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