beyond this size will be clipped 1120 */ 1121 void memory_region_init(MemoryRegion *mr, 1122 Object *owner, 1123 const char *name, 1124 uint64_t size); 1125 1126 /** 1127 * memory_region_ref: Add 1 to a memory region's reference count 1128 * 1129 * Whenever memory regions are accessed outside the BQL, they need to be 1130 * preserved against hot-unplug. MemoryRegions actually do not have their 1131 * own reference count; they piggyback on a QOM object, their "owner". 1132 * This function adds a reference to the owner. 1133 * 1134 * All MemoryRegions must have an owner if they can disappear, even if the 1135 * device they belong to operates exclusively under the BQL. This is because 1136 * the region could be returned at any time by memory_region_find, and this 1137 * is usually under guest control. 1138 * 1139 * @mr: the #MemoryRegion 1140 */ 1141 void memory_region_ref(MemoryRegion *mr); 1142 1143 /** 1144 * memory_region_unref: Remove 1 to a memory region's reference count 1145 * 1146 * Whenever memory regions are accessed outside the BQL, they need to be 1147 * preserved against hot-unplug. MemoryRegions actually do not have their 1148 * own reference count; they piggyback on a QOM object, their "owner". 1149 * This function removes a reference to the owner and possibly destroys it. 1150 * 1151 * @mr: the #MemoryRegion 1152 */ 1153 void memory_region_unref(MemoryRegion *mr); 1154 1155 /** 1156 * memory_region_init_io: Initialize an I/O memory region. 1157 * 1158 * Accesses into the region will cause the callbacks in @ops to be called. 1159 * if @size is nonzero, subregions will be clipped to @size. 1160 * 1161 * @mr: the #MemoryRegion to be initialized. 1162 * @owner: the object that tracks the region's reference count 1163 * @ops: a structure containing read and write callbacks to be used when 1164 * I/O is performed on the region. 1165 * @opaque: passed to the read and write callbacks of the @ops structure. 1166 * @name: used for debugging; not visible to the user or ABI 1167 * @size: size of the region. 1168 */ 1169 void memory_region_init_io(MemoryRegion *mr, 1170 Object *owner, 1171 const MemoryRegionOps *ops, 1172 void *opaque, 1173 const char *name, 1174 uint64_t size); 1175 1176 /** 1177 * memory_region_init_ram_nomigrate: Initialize RAM memory region. Accesses 1178 * into the region will modify memory 1179 * directly. 1180 * 1181 * @mr: the #MemoryRegion to be initialized. 1182 * @owner: the object that tracks the region's reference count 1183 * @name: Region name, becomes part of RAMBlock name used in migration stream 1184 * must be unique within any device 1185 * @size: size of the region. 1186 * @errp: pointer to Error*, to store an error if it happens. 1187 * 1188 * Note that this function does not do anything to cause the data in the 1189 * RAM memory region to be migrated; that is the responsibility of the caller. 1190 */ 1191 void memory_region_init_ram_nomigrate(MemoryRegion *mr, 1192 Object *owner, 1193 const char *name, 1194 uint64_t size, 1195 Error **errp); 1196 1197 /** 1198 * memory_region_init_ram_flags_nomigrate: Initialize RAM memory region. 1199 * Accesses into the region will 1200 * modify memory directly. 1201 * 1202 * @mr: the #MemoryRegion to be initialized. 1203 * @owner: the object that tracks the region's reference count 1204 * @name: Region name, becomes part of RAMBlock name used in migration stream 1205 * must be unique within any device 1206 * @size: size of the region. 1207 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_NORESERVE. 1208 * @errp: pointer to Error*, to store an error if it happens. 1209 * 1210 * Note that this function does not do anything to cause the data in the 1211 * RAM memory region to be migrated; that is the responsibility of the caller. 1212 */ 1213 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr, 1214 Object *owner, 1215 const char *name, 1216 uint64_t size, 1217 uint32_t ram_flags, 1218 Error **errp); 1219 1220 /** 1221 * memory_region_init_resizeable_ram: Initialize memory region with resizeable 1222 * RAM. Accesses into the region will 1223 * modify memory directly. Only an initial 1224 * portion of this RAM is actually used. 1225 * Changing the size while migrating 1226 * can result in the migration being 1227 * canceled. 1228 * 1229 * @mr: the #MemoryRegion to be initialized. 1230 * @owner: the object that tracks the region's reference count 1231 * @name: Region name, becomes part of RAMBlock name used in migration stream 1232 * must be unique within any device 1233 * @size: used size of the region. 1234 * @max_size: max size of the region. 1235 * @resized: callback to notify owner about used size change. 1236 * @errp: pointer to Error*, to store an error if it happens. 1237 * 1238 * Note that this function does not do anything to cause the data in the 1239 * RAM memory region to be migrated; that is the responsibility of the caller. 1240 */ 1241 void memory_region_init_resizeable_ram(MemoryRegion *mr, 1242 Object *owner, 1243 const char *name, 1244 uint64_t size, 1245 uint64_t max_size, 1246 void (*resized)(const char*, 1247 uint64_t length, 1248 void *host), 1249 Error **errp); 1250 #ifdef CONFIG_POSIX 1251 1252 /** 1253 * memory_region_init_ram_from_file: Initialize RAM memory region with a 1254 * mmap-ed backend. 1255 * 1256 * @mr: the #MemoryRegion to be initialized. 1257 * @owner: the object that tracks the region's reference count 1258 * @name: Region name, becomes part of RAMBlock name used in migration stream 1259 * must be unique within any device 1260 * @size: size of the region. 1261 * @align: alignment of the region base address; if 0, the default alignment 1262 * (getpagesize()) will be used. 1263 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM, 1264 * RAM_NORESERVE, 1265 * @path: the path in which to allocate the RAM. 1266 * @readonly: true to open @path for reading, false for read/write. 1267 * @errp: pointer to Error*, to store an error if it happens. 1268 * 1269 * Note that this function does not do anything to cause the data in the 1270 * RAM memory region to be migrated; that is the responsibility of the caller. 1271 */ 1272 void memory_region_init_ram_from_file(MemoryRegion *mr, 1273 Object *owner, 1274 const char *name, 1275 uint64_t size, 1276 uint64_t align, 1277 uint32_t ram_flags, 1278 const char *path, 1279 bool readonly, 1280 Error **errp); 1281 1282 /** 1283 * memory_region_init_ram_from_fd: Initialize RAM memory region with a 1284 * mmap-ed backend. 1285 * 1286 * @mr: the #MemoryRegion to be initialized. 1287 * @owner: the object that tracks the region's reference count 1288 * @name: the name of the region. 1289 * @size: size of the region. 1290 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM, 1291 * RAM_NORESERVE, RAM_PROTECTED. 1292 * @fd: the fd to mmap. 1293 * @offset: offset within the file referenced by fd 1294 * @errp: pointer to Error*, to store an error if it happens. 1295 * 1296 * Note that this function does not do anything to cause the data in the 1297 * RAM memory region to be migrated; that is the responsibility of the caller. 1298 */ 1299 void memory_region_init_ram_from_fd(MemoryRegion *mr, 1300 Object *owner, 1301 const char *name, 1302 uint64_t size, 1303 uint32_t ram_flags, 1304 int fd, 1305 ram_addr_t offset, 1306 Error **errp); 1307 #endif 1308 1309 /** 1310 * memory_region_init_ram_ptr: Initialize RAM memory region from a 1311 * user-provided pointer. Accesses into the 1312 * region will modify memory directly. 1313 * 1314 * @mr: the #MemoryRegion to be initialized. 1315 * @owner: the object that tracks the region's reference count 1316 * @name: Region name, becomes part of RAMBlock name used in migration stream 1317 * must be unique within any device 1318 * @size: size of the region. 1319 * @ptr: memory to be mapped; must contain at least @size bytes. 1320 * 1321 * Note that this function does not do anything to cause the data in the 1322 * RAM memory region to be migrated; that is the responsibility of the caller. 1323 */ 1324 void memory_region_init_ram_ptr(MemoryRegion *mr, 1325 Object *owner, 1326 const char *name, 1327 uint64_t size, 1328 void *ptr); 1329 1330 /** 1331 * memory_region_init_ram_device_ptr: Initialize RAM device memory region from 1332 * a user-provided pointer. 1333 * 1334 * A RAM device represents a mapping to a physical device, such as to a PCI 1335 * MMIO BAR of an vfio-pci assigned device. The memory region may be mapped 1336 * into the VM address space and access to the region will modify memory 1337 * directly. However, the memory region should not be included in a memory 1338 * dump (device may not be enabled/mapped at the time of the dump), and 1339 * operations incompatible with manipulating MMIO should be avoided. Replaces 1340 * skip_dump flag. 1341 * 1342 * @mr: the #MemoryRegion to be initialized. 1343 * @owner: the object that tracks the region's reference count 1344 * @name: the name of the region. 1345 * @size: size of the region. 1346 * @ptr: memory to be mapped; must contain at least @size bytes. 1347 * 1348 * Note that this function does not do anything to cause the data in the 1349 * RAM memory region to be migrated; that is the responsibility of the caller. 1350 * (For RAM device memory regions, migrating the contents rarely makes sense.) 1351 */ 1352 void memory_region_init_ram_device_ptr(MemoryRegion *mr, 1353 Object *owner, 1354 const char *name, 1355 uint64_t size, 1356 void *ptr); 1357 1358 /** 1359 * memory_region_init_alias: Initialize a memory region that aliases all or a 1360 * part of another memory region. 1361 * 1362 * @mr: the #MemoryRegion to be initialized. 1363 * @owner: the object that tracks the region's reference count 1364 * @name: used for debugging; not visible to the user or ABI 1365 * @orig: the region to be referenced; @mr will be equivalent to 1366 * @orig between @offset and @offset + @size - 1. 1367 * @offset: start of the section in @orig to be referenced. 1368 * @size: size of the region. 1369 */ 1370 void memory_region_init_alias(MemoryRegion *mr, 1371 Object *owner, 1372 const char *name, 1373 MemoryRegion *orig, 1374 hwaddr offset, 1375 uint64_t size); 1376 1377 /** 1378 * memory_region_init_rom_nomigrate: Initialize a ROM memory region. 1379 * 1380 * This has the same effect as calling memory_region_init_ram_nomigrate() 1381 * and then marking the resulting region read-only with 1382 * memory_region_set_readonly(). 1383 * 1384 * Note that this function does not do anything to cause the data in the 1385 * RAM side of the memory region to be migrated; that is the responsibility 1386 * of the caller. 1387 * 1388 * @mr: the #MemoryRegion to be initialized. 1389 * @owner: the object that tracks the region's reference count 1390 * @name: Region name, becomes part of RAMBlock name used in migration stream 1391 * must be unique within any device 1392 * @size: size of the region. 1393 * @errp: pointer to Error*, to store an error if it happens. 1394 */ 1395 void memory_region_init_rom_nomigrate(MemoryRegion *mr, 1396 Object *owner, 1397 const char *name, 1398 uint64_t size, 1399 Error **errp); 1400 1401 /** 1402 * memory_region_init_rom_device_nomigrate: Initialize a ROM memory region. 1403 * Writes are handled via callbacks. 1404 * 1405 * Note that this function does not do anything to cause the data in the 1406 * RAM side of the memory region to be migrated; that is the responsibility 1407 * of the caller. 1408 * 1409 * @mr: the #MemoryRegion to be initialized. 1410 * @owner: the object that tracks the region's reference count 1411 * @ops: callbacks for write access handling (must not be NULL). 1412 * @opaque: passed to the read and write callbacks of the @ops structure. 1413 * @name: Region name, becomes part of RAMBlock name used in migration stream 1414 * must be unique within any device 1415 * @size: size of the region. 1416 * @errp: pointer to Error*, to store an error if it happens. 1417 */ 1418 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr, 1419 Object *owner, 1420 const MemoryRegionOps *ops, 1421 void *opaque, 1422 const char *name, 1423 uint64_t size, 1424 Error **errp); 1425 1426 /** 1427 * memory_region_init_iommu: Initialize a memory region of a custom type 1428 * that translates addresses 1429 * 1430 * An IOMMU region translates addresses and forwards accesses to a target 1431 * memory region. 1432 * 1433 * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION. 1434 * @_iommu_mr should be a pointer to enough memory for an instance of 1435 * that subclass, @instance_size is the size of that subclass, and 1436 * @mrtypename is its name. This function will initialize @_iommu_mr as an 1437 * instance of the subclass, and its methods will then be called to handle 1438 * accesses to the memory region. See the documentation of 1439 * #IOMMUMemoryRegionClass for further details. 1440 * 1441 * @_iommu_mr: the #IOMMUMemoryRegion to be initialized 1442 * @instance_size: the IOMMUMemoryRegion subclass instance size 1443 * @mrtypename: the type name of the #IOMMUMemoryRegion 1444 * @owner: the object that tracks the region's reference count 1445 * @name: used for debugging; not visible to the user or ABI 1446 * @size: size of the region. 1447 */ 1448 void memory_region_init_iommu(void *_iommu_mr, 1449 size_t instance_size, 1450 const char *mrtypename, 1451 Object *owner, 1452 const char *name, 1453 uint64_t size); 1454 1455 /** 1456 * memory_region_init_ram - Initialize RAM memory region. Accesses into the 1457 * region will modify memory directly. 1458 * 1459 * @mr: the #MemoryRegion to be initialized 1460 * @owner: the object that tracks the region's reference count (must be 1461 * TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL) 1462 * @name: name of the memory region 1463 * @size: size of the region in bytes 1464 * @errp: pointer to Error*, to store an error if it happens. 1465 * 1466 * This function allocates RAM for a board model or device, and 1467 * arranges for it to be migrated (by calling vmstate_register_ram() 1468 * if @owner is a DeviceState, or vmstate_register_ram_global() if 1469 * @owner is NULL). 1470 * 1471 * TODO: Currently we restrict @owner to being either NULL (for 1472 * global RAM regions with no owner) or devices, so that we can 1473 * give the RAM block a unique name for migration purposes. 1474 * We should lift this restriction and allow arbitrary Objects. 1475 * If you pass a non-NULL non-device @owner then we will assert. 1476 */ 1477 void memory_region_init_ram(MemoryRegion *mr, 1478 Object *owner, 1479 const char *name, 1480 uint64_t size, 1481 Error **errp); 1482 1483 /** 1484 * memory_region_init_rom: Initialize a ROM memory region. 1485 * 1486 * This has the same effect as calling memory_region_init_ram() 1487 * and then marking the resulting region read-only with 1488 * memory_region_set_readonly(). This includes arranging for the 1489 * contents to be migrated. 1490 * 1491 * TODO: Currently we restrict @owner to being either NULL (for 1492 * global RAM regions with no owner) or devices, so that we can 1493 * give the RAM block a unique name for migration purposes. 1494 * We should lift this restriction and allow arbitrary Objects. 1495 * If you pass a non-NULL non-device @owner then we will assert. 1496 * 1497 * @mr: the #MemoryRegion to be initialized. 1498 * @owner: the object that tracks the region's reference count 1499 * @name: Region name, becomes part of RAMBlock name used in migration stream 1500 * must be unique within any device 1501 * @size: size of the region. 1502 * @errp: pointer to Error*, to store an error if it happens. 1503 */ 1504 void memory_region_init_rom(MemoryRegion *mr, 1505 Object *owner, 1506 const char *name, 1507 uint64_t size, 1508 Error **errp); 1509 1510 /** 1511 * memory_region_init_rom_device: Initialize a ROM memory region. 1512 * Writes are handled via callbacks. 1513 * 1514 * This function initializes a memory region backed by RAM for reads 1515 * and callbacks for writes, and arranges for the RAM backing to 1516 * be migrated (by calling vmstate_register_ram() 1517 * if @owner is a DeviceState, or vmstate_register_ram_global() if 1518 * @owner is NULL). 1519 * 1520 * TODO: Currently we restrict @owner to being either NULL (for 1521 * global RAM regions with no owner) or devices, so that we can 1522 * give the RAM block a unique name for migration purposes. 1523 * We should lift this restriction and allow arbitrary Objects. 1524 * If you pass a non-NULL non-device @owner then we will assert. 1525 * 1526 * @mr: the #MemoryRegion to be initialized. 1527 * @owner: the object that tracks the region's reference count 1528 * @ops: callbacks for write access handling (must not be NULL). 1529 * @opaque: passed to the read and write callbacks of the @ops structure. 1530 * @name: Region name, becomes part of RAMBlock name used in migration stream 1531 * must be unique within any device 1532 * @size: size of the region. 1533 * @errp: pointer to Error*, to store an error if it happens. 1534 */ 1535 void memory_region_init_rom_device(MemoryRegion *mr, 1536 Object *owner, 1537 const MemoryRegionOps *ops, 1538 void *opaque, 1539 const char *name, 1540 uint64_t size, 1541 Error **errp); 1542 1543 1544 /** 1545 * memory_region_owner: get a memory region's owner. 1546 * 1547 * @mr: the memory region being queried. 1548 */ 1549 Object *memory_region_owner(MemoryRegion *mr); 1550 1551 /** 1552 * memory_region_size: get a memory region's size. 1553 * 1554 * @mr: the memory region being queried. 1555 */ 1556 uint64_t memory_region_size(MemoryRegion *mr); 1557 1558 /** 1559 * memory_region_is_ram: check whether a memory region is random access 1560 * 1561 * Returns %true if a memory region is random access. 1562 * 1563 * @mr: the memory region being queried 1564 */ 1565 static inline bool memory_region_is_ram(MemoryRegion *mr) 1566 { 1567 return mr->ram; 1568 } 1569 1570 /** 1571 * memory_region_is_ram_device: check whether a memory region is a ram device 1572 * 1573 * Returns %true if a memory region is a device backed ram region 1574 * 1575 * @mr: the memory region being queried 1576 */ 1577 bool memory_region_is_ram_device(MemoryRegion *mr); 1578 1579 /** 1580 * memory_region_is_romd: check whether a memory region is in ROMD mode 1581 * 1582 * Returns %true if a memory region is a ROM device and currently set to allow 1583 * direct reads. 1584 * 1585 * @mr: the memory region being queried 1586 */ 1587 static inline bool memory_region_is_romd(MemoryRegion *mr) 1588 { 1589 return mr->rom_device && mr->romd_mode; 1590 } 1591 1592 /** 1593 * memory_region_is_protected: check whether a memory region is protected 1594 * 1595 * Returns %true if a memory region is protected RAM and cannot be accessed 1596 * via standard mechanisms, e.g. DMA. 1597 * 1598 * @mr: the memory region being queried 1599 */ 1600 bool memory_region_is_protected(MemoryRegion *mr); 1601 1602 /** 1603 * memory_region_get_iommu: check whether a memory region is an iommu 1604 * 1605 * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu, 1606 * otherwise NULL. 1607 * 1608 * @mr: the memory region being queried 1609 */ 1610 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr) 1611 { 1612 if (mr->alias) { 1613 return memory_region_get_iommu(mr->alias); 1614 } 1615 if (mr->is_iommu) { 1616 return (IOMMUMemoryRegion *) mr; 1617 } 1618 return NULL; 1619 } 1620 1621 /** 1622 * memory_region_get_iommu_class_nocheck: returns iommu memory region class 1623 * if an iommu or NULL if not 1624 * 1625 * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu, 1626 * otherwise NULL. This is fast path avoiding QOM checking, use with caution. 1627 * 1628 * @iommu_mr: the memory region being queried 1629 */ 1630 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck( 1631 IOMMUMemoryRegion *iommu_mr) 1632 { 1633 return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class); 1634 } 1635 1636 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL) 1637 1638 /** 1639 * memory_region_iommu_get_min_page_size: get minimum supported page size 1640 * for an iommu 1641 * 1642 * Returns minimum supported page size for an iommu. 1643 * 1644 * @iommu_mr: the memory region being queried 1645 */ 1646 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr); 1647 1648 /** 1649 * memory_region_notify_iommu: notify a change in an IOMMU translation entry. 1650 * 1651 * Note: for any IOMMU implementation, an in-place mapping change 1652 * should be notified with an UNMAP followed by a MAP. 1653 * 1654 * @iommu_mr: the memory region that was changed 1655 * @iommu_idx: the IOMMU index for the translation table which has changed 1656 * @event: TLB event with the new entry in the IOMMU translation table. 1657 * The entry replaces all old entries for the same virtual I/O address 1658 * range. 1659 */ 1660 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr, 1661 int iommu_idx, 1662 IOMMUTLBEvent event); 1663 1664 /** 1665 * memory_region_notify_iommu_one: notify a change in an IOMMU translation 1666 * entry to a single notifier 1667 * 1668 * This works just like memory_region_notify_iommu(), but it only 1669 * notifies a specific notifier, not all of them. 1670 * 1671 * @notifier: the notifier to be notified 1672 * @event: TLB event with the new entry in the IOMMU translation table. 1673 * The entry replaces all old entries for the same virtual I/O address 1674 * range. 1675 */ 1676 void memory_region_notify_iommu_one(IOMMUNotifier *notifier, 1677 IOMMUTLBEvent *event); 1678 1679 /** 1680 * memory_region_register_iommu_notifier: register a notifier for changes to 1681 * IOMMU translation entries. 1682 * 1683 * Returns 0 on success, or a negative errno otherwise. In particular, 1684 * -EINVAL indicates that at least one of the attributes of the notifier 1685 * is not supported (flag/range) by the IOMMU memory region. In case of error 1686 * the error object must be created. 1687 * 1688 * @mr: the memory region to observe 1689 * @n: the IOMMUNotifier to be added; the notify callback receives a 1690 * pointer to an #IOMMUTLBEntry as the opaque value; the pointer 1691 * ceases to be valid on exit from the notifier. 1692 * @errp: pointer to Error*, to store an error if it happens. 1693 */ 1694 int memory_region_register_iommu_notifier(MemoryRegion *mr, 1695 IOMMUNotifier *n, Error **errp); 1696 1697 /** 1698 * memory_region_iommu_replay: replay existing IOMMU translations to 1699 * a notifier with the minimum page granularity returned by 1700 * mr->iommu_ops->get_page_size(). 1701 * 1702 * Note: this is not related to record-and-replay functionality. 1703 * 1704 * @iommu_mr: the memory region to observe 1705 * @n: the notifier to which to replay iommu mappings 1706 */ 1707 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n); 1708 1709 /** 1710 * memory_region_unregister_iommu_notifier: unregister a notifier for 1711 * changes to IOMMU translation entries. 1712 * 1713 * @mr: the memory region which was observed and for which notity_stopped() 1714 * needs to be called 1715 * @n: the notifier to be removed. 1716 */ 1717 void memory_region_unregister_iommu_notifier(MemoryRegion *mr, 1718 IOMMUNotifier *n); 1719 1720 /** 1721 * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is 1722 * defined on the IOMMU. 1723 * 1724 * Returns 0 on success, or a negative errno otherwise. In particular, 1725 * -EINVAL indicates that the IOMMU does not support the requested 1726 * attribute. 1727 * 1728 * @iommu_mr: the memory region 1729 * @attr: the requested attribute 1730 * @data: a pointer to the requested attribute data 1731 */ 1732 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr, 1733 enum IOMMUMemoryRegionAttr attr, 1734 void *data); 1735 1736 /** 1737 * memory_region_iommu_attrs_to_index: return the IOMMU index to 1738 * use for translations with the given memory transaction attributes. 1739 * 1740 * @iommu_mr: the memory region 1741 * @attrs: the memory transaction attributes 1742 */ 1743 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr, 1744 MemTxAttrs attrs); 1745 1746 /** 1747 * memory_region_iommu_num_indexes: return the total number of IOMMU 1748 * indexes that this IOMMU supports. 1749 * 1750 * @iommu_mr: the memory region 1751 */ 1752 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr); 1753 1754 /** 1755 * memory_region_iommu_set_page_size_mask: set the supported page 1756 * sizes for a given IOMMU memory region 1757 * 1758 * @iommu_mr: IOMMU memory region 1759 * @page_size_mask: supported page size mask 1760 * @errp: pointer to Error*, to store an error if it happens. 1761 */ 1762 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr, 1763 uint64_t page_size_mask, 1764 Error **errp); 1765 1766 /** 1767 * memory_region_name: get a memory region's name 1768 * 1769 * Returns the string that was used to initialize the memory region. 1770 * 1771 * @mr: the memory region being queried 1772 */ 1773 const char *memory_region_name(const MemoryRegion *mr); 1774 1775 /** 1776 * memory_region_is_logging: return whether a memory region is logging writes 1777 * 1778 * Returns %true if the memory region is logging writes for the given client 1779 * 1780 * @mr: the memory region being queried 1781 * @client: the client being queried 1782 */ 1783 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client); 1784 1785 /** 1786 * memory_region_get_dirty_log_mask: return the clients for which a 1787 * memory region is logging writes. 1788 * 1789 * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants 1790 * are the bit indices. 1791 * 1792 * @mr: the memory region being queried 1793 */ 1794 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr); 1795 1796 /** 1797 * memory_region_is_rom: check whether a memory region is ROM 1798 * 1799 * Returns %true if a memory region is read-only memory. 1800 * 1801 * @mr: the memory region being queried 1802 */ 1803 static inline bool memory_region_is_rom(MemoryRegion *mr) 1804 { 1805 return mr->ram && mr->readonly; 1806 } 1807 1808 /** 1809 * memory_region_is_nonvolatile: check whether a memory region is non-volatile 1810 * 1811 * Returns %true is a memory region is non-volatile memory. 1812 * 1813 * @mr: the memory region being queried 1814 */ 1815 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr) 1816 { 1817 return mr->nonvolatile; 1818 } 1819 1820 /** 1821 * memory_region_get_fd: Get a file descriptor backing a RAM memory region. 1822 * 1823 * Returns a file descriptor backing a file-based RAM memory region, 1824 * or -1 if the region is not a file-based RAM memory region. 1825 * 1826 * @mr: the RAM or alias memory region being queried. 1827 */ 1828 int memory_region_get_fd(MemoryRegion *mr); 1829 1830 /** 1831 * memory_region_from_host: Convert a pointer into a RAM memory region 1832 * and an offset within it. 1833 * 1834 * Given a host pointer inside a RAM memory region (created with 1835 * memory_region_init_ram() or memory_region_init_ram_ptr()), return 1836 * the MemoryRegion and the offset within it. 1837 * 1838 * Use with care; by the time this function returns, the returned pointer is 1839 * not protected by RCU anymore. If the caller is not within an RCU critical 1840 * section and does not hold the iothread lock, it must have other means of 1841 * protecting the pointer, such as a reference to the region that includes 1842 * the incoming ram_addr_t. 1843 * 1844 * @ptr: the host pointer to be converted 1845 * @offset: the offset within memory region 1846 */ 1847 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset); 1848 1849 /** 1850 * memory_region_get_ram_ptr: Get a pointer into a RAM memory region. 1851 * 1852 * Returns a host pointer to a RAM memory region (created with 1853 * memory_region_init_ram() or memory_region_init_ram_ptr()). 1854 * 1855 * Use with care; by the time this function returns, the returned pointer is 1856 * not protected by RCU anymore. If the caller is not within an RCU critical 1857 * section and does not hold the iothread lock, it must have other means of 1858 * protecting the pointer, such as a reference to the region that includes 1859 * the incoming ram_addr_t. 1860 * 1861 * @mr: the memory region being queried. 1862 */ 1863 void *memory_region_get_ram_ptr(MemoryRegion *mr); 1864 1865 /* memory_region_ram_resize: Resize a RAM region. 1866 * 1867 * Resizing RAM while migrating can result in the migration being canceled. 1868 * Care has to be taken if the guest might have already detected the memory. 1869 * 1870 * @mr: a memory region created with @memory_region_init_resizeable_ram. 1871 * @newsize: the new size the region 1872 * @errp: pointer to Error*, to store an error if it happens. 1873 */ 1874 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, 1875 Error **errp); 1876 1877 /** 1878 * memory_region_msync: Synchronize selected address range of 1879 * a memory mapped region 1880 * 1881 * @mr: the memory region to be msync 1882 * @addr: the initial address of the range to be sync 1883 * @size: the size of the range to be sync 1884 */ 1885 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size); 1886 1887 /** 1888 * memory_region_writeback: Trigger cache writeback for 1889 * selected address range 1890 * 1891 * @mr: the memory region to be updated 1892 * @addr: the initial address of the range to be written back 1893 * @size: the size of the range to be written back 1894 */ 1895 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size); 1896 1897 /** 1898 * memory_region_set_log: Turn dirty logging on or off for a region. 1899 * 1900 * Turns dirty logging on or off for a specified client (display, migration). 1901 * Only meaningful for RAM regions. 1902 * 1903 * @mr: the memory region being updated. 1904 * @log: whether dirty logging is to be enabled or disabled. 1905 * @client: the user of the logging information; %DIRTY_MEMORY_VGA only. 1906 */ 1907 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client); 1908 1909 /** 1910 * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region. 1911 * 1912 * Marks a range of bytes as dirty, after it has been dirtied outside 1913 * guest code. 1914 * 1915 * @mr: the memory region being dirtied. 1916 * @addr: the address (relative to the start of the region) being dirtied. 1917 * @size: size of the range being dirtied. 1918 */ 1919 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr, 1920 hwaddr size); 1921 1922 /** 1923 * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range 1924 * 1925 * This function is called when the caller wants to clear the remote 1926 * dirty bitmap of a memory range within the memory region. This can 1927 * be used by e.g. KVM to manually clear dirty log when 1928 * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host 1929 * kernel. 1930 * 1931 * @mr: the memory region to clear the dirty log upon 1932 * @start: start address offset within the memory region 1933 * @len: length of the memory region to clear dirty bitmap 1934 */ 1935 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start, 1936 hwaddr len); 1937 1938 /** 1939 * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty 1940 * bitmap and clear it. 1941 * 1942 * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and 1943 * returns the snapshot. The snapshot can then be used to query dirty 1944 * status, using memory_region_snapshot_get_dirty. Snapshotting allows 1945 * querying the same page multiple times, which is especially useful for 1946 * display updates where the scanlines often are not page aligned. 1947 * 1948 * The dirty bitmap region which gets copyed into the snapshot (and 1949 * cleared afterwards) can be larger than requested. The boundaries 1950 * are rounded up/down so complete bitmap longs (covering 64 pages on 1951 * 64bit hosts) can be copied over into the bitmap snapshot. Which 1952 * isn't a problem for display updates as the extra pages are outside 1953 * the visible area, and in case the visible area changes a full 1954 * display redraw is due anyway. Should other use cases for this 1955 * function emerge we might have to revisit this implementation 1956 * detail. 1957 * 1958 * Use g_free to release DirtyBitmapSnapshot. 1959 * 1960 * @mr: the memory region being queried. 1961 * @addr: the address (relative to the start of the region) being queried. 1962 * @size: the size of the range being queried. 1963 * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA. 1964 */ 1965 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr, 1966 hwaddr addr, 1967 hwaddr size, 1968 unsigned client); 1969 1970 /** 1971 * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty 1972 * in the specified dirty bitmap snapshot. 1973 * 1974 * @mr: the memory region being queried. 1975 * @snap: the dirty bitmap snapshot 1976 * @addr: the address (relative to the start of the region) being queried. 1977 * @size: the size of the range being queried. 1978 */ 1979 bool memory_region_snapshot_get_dirty(MemoryRegion *mr, 1980 DirtyBitmapSnapshot *snap, 1981 hwaddr addr, hwaddr size); 1982 1983 /** 1984 * memory_region_reset_dirty: Mark a range of pages as clean, for a specified 1985 * client. 1986 * 1987 * Marks a range of pages as no longer dirty. 1988 * 1989 * @mr: the region being updated. 1990 * @addr: the start of the subrange being cleaned. 1991 * @size: the size of the subrange being cleaned. 1992 * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or 1993 * %DIRTY_MEMORY_VGA. 1994 */ 1995 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr, 1996 hwaddr size, unsigned client); 1997 1998 /** 1999 * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate 2000 * TBs (for self-modifying code). 2001 * 2002 * The MemoryRegionOps->write() callback of a ROM device must use this function 2003 * to mark byte ranges that have been modified internally, such as by directly 2004 * accessing the memory returned by memory_region_get_ram_ptr(). 2005 * 2006 * This function marks the range dirty and invalidates TBs so that TCG can 2007 * detect self-modifying code. 2008 * 2009 * @mr: the region being flushed. 2010 * @addr: the start, relative to the start of the region, of the range being 2011 * flushed. 2012 * @size: the size, in bytes, of the range being flushed. 2013 */ 2014 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size); 2015 2016 /** 2017 * memory_region_set_readonly: Turn a memory region read-only (or read-write) 2018 * 2019 * Allows a memory region to be marked as read-only (turning it into a ROM). 2020 * only useful on RAM regions. 2021 * 2022 * @mr: the region being updated. 2023 * @readonly: whether rhe region is to be ROM or RAM. 2024 */ 2025 void memory_region_set_readonly(MemoryRegion *mr, bool readonly); 2026 2027 /** 2028 * memory_region_set_nonvolatile: Turn a memory region non-volatile 2029 * 2030 * Allows a memory region to be marked as non-volatile. 2031 * only useful on RAM regions. 2032 * 2033 * @mr: the region being updated. 2034 * @nonvolatile: whether rhe region is to be non-volatile. 2035 */ 2036 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile); 2037 2038 /** 2039 * memory_region_rom_device_set_romd: enable/disable ROMD mode 2040 * 2041 * Allows a ROM device (initialized with memory_region_init_rom_device() to 2042 * set to ROMD mode (default) or MMIO mode. When it is in ROMD mode, the 2043 * device is mapped to guest memory and satisfies read access directly. 2044 * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function. 2045 * Writes are always handled by the #MemoryRegion.write function. 2046 * 2047 * @mr: the memory region to be updated 2048 * @romd_mode: %true to put the region into ROMD mode 2049 */ 2050 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode); 2051 2052 /** 2053 * memory_region_set_coalescing: Enable memory coalescing for the region. 2054 * 2055 * Enabled writes to a region to be queued for later processing. MMIO ->write 2056 * callbacks may be delayed until a non-coalesced MMIO is issued. 2057 * Only useful for IO regions. Roughly similar to write-combining hardware. 2058 * 2059 * @mr: the memory region to be write coalesced 2060 */ 2061 void memory_region_set_coalescing(MemoryRegion *mr); 2062 2063 /** 2064 * memory_region_add_coalescing: Enable memory coalescing for a sub-range of 2065 * a region. 2066 * 2067 * Like memory_region_set_coalescing(), but works on a sub-range of a region. 2068 * Multiple calls can be issued coalesced disjoint ranges. 2069 * 2070 * @mr: the memory region to be updated. 2071 * @offset: the start of the range within the region to be coalesced. 2072 * @size: the size of the subrange to be coalesced. 2073 */ 2074 void memory_region_add_coalescing(MemoryRegion *mr, 2075 hwaddr offset, 2076 uint64_t size); 2077 2078 /** 2079 * memory_region_clear_coalescing: Disable MMIO coalescing for the region. 2080 * 2081 * Disables any coalescing caused by memory_region_set_coalescing() or 2082 * memory_region_add_coalescing(). Roughly equivalent to uncacheble memory 2083 * hardware. 2084 * 2085 * @mr: the memory region to be updated. 2086 */ 2087 void memory_region_clear_coalescing(MemoryRegion *mr); 2088 2089 /** 2090 * memory_region_set_flush_coalesced: Enforce memory coalescing flush before 2091 * accesses. 2092 * 2093 * Ensure that pending coalesced MMIO request are flushed before the memory 2094 * region is accessed. This property is automatically enabled for all regions 2095 * passed to memory_region_set_coalescing() and memory_region_add_coalescing(). 2096 * 2097 * @mr: the memory region to be updated. 2098 */ 2099 void memory_region_set_flush_coalesced(MemoryRegion *mr); 2100 2101 /** 2102 * memory_region_clear_flush_coalesced: Disable memory coalescing flush before 2103 * accesses. 2104 * 2105 * Clear the automatic coalesced MMIO flushing enabled via 2106 * memory_region_set_flush_coalesced. Note that this service has no effect on 2107 * memory regions that have MMIO coalescing enabled for themselves. For them, 2108 * automatic flushing will stop once coalescing is disabled. 2109 * 2110 * @mr: the memory region to be updated. 2111 */ 2112 void memory_region_clear_flush_coalesced(MemoryRegion *mr); 2113 2114 /** 2115 * memory_region_add_eventfd: Request an eventfd to be triggered when a word 2116 * is written to a location. 2117 * 2118 * Marks a word in an IO region (initialized with memory_region_init_io()) 2119 * as a trigger for an eventfd event. The I/O callback will not be called. 2120 * The caller must be prepared to handle failure (that is, take the required 2121 * action if the callback _is_ called). 2122 * 2123 * @mr: the memory region being updated. 2124 * @addr: the address within @mr that is to be monitored 2125 * @size: the size of the access to trigger the eventfd 2126 * @match_data: whether to match against @data, instead of just @addr 2127 * @data: the data to match against the guest write 2128 * @e: event notifier to be triggered when @addr, @size, and @data all match. 2129 **/ 2130 void memory_region_add_eventfd(MemoryRegion *mr, 2131 hwaddr addr, 2132 unsigned size, 2133 bool match_data, 2134 uint64_t data, 2135 EventNotifier *e); 2136 2137 /** 2138 * memory_region_del_eventfd: Cancel an eventfd. 2139 * 2140 * Cancels an eventfd trigger requested by a previous 2141 * memory_region_add_eventfd() call. 2142 * 2143 * @mr: the memory region being updated. 2144 * @addr: the address within @mr that is to be monitored 2145 * @size: the size of the access to trigger the eventfd 2146 * @match_data: whether to match against @data, instead of just @addr 2147 * @data: the data to match against the guest write 2148 * @e: event notifier to be triggered when @addr, @size, and @data all match. 2149 */ 2150 void memory_region_del_eventfd(MemoryRegion *mr, 2151 hwaddr addr, 2152 unsigned size, 2153 bool match_data, 2154 uint64_t data, 2155 EventNotifier *e); 2156 2157 /** 2158 * memory_region_add_subregion: Add a subregion to a container. 2159 * 2160 * Adds a subregion at @offset. The subregion may not overlap with other 2161 * subregions (except for those explicitly marked as overlapping). A region 2162 * may only be added once as a subregion (unless removed with 2163 * memory_region_del_subregion()); use memory_region_init_alias() if you 2164 * want a region to be a subregion in multiple locations. 2165 * 2166 * @mr: the region to contain the new subregion; must be a container 2167 * initialized with memory_region_init(). 2168 * @offset: the offset relative to @mr where @subregion is added. 2169 * @subregion: the subregion to be added. 2170 */ 2171 void memory_region_add_subregion(MemoryRegion *mr, 2172 hwaddr offset, 2173 MemoryRegion *subregion); 2174 /** 2175 * memory_region_add_subregion_overlap: Add a subregion to a container 2176 * with overlap. 2177 * 2178 * Adds a subregion at @offset. The subregion may overlap with other 2179 * subregions. Conflicts are resolved by having a higher @priority hide a 2180 * lower @priority. Subregions without priority are taken as @priority 0. 2181 * A region may only be added once as a subregion (unless removed with 2182 * memory_region_del_subregion()); use memory_region_init_alias() if you 2183 * want a region to be a subregion in multiple locations. 2184 * 2185 * @mr: the region to contain the new subregion; must be a container 2186 * initialized with memory_region_init(). 2187 * @offset: the offset relative to @mr where @subregion is added. 2188 * @subregion: the subregion to be added. 2189 * @priority: used for resolving overlaps; highest priority wins. 2190 */ 2191 void memory_region_add_subregion_overlap(MemoryRegion *mr, 2192 hwaddr offset, 2193 MemoryRegion *subregion, 2194 int priority); 2195 2196 /** 2197 * memory_region_get_ram_addr: Get the ram address associated with a memory 2198 * region 2199 * 2200 * @mr: the region to be queried 2201 */ 2202 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr); 2203 2204 uint64_t memory_region_get_alignment(const MemoryRegion *mr); 2205 /** 2206 * memory_region_del_subregion: Remove a subregion. 2207 * 2208 * Removes a subregion from its container. 2209 * 2210 * @mr: the container to be updated. 2211 * @subregion: the region being removed; must be a current subregion of @mr. 2212 */ 2213 void memory_region_del_subregion(MemoryRegion *mr, 2214 MemoryRegion *subregion); 2215 2216 /* 2217 * memory_region_set_enabled: dynamically enable or disable a region 2218 * 2219 * Enables or disables a memory region. A disabled memory region 2220 * ignores all accesses to itself and its subregions. It does not 2221 * obscure sibling subregions with lower priority - it simply behaves as 2222 * if it was removed from the hierarchy. 2223 * 2224 * Regions default to being enabled. 2225 * 2226 * @mr: the region to be updated 2227 * @enabled: whether to enable or disable the region 2228 */ 2229 void memory_region_set_enabled(MemoryRegion *mr, bool enabled); 2230 2231 /* 2232 * memory_region_set_address: dynamically update the address of a region 2233 * 2234 * Dynamically updates the address of a region, relative to its container. 2235 * May be used on regions are currently part of a memory hierarchy. 2236 * 2237 * @mr: the region to be updated 2238 * @addr: new address, relative to container region 2239 */ 2240 void memory_region_set_address(MemoryRegion *mr, hwaddr addr); 2241 2242 /* 2243 * memory_region_set_size: dynamically update the size of a region. 2244 * 2245 * Dynamically updates the size of a region. 2246 * 2247 * @mr: the region to be updated 2248 * @size: used size of the region. 2249 */ 2250 void memory_region_set_size(MemoryRegion *mr, uint64_t size); 2251 2252 /* 2253 * memory_region_set_alias_offset: dynamically update a memory alias's offset 2254 * 2255 * Dynamically updates the offset into the target region that an alias points 2256 * to, as if the fourth argument to memory_region_init_alias() has changed. 2257 * 2258 * @mr: the #MemoryRegion to be updated; should be an alias. 2259 * @offset: the new offset into the target memory region 2260 */ 2261 void memory_region_set_alias_offset(MemoryRegion *mr, 2262 hwaddr offset); 2263 2264 /** 2265 * memory_region_present: checks if an address relative to a @container 2266 * translates into #MemoryRegion within @container 2267 * 2268 * Answer whether a #MemoryRegion within @container covers the address 2269 * @addr. 2270 * 2271 * @container: a #MemoryRegion within which @addr is a relative address 2272 * @addr: the area within @container to be searched 2273 */ 2274 bool memory_region_present(MemoryRegion *container, hwaddr addr); 2275 2276 /** 2277 * memory_region_is_mapped: returns true if #MemoryRegion is mapped 2278 * into any address space. 2279 * 2280 * @mr: a #MemoryRegion which should be checked if it's mapped 2281 */ 2282 bool memory_region_is_mapped(MemoryRegion *mr); 2283 2284 /** 2285 * memory_region_get_ram_discard_manager: get the #RamDiscardManager for a 2286 * #MemoryRegion 2287 * 2288 * The #RamDiscardManager cannot change while a memory region is mapped. 2289 * 2290 * @mr: the #MemoryRegion 2291 */ 2292 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr); 2293 2294 /** 2295 * memory_region_has_ram_discard_manager: check whether a #MemoryRegion has a 2296 * #RamDiscardManager assigned 2297 * 2298 * @mr: the #MemoryRegion 2299 */ 2300 static inline bool memory_region_has_ram_discard_manager(MemoryRegion *mr) 2301 { 2302 return !!memory_region_get_ram_discard_manager(mr); 2303 } 2304 2305 /** 2306 * memory_region_set_ram_discard_manager: set the #RamDiscardManager for a 2307 * #MemoryRegion 2308 * 2309 * This function must not be called for a mapped #MemoryRegion, a #MemoryRegion 2310 * that does not cover RAM, or a #MemoryRegion that already has a 2311 * #RamDiscardManager assigned. 2312 * 2313 * @mr: the #MemoryRegion 2314 * @rdm: #RamDiscardManager to set 2315 */ 2316 void memory_region_set_ram_discard_manager(MemoryRegion *mr, 2317 RamDiscardManager *rdm); 2318 2319 /** 2320 * memory_region_find: translate an address/size relative to a 2321 * MemoryRegion into a #MemoryRegionSection. 2322 * 2323 * Locates the first #MemoryRegion within @mr that overlaps the range 2324 * given by @addr and @size. 2325 * 2326 * Returns a #MemoryRegionSection that describes a contiguous overlap. 2327 * It will have the following characteristics: 2328 * - @size = 0 iff no overlap was found 2329 * - @mr is non-%NULL iff an overlap was found 2330 * 2331 * Remember that in the return value the @offset_within_region is 2332 * relative to the returned region (in the .@mr field), not to the 2333 * @mr argument. 2334 * 2335 * Similarly, the .@offset_within_address_space is relative to the 2336 * address space that contains both regions, the passed and the 2337 * returned one. However, in the special case where the @mr argument 2338 * has no container (and thus is the root of the address space), the 2339 * following will hold: 2340 * - @offset_within_address_space >= @addr 2341 * - @offset_within_address_space + .@size <= @addr + @size 2342 * 2343 * @mr: a MemoryRegion within which @addr is a relative address 2344 * @addr: start of the area within @as to be searched 2345 * @size: size of the area to be searched 2346 */ 2347 MemoryRegionSection memory_region_find(MemoryRegion *mr, 2348 hwaddr addr, uint64_t size); 2349 2350 /** 2351 * memory_global_dirty_log_sync: synchronize the dirty log for all memory 2352 * 2353 * Synchronizes the dirty page log for all address spaces. 2354 */ 2355 void memory_global_dirty_log_sync(void); 2356 2357 /** 2358 * memory_global_dirty_log_sync: synchronize the dirty log for all memory 2359 * 2360 * Synchronizes the vCPUs with a thread that is reading the dirty bitmap. 2361 * This function must be called after the dirty log bitmap is cleared, and 2362 * before dirty guest memory pages are read. If you are using 2363 * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes 2364 * care of doing this. 2365 */ 2366 void memory_global_after_dirty_log_sync(void); 2367 2368 /** 2369 * memory_region_transaction_begin: Start a transaction. 2370 * 2371 * During a transaction, changes will be accumulated and made visible 2372 * only when the transaction ends (is committed). 2373 */ 2374 void memory_region_transaction_begin(void); 2375 2376 /** 2377 * memory_region_transaction_commit: Commit a transaction and make changes 2378 * visible to the guest. 2379 */ 2380 void memory_region_transaction_commit(void); 2381 2382 /** 2383 * memory_listener_register: register callbacks to be called when memory 2384 * sections are mapped or unmapped into an address 2385 * space 2386 * 2387 * @listener: an object containing the callbacks to be called 2388 * @filter: if non-%NULL, only regions in this address space will be observed 2389 */ 2390 void memory_listener_register(MemoryListener *listener, AddressSpace *filter); 2391 2392 /** 2393 * memory_listener_unregister: undo the effect of memory_listener_register() 2394 * 2395 * @listener: an object containing the callbacks to be removed 2396 */ 2397 void memory_listener_unregister(MemoryListener *listener); 2398 2399 /** 2400 * memory_global_dirty_log_start: begin dirty logging for all regions 2401 * 2402 * @flags: purpose of starting dirty log, migration or dirty rate 2403 */ 2404 void memory_global_dirty_log_start(unsigned int flags); 2405 2406 /** 2407 * memory_global_dirty_log_stop: end dirty logging for all regions 2408 * 2409 * @flags: purpose of stopping dirty log, migration or dirty rate 2410 */ 2411 void memory_global_dirty_log_stop(unsigned int flags); 2412 2413 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled); 2414 2415 /** 2416 * memory_region_dispatch_read: perform a read directly to the specified 2417 * MemoryRegion. 2418 * 2419 * @mr: #MemoryRegion to access 2420 * @addr: address within that region 2421 * @pval: pointer to uint64_t which the data is written to 2422 * @op: size, sign, and endianness of the memory operation 2423 * @attrs: memory transaction attributes to use for the access 2424 */ 2425 MemTxResult memory_region_dispatch_read(MemoryRegion *mr, 2426 hwaddr addr, 2427 uint64_t *pval, 2428 MemOp op, 2429 MemTxAttrs attrs); 2430 /** 2431 * memory_region_dispatch_write: perform a write directly to the specified 2432 * MemoryRegion. 2433 * 2434 * @mr: #MemoryRegion to access 2435 * @addr: address within that region 2436 * @data: data to write 2437 * @op: size, sign, and endianness of the memory operation 2438 * @attrs: memory transaction attributes to use for the access 2439 */ 2440 MemTxResult memory_region_dispatch_write(MemoryRegion *mr, 2441 hwaddr addr, 2442 uint64_t data, 2443 MemOp op, 2444 MemTxAttrs attrs); 2445 2446 /** 2447 * address_space_init: initializes an address space 2448 * 2449 * @as: an uninitialized #AddressSpace 2450 * @root: a #MemoryRegion that routes addresses for the address space 2451 * @name: an address space name. The name is only used for debugging 2452 * output. 2453 */ 2454 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name); 2455 2456 /** 2457 * address_space_destroy: destroy an address space 2458 * 2459 * Releases all resources associated with an address space. After an address space 2460 * is destroyed, its root memory region (given by address_space_init()) may be destroyed 2461 * as well. 2462 * 2463 * @as: address space to be destroyed 2464 */ 2465 void address_space_destroy(AddressSpace *as); 2466 2467 /** 2468 * address_space_remove_listeners: unregister all listeners of an address space 2469 * 2470 * Removes all callbacks previously registered with memory_listener_register() 2471 * for @as. 2472 * 2473 * @as: an initialized #AddressSpace 2474 */ 2475 void address_space_remove_listeners(AddressSpace *as); 2476 2477 /** 2478 * address_space_rw: read from or write to an address space. 2479 * 2480 * Return a MemTxResult indicating whether the operation succeeded 2481 * or failed (eg unassigned memory, device rejected the transaction, 2482 * IOMMU fault). 2483 * 2484 * @as: #AddressSpace to be accessed 2485 * @addr: address within that address space 2486 * @attrs: memory transaction attributes 2487 * @buf: buffer with the data transferred 2488 * @len: the number of bytes to read or write 2489 * @is_write: indicates the transfer direction 2490 */ 2491 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, 2492 MemTxAttrs attrs, void *buf, 2493 hwaddr len, bool is_write); 2494 2495 /** 2496 * address_space_write: write to address space. 2497 * 2498 * Return a MemTxResult indicating whether the operation succeeded 2499 * or failed (eg unassigned memory, device rejected the transaction, 2500 * IOMMU fault). 2501 * 2502 * @as: #AddressSpace to be accessed 2503 * @addr: address within that address space 2504 * @attrs: memory transaction attributes 2505 * @buf: buffer with the data transferred 2506 * @len: the number of bytes to write 2507 */ 2508 MemTxResult address_space_write(AddressSpace *as, hwaddr addr, 2509 MemTxAttrs attrs, 2510 const void *buf, hwaddr len); 2511 2512 /** 2513 * address_space_write_rom: write to address space, including ROM. 2514 * 2515 * This function writes to the specified address space, but will 2516 * write data to both ROM and RAM. This is used for non-guest 2517 * writes like writes from the gdb debug stub or initial loading 2518 * of ROM contents. 2519 * 2520 * Note that portions of the write which attempt to write data to 2521 * a device will be silently ignored -- only real RAM and ROM will 2522 * be written to. 2523 * 2524 * Return a MemTxResult indicating whether the operation succeeded 2525 * or failed (eg unassigned memory, device rejected the transaction, 2526 * IOMMU fault). 2527 * 2528 * @as: #AddressSpace to be accessed 2529 * @addr: address within that address space 2530 * @attrs: memory transaction attributes 2531 * @buf: buffer with the data transferred 2532 * @len: the number of bytes to write 2533 */ 2534 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr, 2535 MemTxAttrs attrs, 2536 const void *buf, hwaddr len); 2537 2538 /* address_space_ld*: load from an address space 2539 * address_space_st*: store to an address space 2540 * 2541 * These functions perform a load or store of the byte, word, 2542 * longword or quad to the specified address within the AddressSpace. 2543 * The _le suffixed functions treat the data as little endian; 2544 * _be indicates big endian; no suffix indicates "same endianness 2545 * as guest CPU". 2546 * 2547 * The "guest CPU endianness" accessors are deprecated for use outside 2548 * target-* code; devices should be CPU-agnostic and use either the LE 2549 * or the BE accessors. 2550 * 2551 * @as #AddressSpace to be accessed 2552 * @addr: address within that address space 2553 * @val: data value, for stores 2554 * @attrs: memory transaction attributes 2555 * @result: location to write the success/failure of the transaction; 2556 * if NULL, this information is discarded 2557 */ 2558 2559 #define SUFFIX 2560 #define ARG1 as 2561 #define ARG1_DECL AddressSpace *as 2562 #include "exec/memory_ldst.h.inc" 2563 2564 #define SUFFIX 2565 #define ARG1 as 2566 #define ARG1_DECL AddressSpace *as 2567 #include "exec/memory_ldst_phys.h.inc" 2568 2569 struct MemoryRegionCache { 2570 void *ptr; 2571 hwaddr xlat; 2572 hwaddr len; 2573 FlatView *fv; 2574 MemoryRegionSection mrs; 2575 bool is_write; 2576 }; 2577 2578 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL }) 2579 2580 2581 /* address_space_ld*_cached: load from a cached #MemoryRegion 2582 * address_space_st*_cached: store into a cached #MemoryRegion 2583 * 2584 * These functions perform a load or store of the byte, word, 2585 * longword or quad to the specified address. The address is 2586 * a physical address in the AddressSpace, but it must lie within 2587 * a #MemoryRegion that was mapped with address_space_cache_init. 2588 * 2589 * The _le suffixed functions treat the data as little endian; 2590 * _be indicates big endian; no suffix indicates "same endianness 2591 * as guest CPU". 2592 * 2593 * The "guest CPU endianness" accessors are deprecated for use outside 2594 * target-* code; devices should be CPU-agnostic and use either the LE 2595 * or the BE accessors. 2596 * 2597 * @cache: previously initialized #MemoryRegionCache to be accessed 2598 * @addr: address within the address space 2599 * @val: data value, for stores 2600 * @attrs: memory transaction attributes 2601 * @result: location to write the success/failure of the transaction; 2602 * if NULL, this information is discarded 2603 */ 2604 2605 #define SUFFIX _cached_slow 2606 #define ARG1 cache 2607 #define ARG1_DECL MemoryRegionCache *cache 2608 #include "exec/memory_ldst.h.inc" 2609 2610 /* Inline fast path for direct RAM access. */ 2611 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache, 2612 hwaddr addr, MemTxAttrs attrs, MemTxResult *result) 2613 { 2614 assert(addr < cache->len); 2615 if (likely(cache->ptr)) { 2616 return ldub_p(cache->ptr + addr); 2617 } else { 2618 return address_space_ldub_cached_slow(cache, addr, attrs, result); 2619 } 2620 } 2621 2622 static inline void address_space_stb_cached(MemoryRegionCache *cache, 2623 hwaddr addr, uint8_t val, MemTxAttrs attrs, MemTxResult *result) 2624 { 2625 assert(addr < cache->len); 2626 if (likely(cache->ptr)) { 2627 stb_p(cache->ptr + addr, val); 2628 } else { 2629 address_space_stb_cached_slow(cache, addr, val, attrs, result); 2630 } 2631 } 2632 2633 #define ENDIANNESS _le 2634 #include "exec/memory_ldst_cached.h.inc" 2635 2636 #define ENDIANNESS _be 2637 #include "exec/memory_ldst_cached.h.inc" 2638 2639 #define SUFFIX _cached 2640 #define ARG1 cache 2641 #define ARG1_DECL MemoryRegionCache *cache 2642 #include "exec/memory_ldst_phys.h.inc" 2643 2644 /* address_space_cache_init: prepare for repeated access to a physical 2645 * memory region 2646 * 2647 * @cache: #MemoryRegionCache to be filled 2648 * @as: #AddressSpace to be accessed 2649 * @addr: address within that address space 2650 * @len: length of buffer 2651 * @is_write: indicates the transfer direction 2652 * 2653 * Will only work with RAM, and may map a subset of the requested range by 2654 * returning a value that is less than @len. On failure, return a negative 2655 * errno value. 2656 * 2657 * Because it only works with RAM, this function can be used for 2658 * read-modify-write operations. In this case, is_write should be %true. 2659 * 2660 * Note that addresses passed to the address_space_*_cached functions 2661 * are relative to @addr. 2662 */ 2663 int64_t address_space_cache_init(MemoryRegionCache *cache, 2664 AddressSpace *as, 2665 hwaddr addr, 2666 hwaddr len, 2667 bool is_write); 2668 2669 /** 2670 * address_space_cache_invalidate: complete a write to a #MemoryRegionCache 2671 * 2672 * @cache: The #MemoryRegionCache to operate on. 2673 * @addr: The first physical address that was written, relative to the 2674 * address that was passed to @address_space_cache_init. 2675 * @access_len: The number of bytes that were written starting at @addr. 2676 */ 2677 void address_space_cache_invalidate(MemoryRegionCache *cache, 2678 hwaddr addr, 2679 hwaddr access_len); 2680 2681 /** 2682 * address_space_cache_destroy: free a #MemoryRegionCache 2683 * 2684 * @cache: The #MemoryRegionCache whose memory should be released. 2685 */ 2686 void address_space_cache_destroy(MemoryRegionCache *cache); 2687 2688 /* address_space_get_iotlb_entry: translate an address into an IOTLB 2689 * entry. Should be called from an RCU critical section. 2690 */ 2691 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr, 2692 bool is_write, MemTxAttrs attrs); 2693 2694 /* address_space_translate: translate an address range into an address space 2695 * into a MemoryRegion and an address range into that section. Should be 2696 * called from an RCU critical section, to avoid that the last reference 2697 * to the returned region disappears after address_space_translate returns. 2698 * 2699 * @fv: #FlatView to be accessed 2700 * @addr: address within that address space 2701 * @xlat: pointer to address within the returned memory region section's 2702 * #MemoryRegion. 2703 * @len: pointer to length 2704 * @is_write: indicates the transfer direction 2705 * @attrs: memory attributes 2706 */ 2707 MemoryRegion *flatview_translate(FlatView *fv, 2708 hwaddr addr, hwaddr *xlat, 2709 hwaddr *len, bool is_write, 2710 MemTxAttrs attrs); 2711 2712 static inline MemoryRegion *address_space_translate(AddressSpace *as, 2713 hwaddr addr, hwaddr *xlat, 2714 hwaddr *len, bool is_write, 2715 MemTxAttrs attrs) 2716 { 2717 return flatview_translate(address_space_to_flatview(as), 2718 addr, xlat, len, is_write, attrs); 2719 } 2720 2721 /* address_space_access_valid: check for validity of accessing an address 2722 * space range 2723 * 2724 * Check whether memory is assigned to the given address space range, and 2725 * access is permitted by any IOMMU regions that are active for the address 2726 * space. 2727 * 2728 * For now, addr and len should be aligned to a page size. This limitation 2729 * will be lifted in the future. 2730 * 2731 * @as: #AddressSpace to be accessed 2732 * @addr: address within that address space 2733 * @len: length of the area to be checked 2734 * @is_write: indicates the transfer direction 2735 * @attrs: memory attributes 2736 */ 2737 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len, 2738 bool is_write, MemTxAttrs attrs); 2739 2740 /* address_space_map: map a physical memory region into a host virtual address 2741 * 2742 * May map a subset of the requested range, given by and returned in @plen. 2743 * May return %NULL and set *@plen to zero(0), if resources needed to perform 2744 * the mapping are exhausted. 2745 * Use only for reads OR writes - not for read-modify-write operations. 2746 * Use cpu_register_map_client() to know when retrying the map operation is 2747 * likely to succeed. 2748 * 2749 * @as: #AddressSpace to be accessed 2750 * @addr: address within that address space 2751 * @plen: pointer to length of buffer; updated on return 2752 * @is_write: indicates the transfer direction 2753 * @attrs: memory attributes 2754 */ 2755 void *address_space_map(AddressSpace *as, hwaddr addr, 2756 hwaddr *plen, bool is_write, MemTxAttrs attrs); 2757 2758 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map() 2759 * 2760 * Will also mark the memory as dirty if @is_write == %true. @access_len gives 2761 * the amount of memory that was actually read or written by the caller. 2762 * 2763 * @as: #AddressSpace used 2764 * @buffer: host pointer as returned by address_space_map() 2765 * @len: buffer length as returned by address_space_map() 2766 * @access_len: amount of data actually transferred 2767 * @is_write: indicates the transfer direction 2768 */ 2769 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len, 2770 bool is_write, hwaddr access_len); 2771 2772 2773 /* Internal functions, part of the implementation of address_space_read. */ 2774 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr, 2775 MemTxAttrs attrs, void *buf, hwaddr len); 2776 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr, 2777 MemTxAttrs attrs, void *buf, 2778 hwaddr len, hwaddr addr1, hwaddr l, 2779 MemoryRegion *mr); 2780 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr); 2781 2782 /* Internal functions, part of the implementation of address_space_read_cached 2783 * and address_space_write_cached. */ 2784 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache, 2785 hwaddr addr, void *buf, hwaddr len); 2786 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache, 2787 hwaddr addr, const void *buf, 2788 hwaddr len); 2789 2790 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write) 2791 { 2792 if (is_write) { 2793 return memory_region_is_ram(mr) && !mr->readonly && 2794 !mr->rom_device && !memory_region_is_ram_device(mr); 2795 } else { 2796 return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) || 2797 memory_region_is_romd(mr); 2798 } 2799 } 2800 2801 /** 2802 * address_space_read: read from an address space. 2803 * 2804 * Return a MemTxResult indicating whether the operation succeeded 2805 * or failed (eg unassigned memory, device rejected the transaction, 2806 * IOMMU fault). Called within RCU critical section. 2807 * 2808 * @as: #AddressSpace to be accessed 2809 * @addr: address within that address space 2810 * @attrs: memory transaction attributes 2811 * @buf: buffer with the data transferred 2812 * @len: length of the data transferred 2813 */ 2814 static inline __attribute__((__always_inline__)) 2815 MemTxResult address_space_read(AddressSpace *as, hwaddr addr, 2816 MemTxAttrs attrs, void *buf, 2817 hwaddr len) 2818 { 2819 MemTxResult result = MEMTX_OK; 2820 hwaddr l, addr1; 2821 void *ptr; 2822 MemoryRegion *mr; 2823 FlatView *fv; 2824 2825 if (__builtin_constant_p(len)) { 2826 if (len) { 2827 RCU_READ_LOCK_GUARD(); 2828 fv = address_space_to_flatview(as); 2829 l = len; 2830 mr = flatview_translate(fv, addr, &addr1, &l, false, attrs); 2831 if (len == l && memory_access_is_direct(mr, false)) { 2832 ptr = qemu_map_ram_ptr(mr->ram_block, addr1); 2833 memcpy(buf, ptr, len); 2834 } else { 2835 result = flatview_read_continue(fv, addr, attrs, buf, len, 2836 addr1, l, mr); 2837 } 2838 } 2839 } else { 2840 result = address_space_read_full(as, addr, attrs, buf, len); 2841 } 2842 return result; 2843 } 2844 2845 /** 2846 * address_space_read_cached: read from 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_read_cached(MemoryRegionCache *cache, hwaddr addr, 2855 void *buf, hwaddr len) 2856 { 2857 assert(addr < cache->len && len <= cache->len - addr); 2858 fuzz_dma_read_cb(cache->xlat + addr, len, cache->mrs.mr); 2859 if (likely(cache->ptr)) { 2860 memcpy(buf, cache->ptr + addr, len); 2861 return MEMTX_OK; 2862 } else { 2863 return address_space_read_cached_slow(cache, addr, buf, len); 2864 } 2865 } 2866 2867 /** 2868 * address_space_write_cached: write to a cached RAM region 2869 * 2870 * @cache: Cached region to be addressed 2871 * @addr: address relative to the base of the RAM region 2872 * @buf: buffer with the data transferred 2873 * @len: length of the data transferred 2874 */ 2875 static inline MemTxResult 2876 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr, 2877 const void *buf, hwaddr len) 2878 { 2879 assert(addr < cache->len && len <= cache->len - addr); 2880 if (likely(cache->ptr)) { 2881 memcpy(cache->ptr + addr, buf, len); 2882 return MEMTX_OK; 2883 } else { 2884 return address_space_write_cached_slow(cache, addr, buf, len); 2885 } 2886 } 2887 2888 #ifdef NEED_CPU_H 2889 /* enum device_endian to MemOp. */ 2890 static inline MemOp devend_memop(enum device_endian end) 2891 { 2892 QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN && 2893 DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN); 2894 2895 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN) 2896 /* Swap if non-host endianness or native (target) endianness */ 2897 return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP; 2898 #else 2899 const int non_host_endianness = 2900 DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN; 2901 2902 /* In this case, native (target) endianness needs no swap. */ 2903 return (end == non_host_endianness) ? MO_BSWAP : 0; 2904 #endif 2905 } 2906 #endif 2907 2908 /* 2909 * Inhibit technologies that require discarding of pages in RAM blocks, e.g., 2910 * to manage the actual amount of memory consumed by the VM (then, the memory 2911 * provided by RAM blocks might be bigger than the desired memory consumption). 2912 * This *must* be set if: 2913 * - Discarding parts of a RAM blocks does not result in the change being 2914 * reflected in the VM and the pages getting freed. 2915 * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous 2916 * discards blindly. 2917 * - Discarding parts of a RAM blocks will result in integrity issues (e.g., 2918 * encrypted VMs). 2919 * Technologies that only temporarily pin the current working set of a 2920 * driver are fine, because we don't expect such pages to be discarded 2921 * (esp. based on guest action like balloon inflation). 2922 * 2923 * This is *not* to be used to protect from concurrent discards (esp., 2924 * postcopy). 2925 * 2926 * Returns 0 if successful. Returns -EBUSY if a technology that relies on 2927 * discards to work reliably is active. 2928 */ 2929 int ram_block_discard_disable(bool state); 2930 2931 /* 2932 * See ram_block_discard_disable(): only disable uncoordinated discards, 2933 * keeping coordinated discards (via the RamDiscardManager) enabled. 2934 */ 2935 int ram_block_uncoordinated_discard_disable(bool state); 2936 2937 /* 2938 * Inhibit technologies that disable discarding of pages in RAM blocks. 2939 * 2940 * Returns 0 if successful. Returns -EBUSY if discards are already set to 2941 * broken. 2942 */ 2943 int ram_block_discard_require(bool state); 2944 2945 /* 2946 * See ram_block_discard_require(): only inhibit technologies that disable 2947 * uncoordinated discarding of pages in RAM blocks, allowing co-existance with 2948 * technologies that only inhibit uncoordinated discards (via the 2949 * RamDiscardManager). 2950 */ 2951 int ram_block_coordinated_discard_require(bool state); 2952 2953 /* 2954 * Test if any discarding of memory in ram blocks is disabled. 2955 */ 2956 bool ram_block_discard_is_disabled(void); 2957 2958 /* 2959 * Test if any discarding of memory in ram blocks is required to work reliably. 2960 */ 2961 bool ram_block_discard_is_required(void); 2962 2963 #endif 2964 2965 #endif 2966