xref: /openbmc/qemu/include/exec/memory.h (revision 1580b897)
1 /*
2  * Physical memory management API
3  *
4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
5  *
6  * Authors:
7  *  Avi Kivity <avi@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #ifndef MEMORY_H
15 #define MEMORY_H
16 
17 #ifndef CONFIG_USER_ONLY
18 
19 #include "exec/cpu-common.h"
20 #include "exec/hwaddr.h"
21 #include "exec/memattrs.h"
22 #include "exec/memop.h"
23 #include "exec/ramlist.h"
24 #include "qemu/bswap.h"
25 #include "qemu/queue.h"
26 #include "qemu/int128.h"
27 #include "qemu/notify.h"
28 #include "qom/object.h"
29 #include "qemu/rcu.h"
30 
31 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
32 
33 #define MAX_PHYS_ADDR_SPACE_BITS 62
34 #define MAX_PHYS_ADDR            (((hwaddr)1 << MAX_PHYS_ADDR_SPACE_BITS) - 1)
35 
36 #define TYPE_MEMORY_REGION "memory-region"
37 DECLARE_INSTANCE_CHECKER(MemoryRegion, MEMORY_REGION,
38                          TYPE_MEMORY_REGION)
39 
40 #define TYPE_IOMMU_MEMORY_REGION "iommu-memory-region"
41 typedef struct IOMMUMemoryRegionClass IOMMUMemoryRegionClass;
42 DECLARE_OBJ_CHECKERS(IOMMUMemoryRegion, IOMMUMemoryRegionClass,
43                      IOMMU_MEMORY_REGION, TYPE_IOMMU_MEMORY_REGION)
44 
45 #define TYPE_RAM_DISCARD_MANAGER "qemu:ram-discard-manager"
46 typedef struct RamDiscardManagerClass RamDiscardManagerClass;
47 typedef struct RamDiscardManager RamDiscardManager;
48 DECLARE_OBJ_CHECKERS(RamDiscardManager, RamDiscardManagerClass,
49                      RAM_DISCARD_MANAGER, TYPE_RAM_DISCARD_MANAGER);
50 
51 #ifdef CONFIG_FUZZ
52 void fuzz_dma_read_cb(size_t addr,
53                       size_t len,
54                       MemoryRegion *mr);
55 #else
56 static inline void fuzz_dma_read_cb(size_t addr,
57                                     size_t len,
58                                     MemoryRegion *mr)
59 {
60     /* Do Nothing */
61 }
62 #endif
63 
64 extern bool global_dirty_log;
65 
66 typedef struct MemoryRegionOps MemoryRegionOps;
67 
68 struct ReservedRegion {
69     hwaddr low;
70     hwaddr high;
71     unsigned type;
72 };
73 
74 /**
75  * struct MemoryRegionSection: describes a fragment of a #MemoryRegion
76  *
77  * @mr: the region, or %NULL if empty
78  * @fv: the flat view of the address space the region is mapped in
79  * @offset_within_region: the beginning of the section, relative to @mr's start
80  * @size: the size of the section; will not exceed @mr's boundaries
81  * @offset_within_address_space: the address of the first byte of the section
82  *     relative to the region's address space
83  * @readonly: writes to this section are ignored
84  * @nonvolatile: this section is non-volatile
85  */
86 struct MemoryRegionSection {
87     Int128 size;
88     MemoryRegion *mr;
89     FlatView *fv;
90     hwaddr offset_within_region;
91     hwaddr offset_within_address_space;
92     bool readonly;
93     bool nonvolatile;
94 };
95 
96 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
97 
98 /* See address_space_translate: bit 0 is read, bit 1 is write.  */
99 typedef enum {
100     IOMMU_NONE = 0,
101     IOMMU_RO   = 1,
102     IOMMU_WO   = 2,
103     IOMMU_RW   = 3,
104 } IOMMUAccessFlags;
105 
106 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0))
107 
108 struct IOMMUTLBEntry {
109     AddressSpace    *target_as;
110     hwaddr           iova;
111     hwaddr           translated_addr;
112     hwaddr           addr_mask;  /* 0xfff = 4k translation */
113     IOMMUAccessFlags perm;
114 };
115 
116 /*
117  * Bitmap for different IOMMUNotifier capabilities. Each notifier can
118  * register with one or multiple IOMMU Notifier capability bit(s).
119  */
120 typedef enum {
121     IOMMU_NOTIFIER_NONE = 0,
122     /* Notify cache invalidations */
123     IOMMU_NOTIFIER_UNMAP = 0x1,
124     /* Notify entry changes (newly created entries) */
125     IOMMU_NOTIFIER_MAP = 0x2,
126     /* Notify changes on device IOTLB entries */
127     IOMMU_NOTIFIER_DEVIOTLB_UNMAP = 0x04,
128 } IOMMUNotifierFlag;
129 
130 #define IOMMU_NOTIFIER_IOTLB_EVENTS (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
131 #define IOMMU_NOTIFIER_DEVIOTLB_EVENTS IOMMU_NOTIFIER_DEVIOTLB_UNMAP
132 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_IOTLB_EVENTS | \
133                             IOMMU_NOTIFIER_DEVIOTLB_EVENTS)
134 
135 struct IOMMUNotifier;
136 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier,
137                             IOMMUTLBEntry *data);
138 
139 struct IOMMUNotifier {
140     IOMMUNotify notify;
141     IOMMUNotifierFlag notifier_flags;
142     /* Notify for address space range start <= addr <= end */
143     hwaddr start;
144     hwaddr end;
145     int iommu_idx;
146     QLIST_ENTRY(IOMMUNotifier) node;
147 };
148 typedef struct IOMMUNotifier IOMMUNotifier;
149 
150 typedef struct IOMMUTLBEvent {
151     IOMMUNotifierFlag type;
152     IOMMUTLBEntry entry;
153 } IOMMUTLBEvent;
154 
155 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
156 #define RAM_PREALLOC   (1 << 0)
157 
158 /* RAM is mmap-ed with MAP_SHARED */
159 #define RAM_SHARED     (1 << 1)
160 
161 /* Only a portion of RAM (used_length) is actually used, and migrated.
162  * Resizing RAM while migrating can result in the migration being canceled.
163  */
164 #define RAM_RESIZEABLE (1 << 2)
165 
166 /* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically
167  * zero the page and wake waiting processes.
168  * (Set during postcopy)
169  */
170 #define RAM_UF_ZEROPAGE (1 << 3)
171 
172 /* RAM can be migrated */
173 #define RAM_MIGRATABLE (1 << 4)
174 
175 /* RAM is a persistent kind memory */
176 #define RAM_PMEM (1 << 5)
177 
178 
179 /*
180  * UFFDIO_WRITEPROTECT is used on this RAMBlock to
181  * support 'write-tracking' migration type.
182  * Implies ram_state->ram_wt_enabled.
183  */
184 #define RAM_UF_WRITEPROTECT (1 << 6)
185 
186 /*
187  * RAM is mmap-ed with MAP_NORESERVE. When set, reserving swap space (or huge
188  * pages if applicable) is skipped: will bail out if not supported. When not
189  * set, the OS will do the reservation, if supported for the memory type.
190  */
191 #define RAM_NORESERVE (1 << 7)
192 
193 static inline void iommu_notifier_init(IOMMUNotifier *n, IOMMUNotify fn,
194                                        IOMMUNotifierFlag flags,
195                                        hwaddr start, hwaddr end,
196                                        int iommu_idx)
197 {
198     n->notify = fn;
199     n->notifier_flags = flags;
200     n->start = start;
201     n->end = end;
202     n->iommu_idx = iommu_idx;
203 }
204 
205 /*
206  * Memory region callbacks
207  */
208 struct MemoryRegionOps {
209     /* Read from the memory region. @addr is relative to @mr; @size is
210      * in bytes. */
211     uint64_t (*read)(void *opaque,
212                      hwaddr addr,
213                      unsigned size);
214     /* Write to the memory region. @addr is relative to @mr; @size is
215      * in bytes. */
216     void (*write)(void *opaque,
217                   hwaddr addr,
218                   uint64_t data,
219                   unsigned size);
220 
221     MemTxResult (*read_with_attrs)(void *opaque,
222                                    hwaddr addr,
223                                    uint64_t *data,
224                                    unsigned size,
225                                    MemTxAttrs attrs);
226     MemTxResult (*write_with_attrs)(void *opaque,
227                                     hwaddr addr,
228                                     uint64_t data,
229                                     unsigned size,
230                                     MemTxAttrs attrs);
231 
232     enum device_endian endianness;
233     /* Guest-visible constraints: */
234     struct {
235         /* If nonzero, specify bounds on access sizes beyond which a machine
236          * check is thrown.
237          */
238         unsigned min_access_size;
239         unsigned max_access_size;
240         /* If true, unaligned accesses are supported.  Otherwise unaligned
241          * accesses throw machine checks.
242          */
243          bool unaligned;
244         /*
245          * If present, and returns #false, the transaction is not accepted
246          * by the device (and results in machine dependent behaviour such
247          * as a machine check exception).
248          */
249         bool (*accepts)(void *opaque, hwaddr addr,
250                         unsigned size, bool is_write,
251                         MemTxAttrs attrs);
252     } valid;
253     /* Internal implementation constraints: */
254     struct {
255         /* If nonzero, specifies the minimum size implemented.  Smaller sizes
256          * will be rounded upwards and a partial result will be returned.
257          */
258         unsigned min_access_size;
259         /* If nonzero, specifies the maximum size implemented.  Larger sizes
260          * will be done as a series of accesses with smaller sizes.
261          */
262         unsigned max_access_size;
263         /* If true, unaligned accesses are supported.  Otherwise all accesses
264          * are converted to (possibly multiple) naturally aligned accesses.
265          */
266         bool unaligned;
267     } impl;
268 };
269 
270 typedef struct MemoryRegionClass {
271     /* private */
272     ObjectClass parent_class;
273 } MemoryRegionClass;
274 
275 
276 enum IOMMUMemoryRegionAttr {
277     IOMMU_ATTR_SPAPR_TCE_FD
278 };
279 
280 /*
281  * IOMMUMemoryRegionClass:
282  *
283  * All IOMMU implementations need to subclass TYPE_IOMMU_MEMORY_REGION
284  * and provide an implementation of at least the @translate method here
285  * to handle requests to the memory region. Other methods are optional.
286  *
287  * The IOMMU implementation must use the IOMMU notifier infrastructure
288  * to report whenever mappings are changed, by calling
289  * memory_region_notify_iommu() (or, if necessary, by calling
290  * memory_region_notify_iommu_one() for each registered notifier).
291  *
292  * Conceptually an IOMMU provides a mapping from input address
293  * to an output TLB entry. If the IOMMU is aware of memory transaction
294  * attributes and the output TLB entry depends on the transaction
295  * attributes, we represent this using IOMMU indexes. Each index
296  * selects a particular translation table that the IOMMU has:
297  *
298  *   @attrs_to_index returns the IOMMU index for a set of transaction attributes
299  *
300  *   @translate takes an input address and an IOMMU index
301  *
302  * and the mapping returned can only depend on the input address and the
303  * IOMMU index.
304  *
305  * Most IOMMUs don't care about the transaction attributes and support
306  * only a single IOMMU index. A more complex IOMMU might have one index
307  * for secure transactions and one for non-secure transactions.
308  */
309 struct IOMMUMemoryRegionClass {
310     /* private: */
311     MemoryRegionClass parent_class;
312 
313     /* public: */
314     /**
315      * @translate:
316      *
317      * Return a TLB entry that contains a given address.
318      *
319      * The IOMMUAccessFlags indicated via @flag are optional and may
320      * be specified as IOMMU_NONE to indicate that the caller needs
321      * the full translation information for both reads and writes. If
322      * the access flags are specified then the IOMMU implementation
323      * may use this as an optimization, to stop doing a page table
324      * walk as soon as it knows that the requested permissions are not
325      * allowed. If IOMMU_NONE is passed then the IOMMU must do the
326      * full page table walk and report the permissions in the returned
327      * IOMMUTLBEntry. (Note that this implies that an IOMMU may not
328      * return different mappings for reads and writes.)
329      *
330      * The returned information remains valid while the caller is
331      * holding the big QEMU lock or is inside an RCU critical section;
332      * if the caller wishes to cache the mapping beyond that it must
333      * register an IOMMU notifier so it can invalidate its cached
334      * information when the IOMMU mapping changes.
335      *
336      * @iommu: the IOMMUMemoryRegion
337      *
338      * @hwaddr: address to be translated within the memory region
339      *
340      * @flag: requested access permission
341      *
342      * @iommu_idx: IOMMU index for the translation
343      */
344     IOMMUTLBEntry (*translate)(IOMMUMemoryRegion *iommu, hwaddr addr,
345                                IOMMUAccessFlags flag, int iommu_idx);
346     /**
347      * @get_min_page_size:
348      *
349      * Returns minimum supported page size in bytes.
350      *
351      * If this method is not provided then the minimum is assumed to
352      * be TARGET_PAGE_SIZE.
353      *
354      * @iommu: the IOMMUMemoryRegion
355      */
356     uint64_t (*get_min_page_size)(IOMMUMemoryRegion *iommu);
357     /**
358      * @notify_flag_changed:
359      *
360      * Called when IOMMU Notifier flag changes (ie when the set of
361      * events which IOMMU users are requesting notification for changes).
362      * Optional method -- need not be provided if the IOMMU does not
363      * need to know exactly which events must be notified.
364      *
365      * @iommu: the IOMMUMemoryRegion
366      *
367      * @old_flags: events which previously needed to be notified
368      *
369      * @new_flags: events which now need to be notified
370      *
371      * Returns 0 on success, or a negative errno; in particular
372      * returns -EINVAL if the new flag bitmap is not supported by the
373      * IOMMU memory region. In case of failure, the error object
374      * must be created
375      */
376     int (*notify_flag_changed)(IOMMUMemoryRegion *iommu,
377                                IOMMUNotifierFlag old_flags,
378                                IOMMUNotifierFlag new_flags,
379                                Error **errp);
380     /**
381      * @replay:
382      *
383      * Called to handle memory_region_iommu_replay().
384      *
385      * The default implementation of memory_region_iommu_replay() is to
386      * call the IOMMU translate method for every page in the address space
387      * with flag == IOMMU_NONE and then call the notifier if translate
388      * returns a valid mapping. If this method is implemented then it
389      * overrides the default behaviour, and must provide the full semantics
390      * of memory_region_iommu_replay(), by calling @notifier for every
391      * translation present in the IOMMU.
392      *
393      * Optional method -- an IOMMU only needs to provide this method
394      * if the default is inefficient or produces undesirable side effects.
395      *
396      * Note: this is not related to record-and-replay functionality.
397      */
398     void (*replay)(IOMMUMemoryRegion *iommu, IOMMUNotifier *notifier);
399 
400     /**
401      * @get_attr:
402      *
403      * Get IOMMU misc attributes. This is an optional method that
404      * can be used to allow users of the IOMMU to get implementation-specific
405      * information. The IOMMU implements this method to handle calls
406      * by IOMMU users to memory_region_iommu_get_attr() by filling in
407      * the arbitrary data pointer for any IOMMUMemoryRegionAttr values that
408      * the IOMMU supports. If the method is unimplemented then
409      * memory_region_iommu_get_attr() will always return -EINVAL.
410      *
411      * @iommu: the IOMMUMemoryRegion
412      *
413      * @attr: attribute being queried
414      *
415      * @data: memory to fill in with the attribute data
416      *
417      * Returns 0 on success, or a negative errno; in particular
418      * returns -EINVAL for unrecognized or unimplemented attribute types.
419      */
420     int (*get_attr)(IOMMUMemoryRegion *iommu, enum IOMMUMemoryRegionAttr attr,
421                     void *data);
422 
423     /**
424      * @attrs_to_index:
425      *
426      * Return the IOMMU index to use for a given set of transaction attributes.
427      *
428      * Optional method: if an IOMMU only supports a single IOMMU index then
429      * the default implementation of memory_region_iommu_attrs_to_index()
430      * will return 0.
431      *
432      * The indexes supported by an IOMMU must be contiguous, starting at 0.
433      *
434      * @iommu: the IOMMUMemoryRegion
435      * @attrs: memory transaction attributes
436      */
437     int (*attrs_to_index)(IOMMUMemoryRegion *iommu, MemTxAttrs attrs);
438 
439     /**
440      * @num_indexes:
441      *
442      * Return the number of IOMMU indexes this IOMMU supports.
443      *
444      * Optional method: if this method is not provided, then
445      * memory_region_iommu_num_indexes() will return 1, indicating that
446      * only a single IOMMU index is supported.
447      *
448      * @iommu: the IOMMUMemoryRegion
449      */
450     int (*num_indexes)(IOMMUMemoryRegion *iommu);
451 
452     /**
453      * @iommu_set_page_size_mask:
454      *
455      * Restrict the page size mask that can be supported with a given IOMMU
456      * memory region. Used for example to propagate host physical IOMMU page
457      * size mask limitations to the virtual IOMMU.
458      *
459      * Optional method: if this method is not provided, then the default global
460      * page mask is used.
461      *
462      * @iommu: the IOMMUMemoryRegion
463      *
464      * @page_size_mask: a bitmask of supported page sizes. At least one bit,
465      * representing the smallest page size, must be set. Additional set bits
466      * represent supported block sizes. For example a host physical IOMMU that
467      * uses page tables with a page size of 4kB, and supports 2MB and 4GB
468      * blocks, will set mask 0x40201000. A granule of 4kB with indiscriminate
469      * block sizes is specified with mask 0xfffffffffffff000.
470      *
471      * Returns 0 on success, or a negative error. In case of failure, the error
472      * object must be created.
473      */
474      int (*iommu_set_page_size_mask)(IOMMUMemoryRegion *iommu,
475                                      uint64_t page_size_mask,
476                                      Error **errp);
477 };
478 
479 typedef struct RamDiscardListener RamDiscardListener;
480 typedef int (*NotifyRamPopulate)(RamDiscardListener *rdl,
481                                  MemoryRegionSection *section);
482 typedef void (*NotifyRamDiscard)(RamDiscardListener *rdl,
483                                  MemoryRegionSection *section);
484 
485 struct RamDiscardListener {
486     /*
487      * @notify_populate:
488      *
489      * Notification that previously discarded memory is about to get populated.
490      * Listeners are able to object. If any listener objects, already
491      * successfully notified listeners are notified about a discard again.
492      *
493      * @rdl: the #RamDiscardListener getting notified
494      * @section: the #MemoryRegionSection to get populated. The section
495      *           is aligned within the memory region to the minimum granularity
496      *           unless it would exceed the registered section.
497      *
498      * Returns 0 on success. If the notification is rejected by the listener,
499      * an error is returned.
500      */
501     NotifyRamPopulate notify_populate;
502 
503     /*
504      * @notify_discard:
505      *
506      * Notification that previously populated memory was discarded successfully
507      * and listeners should drop all references to such memory and prevent
508      * new population (e.g., unmap).
509      *
510      * @rdl: the #RamDiscardListener getting notified
511      * @section: the #MemoryRegionSection to get populated. The section
512      *           is aligned within the memory region to the minimum granularity
513      *           unless it would exceed the registered section.
514      */
515     NotifyRamDiscard notify_discard;
516 
517     /*
518      * @double_discard_supported:
519      *
520      * The listener suppors getting @notify_discard notifications that span
521      * already discarded parts.
522      */
523     bool double_discard_supported;
524 
525     MemoryRegionSection *section;
526     QLIST_ENTRY(RamDiscardListener) next;
527 };
528 
529 static inline void ram_discard_listener_init(RamDiscardListener *rdl,
530                                              NotifyRamPopulate populate_fn,
531                                              NotifyRamDiscard discard_fn,
532                                              bool double_discard_supported)
533 {
534     rdl->notify_populate = populate_fn;
535     rdl->notify_discard = discard_fn;
536     rdl->double_discard_supported = double_discard_supported;
537 }
538 
539 typedef int (*ReplayRamPopulate)(MemoryRegionSection *section, void *opaque);
540 
541 /*
542  * RamDiscardManagerClass:
543  *
544  * A #RamDiscardManager coordinates which parts of specific RAM #MemoryRegion
545  * regions are currently populated to be used/accessed by the VM, notifying
546  * after parts were discarded (freeing up memory) and before parts will be
547  * populated (consuming memory), to be used/acessed by the VM.
548  *
549  * A #RamDiscardManager can only be set for a RAM #MemoryRegion while the
550  * #MemoryRegion isn't mapped yet; it cannot change while the #MemoryRegion is
551  * mapped.
552  *
553  * The #RamDiscardManager is intended to be used by technologies that are
554  * incompatible with discarding of RAM (e.g., VFIO, which may pin all
555  * memory inside a #MemoryRegion), and require proper coordination to only
556  * map the currently populated parts, to hinder parts that are expected to
557  * remain discarded from silently getting populated and consuming memory.
558  * Technologies that support discarding of RAM don't have to bother and can
559  * simply map the whole #MemoryRegion.
560  *
561  * An example #RamDiscardManager is virtio-mem, which logically (un)plugs
562  * memory within an assigned RAM #MemoryRegion, coordinated with the VM.
563  * Logically unplugging memory consists of discarding RAM. The VM agreed to not
564  * access unplugged (discarded) memory - especially via DMA. virtio-mem will
565  * properly coordinate with listeners before memory is plugged (populated),
566  * and after memory is unplugged (discarded).
567  *
568  * Listeners are called in multiples of the minimum granularity (unless it
569  * would exceed the registered range) and changes are aligned to the minimum
570  * granularity within the #MemoryRegion. Listeners have to prepare for memory
571  * becomming discarded in a different granularity than it was populated and the
572  * other way around.
573  */
574 struct RamDiscardManagerClass {
575     /* private */
576     InterfaceClass parent_class;
577 
578     /* public */
579 
580     /**
581      * @get_min_granularity:
582      *
583      * Get the minimum granularity in which listeners will get notified
584      * about changes within the #MemoryRegion via the #RamDiscardManager.
585      *
586      * @rdm: the #RamDiscardManager
587      * @mr: the #MemoryRegion
588      *
589      * Returns the minimum granularity.
590      */
591     uint64_t (*get_min_granularity)(const RamDiscardManager *rdm,
592                                     const MemoryRegion *mr);
593 
594     /**
595      * @is_populated:
596      *
597      * Check whether the given #MemoryRegionSection is completely populated
598      * (i.e., no parts are currently discarded) via the #RamDiscardManager.
599      * There are no alignment requirements.
600      *
601      * @rdm: the #RamDiscardManager
602      * @section: the #MemoryRegionSection
603      *
604      * Returns whether the given range is completely populated.
605      */
606     bool (*is_populated)(const RamDiscardManager *rdm,
607                          const MemoryRegionSection *section);
608 
609     /**
610      * @replay_populated:
611      *
612      * Call the #ReplayRamPopulate callback for all populated parts within the
613      * #MemoryRegionSection via the #RamDiscardManager.
614      *
615      * In case any call fails, no further calls are made.
616      *
617      * @rdm: the #RamDiscardManager
618      * @section: the #MemoryRegionSection
619      * @replay_fn: the #ReplayRamPopulate callback
620      * @opaque: pointer to forward to the callback
621      *
622      * Returns 0 on success, or a negative error if any notification failed.
623      */
624     int (*replay_populated)(const RamDiscardManager *rdm,
625                             MemoryRegionSection *section,
626                             ReplayRamPopulate replay_fn, void *opaque);
627 
628     /**
629      * @register_listener:
630      *
631      * Register a #RamDiscardListener for the given #MemoryRegionSection and
632      * immediately notify the #RamDiscardListener about all populated parts
633      * within the #MemoryRegionSection via the #RamDiscardManager.
634      *
635      * In case any notification fails, no further notifications are triggered
636      * and an error is logged.
637      *
638      * @rdm: the #RamDiscardManager
639      * @rdl: the #RamDiscardListener
640      * @section: the #MemoryRegionSection
641      */
642     void (*register_listener)(RamDiscardManager *rdm,
643                               RamDiscardListener *rdl,
644                               MemoryRegionSection *section);
645 
646     /**
647      * @unregister_listener:
648      *
649      * Unregister a previously registered #RamDiscardListener via the
650      * #RamDiscardManager after notifying the #RamDiscardListener about all
651      * populated parts becoming unpopulated within the registered
652      * #MemoryRegionSection.
653      *
654      * @rdm: the #RamDiscardManager
655      * @rdl: the #RamDiscardListener
656      */
657     void (*unregister_listener)(RamDiscardManager *rdm,
658                                 RamDiscardListener *rdl);
659 };
660 
661 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
662                                                  const MemoryRegion *mr);
663 
664 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
665                                       const MemoryRegionSection *section);
666 
667 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
668                                          MemoryRegionSection *section,
669                                          ReplayRamPopulate replay_fn,
670                                          void *opaque);
671 
672 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
673                                            RamDiscardListener *rdl,
674                                            MemoryRegionSection *section);
675 
676 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
677                                              RamDiscardListener *rdl);
678 
679 typedef struct CoalescedMemoryRange CoalescedMemoryRange;
680 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd;
681 
682 /** MemoryRegion:
683  *
684  * A struct representing a memory region.
685  */
686 struct MemoryRegion {
687     Object parent_obj;
688 
689     /* private: */
690 
691     /* The following fields should fit in a cache line */
692     bool romd_mode;
693     bool ram;
694     bool subpage;
695     bool readonly; /* For RAM regions */
696     bool nonvolatile;
697     bool rom_device;
698     bool flush_coalesced_mmio;
699     uint8_t dirty_log_mask;
700     bool is_iommu;
701     RAMBlock *ram_block;
702     Object *owner;
703 
704     const MemoryRegionOps *ops;
705     void *opaque;
706     MemoryRegion *container;
707     Int128 size;
708     hwaddr addr;
709     void (*destructor)(MemoryRegion *mr);
710     uint64_t align;
711     bool terminates;
712     bool ram_device;
713     bool enabled;
714     bool warning_printed; /* For reservations */
715     uint8_t vga_logging_count;
716     MemoryRegion *alias;
717     hwaddr alias_offset;
718     int32_t priority;
719     QTAILQ_HEAD(, MemoryRegion) subregions;
720     QTAILQ_ENTRY(MemoryRegion) subregions_link;
721     QTAILQ_HEAD(, CoalescedMemoryRange) coalesced;
722     const char *name;
723     unsigned ioeventfd_nb;
724     MemoryRegionIoeventfd *ioeventfds;
725     RamDiscardManager *rdm; /* Only for RAM */
726 };
727 
728 struct IOMMUMemoryRegion {
729     MemoryRegion parent_obj;
730 
731     QLIST_HEAD(, IOMMUNotifier) iommu_notify;
732     IOMMUNotifierFlag iommu_notify_flags;
733 };
734 
735 #define IOMMU_NOTIFIER_FOREACH(n, mr) \
736     QLIST_FOREACH((n), &(mr)->iommu_notify, node)
737 
738 /**
739  * struct MemoryListener: callbacks structure for updates to the physical memory map
740  *
741  * Allows a component to adjust to changes in the guest-visible memory map.
742  * Use with memory_listener_register() and memory_listener_unregister().
743  */
744 struct MemoryListener {
745     /**
746      * @begin:
747      *
748      * Called at the beginning of an address space update transaction.
749      * Followed by calls to #MemoryListener.region_add(),
750      * #MemoryListener.region_del(), #MemoryListener.region_nop(),
751      * #MemoryListener.log_start() and #MemoryListener.log_stop() in
752      * increasing address order.
753      *
754      * @listener: The #MemoryListener.
755      */
756     void (*begin)(MemoryListener *listener);
757 
758     /**
759      * @commit:
760      *
761      * Called at the end of an address space update transaction,
762      * after the last call to #MemoryListener.region_add(),
763      * #MemoryListener.region_del() or #MemoryListener.region_nop(),
764      * #MemoryListener.log_start() and #MemoryListener.log_stop().
765      *
766      * @listener: The #MemoryListener.
767      */
768     void (*commit)(MemoryListener *listener);
769 
770     /**
771      * @region_add:
772      *
773      * Called during an address space update transaction,
774      * for a section of the address space that is new in this address space
775      * space since the last transaction.
776      *
777      * @listener: The #MemoryListener.
778      * @section: The new #MemoryRegionSection.
779      */
780     void (*region_add)(MemoryListener *listener, MemoryRegionSection *section);
781 
782     /**
783      * @region_del:
784      *
785      * Called during an address space update transaction,
786      * for a section of the address space that has disappeared in the address
787      * space since the last transaction.
788      *
789      * @listener: The #MemoryListener.
790      * @section: The old #MemoryRegionSection.
791      */
792     void (*region_del)(MemoryListener *listener, MemoryRegionSection *section);
793 
794     /**
795      * @region_nop:
796      *
797      * Called during an address space update transaction,
798      * for a section of the address space that is in the same place in the address
799      * space as in the last transaction.
800      *
801      * @listener: The #MemoryListener.
802      * @section: The #MemoryRegionSection.
803      */
804     void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section);
805 
806     /**
807      * @log_start:
808      *
809      * Called during an address space update transaction, after
810      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
811      * #MemoryListener.region_nop(), if dirty memory logging clients have
812      * become active since the last transaction.
813      *
814      * @listener: The #MemoryListener.
815      * @section: The #MemoryRegionSection.
816      * @old: A bitmap of dirty memory logging clients that were active in
817      * the previous transaction.
818      * @new: A bitmap of dirty memory logging clients that are active in
819      * the current transaction.
820      */
821     void (*log_start)(MemoryListener *listener, MemoryRegionSection *section,
822                       int old, int new);
823 
824     /**
825      * @log_stop:
826      *
827      * Called during an address space update transaction, after
828      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
829      * #MemoryListener.region_nop() and possibly after
830      * #MemoryListener.log_start(), if dirty memory logging clients have
831      * become inactive since the last transaction.
832      *
833      * @listener: The #MemoryListener.
834      * @section: The #MemoryRegionSection.
835      * @old: A bitmap of dirty memory logging clients that were active in
836      * the previous transaction.
837      * @new: A bitmap of dirty memory logging clients that are active in
838      * the current transaction.
839      */
840     void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section,
841                      int old, int new);
842 
843     /**
844      * @log_sync:
845      *
846      * Called by memory_region_snapshot_and_clear_dirty() and
847      * memory_global_dirty_log_sync(), before accessing QEMU's "official"
848      * copy of the dirty memory bitmap for a #MemoryRegionSection.
849      *
850      * @listener: The #MemoryListener.
851      * @section: The #MemoryRegionSection.
852      */
853     void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section);
854 
855     /**
856      * @log_sync_global:
857      *
858      * This is the global version of @log_sync when the listener does
859      * not have a way to synchronize the log with finer granularity.
860      * When the listener registers with @log_sync_global defined, then
861      * its @log_sync must be NULL.  Vice versa.
862      *
863      * @listener: The #MemoryListener.
864      */
865     void (*log_sync_global)(MemoryListener *listener);
866 
867     /**
868      * @log_clear:
869      *
870      * Called before reading the dirty memory bitmap for a
871      * #MemoryRegionSection.
872      *
873      * @listener: The #MemoryListener.
874      * @section: The #MemoryRegionSection.
875      */
876     void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section);
877 
878     /**
879      * @log_global_start:
880      *
881      * Called by memory_global_dirty_log_start(), which
882      * enables the %DIRTY_LOG_MIGRATION client on all memory regions in
883      * the address space.  #MemoryListener.log_global_start() is also
884      * called when a #MemoryListener is added, if global dirty logging is
885      * active at that time.
886      *
887      * @listener: The #MemoryListener.
888      */
889     void (*log_global_start)(MemoryListener *listener);
890 
891     /**
892      * @log_global_stop:
893      *
894      * Called by memory_global_dirty_log_stop(), which
895      * disables the %DIRTY_LOG_MIGRATION client on all memory regions in
896      * the address space.
897      *
898      * @listener: The #MemoryListener.
899      */
900     void (*log_global_stop)(MemoryListener *listener);
901 
902     /**
903      * @log_global_after_sync:
904      *
905      * Called after reading the dirty memory bitmap
906      * for any #MemoryRegionSection.
907      *
908      * @listener: The #MemoryListener.
909      */
910     void (*log_global_after_sync)(MemoryListener *listener);
911 
912     /**
913      * @eventfd_add:
914      *
915      * Called during an address space update transaction,
916      * for a section of the address space that has had a new ioeventfd
917      * registration since the last transaction.
918      *
919      * @listener: The #MemoryListener.
920      * @section: The new #MemoryRegionSection.
921      * @match_data: The @match_data parameter for the new ioeventfd.
922      * @data: The @data parameter for the new ioeventfd.
923      * @e: The #EventNotifier parameter for the new ioeventfd.
924      */
925     void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section,
926                         bool match_data, uint64_t data, EventNotifier *e);
927 
928     /**
929      * @eventfd_del:
930      *
931      * Called during an address space update transaction,
932      * for a section of the address space that has dropped an ioeventfd
933      * registration since the last transaction.
934      *
935      * @listener: The #MemoryListener.
936      * @section: The new #MemoryRegionSection.
937      * @match_data: The @match_data parameter for the dropped ioeventfd.
938      * @data: The @data parameter for the dropped ioeventfd.
939      * @e: The #EventNotifier parameter for the dropped ioeventfd.
940      */
941     void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section,
942                         bool match_data, uint64_t data, EventNotifier *e);
943 
944     /**
945      * @coalesced_io_add:
946      *
947      * Called during an address space update transaction,
948      * for a section of the address space that has had a new coalesced
949      * MMIO range registration since the last transaction.
950      *
951      * @listener: The #MemoryListener.
952      * @section: The new #MemoryRegionSection.
953      * @addr: The starting address for the coalesced MMIO range.
954      * @len: The length of the coalesced MMIO range.
955      */
956     void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section,
957                                hwaddr addr, hwaddr len);
958 
959     /**
960      * @coalesced_io_del:
961      *
962      * Called during an address space update transaction,
963      * for a section of the address space that has dropped a coalesced
964      * MMIO range since the last transaction.
965      *
966      * @listener: The #MemoryListener.
967      * @section: The new #MemoryRegionSection.
968      * @addr: The starting address for the coalesced MMIO range.
969      * @len: The length of the coalesced MMIO range.
970      */
971     void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section,
972                                hwaddr addr, hwaddr len);
973     /**
974      * @priority:
975      *
976      * Govern the order in which memory listeners are invoked. Lower priorities
977      * are invoked earlier for "add" or "start" callbacks, and later for "delete"
978      * or "stop" callbacks.
979      */
980     unsigned priority;
981 
982     /* private: */
983     AddressSpace *address_space;
984     QTAILQ_ENTRY(MemoryListener) link;
985     QTAILQ_ENTRY(MemoryListener) link_as;
986 };
987 
988 /**
989  * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects
990  */
991 struct AddressSpace {
992     /* private: */
993     struct rcu_head rcu;
994     char *name;
995     MemoryRegion *root;
996 
997     /* Accessed via RCU.  */
998     struct FlatView *current_map;
999 
1000     int ioeventfd_nb;
1001     struct MemoryRegionIoeventfd *ioeventfds;
1002     QTAILQ_HEAD(, MemoryListener) listeners;
1003     QTAILQ_ENTRY(AddressSpace) address_spaces_link;
1004 };
1005 
1006 typedef struct AddressSpaceDispatch AddressSpaceDispatch;
1007 typedef struct FlatRange FlatRange;
1008 
1009 /* Flattened global view of current active memory hierarchy.  Kept in sorted
1010  * order.
1011  */
1012 struct FlatView {
1013     struct rcu_head rcu;
1014     unsigned ref;
1015     FlatRange *ranges;
1016     unsigned nr;
1017     unsigned nr_allocated;
1018     struct AddressSpaceDispatch *dispatch;
1019     MemoryRegion *root;
1020 };
1021 
1022 static inline FlatView *address_space_to_flatview(AddressSpace *as)
1023 {
1024     return qatomic_rcu_read(&as->current_map);
1025 }
1026 
1027 /**
1028  * typedef flatview_cb: callback for flatview_for_each_range()
1029  *
1030  * @start: start address of the range within the FlatView
1031  * @len: length of the range in bytes
1032  * @mr: MemoryRegion covering this range
1033  * @offset_in_region: offset of the first byte of the range within @mr
1034  * @opaque: data pointer passed to flatview_for_each_range()
1035  *
1036  * Returns: true to stop the iteration, false to keep going.
1037  */
1038 typedef bool (*flatview_cb)(Int128 start,
1039                             Int128 len,
1040                             const MemoryRegion *mr,
1041                             hwaddr offset_in_region,
1042                             void *opaque);
1043 
1044 /**
1045  * flatview_for_each_range: Iterate through a FlatView
1046  * @fv: the FlatView to iterate through
1047  * @cb: function to call for each range
1048  * @opaque: opaque data pointer to pass to @cb
1049  *
1050  * A FlatView is made up of a list of non-overlapping ranges, each of
1051  * which is a slice of a MemoryRegion. This function iterates through
1052  * each range in @fv, calling @cb. The callback function can terminate
1053  * iteration early by returning 'true'.
1054  */
1055 void flatview_for_each_range(FlatView *fv, flatview_cb cb, void *opaque);
1056 
1057 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a,
1058                                           MemoryRegionSection *b)
1059 {
1060     return a->mr == b->mr &&
1061            a->fv == b->fv &&
1062            a->offset_within_region == b->offset_within_region &&
1063            a->offset_within_address_space == b->offset_within_address_space &&
1064            int128_eq(a->size, b->size) &&
1065            a->readonly == b->readonly &&
1066            a->nonvolatile == b->nonvolatile;
1067 }
1068 
1069 /**
1070  * memory_region_section_new_copy: Copy a memory region section
1071  *
1072  * Allocate memory for a new copy, copy the memory region section, and
1073  * properly take a reference on all relevant members.
1074  *
1075  * @s: the #MemoryRegionSection to copy
1076  */
1077 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s);
1078 
1079 /**
1080  * memory_region_section_new_copy: Free a copied memory region section
1081  *
1082  * Free a copy of a memory section created via memory_region_section_new_copy().
1083  * properly dropping references on all relevant members.
1084  *
1085  * @s: the #MemoryRegionSection to copy
1086  */
1087 void memory_region_section_free_copy(MemoryRegionSection *s);
1088 
1089 /**
1090  * memory_region_init: Initialize a memory region
1091  *
1092  * The region typically acts as a container for other memory regions.  Use
1093  * memory_region_add_subregion() to add subregions.
1094  *
1095  * @mr: the #MemoryRegion to be initialized
1096  * @owner: the object that tracks the region's reference count
1097  * @name: used for debugging; not visible to the user or ABI
1098  * @size: size of the region; any subregions beyond this size will be clipped
1099  */
1100 void memory_region_init(MemoryRegion *mr,
1101                         Object *owner,
1102                         const char *name,
1103                         uint64_t size);
1104 
1105 /**
1106  * memory_region_ref: Add 1 to a memory region's reference count
1107  *
1108  * Whenever memory regions are accessed outside the BQL, they need to be
1109  * preserved against hot-unplug.  MemoryRegions actually do not have their
1110  * own reference count; they piggyback on a QOM object, their "owner".
1111  * This function adds a reference to the owner.
1112  *
1113  * All MemoryRegions must have an owner if they can disappear, even if the
1114  * device they belong to operates exclusively under the BQL.  This is because
1115  * the region could be returned at any time by memory_region_find, and this
1116  * is usually under guest control.
1117  *
1118  * @mr: the #MemoryRegion
1119  */
1120 void memory_region_ref(MemoryRegion *mr);
1121 
1122 /**
1123  * memory_region_unref: Remove 1 to a memory region's reference count
1124  *
1125  * Whenever memory regions are accessed outside the BQL, they need to be
1126  * preserved against hot-unplug.  MemoryRegions actually do not have their
1127  * own reference count; they piggyback on a QOM object, their "owner".
1128  * This function removes a reference to the owner and possibly destroys it.
1129  *
1130  * @mr: the #MemoryRegion
1131  */
1132 void memory_region_unref(MemoryRegion *mr);
1133 
1134 /**
1135  * memory_region_init_io: Initialize an I/O memory region.
1136  *
1137  * Accesses into the region will cause the callbacks in @ops to be called.
1138  * if @size is nonzero, subregions will be clipped to @size.
1139  *
1140  * @mr: the #MemoryRegion to be initialized.
1141  * @owner: the object that tracks the region's reference count
1142  * @ops: a structure containing read and write callbacks to be used when
1143  *       I/O is performed on the region.
1144  * @opaque: passed to the read and write callbacks of the @ops structure.
1145  * @name: used for debugging; not visible to the user or ABI
1146  * @size: size of the region.
1147  */
1148 void memory_region_init_io(MemoryRegion *mr,
1149                            Object *owner,
1150                            const MemoryRegionOps *ops,
1151                            void *opaque,
1152                            const char *name,
1153                            uint64_t size);
1154 
1155 /**
1156  * memory_region_init_ram_nomigrate:  Initialize RAM memory region.  Accesses
1157  *                                    into the region will modify memory
1158  *                                    directly.
1159  *
1160  * @mr: the #MemoryRegion to be initialized.
1161  * @owner: the object that tracks the region's reference count
1162  * @name: Region name, becomes part of RAMBlock name used in migration stream
1163  *        must be unique within any device
1164  * @size: size of the region.
1165  * @errp: pointer to Error*, to store an error if it happens.
1166  *
1167  * Note that this function does not do anything to cause the data in the
1168  * RAM memory region to be migrated; that is the responsibility of the caller.
1169  */
1170 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1171                                       Object *owner,
1172                                       const char *name,
1173                                       uint64_t size,
1174                                       Error **errp);
1175 
1176 /**
1177  * memory_region_init_ram_flags_nomigrate:  Initialize RAM memory region.
1178  *                                          Accesses into the region will
1179  *                                          modify memory directly.
1180  *
1181  * @mr: the #MemoryRegion to be initialized.
1182  * @owner: the object that tracks the region's reference count
1183  * @name: Region name, becomes part of RAMBlock name used in migration stream
1184  *        must be unique within any device
1185  * @size: size of the region.
1186  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_NORESERVE.
1187  * @errp: pointer to Error*, to store an error if it happens.
1188  *
1189  * Note that this function does not do anything to cause the data in the
1190  * RAM memory region to be migrated; that is the responsibility of the caller.
1191  */
1192 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1193                                             Object *owner,
1194                                             const char *name,
1195                                             uint64_t size,
1196                                             uint32_t ram_flags,
1197                                             Error **errp);
1198 
1199 /**
1200  * memory_region_init_resizeable_ram:  Initialize memory region with resizeable
1201  *                                     RAM.  Accesses into the region will
1202  *                                     modify memory directly.  Only an initial
1203  *                                     portion of this RAM is actually used.
1204  *                                     Changing the size while migrating
1205  *                                     can result in the migration being
1206  *                                     canceled.
1207  *
1208  * @mr: the #MemoryRegion to be initialized.
1209  * @owner: the object that tracks the region's reference count
1210  * @name: Region name, becomes part of RAMBlock name used in migration stream
1211  *        must be unique within any device
1212  * @size: used size of the region.
1213  * @max_size: max size of the region.
1214  * @resized: callback to notify owner about used size change.
1215  * @errp: pointer to Error*, to store an error if it happens.
1216  *
1217  * Note that this function does not do anything to cause the data in the
1218  * RAM memory region to be migrated; that is the responsibility of the caller.
1219  */
1220 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1221                                        Object *owner,
1222                                        const char *name,
1223                                        uint64_t size,
1224                                        uint64_t max_size,
1225                                        void (*resized)(const char*,
1226                                                        uint64_t length,
1227                                                        void *host),
1228                                        Error **errp);
1229 #ifdef CONFIG_POSIX
1230 
1231 /**
1232  * memory_region_init_ram_from_file:  Initialize RAM memory region with a
1233  *                                    mmap-ed backend.
1234  *
1235  * @mr: the #MemoryRegion to be initialized.
1236  * @owner: the object that tracks the region's reference count
1237  * @name: Region name, becomes part of RAMBlock name used in migration stream
1238  *        must be unique within any device
1239  * @size: size of the region.
1240  * @align: alignment of the region base address; if 0, the default alignment
1241  *         (getpagesize()) will be used.
1242  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1243  *             RAM_NORESERVE,
1244  * @path: the path in which to allocate the RAM.
1245  * @readonly: true to open @path for reading, false for read/write.
1246  * @errp: pointer to Error*, to store an error if it happens.
1247  *
1248  * Note that this function does not do anything to cause the data in the
1249  * RAM memory region to be migrated; that is the responsibility of the caller.
1250  */
1251 void memory_region_init_ram_from_file(MemoryRegion *mr,
1252                                       Object *owner,
1253                                       const char *name,
1254                                       uint64_t size,
1255                                       uint64_t align,
1256                                       uint32_t ram_flags,
1257                                       const char *path,
1258                                       bool readonly,
1259                                       Error **errp);
1260 
1261 /**
1262  * memory_region_init_ram_from_fd:  Initialize RAM memory region with a
1263  *                                  mmap-ed backend.
1264  *
1265  * @mr: the #MemoryRegion to be initialized.
1266  * @owner: the object that tracks the region's reference count
1267  * @name: the name of the region.
1268  * @size: size of the region.
1269  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1270  *             RAM_NORESERVE.
1271  * @fd: the fd to mmap.
1272  * @offset: offset within the file referenced by fd
1273  * @errp: pointer to Error*, to store an error if it happens.
1274  *
1275  * Note that this function does not do anything to cause the data in the
1276  * RAM memory region to be migrated; that is the responsibility of the caller.
1277  */
1278 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1279                                     Object *owner,
1280                                     const char *name,
1281                                     uint64_t size,
1282                                     uint32_t ram_flags,
1283                                     int fd,
1284                                     ram_addr_t offset,
1285                                     Error **errp);
1286 #endif
1287 
1288 /**
1289  * memory_region_init_ram_ptr:  Initialize RAM memory region from a
1290  *                              user-provided pointer.  Accesses into the
1291  *                              region will modify memory directly.
1292  *
1293  * @mr: the #MemoryRegion to be initialized.
1294  * @owner: the object that tracks the region's reference count
1295  * @name: Region name, becomes part of RAMBlock name used in migration stream
1296  *        must be unique within any device
1297  * @size: size of the region.
1298  * @ptr: memory to be mapped; must contain at least @size bytes.
1299  *
1300  * Note that this function does not do anything to cause the data in the
1301  * RAM memory region to be migrated; that is the responsibility of the caller.
1302  */
1303 void memory_region_init_ram_ptr(MemoryRegion *mr,
1304                                 Object *owner,
1305                                 const char *name,
1306                                 uint64_t size,
1307                                 void *ptr);
1308 
1309 /**
1310  * memory_region_init_ram_device_ptr:  Initialize RAM device memory region from
1311  *                                     a user-provided pointer.
1312  *
1313  * A RAM device represents a mapping to a physical device, such as to a PCI
1314  * MMIO BAR of an vfio-pci assigned device.  The memory region may be mapped
1315  * into the VM address space and access to the region will modify memory
1316  * directly.  However, the memory region should not be included in a memory
1317  * dump (device may not be enabled/mapped at the time of the dump), and
1318  * operations incompatible with manipulating MMIO should be avoided.  Replaces
1319  * skip_dump flag.
1320  *
1321  * @mr: the #MemoryRegion to be initialized.
1322  * @owner: the object that tracks the region's reference count
1323  * @name: the name of the region.
1324  * @size: size of the region.
1325  * @ptr: memory to be mapped; must contain at least @size bytes.
1326  *
1327  * Note that this function does not do anything to cause the data in the
1328  * RAM memory region to be migrated; that is the responsibility of the caller.
1329  * (For RAM device memory regions, migrating the contents rarely makes sense.)
1330  */
1331 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1332                                        Object *owner,
1333                                        const char *name,
1334                                        uint64_t size,
1335                                        void *ptr);
1336 
1337 /**
1338  * memory_region_init_alias: Initialize a memory region that aliases all or a
1339  *                           part of another memory region.
1340  *
1341  * @mr: the #MemoryRegion to be initialized.
1342  * @owner: the object that tracks the region's reference count
1343  * @name: used for debugging; not visible to the user or ABI
1344  * @orig: the region to be referenced; @mr will be equivalent to
1345  *        @orig between @offset and @offset + @size - 1.
1346  * @offset: start of the section in @orig to be referenced.
1347  * @size: size of the region.
1348  */
1349 void memory_region_init_alias(MemoryRegion *mr,
1350                               Object *owner,
1351                               const char *name,
1352                               MemoryRegion *orig,
1353                               hwaddr offset,
1354                               uint64_t size);
1355 
1356 /**
1357  * memory_region_init_rom_nomigrate: Initialize a ROM memory region.
1358  *
1359  * This has the same effect as calling memory_region_init_ram_nomigrate()
1360  * and then marking the resulting region read-only with
1361  * memory_region_set_readonly().
1362  *
1363  * Note that this function does not do anything to cause the data in the
1364  * RAM side of the memory region to be migrated; that is the responsibility
1365  * of the caller.
1366  *
1367  * @mr: the #MemoryRegion to be initialized.
1368  * @owner: the object that tracks the region's reference count
1369  * @name: Region name, becomes part of RAMBlock name used in migration stream
1370  *        must be unique within any device
1371  * @size: size of the region.
1372  * @errp: pointer to Error*, to store an error if it happens.
1373  */
1374 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1375                                       Object *owner,
1376                                       const char *name,
1377                                       uint64_t size,
1378                                       Error **errp);
1379 
1380 /**
1381  * memory_region_init_rom_device_nomigrate:  Initialize a ROM memory region.
1382  *                                 Writes are handled via callbacks.
1383  *
1384  * Note that this function does not do anything to cause the data in the
1385  * RAM side of the memory region to be migrated; that is the responsibility
1386  * of the caller.
1387  *
1388  * @mr: the #MemoryRegion to be initialized.
1389  * @owner: the object that tracks the region's reference count
1390  * @ops: callbacks for write access handling (must not be NULL).
1391  * @opaque: passed to the read and write callbacks of the @ops structure.
1392  * @name: Region name, becomes part of RAMBlock name used in migration stream
1393  *        must be unique within any device
1394  * @size: size of the region.
1395  * @errp: pointer to Error*, to store an error if it happens.
1396  */
1397 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1398                                              Object *owner,
1399                                              const MemoryRegionOps *ops,
1400                                              void *opaque,
1401                                              const char *name,
1402                                              uint64_t size,
1403                                              Error **errp);
1404 
1405 /**
1406  * memory_region_init_iommu: Initialize a memory region of a custom type
1407  * that translates addresses
1408  *
1409  * An IOMMU region translates addresses and forwards accesses to a target
1410  * memory region.
1411  *
1412  * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION.
1413  * @_iommu_mr should be a pointer to enough memory for an instance of
1414  * that subclass, @instance_size is the size of that subclass, and
1415  * @mrtypename is its name. This function will initialize @_iommu_mr as an
1416  * instance of the subclass, and its methods will then be called to handle
1417  * accesses to the memory region. See the documentation of
1418  * #IOMMUMemoryRegionClass for further details.
1419  *
1420  * @_iommu_mr: the #IOMMUMemoryRegion to be initialized
1421  * @instance_size: the IOMMUMemoryRegion subclass instance size
1422  * @mrtypename: the type name of the #IOMMUMemoryRegion
1423  * @owner: the object that tracks the region's reference count
1424  * @name: used for debugging; not visible to the user or ABI
1425  * @size: size of the region.
1426  */
1427 void memory_region_init_iommu(void *_iommu_mr,
1428                               size_t instance_size,
1429                               const char *mrtypename,
1430                               Object *owner,
1431                               const char *name,
1432                               uint64_t size);
1433 
1434 /**
1435  * memory_region_init_ram - Initialize RAM memory region.  Accesses into the
1436  *                          region will modify memory directly.
1437  *
1438  * @mr: the #MemoryRegion to be initialized
1439  * @owner: the object that tracks the region's reference count (must be
1440  *         TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL)
1441  * @name: name of the memory region
1442  * @size: size of the region in bytes
1443  * @errp: pointer to Error*, to store an error if it happens.
1444  *
1445  * This function allocates RAM for a board model or device, and
1446  * arranges for it to be migrated (by calling vmstate_register_ram()
1447  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1448  * @owner is NULL).
1449  *
1450  * TODO: Currently we restrict @owner to being either NULL (for
1451  * global RAM regions with no owner) or devices, so that we can
1452  * give the RAM block a unique name for migration purposes.
1453  * We should lift this restriction and allow arbitrary Objects.
1454  * If you pass a non-NULL non-device @owner then we will assert.
1455  */
1456 void memory_region_init_ram(MemoryRegion *mr,
1457                             Object *owner,
1458                             const char *name,
1459                             uint64_t size,
1460                             Error **errp);
1461 
1462 /**
1463  * memory_region_init_rom: Initialize a ROM memory region.
1464  *
1465  * This has the same effect as calling memory_region_init_ram()
1466  * and then marking the resulting region read-only with
1467  * memory_region_set_readonly(). This includes arranging for the
1468  * contents to be migrated.
1469  *
1470  * TODO: Currently we restrict @owner to being either NULL (for
1471  * global RAM regions with no owner) or devices, so that we can
1472  * give the RAM block a unique name for migration purposes.
1473  * We should lift this restriction and allow arbitrary Objects.
1474  * If you pass a non-NULL non-device @owner then we will assert.
1475  *
1476  * @mr: the #MemoryRegion to be initialized.
1477  * @owner: the object that tracks the region's reference count
1478  * @name: Region name, becomes part of RAMBlock name used in migration stream
1479  *        must be unique within any device
1480  * @size: size of the region.
1481  * @errp: pointer to Error*, to store an error if it happens.
1482  */
1483 void memory_region_init_rom(MemoryRegion *mr,
1484                             Object *owner,
1485                             const char *name,
1486                             uint64_t size,
1487                             Error **errp);
1488 
1489 /**
1490  * memory_region_init_rom_device:  Initialize a ROM memory region.
1491  *                                 Writes are handled via callbacks.
1492  *
1493  * This function initializes a memory region backed by RAM for reads
1494  * and callbacks for writes, and arranges for the RAM backing to
1495  * be migrated (by calling vmstate_register_ram()
1496  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1497  * @owner is NULL).
1498  *
1499  * TODO: Currently we restrict @owner to being either NULL (for
1500  * global RAM regions with no owner) or devices, so that we can
1501  * give the RAM block a unique name for migration purposes.
1502  * We should lift this restriction and allow arbitrary Objects.
1503  * If you pass a non-NULL non-device @owner then we will assert.
1504  *
1505  * @mr: the #MemoryRegion to be initialized.
1506  * @owner: the object that tracks the region's reference count
1507  * @ops: callbacks for write access handling (must not be NULL).
1508  * @opaque: passed to the read and write callbacks of the @ops structure.
1509  * @name: Region name, becomes part of RAMBlock name used in migration stream
1510  *        must be unique within any device
1511  * @size: size of the region.
1512  * @errp: pointer to Error*, to store an error if it happens.
1513  */
1514 void memory_region_init_rom_device(MemoryRegion *mr,
1515                                    Object *owner,
1516                                    const MemoryRegionOps *ops,
1517                                    void *opaque,
1518                                    const char *name,
1519                                    uint64_t size,
1520                                    Error **errp);
1521 
1522 
1523 /**
1524  * memory_region_owner: get a memory region's owner.
1525  *
1526  * @mr: the memory region being queried.
1527  */
1528 Object *memory_region_owner(MemoryRegion *mr);
1529 
1530 /**
1531  * memory_region_size: get a memory region's size.
1532  *
1533  * @mr: the memory region being queried.
1534  */
1535 uint64_t memory_region_size(MemoryRegion *mr);
1536 
1537 /**
1538  * memory_region_is_ram: check whether a memory region is random access
1539  *
1540  * Returns %true if a memory region is random access.
1541  *
1542  * @mr: the memory region being queried
1543  */
1544 static inline bool memory_region_is_ram(MemoryRegion *mr)
1545 {
1546     return mr->ram;
1547 }
1548 
1549 /**
1550  * memory_region_is_ram_device: check whether a memory region is a ram device
1551  *
1552  * Returns %true if a memory region is a device backed ram region
1553  *
1554  * @mr: the memory region being queried
1555  */
1556 bool memory_region_is_ram_device(MemoryRegion *mr);
1557 
1558 /**
1559  * memory_region_is_romd: check whether a memory region is in ROMD mode
1560  *
1561  * Returns %true if a memory region is a ROM device and currently set to allow
1562  * direct reads.
1563  *
1564  * @mr: the memory region being queried
1565  */
1566 static inline bool memory_region_is_romd(MemoryRegion *mr)
1567 {
1568     return mr->rom_device && mr->romd_mode;
1569 }
1570 
1571 /**
1572  * memory_region_get_iommu: check whether a memory region is an iommu
1573  *
1574  * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu,
1575  * otherwise NULL.
1576  *
1577  * @mr: the memory region being queried
1578  */
1579 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr)
1580 {
1581     if (mr->alias) {
1582         return memory_region_get_iommu(mr->alias);
1583     }
1584     if (mr->is_iommu) {
1585         return (IOMMUMemoryRegion *) mr;
1586     }
1587     return NULL;
1588 }
1589 
1590 /**
1591  * memory_region_get_iommu_class_nocheck: returns iommu memory region class
1592  *   if an iommu or NULL if not
1593  *
1594  * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu,
1595  * otherwise NULL. This is fast path avoiding QOM checking, use with caution.
1596  *
1597  * @iommu_mr: the memory region being queried
1598  */
1599 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck(
1600         IOMMUMemoryRegion *iommu_mr)
1601 {
1602     return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class);
1603 }
1604 
1605 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL)
1606 
1607 /**
1608  * memory_region_iommu_get_min_page_size: get minimum supported page size
1609  * for an iommu
1610  *
1611  * Returns minimum supported page size for an iommu.
1612  *
1613  * @iommu_mr: the memory region being queried
1614  */
1615 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr);
1616 
1617 /**
1618  * memory_region_notify_iommu: notify a change in an IOMMU translation entry.
1619  *
1620  * Note: for any IOMMU implementation, an in-place mapping change
1621  * should be notified with an UNMAP followed by a MAP.
1622  *
1623  * @iommu_mr: the memory region that was changed
1624  * @iommu_idx: the IOMMU index for the translation table which has changed
1625  * @event: TLB event with the new entry in the IOMMU translation table.
1626  *         The entry replaces all old entries for the same virtual I/O address
1627  *         range.
1628  */
1629 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1630                                 int iommu_idx,
1631                                 IOMMUTLBEvent event);
1632 
1633 /**
1634  * memory_region_notify_iommu_one: notify a change in an IOMMU translation
1635  *                           entry to a single notifier
1636  *
1637  * This works just like memory_region_notify_iommu(), but it only
1638  * notifies a specific notifier, not all of them.
1639  *
1640  * @notifier: the notifier to be notified
1641  * @event: TLB event with the new entry in the IOMMU translation table.
1642  *         The entry replaces all old entries for the same virtual I/O address
1643  *         range.
1644  */
1645 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1646                                     IOMMUTLBEvent *event);
1647 
1648 /**
1649  * memory_region_register_iommu_notifier: register a notifier for changes to
1650  * IOMMU translation entries.
1651  *
1652  * Returns 0 on success, or a negative errno otherwise. In particular,
1653  * -EINVAL indicates that at least one of the attributes of the notifier
1654  * is not supported (flag/range) by the IOMMU memory region. In case of error
1655  * the error object must be created.
1656  *
1657  * @mr: the memory region to observe
1658  * @n: the IOMMUNotifier to be added; the notify callback receives a
1659  *     pointer to an #IOMMUTLBEntry as the opaque value; the pointer
1660  *     ceases to be valid on exit from the notifier.
1661  * @errp: pointer to Error*, to store an error if it happens.
1662  */
1663 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1664                                           IOMMUNotifier *n, Error **errp);
1665 
1666 /**
1667  * memory_region_iommu_replay: replay existing IOMMU translations to
1668  * a notifier with the minimum page granularity returned by
1669  * mr->iommu_ops->get_page_size().
1670  *
1671  * Note: this is not related to record-and-replay functionality.
1672  *
1673  * @iommu_mr: the memory region to observe
1674  * @n: the notifier to which to replay iommu mappings
1675  */
1676 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
1677 
1678 /**
1679  * memory_region_unregister_iommu_notifier: unregister a notifier for
1680  * changes to IOMMU translation entries.
1681  *
1682  * @mr: the memory region which was observed and for which notity_stopped()
1683  *      needs to be called
1684  * @n: the notifier to be removed.
1685  */
1686 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1687                                              IOMMUNotifier *n);
1688 
1689 /**
1690  * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is
1691  * defined on the IOMMU.
1692  *
1693  * Returns 0 on success, or a negative errno otherwise. In particular,
1694  * -EINVAL indicates that the IOMMU does not support the requested
1695  * attribute.
1696  *
1697  * @iommu_mr: the memory region
1698  * @attr: the requested attribute
1699  * @data: a pointer to the requested attribute data
1700  */
1701 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1702                                  enum IOMMUMemoryRegionAttr attr,
1703                                  void *data);
1704 
1705 /**
1706  * memory_region_iommu_attrs_to_index: return the IOMMU index to
1707  * use for translations with the given memory transaction attributes.
1708  *
1709  * @iommu_mr: the memory region
1710  * @attrs: the memory transaction attributes
1711  */
1712 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1713                                        MemTxAttrs attrs);
1714 
1715 /**
1716  * memory_region_iommu_num_indexes: return the total number of IOMMU
1717  * indexes that this IOMMU supports.
1718  *
1719  * @iommu_mr: the memory region
1720  */
1721 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr);
1722 
1723 /**
1724  * memory_region_iommu_set_page_size_mask: set the supported page
1725  * sizes for a given IOMMU memory region
1726  *
1727  * @iommu_mr: IOMMU memory region
1728  * @page_size_mask: supported page size mask
1729  * @errp: pointer to Error*, to store an error if it happens.
1730  */
1731 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1732                                            uint64_t page_size_mask,
1733                                            Error **errp);
1734 
1735 /**
1736  * memory_region_name: get a memory region's name
1737  *
1738  * Returns the string that was used to initialize the memory region.
1739  *
1740  * @mr: the memory region being queried
1741  */
1742 const char *memory_region_name(const MemoryRegion *mr);
1743 
1744 /**
1745  * memory_region_is_logging: return whether a memory region is logging writes
1746  *
1747  * Returns %true if the memory region is logging writes for the given client
1748  *
1749  * @mr: the memory region being queried
1750  * @client: the client being queried
1751  */
1752 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client);
1753 
1754 /**
1755  * memory_region_get_dirty_log_mask: return the clients for which a
1756  * memory region is logging writes.
1757  *
1758  * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants
1759  * are the bit indices.
1760  *
1761  * @mr: the memory region being queried
1762  */
1763 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr);
1764 
1765 /**
1766  * memory_region_is_rom: check whether a memory region is ROM
1767  *
1768  * Returns %true if a memory region is read-only memory.
1769  *
1770  * @mr: the memory region being queried
1771  */
1772 static inline bool memory_region_is_rom(MemoryRegion *mr)
1773 {
1774     return mr->ram && mr->readonly;
1775 }
1776 
1777 /**
1778  * memory_region_is_nonvolatile: check whether a memory region is non-volatile
1779  *
1780  * Returns %true is a memory region is non-volatile memory.
1781  *
1782  * @mr: the memory region being queried
1783  */
1784 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr)
1785 {
1786     return mr->nonvolatile;
1787 }
1788 
1789 /**
1790  * memory_region_get_fd: Get a file descriptor backing a RAM memory region.
1791  *
1792  * Returns a file descriptor backing a file-based RAM memory region,
1793  * or -1 if the region is not a file-based RAM memory region.
1794  *
1795  * @mr: the RAM or alias memory region being queried.
1796  */
1797 int memory_region_get_fd(MemoryRegion *mr);
1798 
1799 /**
1800  * memory_region_from_host: Convert a pointer into a RAM memory region
1801  * and an offset within it.
1802  *
1803  * Given a host pointer inside a RAM memory region (created with
1804  * memory_region_init_ram() or memory_region_init_ram_ptr()), return
1805  * the MemoryRegion and the offset within it.
1806  *
1807  * Use with care; by the time this function returns, the returned pointer is
1808  * not protected by RCU anymore.  If the caller is not within an RCU critical
1809  * section and does not hold the iothread lock, it must have other means of
1810  * protecting the pointer, such as a reference to the region that includes
1811  * the incoming ram_addr_t.
1812  *
1813  * @ptr: the host pointer to be converted
1814  * @offset: the offset within memory region
1815  */
1816 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset);
1817 
1818 /**
1819  * memory_region_get_ram_ptr: Get a pointer into a RAM memory region.
1820  *
1821  * Returns a host pointer to a RAM memory region (created with
1822  * memory_region_init_ram() or memory_region_init_ram_ptr()).
1823  *
1824  * Use with care; by the time this function returns, the returned pointer is
1825  * not protected by RCU anymore.  If the caller is not within an RCU critical
1826  * section and does not hold the iothread lock, it must have other means of
1827  * protecting the pointer, such as a reference to the region that includes
1828  * the incoming ram_addr_t.
1829  *
1830  * @mr: the memory region being queried.
1831  */
1832 void *memory_region_get_ram_ptr(MemoryRegion *mr);
1833 
1834 /* memory_region_ram_resize: Resize a RAM region.
1835  *
1836  * Resizing RAM while migrating can result in the migration being canceled.
1837  * Care has to be taken if the guest might have already detected the memory.
1838  *
1839  * @mr: a memory region created with @memory_region_init_resizeable_ram.
1840  * @newsize: the new size the region
1841  * @errp: pointer to Error*, to store an error if it happens.
1842  */
1843 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize,
1844                               Error **errp);
1845 
1846 /**
1847  * memory_region_msync: Synchronize selected address range of
1848  * a memory mapped region
1849  *
1850  * @mr: the memory region to be msync
1851  * @addr: the initial address of the range to be sync
1852  * @size: the size of the range to be sync
1853  */
1854 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size);
1855 
1856 /**
1857  * memory_region_writeback: Trigger cache writeback for
1858  * selected address range
1859  *
1860  * @mr: the memory region to be updated
1861  * @addr: the initial address of the range to be written back
1862  * @size: the size of the range to be written back
1863  */
1864 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size);
1865 
1866 /**
1867  * memory_region_set_log: Turn dirty logging on or off for a region.
1868  *
1869  * Turns dirty logging on or off for a specified client (display, migration).
1870  * Only meaningful for RAM regions.
1871  *
1872  * @mr: the memory region being updated.
1873  * @log: whether dirty logging is to be enabled or disabled.
1874  * @client: the user of the logging information; %DIRTY_MEMORY_VGA only.
1875  */
1876 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client);
1877 
1878 /**
1879  * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region.
1880  *
1881  * Marks a range of bytes as dirty, after it has been dirtied outside
1882  * guest code.
1883  *
1884  * @mr: the memory region being dirtied.
1885  * @addr: the address (relative to the start of the region) being dirtied.
1886  * @size: size of the range being dirtied.
1887  */
1888 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1889                              hwaddr size);
1890 
1891 /**
1892  * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range
1893  *
1894  * This function is called when the caller wants to clear the remote
1895  * dirty bitmap of a memory range within the memory region.  This can
1896  * be used by e.g. KVM to manually clear dirty log when
1897  * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host
1898  * kernel.
1899  *
1900  * @mr:     the memory region to clear the dirty log upon
1901  * @start:  start address offset within the memory region
1902  * @len:    length of the memory region to clear dirty bitmap
1903  */
1904 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
1905                                       hwaddr len);
1906 
1907 /**
1908  * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty
1909  *                                         bitmap and clear it.
1910  *
1911  * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and
1912  * returns the snapshot.  The snapshot can then be used to query dirty
1913  * status, using memory_region_snapshot_get_dirty.  Snapshotting allows
1914  * querying the same page multiple times, which is especially useful for
1915  * display updates where the scanlines often are not page aligned.
1916  *
1917  * The dirty bitmap region which gets copyed into the snapshot (and
1918  * cleared afterwards) can be larger than requested.  The boundaries
1919  * are rounded up/down so complete bitmap longs (covering 64 pages on
1920  * 64bit hosts) can be copied over into the bitmap snapshot.  Which
1921  * isn't a problem for display updates as the extra pages are outside
1922  * the visible area, and in case the visible area changes a full
1923  * display redraw is due anyway.  Should other use cases for this
1924  * function emerge we might have to revisit this implementation
1925  * detail.
1926  *
1927  * Use g_free to release DirtyBitmapSnapshot.
1928  *
1929  * @mr: the memory region being queried.
1930  * @addr: the address (relative to the start of the region) being queried.
1931  * @size: the size of the range being queried.
1932  * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA.
1933  */
1934 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
1935                                                             hwaddr addr,
1936                                                             hwaddr size,
1937                                                             unsigned client);
1938 
1939 /**
1940  * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty
1941  *                                   in the specified dirty bitmap snapshot.
1942  *
1943  * @mr: the memory region being queried.
1944  * @snap: the dirty bitmap snapshot
1945  * @addr: the address (relative to the start of the region) being queried.
1946  * @size: the size of the range being queried.
1947  */
1948 bool memory_region_snapshot_get_dirty(MemoryRegion *mr,
1949                                       DirtyBitmapSnapshot *snap,
1950                                       hwaddr addr, hwaddr size);
1951 
1952 /**
1953  * memory_region_reset_dirty: Mark a range of pages as clean, for a specified
1954  *                            client.
1955  *
1956  * Marks a range of pages as no longer dirty.
1957  *
1958  * @mr: the region being updated.
1959  * @addr: the start of the subrange being cleaned.
1960  * @size: the size of the subrange being cleaned.
1961  * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or
1962  *          %DIRTY_MEMORY_VGA.
1963  */
1964 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1965                                hwaddr size, unsigned client);
1966 
1967 /**
1968  * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate
1969  *                                 TBs (for self-modifying code).
1970  *
1971  * The MemoryRegionOps->write() callback of a ROM device must use this function
1972  * to mark byte ranges that have been modified internally, such as by directly
1973  * accessing the memory returned by memory_region_get_ram_ptr().
1974  *
1975  * This function marks the range dirty and invalidates TBs so that TCG can
1976  * detect self-modifying code.
1977  *
1978  * @mr: the region being flushed.
1979  * @addr: the start, relative to the start of the region, of the range being
1980  *        flushed.
1981  * @size: the size, in bytes, of the range being flushed.
1982  */
1983 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size);
1984 
1985 /**
1986  * memory_region_set_readonly: Turn a memory region read-only (or read-write)
1987  *
1988  * Allows a memory region to be marked as read-only (turning it into a ROM).
1989  * only useful on RAM regions.
1990  *
1991  * @mr: the region being updated.
1992  * @readonly: whether rhe region is to be ROM or RAM.
1993  */
1994 void memory_region_set_readonly(MemoryRegion *mr, bool readonly);
1995 
1996 /**
1997  * memory_region_set_nonvolatile: Turn a memory region non-volatile
1998  *
1999  * Allows a memory region to be marked as non-volatile.
2000  * only useful on RAM regions.
2001  *
2002  * @mr: the region being updated.
2003  * @nonvolatile: whether rhe region is to be non-volatile.
2004  */
2005 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile);
2006 
2007 /**
2008  * memory_region_rom_device_set_romd: enable/disable ROMD mode
2009  *
2010  * Allows a ROM device (initialized with memory_region_init_rom_device() to
2011  * set to ROMD mode (default) or MMIO mode.  When it is in ROMD mode, the
2012  * device is mapped to guest memory and satisfies read access directly.
2013  * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function.
2014  * Writes are always handled by the #MemoryRegion.write function.
2015  *
2016  * @mr: the memory region to be updated
2017  * @romd_mode: %true to put the region into ROMD mode
2018  */
2019 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode);
2020 
2021 /**
2022  * memory_region_set_coalescing: Enable memory coalescing for the region.
2023  *
2024  * Enabled writes to a region to be queued for later processing. MMIO ->write
2025  * callbacks may be delayed until a non-coalesced MMIO is issued.
2026  * Only useful for IO regions.  Roughly similar to write-combining hardware.
2027  *
2028  * @mr: the memory region to be write coalesced
2029  */
2030 void memory_region_set_coalescing(MemoryRegion *mr);
2031 
2032 /**
2033  * memory_region_add_coalescing: Enable memory coalescing for a sub-range of
2034  *                               a region.
2035  *
2036  * Like memory_region_set_coalescing(), but works on a sub-range of a region.
2037  * Multiple calls can be issued coalesced disjoint ranges.
2038  *
2039  * @mr: the memory region to be updated.
2040  * @offset: the start of the range within the region to be coalesced.
2041  * @size: the size of the subrange to be coalesced.
2042  */
2043 void memory_region_add_coalescing(MemoryRegion *mr,
2044                                   hwaddr offset,
2045                                   uint64_t size);
2046 
2047 /**
2048  * memory_region_clear_coalescing: Disable MMIO coalescing for the region.
2049  *
2050  * Disables any coalescing caused by memory_region_set_coalescing() or
2051  * memory_region_add_coalescing().  Roughly equivalent to uncacheble memory
2052  * hardware.
2053  *
2054  * @mr: the memory region to be updated.
2055  */
2056 void memory_region_clear_coalescing(MemoryRegion *mr);
2057 
2058 /**
2059  * memory_region_set_flush_coalesced: Enforce memory coalescing flush before
2060  *                                    accesses.
2061  *
2062  * Ensure that pending coalesced MMIO request are flushed before the memory
2063  * region is accessed. This property is automatically enabled for all regions
2064  * passed to memory_region_set_coalescing() and memory_region_add_coalescing().
2065  *
2066  * @mr: the memory region to be updated.
2067  */
2068 void memory_region_set_flush_coalesced(MemoryRegion *mr);
2069 
2070 /**
2071  * memory_region_clear_flush_coalesced: Disable memory coalescing flush before
2072  *                                      accesses.
2073  *
2074  * Clear the automatic coalesced MMIO flushing enabled via
2075  * memory_region_set_flush_coalesced. Note that this service has no effect on
2076  * memory regions that have MMIO coalescing enabled for themselves. For them,
2077  * automatic flushing will stop once coalescing is disabled.
2078  *
2079  * @mr: the memory region to be updated.
2080  */
2081 void memory_region_clear_flush_coalesced(MemoryRegion *mr);
2082 
2083 /**
2084  * memory_region_add_eventfd: Request an eventfd to be triggered when a word
2085  *                            is written to a location.
2086  *
2087  * Marks a word in an IO region (initialized with memory_region_init_io())
2088  * as a trigger for an eventfd event.  The I/O callback will not be called.
2089  * The caller must be prepared to handle failure (that is, take the required
2090  * action if the callback _is_ called).
2091  *
2092  * @mr: the memory region being updated.
2093  * @addr: the address within @mr that is to be monitored
2094  * @size: the size of the access to trigger the eventfd
2095  * @match_data: whether to match against @data, instead of just @addr
2096  * @data: the data to match against the guest write
2097  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2098  **/
2099 void memory_region_add_eventfd(MemoryRegion *mr,
2100                                hwaddr addr,
2101                                unsigned size,
2102                                bool match_data,
2103                                uint64_t data,
2104                                EventNotifier *e);
2105 
2106 /**
2107  * memory_region_del_eventfd: Cancel an eventfd.
2108  *
2109  * Cancels an eventfd trigger requested by a previous
2110  * memory_region_add_eventfd() call.
2111  *
2112  * @mr: the memory region being updated.
2113  * @addr: the address within @mr that is to be monitored
2114  * @size: the size of the access to trigger the eventfd
2115  * @match_data: whether to match against @data, instead of just @addr
2116  * @data: the data to match against the guest write
2117  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2118  */
2119 void memory_region_del_eventfd(MemoryRegion *mr,
2120                                hwaddr addr,
2121                                unsigned size,
2122                                bool match_data,
2123                                uint64_t data,
2124                                EventNotifier *e);
2125 
2126 /**
2127  * memory_region_add_subregion: Add a subregion to a container.
2128  *
2129  * Adds a subregion at @offset.  The subregion may not overlap with other
2130  * subregions (except for those explicitly marked as overlapping).  A region
2131  * may only be added once as a subregion (unless removed with
2132  * memory_region_del_subregion()); use memory_region_init_alias() if you
2133  * want a region to be a subregion in multiple locations.
2134  *
2135  * @mr: the region to contain the new subregion; must be a container
2136  *      initialized with memory_region_init().
2137  * @offset: the offset relative to @mr where @subregion is added.
2138  * @subregion: the subregion to be added.
2139  */
2140 void memory_region_add_subregion(MemoryRegion *mr,
2141                                  hwaddr offset,
2142                                  MemoryRegion *subregion);
2143 /**
2144  * memory_region_add_subregion_overlap: Add a subregion to a container
2145  *                                      with overlap.
2146  *
2147  * Adds a subregion at @offset.  The subregion may overlap with other
2148  * subregions.  Conflicts are resolved by having a higher @priority hide a
2149  * lower @priority. Subregions without priority are taken as @priority 0.
2150  * A region may only be added once as a subregion (unless removed with
2151  * memory_region_del_subregion()); use memory_region_init_alias() if you
2152  * want a region to be a subregion in multiple locations.
2153  *
2154  * @mr: the region to contain the new subregion; must be a container
2155  *      initialized with memory_region_init().
2156  * @offset: the offset relative to @mr where @subregion is added.
2157  * @subregion: the subregion to be added.
2158  * @priority: used for resolving overlaps; highest priority wins.
2159  */
2160 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2161                                          hwaddr offset,
2162                                          MemoryRegion *subregion,
2163                                          int priority);
2164 
2165 /**
2166  * memory_region_get_ram_addr: Get the ram address associated with a memory
2167  *                             region
2168  *
2169  * @mr: the region to be queried
2170  */
2171 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr);
2172 
2173 uint64_t memory_region_get_alignment(const MemoryRegion *mr);
2174 /**
2175  * memory_region_del_subregion: Remove a subregion.
2176  *
2177  * Removes a subregion from its container.
2178  *
2179  * @mr: the container to be updated.
2180  * @subregion: the region being removed; must be a current subregion of @mr.
2181  */
2182 void memory_region_del_subregion(MemoryRegion *mr,
2183                                  MemoryRegion *subregion);
2184 
2185 /*
2186  * memory_region_set_enabled: dynamically enable or disable a region
2187  *
2188  * Enables or disables a memory region.  A disabled memory region
2189  * ignores all accesses to itself and its subregions.  It does not
2190  * obscure sibling subregions with lower priority - it simply behaves as
2191  * if it was removed from the hierarchy.
2192  *
2193  * Regions default to being enabled.
2194  *
2195  * @mr: the region to be updated
2196  * @enabled: whether to enable or disable the region
2197  */
2198 void memory_region_set_enabled(MemoryRegion *mr, bool enabled);
2199 
2200 /*
2201  * memory_region_set_address: dynamically update the address of a region
2202  *
2203  * Dynamically updates the address of a region, relative to its container.
2204  * May be used on regions are currently part of a memory hierarchy.
2205  *
2206  * @mr: the region to be updated
2207  * @addr: new address, relative to container region
2208  */
2209 void memory_region_set_address(MemoryRegion *mr, hwaddr addr);
2210 
2211 /*
2212  * memory_region_set_size: dynamically update the size of a region.
2213  *
2214  * Dynamically updates the size of a region.
2215  *
2216  * @mr: the region to be updated
2217  * @size: used size of the region.
2218  */
2219 void memory_region_set_size(MemoryRegion *mr, uint64_t size);
2220 
2221 /*
2222  * memory_region_set_alias_offset: dynamically update a memory alias's offset
2223  *
2224  * Dynamically updates the offset into the target region that an alias points
2225  * to, as if the fourth argument to memory_region_init_alias() has changed.
2226  *
2227  * @mr: the #MemoryRegion to be updated; should be an alias.
2228  * @offset: the new offset into the target memory region
2229  */
2230 void memory_region_set_alias_offset(MemoryRegion *mr,
2231                                     hwaddr offset);
2232 
2233 /**
2234  * memory_region_present: checks if an address relative to a @container
2235  * translates into #MemoryRegion within @container
2236  *
2237  * Answer whether a #MemoryRegion within @container covers the address
2238  * @addr.
2239  *
2240  * @container: a #MemoryRegion within which @addr is a relative address
2241  * @addr: the area within @container to be searched
2242  */
2243 bool memory_region_present(MemoryRegion *container, hwaddr addr);
2244 
2245 /**
2246  * memory_region_is_mapped: returns true if #MemoryRegion is mapped
2247  * into any address space.
2248  *
2249  * @mr: a #MemoryRegion which should be checked if it's mapped
2250  */
2251 bool memory_region_is_mapped(MemoryRegion *mr);
2252 
2253 /**
2254  * memory_region_get_ram_discard_manager: get the #RamDiscardManager for a
2255  * #MemoryRegion
2256  *
2257  * The #RamDiscardManager cannot change while a memory region is mapped.
2258  *
2259  * @mr: the #MemoryRegion
2260  */
2261 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr);
2262 
2263 /**
2264  * memory_region_has_ram_discard_manager: check whether a #MemoryRegion has a
2265  * #RamDiscardManager assigned
2266  *
2267  * @mr: the #MemoryRegion
2268  */
2269 static inline bool memory_region_has_ram_discard_manager(MemoryRegion *mr)
2270 {
2271     return !!memory_region_get_ram_discard_manager(mr);
2272 }
2273 
2274 /**
2275  * memory_region_set_ram_discard_manager: set the #RamDiscardManager for a
2276  * #MemoryRegion
2277  *
2278  * This function must not be called for a mapped #MemoryRegion, a #MemoryRegion
2279  * that does not cover RAM, or a #MemoryRegion that already has a
2280  * #RamDiscardManager assigned.
2281  *
2282  * @mr: the #MemoryRegion
2283  * @rdm: #RamDiscardManager to set
2284  */
2285 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2286                                            RamDiscardManager *rdm);
2287 
2288 /**
2289  * memory_region_find: translate an address/size relative to a
2290  * MemoryRegion into a #MemoryRegionSection.
2291  *
2292  * Locates the first #MemoryRegion within @mr that overlaps the range
2293  * given by @addr and @size.
2294  *
2295  * Returns a #MemoryRegionSection that describes a contiguous overlap.
2296  * It will have the following characteristics:
2297  * - @size = 0 iff no overlap was found
2298  * - @mr is non-%NULL iff an overlap was found
2299  *
2300  * Remember that in the return value the @offset_within_region is
2301  * relative to the returned region (in the .@mr field), not to the
2302  * @mr argument.
2303  *
2304  * Similarly, the .@offset_within_address_space is relative to the
2305  * address space that contains both regions, the passed and the
2306  * returned one.  However, in the special case where the @mr argument
2307  * has no container (and thus is the root of the address space), the
2308  * following will hold:
2309  * - @offset_within_address_space >= @addr
2310  * - @offset_within_address_space + .@size <= @addr + @size
2311  *
2312  * @mr: a MemoryRegion within which @addr is a relative address
2313  * @addr: start of the area within @as to be searched
2314  * @size: size of the area to be searched
2315  */
2316 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2317                                        hwaddr addr, uint64_t size);
2318 
2319 /**
2320  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2321  *
2322  * Synchronizes the dirty page log for all address spaces.
2323  */
2324 void memory_global_dirty_log_sync(void);
2325 
2326 /**
2327  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2328  *
2329  * Synchronizes the vCPUs with a thread that is reading the dirty bitmap.
2330  * This function must be called after the dirty log bitmap is cleared, and
2331  * before dirty guest memory pages are read.  If you are using
2332  * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes
2333  * care of doing this.
2334  */
2335 void memory_global_after_dirty_log_sync(void);
2336 
2337 /**
2338  * memory_region_transaction_begin: Start a transaction.
2339  *
2340  * During a transaction, changes will be accumulated and made visible
2341  * only when the transaction ends (is committed).
2342  */
2343 void memory_region_transaction_begin(void);
2344 
2345 /**
2346  * memory_region_transaction_commit: Commit a transaction and make changes
2347  *                                   visible to the guest.
2348  */
2349 void memory_region_transaction_commit(void);
2350 
2351 /**
2352  * memory_listener_register: register callbacks to be called when memory
2353  *                           sections are mapped or unmapped into an address
2354  *                           space
2355  *
2356  * @listener: an object containing the callbacks to be called
2357  * @filter: if non-%NULL, only regions in this address space will be observed
2358  */
2359 void memory_listener_register(MemoryListener *listener, AddressSpace *filter);
2360 
2361 /**
2362  * memory_listener_unregister: undo the effect of memory_listener_register()
2363  *
2364  * @listener: an object containing the callbacks to be removed
2365  */
2366 void memory_listener_unregister(MemoryListener *listener);
2367 
2368 /**
2369  * memory_global_dirty_log_start: begin dirty logging for all regions
2370  */
2371 void memory_global_dirty_log_start(void);
2372 
2373 /**
2374  * memory_global_dirty_log_stop: end dirty logging for all regions
2375  */
2376 void memory_global_dirty_log_stop(void);
2377 
2378 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled);
2379 
2380 /**
2381  * memory_region_dispatch_read: perform a read directly to the specified
2382  * MemoryRegion.
2383  *
2384  * @mr: #MemoryRegion to access
2385  * @addr: address within that region
2386  * @pval: pointer to uint64_t which the data is written to
2387  * @op: size, sign, and endianness of the memory operation
2388  * @attrs: memory transaction attributes to use for the access
2389  */
2390 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
2391                                         hwaddr addr,
2392                                         uint64_t *pval,
2393                                         MemOp op,
2394                                         MemTxAttrs attrs);
2395 /**
2396  * memory_region_dispatch_write: perform a write directly to the specified
2397  * MemoryRegion.
2398  *
2399  * @mr: #MemoryRegion to access
2400  * @addr: address within that region
2401  * @data: data to write
2402  * @op: size, sign, and endianness of the memory operation
2403  * @attrs: memory transaction attributes to use for the access
2404  */
2405 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
2406                                          hwaddr addr,
2407                                          uint64_t data,
2408                                          MemOp op,
2409                                          MemTxAttrs attrs);
2410 
2411 /**
2412  * address_space_init: initializes an address space
2413  *
2414  * @as: an uninitialized #AddressSpace
2415  * @root: a #MemoryRegion that routes addresses for the address space
2416  * @name: an address space name.  The name is only used for debugging
2417  *        output.
2418  */
2419 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name);
2420 
2421 /**
2422  * address_space_destroy: destroy an address space
2423  *
2424  * Releases all resources associated with an address space.  After an address space
2425  * is destroyed, its root memory region (given by address_space_init()) may be destroyed
2426  * as well.
2427  *
2428  * @as: address space to be destroyed
2429  */
2430 void address_space_destroy(AddressSpace *as);
2431 
2432 /**
2433  * address_space_remove_listeners: unregister all listeners of an address space
2434  *
2435  * Removes all callbacks previously registered with memory_listener_register()
2436  * for @as.
2437  *
2438  * @as: an initialized #AddressSpace
2439  */
2440 void address_space_remove_listeners(AddressSpace *as);
2441 
2442 /**
2443  * address_space_rw: read from or write to an address space.
2444  *
2445  * Return a MemTxResult indicating whether the operation succeeded
2446  * or failed (eg unassigned memory, device rejected the transaction,
2447  * IOMMU fault).
2448  *
2449  * @as: #AddressSpace to be accessed
2450  * @addr: address within that address space
2451  * @attrs: memory transaction attributes
2452  * @buf: buffer with the data transferred
2453  * @len: the number of bytes to read or write
2454  * @is_write: indicates the transfer direction
2455  */
2456 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
2457                              MemTxAttrs attrs, void *buf,
2458                              hwaddr len, bool is_write);
2459 
2460 /**
2461  * address_space_write: write to address space.
2462  *
2463  * Return a MemTxResult indicating whether the operation succeeded
2464  * or failed (eg unassigned memory, device rejected the transaction,
2465  * IOMMU fault).
2466  *
2467  * @as: #AddressSpace to be accessed
2468  * @addr: address within that address space
2469  * @attrs: memory transaction attributes
2470  * @buf: buffer with the data transferred
2471  * @len: the number of bytes to write
2472  */
2473 MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
2474                                 MemTxAttrs attrs,
2475                                 const void *buf, hwaddr len);
2476 
2477 /**
2478  * address_space_write_rom: write to address space, including ROM.
2479  *
2480  * This function writes to the specified address space, but will
2481  * write data to both ROM and RAM. This is used for non-guest
2482  * writes like writes from the gdb debug stub or initial loading
2483  * of ROM contents.
2484  *
2485  * Note that portions of the write which attempt to write data to
2486  * a device will be silently ignored -- only real RAM and ROM will
2487  * be written to.
2488  *
2489  * Return a MemTxResult indicating whether the operation succeeded
2490  * or failed (eg unassigned memory, device rejected the transaction,
2491  * IOMMU fault).
2492  *
2493  * @as: #AddressSpace to be accessed
2494  * @addr: address within that address space
2495  * @attrs: memory transaction attributes
2496  * @buf: buffer with the data transferred
2497  * @len: the number of bytes to write
2498  */
2499 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr,
2500                                     MemTxAttrs attrs,
2501                                     const void *buf, hwaddr len);
2502 
2503 /* address_space_ld*: load from an address space
2504  * address_space_st*: store to an address space
2505  *
2506  * These functions perform a load or store of the byte, word,
2507  * longword or quad to the specified address within the AddressSpace.
2508  * The _le suffixed functions treat the data as little endian;
2509  * _be indicates big endian; no suffix indicates "same endianness
2510  * as guest CPU".
2511  *
2512  * The "guest CPU endianness" accessors are deprecated for use outside
2513  * target-* code; devices should be CPU-agnostic and use either the LE
2514  * or the BE accessors.
2515  *
2516  * @as #AddressSpace to be accessed
2517  * @addr: address within that address space
2518  * @val: data value, for stores
2519  * @attrs: memory transaction attributes
2520  * @result: location to write the success/failure of the transaction;
2521  *   if NULL, this information is discarded
2522  */
2523 
2524 #define SUFFIX
2525 #define ARG1         as
2526 #define ARG1_DECL    AddressSpace *as
2527 #include "exec/memory_ldst.h.inc"
2528 
2529 #define SUFFIX
2530 #define ARG1         as
2531 #define ARG1_DECL    AddressSpace *as
2532 #include "exec/memory_ldst_phys.h.inc"
2533 
2534 struct MemoryRegionCache {
2535     void *ptr;
2536     hwaddr xlat;
2537     hwaddr len;
2538     FlatView *fv;
2539     MemoryRegionSection mrs;
2540     bool is_write;
2541 };
2542 
2543 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL })
2544 
2545 
2546 /* address_space_ld*_cached: load from a cached #MemoryRegion
2547  * address_space_st*_cached: store into a cached #MemoryRegion
2548  *
2549  * These functions perform a load or store of the byte, word,
2550  * longword or quad to the specified address.  The address is
2551  * a physical address in the AddressSpace, but it must lie within
2552  * a #MemoryRegion that was mapped with address_space_cache_init.
2553  *
2554  * The _le suffixed functions treat the data as little endian;
2555  * _be indicates big endian; no suffix indicates "same endianness
2556  * as guest CPU".
2557  *
2558  * The "guest CPU endianness" accessors are deprecated for use outside
2559  * target-* code; devices should be CPU-agnostic and use either the LE
2560  * or the BE accessors.
2561  *
2562  * @cache: previously initialized #MemoryRegionCache to be accessed
2563  * @addr: address within the address space
2564  * @val: data value, for stores
2565  * @attrs: memory transaction attributes
2566  * @result: location to write the success/failure of the transaction;
2567  *   if NULL, this information is discarded
2568  */
2569 
2570 #define SUFFIX       _cached_slow
2571 #define ARG1         cache
2572 #define ARG1_DECL    MemoryRegionCache *cache
2573 #include "exec/memory_ldst.h.inc"
2574 
2575 /* Inline fast path for direct RAM access.  */
2576 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache,
2577     hwaddr addr, MemTxAttrs attrs, MemTxResult *result)
2578 {
2579     assert(addr < cache->len);
2580     if (likely(cache->ptr)) {
2581         return ldub_p(cache->ptr + addr);
2582     } else {
2583         return address_space_ldub_cached_slow(cache, addr, attrs, result);
2584     }
2585 }
2586 
2587 static inline void address_space_stb_cached(MemoryRegionCache *cache,
2588     hwaddr addr, uint8_t val, MemTxAttrs attrs, MemTxResult *result)
2589 {
2590     assert(addr < cache->len);
2591     if (likely(cache->ptr)) {
2592         stb_p(cache->ptr + addr, val);
2593     } else {
2594         address_space_stb_cached_slow(cache, addr, val, attrs, result);
2595     }
2596 }
2597 
2598 #define ENDIANNESS   _le
2599 #include "exec/memory_ldst_cached.h.inc"
2600 
2601 #define ENDIANNESS   _be
2602 #include "exec/memory_ldst_cached.h.inc"
2603 
2604 #define SUFFIX       _cached
2605 #define ARG1         cache
2606 #define ARG1_DECL    MemoryRegionCache *cache
2607 #include "exec/memory_ldst_phys.h.inc"
2608 
2609 /* address_space_cache_init: prepare for repeated access to a physical
2610  * memory region
2611  *
2612  * @cache: #MemoryRegionCache to be filled
2613  * @as: #AddressSpace to be accessed
2614  * @addr: address within that address space
2615  * @len: length of buffer
2616  * @is_write: indicates the transfer direction
2617  *
2618  * Will only work with RAM, and may map a subset of the requested range by
2619  * returning a value that is less than @len.  On failure, return a negative
2620  * errno value.
2621  *
2622  * Because it only works with RAM, this function can be used for
2623  * read-modify-write operations.  In this case, is_write should be %true.
2624  *
2625  * Note that addresses passed to the address_space_*_cached functions
2626  * are relative to @addr.
2627  */
2628 int64_t address_space_cache_init(MemoryRegionCache *cache,
2629                                  AddressSpace *as,
2630                                  hwaddr addr,
2631                                  hwaddr len,
2632                                  bool is_write);
2633 
2634 /**
2635  * address_space_cache_invalidate: complete a write to a #MemoryRegionCache
2636  *
2637  * @cache: The #MemoryRegionCache to operate on.
2638  * @addr: The first physical address that was written, relative to the
2639  * address that was passed to @address_space_cache_init.
2640  * @access_len: The number of bytes that were written starting at @addr.
2641  */
2642 void address_space_cache_invalidate(MemoryRegionCache *cache,
2643                                     hwaddr addr,
2644                                     hwaddr access_len);
2645 
2646 /**
2647  * address_space_cache_destroy: free a #MemoryRegionCache
2648  *
2649  * @cache: The #MemoryRegionCache whose memory should be released.
2650  */
2651 void address_space_cache_destroy(MemoryRegionCache *cache);
2652 
2653 /* address_space_get_iotlb_entry: translate an address into an IOTLB
2654  * entry. Should be called from an RCU critical section.
2655  */
2656 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
2657                                             bool is_write, MemTxAttrs attrs);
2658 
2659 /* address_space_translate: translate an address range into an address space
2660  * into a MemoryRegion and an address range into that section.  Should be
2661  * called from an RCU critical section, to avoid that the last reference
2662  * to the returned region disappears after address_space_translate returns.
2663  *
2664  * @fv: #FlatView to be accessed
2665  * @addr: address within that address space
2666  * @xlat: pointer to address within the returned memory region section's
2667  * #MemoryRegion.
2668  * @len: pointer to length
2669  * @is_write: indicates the transfer direction
2670  * @attrs: memory attributes
2671  */
2672 MemoryRegion *flatview_translate(FlatView *fv,
2673                                  hwaddr addr, hwaddr *xlat,
2674                                  hwaddr *len, bool is_write,
2675                                  MemTxAttrs attrs);
2676 
2677 static inline MemoryRegion *address_space_translate(AddressSpace *as,
2678                                                     hwaddr addr, hwaddr *xlat,
2679                                                     hwaddr *len, bool is_write,
2680                                                     MemTxAttrs attrs)
2681 {
2682     return flatview_translate(address_space_to_flatview(as),
2683                               addr, xlat, len, is_write, attrs);
2684 }
2685 
2686 /* address_space_access_valid: check for validity of accessing an address
2687  * space range
2688  *
2689  * Check whether memory is assigned to the given address space range, and
2690  * access is permitted by any IOMMU regions that are active for the address
2691  * space.
2692  *
2693  * For now, addr and len should be aligned to a page size.  This limitation
2694  * will be lifted in the future.
2695  *
2696  * @as: #AddressSpace to be accessed
2697  * @addr: address within that address space
2698  * @len: length of the area to be checked
2699  * @is_write: indicates the transfer direction
2700  * @attrs: memory attributes
2701  */
2702 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len,
2703                                 bool is_write, MemTxAttrs attrs);
2704 
2705 /* address_space_map: map a physical memory region into a host virtual address
2706  *
2707  * May map a subset of the requested range, given by and returned in @plen.
2708  * May return %NULL and set *@plen to zero(0), if resources needed to perform
2709  * the mapping are exhausted.
2710  * Use only for reads OR writes - not for read-modify-write operations.
2711  * Use cpu_register_map_client() to know when retrying the map operation is
2712  * likely to succeed.
2713  *
2714  * @as: #AddressSpace to be accessed
2715  * @addr: address within that address space
2716  * @plen: pointer to length of buffer; updated on return
2717  * @is_write: indicates the transfer direction
2718  * @attrs: memory attributes
2719  */
2720 void *address_space_map(AddressSpace *as, hwaddr addr,
2721                         hwaddr *plen, bool is_write, MemTxAttrs attrs);
2722 
2723 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map()
2724  *
2725  * Will also mark the memory as dirty if @is_write == %true.  @access_len gives
2726  * the amount of memory that was actually read or written by the caller.
2727  *
2728  * @as: #AddressSpace used
2729  * @buffer: host pointer as returned by address_space_map()
2730  * @len: buffer length as returned by address_space_map()
2731  * @access_len: amount of data actually transferred
2732  * @is_write: indicates the transfer direction
2733  */
2734 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2735                          bool is_write, hwaddr access_len);
2736 
2737 
2738 /* Internal functions, part of the implementation of address_space_read.  */
2739 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
2740                                     MemTxAttrs attrs, void *buf, hwaddr len);
2741 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
2742                                    MemTxAttrs attrs, void *buf,
2743                                    hwaddr len, hwaddr addr1, hwaddr l,
2744                                    MemoryRegion *mr);
2745 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr);
2746 
2747 /* Internal functions, part of the implementation of address_space_read_cached
2748  * and address_space_write_cached.  */
2749 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache,
2750                                            hwaddr addr, void *buf, hwaddr len);
2751 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache,
2752                                             hwaddr addr, const void *buf,
2753                                             hwaddr len);
2754 
2755 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
2756 {
2757     if (is_write) {
2758         return memory_region_is_ram(mr) && !mr->readonly &&
2759                !mr->rom_device && !memory_region_is_ram_device(mr);
2760     } else {
2761         return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) ||
2762                memory_region_is_romd(mr);
2763     }
2764 }
2765 
2766 /**
2767  * address_space_read: read from an address space.
2768  *
2769  * Return a MemTxResult indicating whether the operation succeeded
2770  * or failed (eg unassigned memory, device rejected the transaction,
2771  * IOMMU fault).  Called within RCU critical section.
2772  *
2773  * @as: #AddressSpace to be accessed
2774  * @addr: address within that address space
2775  * @attrs: memory transaction attributes
2776  * @buf: buffer with the data transferred
2777  * @len: length of the data transferred
2778  */
2779 static inline __attribute__((__always_inline__))
2780 MemTxResult address_space_read(AddressSpace *as, hwaddr addr,
2781                                MemTxAttrs attrs, void *buf,
2782                                hwaddr len)
2783 {
2784     MemTxResult result = MEMTX_OK;
2785     hwaddr l, addr1;
2786     void *ptr;
2787     MemoryRegion *mr;
2788     FlatView *fv;
2789 
2790     if (__builtin_constant_p(len)) {
2791         if (len) {
2792             RCU_READ_LOCK_GUARD();
2793             fv = address_space_to_flatview(as);
2794             l = len;
2795             mr = flatview_translate(fv, addr, &addr1, &l, false, attrs);
2796             if (len == l && memory_access_is_direct(mr, false)) {
2797                 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
2798                 memcpy(buf, ptr, len);
2799             } else {
2800                 result = flatview_read_continue(fv, addr, attrs, buf, len,
2801                                                 addr1, l, mr);
2802             }
2803         }
2804     } else {
2805         result = address_space_read_full(as, addr, attrs, buf, len);
2806     }
2807     return result;
2808 }
2809 
2810 /**
2811  * address_space_read_cached: read from a cached RAM region
2812  *
2813  * @cache: Cached region to be addressed
2814  * @addr: address relative to the base of the RAM region
2815  * @buf: buffer with the data transferred
2816  * @len: length of the data transferred
2817  */
2818 static inline MemTxResult
2819 address_space_read_cached(MemoryRegionCache *cache, hwaddr addr,
2820                           void *buf, hwaddr len)
2821 {
2822     assert(addr < cache->len && len <= cache->len - addr);
2823     fuzz_dma_read_cb(cache->xlat + addr, len, cache->mrs.mr);
2824     if (likely(cache->ptr)) {
2825         memcpy(buf, cache->ptr + addr, len);
2826         return MEMTX_OK;
2827     } else {
2828         return address_space_read_cached_slow(cache, addr, buf, len);
2829     }
2830 }
2831 
2832 /**
2833  * address_space_write_cached: write to a cached RAM region
2834  *
2835  * @cache: Cached region to be addressed
2836  * @addr: address relative to the base of the RAM region
2837  * @buf: buffer with the data transferred
2838  * @len: length of the data transferred
2839  */
2840 static inline MemTxResult
2841 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr,
2842                            const void *buf, hwaddr len)
2843 {
2844     assert(addr < cache->len && len <= cache->len - addr);
2845     if (likely(cache->ptr)) {
2846         memcpy(cache->ptr + addr, buf, len);
2847         return MEMTX_OK;
2848     } else {
2849         return address_space_write_cached_slow(cache, addr, buf, len);
2850     }
2851 }
2852 
2853 #ifdef NEED_CPU_H
2854 /* enum device_endian to MemOp.  */
2855 static inline MemOp devend_memop(enum device_endian end)
2856 {
2857     QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN &&
2858                       DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN);
2859 
2860 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
2861     /* Swap if non-host endianness or native (target) endianness */
2862     return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP;
2863 #else
2864     const int non_host_endianness =
2865         DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN;
2866 
2867     /* In this case, native (target) endianness needs no swap.  */
2868     return (end == non_host_endianness) ? MO_BSWAP : 0;
2869 #endif
2870 }
2871 #endif
2872 
2873 /*
2874  * Inhibit technologies that require discarding of pages in RAM blocks, e.g.,
2875  * to manage the actual amount of memory consumed by the VM (then, the memory
2876  * provided by RAM blocks might be bigger than the desired memory consumption).
2877  * This *must* be set if:
2878  * - Discarding parts of a RAM blocks does not result in the change being
2879  *   reflected in the VM and the pages getting freed.
2880  * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous
2881  *   discards blindly.
2882  * - Discarding parts of a RAM blocks will result in integrity issues (e.g.,
2883  *   encrypted VMs).
2884  * Technologies that only temporarily pin the current working set of a
2885  * driver are fine, because we don't expect such pages to be discarded
2886  * (esp. based on guest action like balloon inflation).
2887  *
2888  * This is *not* to be used to protect from concurrent discards (esp.,
2889  * postcopy).
2890  *
2891  * Returns 0 if successful. Returns -EBUSY if a technology that relies on
2892  * discards to work reliably is active.
2893  */
2894 int ram_block_discard_disable(bool state);
2895 
2896 /*
2897  * See ram_block_discard_disable(): only disable uncoordinated discards,
2898  * keeping coordinated discards (via the RamDiscardManager) enabled.
2899  */
2900 int ram_block_uncoordinated_discard_disable(bool state);
2901 
2902 /*
2903  * Inhibit technologies that disable discarding of pages in RAM blocks.
2904  *
2905  * Returns 0 if successful. Returns -EBUSY if discards are already set to
2906  * broken.
2907  */
2908 int ram_block_discard_require(bool state);
2909 
2910 /*
2911  * See ram_block_discard_require(): only inhibit technologies that disable
2912  * uncoordinated discarding of pages in RAM blocks, allowing co-existance with
2913  * technologies that only inhibit uncoordinated discards (via the
2914  * RamDiscardManager).
2915  */
2916 int ram_block_coordinated_discard_require(bool state);
2917 
2918 /*
2919  * Test if any discarding of memory in ram blocks is disabled.
2920  */
2921 bool ram_block_discard_is_disabled(void);
2922 
2923 /*
2924  * Test if any discarding of memory in ram blocks is required to work reliably.
2925  */
2926 bool ram_block_discard_is_required(void);
2927 
2928 #endif
2929 
2930 #endif
2931