xref: /qemu/include/exec/memory.h (revision dbc8221f)
1 /*
2  * Physical memory management API
3  *
4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
5  *
6  * Authors:
7  *  Avi Kivity <avi@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #ifndef MEMORY_H
15 #define MEMORY_H
16 
17 #ifndef CONFIG_USER_ONLY
18 
19 #include "exec/cpu-common.h"
20 #include "exec/hwaddr.h"
21 #include "exec/memattrs.h"
22 #include "exec/memop.h"
23 #include "exec/ramlist.h"
24 #include "qemu/bswap.h"
25 #include "qemu/queue.h"
26 #include "qemu/int128.h"
27 #include "qemu/notify.h"
28 #include "qom/object.h"
29 #include "qemu/rcu.h"
30 
31 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
32 
33 #define MAX_PHYS_ADDR_SPACE_BITS 62
34 #define MAX_PHYS_ADDR            (((hwaddr)1 << MAX_PHYS_ADDR_SPACE_BITS) - 1)
35 
36 #define TYPE_MEMORY_REGION "memory-region"
37 DECLARE_INSTANCE_CHECKER(MemoryRegion, MEMORY_REGION,
38                          TYPE_MEMORY_REGION)
39 
40 #define TYPE_IOMMU_MEMORY_REGION "iommu-memory-region"
41 typedef struct IOMMUMemoryRegionClass IOMMUMemoryRegionClass;
42 DECLARE_OBJ_CHECKERS(IOMMUMemoryRegion, IOMMUMemoryRegionClass,
43                      IOMMU_MEMORY_REGION, TYPE_IOMMU_MEMORY_REGION)
44 
45 #define TYPE_RAM_DISCARD_MANAGER "qemu:ram-discard-manager"
46 typedef struct RamDiscardManagerClass RamDiscardManagerClass;
47 typedef struct RamDiscardManager RamDiscardManager;
48 DECLARE_OBJ_CHECKERS(RamDiscardManager, RamDiscardManagerClass,
49                      RAM_DISCARD_MANAGER, TYPE_RAM_DISCARD_MANAGER);
50 
51 #ifdef CONFIG_FUZZ
52 void fuzz_dma_read_cb(size_t addr,
53                       size_t len,
54                       MemoryRegion *mr);
55 #else
56 static inline void fuzz_dma_read_cb(size_t addr,
57                                     size_t len,
58                                     MemoryRegion *mr)
59 {
60     /* Do Nothing */
61 }
62 #endif
63 
64 extern bool global_dirty_log;
65 
66 typedef struct MemoryRegionOps MemoryRegionOps;
67 
68 struct ReservedRegion {
69     hwaddr low;
70     hwaddr high;
71     unsigned type;
72 };
73 
74 /**
75  * struct MemoryRegionSection: describes a fragment of a #MemoryRegion
76  *
77  * @mr: the region, or %NULL if empty
78  * @fv: the flat view of the address space the region is mapped in
79  * @offset_within_region: the beginning of the section, relative to @mr's start
80  * @size: the size of the section; will not exceed @mr's boundaries
81  * @offset_within_address_space: the address of the first byte of the section
82  *     relative to the region's address space
83  * @readonly: writes to this section are ignored
84  * @nonvolatile: this section is non-volatile
85  */
86 struct MemoryRegionSection {
87     Int128 size;
88     MemoryRegion *mr;
89     FlatView *fv;
90     hwaddr offset_within_region;
91     hwaddr offset_within_address_space;
92     bool readonly;
93     bool nonvolatile;
94 };
95 
96 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
97 
98 /* See address_space_translate: bit 0 is read, bit 1 is write.  */
99 typedef enum {
100     IOMMU_NONE = 0,
101     IOMMU_RO   = 1,
102     IOMMU_WO   = 2,
103     IOMMU_RW   = 3,
104 } IOMMUAccessFlags;
105 
106 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0))
107 
108 struct IOMMUTLBEntry {
109     AddressSpace    *target_as;
110     hwaddr           iova;
111     hwaddr           translated_addr;
112     hwaddr           addr_mask;  /* 0xfff = 4k translation */
113     IOMMUAccessFlags perm;
114 };
115 
116 /*
117  * Bitmap for different IOMMUNotifier capabilities. Each notifier can
118  * register with one or multiple IOMMU Notifier capability bit(s).
119  */
120 typedef enum {
121     IOMMU_NOTIFIER_NONE = 0,
122     /* Notify cache invalidations */
123     IOMMU_NOTIFIER_UNMAP = 0x1,
124     /* Notify entry changes (newly created entries) */
125     IOMMU_NOTIFIER_MAP = 0x2,
126     /* Notify changes on device IOTLB entries */
127     IOMMU_NOTIFIER_DEVIOTLB_UNMAP = 0x04,
128 } IOMMUNotifierFlag;
129 
130 #define IOMMU_NOTIFIER_IOTLB_EVENTS (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
131 #define IOMMU_NOTIFIER_DEVIOTLB_EVENTS IOMMU_NOTIFIER_DEVIOTLB_UNMAP
132 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_IOTLB_EVENTS | \
133                             IOMMU_NOTIFIER_DEVIOTLB_EVENTS)
134 
135 struct IOMMUNotifier;
136 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier,
137                             IOMMUTLBEntry *data);
138 
139 struct IOMMUNotifier {
140     IOMMUNotify notify;
141     IOMMUNotifierFlag notifier_flags;
142     /* Notify for address space range start <= addr <= end */
143     hwaddr start;
144     hwaddr end;
145     int iommu_idx;
146     QLIST_ENTRY(IOMMUNotifier) node;
147 };
148 typedef struct IOMMUNotifier IOMMUNotifier;
149 
150 typedef struct IOMMUTLBEvent {
151     IOMMUNotifierFlag type;
152     IOMMUTLBEntry entry;
153 } IOMMUTLBEvent;
154 
155 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
156 #define RAM_PREALLOC   (1 << 0)
157 
158 /* RAM is mmap-ed with MAP_SHARED */
159 #define RAM_SHARED     (1 << 1)
160 
161 /* Only a portion of RAM (used_length) is actually used, and migrated.
162  * Resizing RAM while migrating can result in the migration being canceled.
163  */
164 #define RAM_RESIZEABLE (1 << 2)
165 
166 /* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically
167  * zero the page and wake waiting processes.
168  * (Set during postcopy)
169  */
170 #define RAM_UF_ZEROPAGE (1 << 3)
171 
172 /* RAM can be migrated */
173 #define RAM_MIGRATABLE (1 << 4)
174 
175 /* RAM is a persistent kind memory */
176 #define RAM_PMEM (1 << 5)
177 
178 
179 /*
180  * UFFDIO_WRITEPROTECT is used on this RAMBlock to
181  * support 'write-tracking' migration type.
182  * Implies ram_state->ram_wt_enabled.
183  */
184 #define RAM_UF_WRITEPROTECT (1 << 6)
185 
186 /*
187  * RAM is mmap-ed with MAP_NORESERVE. When set, reserving swap space (or huge
188  * pages if applicable) is skipped: will bail out if not supported. When not
189  * set, the OS will do the reservation, if supported for the memory type.
190  */
191 #define RAM_NORESERVE (1 << 7)
192 
193 /* RAM that isn't accessible through normal means. */
194 #define RAM_PROTECTED (1 << 8)
195 
196 static inline void iommu_notifier_init(IOMMUNotifier *n, IOMMUNotify fn,
197                                        IOMMUNotifierFlag flags,
198                                        hwaddr start, hwaddr end,
199                                        int iommu_idx)
200 {
201     n->notify = fn;
202     n->notifier_flags = flags;
203     n->start = start;
204     n->end = end;
205     n->iommu_idx = iommu_idx;
206 }
207 
208 /*
209  * Memory region callbacks
210  */
211 struct MemoryRegionOps {
212     /* Read from the memory region. @addr is relative to @mr; @size is
213      * in bytes. */
214     uint64_t (*read)(void *opaque,
215                      hwaddr addr,
216                      unsigned size);
217     /* Write to the memory region. @addr is relative to @mr; @size is
218      * in bytes. */
219     void (*write)(void *opaque,
220                   hwaddr addr,
221                   uint64_t data,
222                   unsigned size);
223 
224     MemTxResult (*read_with_attrs)(void *opaque,
225                                    hwaddr addr,
226                                    uint64_t *data,
227                                    unsigned size,
228                                    MemTxAttrs attrs);
229     MemTxResult (*write_with_attrs)(void *opaque,
230                                     hwaddr addr,
231                                     uint64_t data,
232                                     unsigned size,
233                                     MemTxAttrs attrs);
234 
235     enum device_endian endianness;
236     /* Guest-visible constraints: */
237     struct {
238         /* If nonzero, specify bounds on access sizes beyond which a machine
239          * check is thrown.
240          */
241         unsigned min_access_size;
242         unsigned max_access_size;
243         /* If true, unaligned accesses are supported.  Otherwise unaligned
244          * accesses throw machine checks.
245          */
246          bool unaligned;
247         /*
248          * If present, and returns #false, the transaction is not accepted
249          * by the device (and results in machine dependent behaviour such
250          * as a machine check exception).
251          */
252         bool (*accepts)(void *opaque, hwaddr addr,
253                         unsigned size, bool is_write,
254                         MemTxAttrs attrs);
255     } valid;
256     /* Internal implementation constraints: */
257     struct {
258         /* If nonzero, specifies the minimum size implemented.  Smaller sizes
259          * will be rounded upwards and a partial result will be returned.
260          */
261         unsigned min_access_size;
262         /* If nonzero, specifies the maximum size implemented.  Larger sizes
263          * will be done as a series of accesses with smaller sizes.
264          */
265         unsigned max_access_size;
266         /* If true, unaligned accesses are supported.  Otherwise all accesses
267          * are converted to (possibly multiple) naturally aligned accesses.
268          */
269         bool unaligned;
270     } impl;
271 };
272 
273 typedef struct MemoryRegionClass {
274     /* private */
275     ObjectClass parent_class;
276 } MemoryRegionClass;
277 
278 
279 enum IOMMUMemoryRegionAttr {
280     IOMMU_ATTR_SPAPR_TCE_FD
281 };
282 
283 /*
284  * IOMMUMemoryRegionClass:
285  *
286  * All IOMMU implementations need to subclass TYPE_IOMMU_MEMORY_REGION
287  * and provide an implementation of at least the @translate method here
288  * to handle requests to the memory region. Other methods are optional.
289  *
290  * The IOMMU implementation must use the IOMMU notifier infrastructure
291  * to report whenever mappings are changed, by calling
292  * memory_region_notify_iommu() (or, if necessary, by calling
293  * memory_region_notify_iommu_one() for each registered notifier).
294  *
295  * Conceptually an IOMMU provides a mapping from input address
296  * to an output TLB entry. If the IOMMU is aware of memory transaction
297  * attributes and the output TLB entry depends on the transaction
298  * attributes, we represent this using IOMMU indexes. Each index
299  * selects a particular translation table that the IOMMU has:
300  *
301  *   @attrs_to_index returns the IOMMU index for a set of transaction attributes
302  *
303  *   @translate takes an input address and an IOMMU index
304  *
305  * and the mapping returned can only depend on the input address and the
306  * IOMMU index.
307  *
308  * Most IOMMUs don't care about the transaction attributes and support
309  * only a single IOMMU index. A more complex IOMMU might have one index
310  * for secure transactions and one for non-secure transactions.
311  */
312 struct IOMMUMemoryRegionClass {
313     /* private: */
314     MemoryRegionClass parent_class;
315 
316     /* public: */
317     /**
318      * @translate:
319      *
320      * Return a TLB entry that contains a given address.
321      *
322      * The IOMMUAccessFlags indicated via @flag are optional and may
323      * be specified as IOMMU_NONE to indicate that the caller needs
324      * the full translation information for both reads and writes. If
325      * the access flags are specified then the IOMMU implementation
326      * may use this as an optimization, to stop doing a page table
327      * walk as soon as it knows that the requested permissions are not
328      * allowed. If IOMMU_NONE is passed then the IOMMU must do the
329      * full page table walk and report the permissions in the returned
330      * IOMMUTLBEntry. (Note that this implies that an IOMMU may not
331      * return different mappings for reads and writes.)
332      *
333      * The returned information remains valid while the caller is
334      * holding the big QEMU lock or is inside an RCU critical section;
335      * if the caller wishes to cache the mapping beyond that it must
336      * register an IOMMU notifier so it can invalidate its cached
337      * information when the IOMMU mapping changes.
338      *
339      * @iommu: the IOMMUMemoryRegion
340      *
341      * @hwaddr: address to be translated within the memory region
342      *
343      * @flag: requested access permission
344      *
345      * @iommu_idx: IOMMU index for the translation
346      */
347     IOMMUTLBEntry (*translate)(IOMMUMemoryRegion *iommu, hwaddr addr,
348                                IOMMUAccessFlags flag, int iommu_idx);
349     /**
350      * @get_min_page_size:
351      *
352      * Returns minimum supported page size in bytes.
353      *
354      * If this method is not provided then the minimum is assumed to
355      * be TARGET_PAGE_SIZE.
356      *
357      * @iommu: the IOMMUMemoryRegion
358      */
359     uint64_t (*get_min_page_size)(IOMMUMemoryRegion *iommu);
360     /**
361      * @notify_flag_changed:
362      *
363      * Called when IOMMU Notifier flag changes (ie when the set of
364      * events which IOMMU users are requesting notification for changes).
365      * Optional method -- need not be provided if the IOMMU does not
366      * need to know exactly which events must be notified.
367      *
368      * @iommu: the IOMMUMemoryRegion
369      *
370      * @old_flags: events which previously needed to be notified
371      *
372      * @new_flags: events which now need to be notified
373      *
374      * Returns 0 on success, or a negative errno; in particular
375      * returns -EINVAL if the new flag bitmap is not supported by the
376      * IOMMU memory region. In case of failure, the error object
377      * must be created
378      */
379     int (*notify_flag_changed)(IOMMUMemoryRegion *iommu,
380                                IOMMUNotifierFlag old_flags,
381                                IOMMUNotifierFlag new_flags,
382                                Error **errp);
383     /**
384      * @replay:
385      *
386      * Called to handle memory_region_iommu_replay().
387      *
388      * The default implementation of memory_region_iommu_replay() is to
389      * call the IOMMU translate method for every page in the address space
390      * with flag == IOMMU_NONE and then call the notifier if translate
391      * returns a valid mapping. If this method is implemented then it
392      * overrides the default behaviour, and must provide the full semantics
393      * of memory_region_iommu_replay(), by calling @notifier for every
394      * translation present in the IOMMU.
395      *
396      * Optional method -- an IOMMU only needs to provide this method
397      * if the default is inefficient or produces undesirable side effects.
398      *
399      * Note: this is not related to record-and-replay functionality.
400      */
401     void (*replay)(IOMMUMemoryRegion *iommu, IOMMUNotifier *notifier);
402 
403     /**
404      * @get_attr:
405      *
406      * Get IOMMU misc attributes. This is an optional method that
407      * can be used to allow users of the IOMMU to get implementation-specific
408      * information. The IOMMU implements this method to handle calls
409      * by IOMMU users to memory_region_iommu_get_attr() by filling in
410      * the arbitrary data pointer for any IOMMUMemoryRegionAttr values that
411      * the IOMMU supports. If the method is unimplemented then
412      * memory_region_iommu_get_attr() will always return -EINVAL.
413      *
414      * @iommu: the IOMMUMemoryRegion
415      *
416      * @attr: attribute being queried
417      *
418      * @data: memory to fill in with the attribute data
419      *
420      * Returns 0 on success, or a negative errno; in particular
421      * returns -EINVAL for unrecognized or unimplemented attribute types.
422      */
423     int (*get_attr)(IOMMUMemoryRegion *iommu, enum IOMMUMemoryRegionAttr attr,
424                     void *data);
425 
426     /**
427      * @attrs_to_index:
428      *
429      * Return the IOMMU index to use for a given set of transaction attributes.
430      *
431      * Optional method: if an IOMMU only supports a single IOMMU index then
432      * the default implementation of memory_region_iommu_attrs_to_index()
433      * will return 0.
434      *
435      * The indexes supported by an IOMMU must be contiguous, starting at 0.
436      *
437      * @iommu: the IOMMUMemoryRegion
438      * @attrs: memory transaction attributes
439      */
440     int (*attrs_to_index)(IOMMUMemoryRegion *iommu, MemTxAttrs attrs);
441 
442     /**
443      * @num_indexes:
444      *
445      * Return the number of IOMMU indexes this IOMMU supports.
446      *
447      * Optional method: if this method is not provided, then
448      * memory_region_iommu_num_indexes() will return 1, indicating that
449      * only a single IOMMU index is supported.
450      *
451      * @iommu: the IOMMUMemoryRegion
452      */
453     int (*num_indexes)(IOMMUMemoryRegion *iommu);
454 
455     /**
456      * @iommu_set_page_size_mask:
457      *
458      * Restrict the page size mask that can be supported with a given IOMMU
459      * memory region. Used for example to propagate host physical IOMMU page
460      * size mask limitations to the virtual IOMMU.
461      *
462      * Optional method: if this method is not provided, then the default global
463      * page mask is used.
464      *
465      * @iommu: the IOMMUMemoryRegion
466      *
467      * @page_size_mask: a bitmask of supported page sizes. At least one bit,
468      * representing the smallest page size, must be set. Additional set bits
469      * represent supported block sizes. For example a host physical IOMMU that
470      * uses page tables with a page size of 4kB, and supports 2MB and 4GB
471      * blocks, will set mask 0x40201000. A granule of 4kB with indiscriminate
472      * block sizes is specified with mask 0xfffffffffffff000.
473      *
474      * Returns 0 on success, or a negative error. In case of failure, the error
475      * object must be created.
476      */
477      int (*iommu_set_page_size_mask)(IOMMUMemoryRegion *iommu,
478                                      uint64_t page_size_mask,
479                                      Error **errp);
480 };
481 
482 typedef struct RamDiscardListener RamDiscardListener;
483 typedef int (*NotifyRamPopulate)(RamDiscardListener *rdl,
484                                  MemoryRegionSection *section);
485 typedef void (*NotifyRamDiscard)(RamDiscardListener *rdl,
486                                  MemoryRegionSection *section);
487 
488 struct RamDiscardListener {
489     /*
490      * @notify_populate:
491      *
492      * Notification that previously discarded memory is about to get populated.
493      * Listeners are able to object. If any listener objects, already
494      * successfully notified listeners are notified about a discard again.
495      *
496      * @rdl: the #RamDiscardListener getting notified
497      * @section: the #MemoryRegionSection to get populated. The section
498      *           is aligned within the memory region to the minimum granularity
499      *           unless it would exceed the registered section.
500      *
501      * Returns 0 on success. If the notification is rejected by the listener,
502      * an error is returned.
503      */
504     NotifyRamPopulate notify_populate;
505 
506     /*
507      * @notify_discard:
508      *
509      * Notification that previously populated memory was discarded successfully
510      * and listeners should drop all references to such memory and prevent
511      * new population (e.g., unmap).
512      *
513      * @rdl: the #RamDiscardListener getting notified
514      * @section: the #MemoryRegionSection to get populated. The section
515      *           is aligned within the memory region to the minimum granularity
516      *           unless it would exceed the registered section.
517      */
518     NotifyRamDiscard notify_discard;
519 
520     /*
521      * @double_discard_supported:
522      *
523      * The listener suppors getting @notify_discard notifications that span
524      * already discarded parts.
525      */
526     bool double_discard_supported;
527 
528     MemoryRegionSection *section;
529     QLIST_ENTRY(RamDiscardListener) next;
530 };
531 
532 static inline void ram_discard_listener_init(RamDiscardListener *rdl,
533                                              NotifyRamPopulate populate_fn,
534                                              NotifyRamDiscard discard_fn,
535                                              bool double_discard_supported)
536 {
537     rdl->notify_populate = populate_fn;
538     rdl->notify_discard = discard_fn;
539     rdl->double_discard_supported = double_discard_supported;
540 }
541 
542 typedef int (*ReplayRamPopulate)(MemoryRegionSection *section, void *opaque);
543 
544 /*
545  * RamDiscardManagerClass:
546  *
547  * A #RamDiscardManager coordinates which parts of specific RAM #MemoryRegion
548  * regions are currently populated to be used/accessed by the VM, notifying
549  * after parts were discarded (freeing up memory) and before parts will be
550  * populated (consuming memory), to be used/acessed by the VM.
551  *
552  * A #RamDiscardManager can only be set for a RAM #MemoryRegion while the
553  * #MemoryRegion isn't mapped yet; it cannot change while the #MemoryRegion is
554  * mapped.
555  *
556  * The #RamDiscardManager is intended to be used by technologies that are
557  * incompatible with discarding of RAM (e.g., VFIO, which may pin all
558  * memory inside a #MemoryRegion), and require proper coordination to only
559  * map the currently populated parts, to hinder parts that are expected to
560  * remain discarded from silently getting populated and consuming memory.
561  * Technologies that support discarding of RAM don't have to bother and can
562  * simply map the whole #MemoryRegion.
563  *
564  * An example #RamDiscardManager is virtio-mem, which logically (un)plugs
565  * memory within an assigned RAM #MemoryRegion, coordinated with the VM.
566  * Logically unplugging memory consists of discarding RAM. The VM agreed to not
567  * access unplugged (discarded) memory - especially via DMA. virtio-mem will
568  * properly coordinate with listeners before memory is plugged (populated),
569  * and after memory is unplugged (discarded).
570  *
571  * Listeners are called in multiples of the minimum granularity (unless it
572  * would exceed the registered range) and changes are aligned to the minimum
573  * granularity within the #MemoryRegion. Listeners have to prepare for memory
574  * becomming discarded in a different granularity than it was populated and the
575  * other way around.
576  */
577 struct RamDiscardManagerClass {
578     /* private */
579     InterfaceClass parent_class;
580 
581     /* public */
582 
583     /**
584      * @get_min_granularity:
585      *
586      * Get the minimum granularity in which listeners will get notified
587      * about changes within the #MemoryRegion via the #RamDiscardManager.
588      *
589      * @rdm: the #RamDiscardManager
590      * @mr: the #MemoryRegion
591      *
592      * Returns the minimum granularity.
593      */
594     uint64_t (*get_min_granularity)(const RamDiscardManager *rdm,
595                                     const MemoryRegion *mr);
596 
597     /**
598      * @is_populated:
599      *
600      * Check whether the given #MemoryRegionSection is completely populated
601      * (i.e., no parts are currently discarded) via the #RamDiscardManager.
602      * There are no alignment requirements.
603      *
604      * @rdm: the #RamDiscardManager
605      * @section: the #MemoryRegionSection
606      *
607      * Returns whether the given range is completely populated.
608      */
609     bool (*is_populated)(const RamDiscardManager *rdm,
610                          const MemoryRegionSection *section);
611 
612     /**
613      * @replay_populated:
614      *
615      * Call the #ReplayRamPopulate callback for all populated parts within the
616      * #MemoryRegionSection via the #RamDiscardManager.
617      *
618      * In case any call fails, no further calls are made.
619      *
620      * @rdm: the #RamDiscardManager
621      * @section: the #MemoryRegionSection
622      * @replay_fn: the #ReplayRamPopulate callback
623      * @opaque: pointer to forward to the callback
624      *
625      * Returns 0 on success, or a negative error if any notification failed.
626      */
627     int (*replay_populated)(const RamDiscardManager *rdm,
628                             MemoryRegionSection *section,
629                             ReplayRamPopulate replay_fn, void *opaque);
630 
631     /**
632      * @register_listener:
633      *
634      * Register a #RamDiscardListener for the given #MemoryRegionSection and
635      * immediately notify the #RamDiscardListener about all populated parts
636      * within the #MemoryRegionSection via the #RamDiscardManager.
637      *
638      * In case any notification fails, no further notifications are triggered
639      * and an error is logged.
640      *
641      * @rdm: the #RamDiscardManager
642      * @rdl: the #RamDiscardListener
643      * @section: the #MemoryRegionSection
644      */
645     void (*register_listener)(RamDiscardManager *rdm,
646                               RamDiscardListener *rdl,
647                               MemoryRegionSection *section);
648 
649     /**
650      * @unregister_listener:
651      *
652      * Unregister a previously registered #RamDiscardListener via the
653      * #RamDiscardManager after notifying the #RamDiscardListener about all
654      * populated parts becoming unpopulated within the registered
655      * #MemoryRegionSection.
656      *
657      * @rdm: the #RamDiscardManager
658      * @rdl: the #RamDiscardListener
659      */
660     void (*unregister_listener)(RamDiscardManager *rdm,
661                                 RamDiscardListener *rdl);
662 };
663 
664 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
665                                                  const MemoryRegion *mr);
666 
667 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
668                                       const MemoryRegionSection *section);
669 
670 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
671                                          MemoryRegionSection *section,
672                                          ReplayRamPopulate replay_fn,
673                                          void *opaque);
674 
675 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
676                                            RamDiscardListener *rdl,
677                                            MemoryRegionSection *section);
678 
679 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
680                                              RamDiscardListener *rdl);
681 
682 typedef struct CoalescedMemoryRange CoalescedMemoryRange;
683 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd;
684 
685 /** MemoryRegion:
686  *
687  * A struct representing a memory region.
688  */
689 struct MemoryRegion {
690     Object parent_obj;
691 
692     /* private: */
693 
694     /* The following fields should fit in a cache line */
695     bool romd_mode;
696     bool ram;
697     bool subpage;
698     bool readonly; /* For RAM regions */
699     bool nonvolatile;
700     bool rom_device;
701     bool flush_coalesced_mmio;
702     uint8_t dirty_log_mask;
703     bool is_iommu;
704     RAMBlock *ram_block;
705     Object *owner;
706 
707     const MemoryRegionOps *ops;
708     void *opaque;
709     MemoryRegion *container;
710     Int128 size;
711     hwaddr addr;
712     void (*destructor)(MemoryRegion *mr);
713     uint64_t align;
714     bool terminates;
715     bool ram_device;
716     bool enabled;
717     bool warning_printed; /* For reservations */
718     uint8_t vga_logging_count;
719     MemoryRegion *alias;
720     hwaddr alias_offset;
721     int32_t priority;
722     QTAILQ_HEAD(, MemoryRegion) subregions;
723     QTAILQ_ENTRY(MemoryRegion) subregions_link;
724     QTAILQ_HEAD(, CoalescedMemoryRange) coalesced;
725     const char *name;
726     unsigned ioeventfd_nb;
727     MemoryRegionIoeventfd *ioeventfds;
728     RamDiscardManager *rdm; /* Only for RAM */
729 };
730 
731 struct IOMMUMemoryRegion {
732     MemoryRegion parent_obj;
733 
734     QLIST_HEAD(, IOMMUNotifier) iommu_notify;
735     IOMMUNotifierFlag iommu_notify_flags;
736 };
737 
738 #define IOMMU_NOTIFIER_FOREACH(n, mr) \
739     QLIST_FOREACH((n), &(mr)->iommu_notify, node)
740 
741 /**
742  * struct MemoryListener: callbacks structure for updates to the physical memory map
743  *
744  * Allows a component to adjust to changes in the guest-visible memory map.
745  * Use with memory_listener_register() and memory_listener_unregister().
746  */
747 struct MemoryListener {
748     /**
749      * @begin:
750      *
751      * Called at the beginning of an address space update transaction.
752      * Followed by calls to #MemoryListener.region_add(),
753      * #MemoryListener.region_del(), #MemoryListener.region_nop(),
754      * #MemoryListener.log_start() and #MemoryListener.log_stop() in
755      * increasing address order.
756      *
757      * @listener: The #MemoryListener.
758      */
759     void (*begin)(MemoryListener *listener);
760 
761     /**
762      * @commit:
763      *
764      * Called at the end of an address space update transaction,
765      * after the last call to #MemoryListener.region_add(),
766      * #MemoryListener.region_del() or #MemoryListener.region_nop(),
767      * #MemoryListener.log_start() and #MemoryListener.log_stop().
768      *
769      * @listener: The #MemoryListener.
770      */
771     void (*commit)(MemoryListener *listener);
772 
773     /**
774      * @region_add:
775      *
776      * Called during an address space update transaction,
777      * for a section of the address space that is new in this address space
778      * space since the last transaction.
779      *
780      * @listener: The #MemoryListener.
781      * @section: The new #MemoryRegionSection.
782      */
783     void (*region_add)(MemoryListener *listener, MemoryRegionSection *section);
784 
785     /**
786      * @region_del:
787      *
788      * Called during an address space update transaction,
789      * for a section of the address space that has disappeared in the address
790      * space since the last transaction.
791      *
792      * @listener: The #MemoryListener.
793      * @section: The old #MemoryRegionSection.
794      */
795     void (*region_del)(MemoryListener *listener, MemoryRegionSection *section);
796 
797     /**
798      * @region_nop:
799      *
800      * Called during an address space update transaction,
801      * for a section of the address space that is in the same place in the address
802      * space as in the last transaction.
803      *
804      * @listener: The #MemoryListener.
805      * @section: The #MemoryRegionSection.
806      */
807     void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section);
808 
809     /**
810      * @log_start:
811      *
812      * Called during an address space update transaction, after
813      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
814      * #MemoryListener.region_nop(), if dirty memory logging clients have
815      * become active since the last transaction.
816      *
817      * @listener: The #MemoryListener.
818      * @section: The #MemoryRegionSection.
819      * @old: A bitmap of dirty memory logging clients that were active in
820      * the previous transaction.
821      * @new: A bitmap of dirty memory logging clients that are active in
822      * the current transaction.
823      */
824     void (*log_start)(MemoryListener *listener, MemoryRegionSection *section,
825                       int old, int new);
826 
827     /**
828      * @log_stop:
829      *
830      * Called during an address space update transaction, after
831      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
832      * #MemoryListener.region_nop() and possibly after
833      * #MemoryListener.log_start(), if dirty memory logging clients have
834      * become inactive since the last transaction.
835      *
836      * @listener: The #MemoryListener.
837      * @section: The #MemoryRegionSection.
838      * @old: A bitmap of dirty memory logging clients that were active in
839      * the previous transaction.
840      * @new: A bitmap of dirty memory logging clients that are active in
841      * the current transaction.
842      */
843     void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section,
844                      int old, int new);
845 
846     /**
847      * @log_sync:
848      *
849      * Called by memory_region_snapshot_and_clear_dirty() and
850      * memory_global_dirty_log_sync(), before accessing QEMU's "official"
851      * copy of the dirty memory bitmap for a #MemoryRegionSection.
852      *
853      * @listener: The #MemoryListener.
854      * @section: The #MemoryRegionSection.
855      */
856     void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section);
857 
858     /**
859      * @log_sync_global:
860      *
861      * This is the global version of @log_sync when the listener does
862      * not have a way to synchronize the log with finer granularity.
863      * When the listener registers with @log_sync_global defined, then
864      * its @log_sync must be NULL.  Vice versa.
865      *
866      * @listener: The #MemoryListener.
867      */
868     void (*log_sync_global)(MemoryListener *listener);
869 
870     /**
871      * @log_clear:
872      *
873      * Called before reading the dirty memory bitmap for a
874      * #MemoryRegionSection.
875      *
876      * @listener: The #MemoryListener.
877      * @section: The #MemoryRegionSection.
878      */
879     void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section);
880 
881     /**
882      * @log_global_start:
883      *
884      * Called by memory_global_dirty_log_start(), which
885      * enables the %DIRTY_LOG_MIGRATION client on all memory regions in
886      * the address space.  #MemoryListener.log_global_start() is also
887      * called when a #MemoryListener is added, if global dirty logging is
888      * active at that time.
889      *
890      * @listener: The #MemoryListener.
891      */
892     void (*log_global_start)(MemoryListener *listener);
893 
894     /**
895      * @log_global_stop:
896      *
897      * Called by memory_global_dirty_log_stop(), which
898      * disables the %DIRTY_LOG_MIGRATION client on all memory regions in
899      * the address space.
900      *
901      * @listener: The #MemoryListener.
902      */
903     void (*log_global_stop)(MemoryListener *listener);
904 
905     /**
906      * @log_global_after_sync:
907      *
908      * Called after reading the dirty memory bitmap
909      * for any #MemoryRegionSection.
910      *
911      * @listener: The #MemoryListener.
912      */
913     void (*log_global_after_sync)(MemoryListener *listener);
914 
915     /**
916      * @eventfd_add:
917      *
918      * Called during an address space update transaction,
919      * for a section of the address space that has had a new ioeventfd
920      * registration since the last transaction.
921      *
922      * @listener: The #MemoryListener.
923      * @section: The new #MemoryRegionSection.
924      * @match_data: The @match_data parameter for the new ioeventfd.
925      * @data: The @data parameter for the new ioeventfd.
926      * @e: The #EventNotifier parameter for the new ioeventfd.
927      */
928     void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section,
929                         bool match_data, uint64_t data, EventNotifier *e);
930 
931     /**
932      * @eventfd_del:
933      *
934      * Called during an address space update transaction,
935      * for a section of the address space that has dropped an ioeventfd
936      * registration since the last transaction.
937      *
938      * @listener: The #MemoryListener.
939      * @section: The new #MemoryRegionSection.
940      * @match_data: The @match_data parameter for the dropped ioeventfd.
941      * @data: The @data parameter for the dropped ioeventfd.
942      * @e: The #EventNotifier parameter for the dropped ioeventfd.
943      */
944     void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section,
945                         bool match_data, uint64_t data, EventNotifier *e);
946 
947     /**
948      * @coalesced_io_add:
949      *
950      * Called during an address space update transaction,
951      * for a section of the address space that has had a new coalesced
952      * MMIO range registration since the last transaction.
953      *
954      * @listener: The #MemoryListener.
955      * @section: The new #MemoryRegionSection.
956      * @addr: The starting address for the coalesced MMIO range.
957      * @len: The length of the coalesced MMIO range.
958      */
959     void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section,
960                                hwaddr addr, hwaddr len);
961 
962     /**
963      * @coalesced_io_del:
964      *
965      * Called during an address space update transaction,
966      * for a section of the address space that has dropped a coalesced
967      * MMIO range since the last transaction.
968      *
969      * @listener: The #MemoryListener.
970      * @section: The new #MemoryRegionSection.
971      * @addr: The starting address for the coalesced MMIO range.
972      * @len: The length of the coalesced MMIO range.
973      */
974     void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section,
975                                hwaddr addr, hwaddr len);
976     /**
977      * @priority:
978      *
979      * Govern the order in which memory listeners are invoked. Lower priorities
980      * are invoked earlier for "add" or "start" callbacks, and later for "delete"
981      * or "stop" callbacks.
982      */
983     unsigned priority;
984 
985     /**
986      * @name:
987      *
988      * Name of the listener.  It can be used in contexts where we'd like to
989      * identify one memory listener with the rest.
990      */
991     const char *name;
992 
993     /* private: */
994     AddressSpace *address_space;
995     QTAILQ_ENTRY(MemoryListener) link;
996     QTAILQ_ENTRY(MemoryListener) link_as;
997 };
998 
999 /**
1000  * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects
1001  */
1002 struct AddressSpace {
1003     /* private: */
1004     struct rcu_head rcu;
1005     char *name;
1006     MemoryRegion *root;
1007 
1008     /* Accessed via RCU.  */
1009     struct FlatView *current_map;
1010 
1011     int ioeventfd_nb;
1012     struct MemoryRegionIoeventfd *ioeventfds;
1013     QTAILQ_HEAD(, MemoryListener) listeners;
1014     QTAILQ_ENTRY(AddressSpace) address_spaces_link;
1015 };
1016 
1017 typedef struct AddressSpaceDispatch AddressSpaceDispatch;
1018 typedef struct FlatRange FlatRange;
1019 
1020 /* Flattened global view of current active memory hierarchy.  Kept in sorted
1021  * order.
1022  */
1023 struct FlatView {
1024     struct rcu_head rcu;
1025     unsigned ref;
1026     FlatRange *ranges;
1027     unsigned nr;
1028     unsigned nr_allocated;
1029     struct AddressSpaceDispatch *dispatch;
1030     MemoryRegion *root;
1031 };
1032 
1033 static inline FlatView *address_space_to_flatview(AddressSpace *as)
1034 {
1035     return qatomic_rcu_read(&as->current_map);
1036 }
1037 
1038 /**
1039  * typedef flatview_cb: callback for flatview_for_each_range()
1040  *
1041  * @start: start address of the range within the FlatView
1042  * @len: length of the range in bytes
1043  * @mr: MemoryRegion covering this range
1044  * @offset_in_region: offset of the first byte of the range within @mr
1045  * @opaque: data pointer passed to flatview_for_each_range()
1046  *
1047  * Returns: true to stop the iteration, false to keep going.
1048  */
1049 typedef bool (*flatview_cb)(Int128 start,
1050                             Int128 len,
1051                             const MemoryRegion *mr,
1052                             hwaddr offset_in_region,
1053                             void *opaque);
1054 
1055 /**
1056  * flatview_for_each_range: Iterate through a FlatView
1057  * @fv: the FlatView to iterate through
1058  * @cb: function to call for each range
1059  * @opaque: opaque data pointer to pass to @cb
1060  *
1061  * A FlatView is made up of a list of non-overlapping ranges, each of
1062  * which is a slice of a MemoryRegion. This function iterates through
1063  * each range in @fv, calling @cb. The callback function can terminate
1064  * iteration early by returning 'true'.
1065  */
1066 void flatview_for_each_range(FlatView *fv, flatview_cb cb, void *opaque);
1067 
1068 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a,
1069                                           MemoryRegionSection *b)
1070 {
1071     return a->mr == b->mr &&
1072            a->fv == b->fv &&
1073            a->offset_within_region == b->offset_within_region &&
1074            a->offset_within_address_space == b->offset_within_address_space &&
1075            int128_eq(a->size, b->size) &&
1076            a->readonly == b->readonly &&
1077            a->nonvolatile == b->nonvolatile;
1078 }
1079 
1080 /**
1081  * memory_region_section_new_copy: Copy a memory region section
1082  *
1083  * Allocate memory for a new copy, copy the memory region section, and
1084  * properly take a reference on all relevant members.
1085  *
1086  * @s: the #MemoryRegionSection to copy
1087  */
1088 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s);
1089 
1090 /**
1091  * memory_region_section_new_copy: Free a copied memory region section
1092  *
1093  * Free a copy of a memory section created via memory_region_section_new_copy().
1094  * properly dropping references on all relevant members.
1095  *
1096  * @s: the #MemoryRegionSection to copy
1097  */
1098 void memory_region_section_free_copy(MemoryRegionSection *s);
1099 
1100 /**
1101  * memory_region_init: Initialize a memory region
1102  *
1103  * The region typically acts as a container for other memory regions.  Use
1104  * memory_region_add_subregion() to add subregions.
1105  *
1106  * @mr: the #MemoryRegion to be initialized
1107  * @owner: the object that tracks the region's reference count
1108  * @name: used for debugging; not visible to the user or ABI
1109  * @size: size of the region; any subregions beyond this size will be clipped
1110  */
1111 void memory_region_init(MemoryRegion *mr,
1112                         Object *owner,
1113                         const char *name,
1114                         uint64_t size);
1115 
1116 /**
1117  * memory_region_ref: Add 1 to a memory region's reference count
1118  *
1119  * Whenever memory regions are accessed outside the BQL, they need to be
1120  * preserved against hot-unplug.  MemoryRegions actually do not have their
1121  * own reference count; they piggyback on a QOM object, their "owner".
1122  * This function adds a reference to the owner.
1123  *
1124  * All MemoryRegions must have an owner if they can disappear, even if the
1125  * device they belong to operates exclusively under the BQL.  This is because
1126  * the region could be returned at any time by memory_region_find, and this
1127  * is usually under guest control.
1128  *
1129  * @mr: the #MemoryRegion
1130  */
1131 void memory_region_ref(MemoryRegion *mr);
1132 
1133 /**
1134  * memory_region_unref: Remove 1 to a memory region's reference count
1135  *
1136  * Whenever memory regions are accessed outside the BQL, they need to be
1137  * preserved against hot-unplug.  MemoryRegions actually do not have their
1138  * own reference count; they piggyback on a QOM object, their "owner".
1139  * This function removes a reference to the owner and possibly destroys it.
1140  *
1141  * @mr: the #MemoryRegion
1142  */
1143 void memory_region_unref(MemoryRegion *mr);
1144 
1145 /**
1146  * memory_region_init_io: Initialize an I/O memory region.
1147  *
1148  * Accesses into the region will cause the callbacks in @ops to be called.
1149  * if @size is nonzero, subregions will be clipped to @size.
1150  *
1151  * @mr: the #MemoryRegion to be initialized.
1152  * @owner: the object that tracks the region's reference count
1153  * @ops: a structure containing read and write callbacks to be used when
1154  *       I/O is performed on the region.
1155  * @opaque: passed to the read and write callbacks of the @ops structure.
1156  * @name: used for debugging; not visible to the user or ABI
1157  * @size: size of the region.
1158  */
1159 void memory_region_init_io(MemoryRegion *mr,
1160                            Object *owner,
1161                            const MemoryRegionOps *ops,
1162                            void *opaque,
1163                            const char *name,
1164                            uint64_t size);
1165 
1166 /**
1167  * memory_region_init_ram_nomigrate:  Initialize RAM memory region.  Accesses
1168  *                                    into the region will modify memory
1169  *                                    directly.
1170  *
1171  * @mr: the #MemoryRegion to be initialized.
1172  * @owner: the object that tracks the region's reference count
1173  * @name: Region name, becomes part of RAMBlock name used in migration stream
1174  *        must be unique within any device
1175  * @size: size of the region.
1176  * @errp: pointer to Error*, to store an error if it happens.
1177  *
1178  * Note that this function does not do anything to cause the data in the
1179  * RAM memory region to be migrated; that is the responsibility of the caller.
1180  */
1181 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1182                                       Object *owner,
1183                                       const char *name,
1184                                       uint64_t size,
1185                                       Error **errp);
1186 
1187 /**
1188  * memory_region_init_ram_flags_nomigrate:  Initialize RAM memory region.
1189  *                                          Accesses into the region will
1190  *                                          modify memory directly.
1191  *
1192  * @mr: the #MemoryRegion to be initialized.
1193  * @owner: the object that tracks the region's reference count
1194  * @name: Region name, becomes part of RAMBlock name used in migration stream
1195  *        must be unique within any device
1196  * @size: size of the region.
1197  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_NORESERVE.
1198  * @errp: pointer to Error*, to store an error if it happens.
1199  *
1200  * Note that this function does not do anything to cause the data in the
1201  * RAM memory region to be migrated; that is the responsibility of the caller.
1202  */
1203 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1204                                             Object *owner,
1205                                             const char *name,
1206                                             uint64_t size,
1207                                             uint32_t ram_flags,
1208                                             Error **errp);
1209 
1210 /**
1211  * memory_region_init_resizeable_ram:  Initialize memory region with resizeable
1212  *                                     RAM.  Accesses into the region will
1213  *                                     modify memory directly.  Only an initial
1214  *                                     portion of this RAM is actually used.
1215  *                                     Changing the size while migrating
1216  *                                     can result in the migration being
1217  *                                     canceled.
1218  *
1219  * @mr: the #MemoryRegion to be initialized.
1220  * @owner: the object that tracks the region's reference count
1221  * @name: Region name, becomes part of RAMBlock name used in migration stream
1222  *        must be unique within any device
1223  * @size: used size of the region.
1224  * @max_size: max size of the region.
1225  * @resized: callback to notify owner about used size change.
1226  * @errp: pointer to Error*, to store an error if it happens.
1227  *
1228  * Note that this function does not do anything to cause the data in the
1229  * RAM memory region to be migrated; that is the responsibility of the caller.
1230  */
1231 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1232                                        Object *owner,
1233                                        const char *name,
1234                                        uint64_t size,
1235                                        uint64_t max_size,
1236                                        void (*resized)(const char*,
1237                                                        uint64_t length,
1238                                                        void *host),
1239                                        Error **errp);
1240 #ifdef CONFIG_POSIX
1241 
1242 /**
1243  * memory_region_init_ram_from_file:  Initialize RAM memory region with a
1244  *                                    mmap-ed backend.
1245  *
1246  * @mr: the #MemoryRegion to be initialized.
1247  * @owner: the object that tracks the region's reference count
1248  * @name: Region name, becomes part of RAMBlock name used in migration stream
1249  *        must be unique within any device
1250  * @size: size of the region.
1251  * @align: alignment of the region base address; if 0, the default alignment
1252  *         (getpagesize()) will be used.
1253  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1254  *             RAM_NORESERVE,
1255  * @path: the path in which to allocate the RAM.
1256  * @readonly: true to open @path for reading, false for read/write.
1257  * @errp: pointer to Error*, to store an error if it happens.
1258  *
1259  * Note that this function does not do anything to cause the data in the
1260  * RAM memory region to be migrated; that is the responsibility of the caller.
1261  */
1262 void memory_region_init_ram_from_file(MemoryRegion *mr,
1263                                       Object *owner,
1264                                       const char *name,
1265                                       uint64_t size,
1266                                       uint64_t align,
1267                                       uint32_t ram_flags,
1268                                       const char *path,
1269                                       bool readonly,
1270                                       Error **errp);
1271 
1272 /**
1273  * memory_region_init_ram_from_fd:  Initialize RAM memory region with a
1274  *                                  mmap-ed backend.
1275  *
1276  * @mr: the #MemoryRegion to be initialized.
1277  * @owner: the object that tracks the region's reference count
1278  * @name: the name of the region.
1279  * @size: size of the region.
1280  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1281  *             RAM_NORESERVE, RAM_PROTECTED.
1282  * @fd: the fd to mmap.
1283  * @offset: offset within the file referenced by fd
1284  * @errp: pointer to Error*, to store an error if it happens.
1285  *
1286  * Note that this function does not do anything to cause the data in the
1287  * RAM memory region to be migrated; that is the responsibility of the caller.
1288  */
1289 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1290                                     Object *owner,
1291                                     const char *name,
1292                                     uint64_t size,
1293                                     uint32_t ram_flags,
1294                                     int fd,
1295                                     ram_addr_t offset,
1296                                     Error **errp);
1297 #endif
1298 
1299 /**
1300  * memory_region_init_ram_ptr:  Initialize RAM memory region from a
1301  *                              user-provided pointer.  Accesses into the
1302  *                              region will modify memory directly.
1303  *
1304  * @mr: the #MemoryRegion to be initialized.
1305  * @owner: the object that tracks the region's reference count
1306  * @name: Region name, becomes part of RAMBlock name used in migration stream
1307  *        must be unique within any device
1308  * @size: size of the region.
1309  * @ptr: memory to be mapped; must contain at least @size bytes.
1310  *
1311  * Note that this function does not do anything to cause the data in the
1312  * RAM memory region to be migrated; that is the responsibility of the caller.
1313  */
1314 void memory_region_init_ram_ptr(MemoryRegion *mr,
1315                                 Object *owner,
1316                                 const char *name,
1317                                 uint64_t size,
1318                                 void *ptr);
1319 
1320 /**
1321  * memory_region_init_ram_device_ptr:  Initialize RAM device memory region from
1322  *                                     a user-provided pointer.
1323  *
1324  * A RAM device represents a mapping to a physical device, such as to a PCI
1325  * MMIO BAR of an vfio-pci assigned device.  The memory region may be mapped
1326  * into the VM address space and access to the region will modify memory
1327  * directly.  However, the memory region should not be included in a memory
1328  * dump (device may not be enabled/mapped at the time of the dump), and
1329  * operations incompatible with manipulating MMIO should be avoided.  Replaces
1330  * skip_dump flag.
1331  *
1332  * @mr: the #MemoryRegion to be initialized.
1333  * @owner: the object that tracks the region's reference count
1334  * @name: the name of the region.
1335  * @size: size of the region.
1336  * @ptr: memory to be mapped; must contain at least @size bytes.
1337  *
1338  * Note that this function does not do anything to cause the data in the
1339  * RAM memory region to be migrated; that is the responsibility of the caller.
1340  * (For RAM device memory regions, migrating the contents rarely makes sense.)
1341  */
1342 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1343                                        Object *owner,
1344                                        const char *name,
1345                                        uint64_t size,
1346                                        void *ptr);
1347 
1348 /**
1349  * memory_region_init_alias: Initialize a memory region that aliases all or a
1350  *                           part of another memory region.
1351  *
1352  * @mr: the #MemoryRegion to be initialized.
1353  * @owner: the object that tracks the region's reference count
1354  * @name: used for debugging; not visible to the user or ABI
1355  * @orig: the region to be referenced; @mr will be equivalent to
1356  *        @orig between @offset and @offset + @size - 1.
1357  * @offset: start of the section in @orig to be referenced.
1358  * @size: size of the region.
1359  */
1360 void memory_region_init_alias(MemoryRegion *mr,
1361                               Object *owner,
1362                               const char *name,
1363                               MemoryRegion *orig,
1364                               hwaddr offset,
1365                               uint64_t size);
1366 
1367 /**
1368  * memory_region_init_rom_nomigrate: Initialize a ROM memory region.
1369  *
1370  * This has the same effect as calling memory_region_init_ram_nomigrate()
1371  * and then marking the resulting region read-only with
1372  * memory_region_set_readonly().
1373  *
1374  * Note that this function does not do anything to cause the data in the
1375  * RAM side of the memory region to be migrated; that is the responsibility
1376  * of the caller.
1377  *
1378  * @mr: the #MemoryRegion to be initialized.
1379  * @owner: the object that tracks the region's reference count
1380  * @name: Region name, becomes part of RAMBlock name used in migration stream
1381  *        must be unique within any device
1382  * @size: size of the region.
1383  * @errp: pointer to Error*, to store an error if it happens.
1384  */
1385 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1386                                       Object *owner,
1387                                       const char *name,
1388                                       uint64_t size,
1389                                       Error **errp);
1390 
1391 /**
1392  * memory_region_init_rom_device_nomigrate:  Initialize a ROM memory region.
1393  *                                 Writes are handled via callbacks.
1394  *
1395  * Note that this function does not do anything to cause the data in the
1396  * RAM side of the memory region to be migrated; that is the responsibility
1397  * of the caller.
1398  *
1399  * @mr: the #MemoryRegion to be initialized.
1400  * @owner: the object that tracks the region's reference count
1401  * @ops: callbacks for write access handling (must not be NULL).
1402  * @opaque: passed to the read and write callbacks of the @ops structure.
1403  * @name: Region name, becomes part of RAMBlock name used in migration stream
1404  *        must be unique within any device
1405  * @size: size of the region.
1406  * @errp: pointer to Error*, to store an error if it happens.
1407  */
1408 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1409                                              Object *owner,
1410                                              const MemoryRegionOps *ops,
1411                                              void *opaque,
1412                                              const char *name,
1413                                              uint64_t size,
1414                                              Error **errp);
1415 
1416 /**
1417  * memory_region_init_iommu: Initialize a memory region of a custom type
1418  * that translates addresses
1419  *
1420  * An IOMMU region translates addresses and forwards accesses to a target
1421  * memory region.
1422  *
1423  * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION.
1424  * @_iommu_mr should be a pointer to enough memory for an instance of
1425  * that subclass, @instance_size is the size of that subclass, and
1426  * @mrtypename is its name. This function will initialize @_iommu_mr as an
1427  * instance of the subclass, and its methods will then be called to handle
1428  * accesses to the memory region. See the documentation of
1429  * #IOMMUMemoryRegionClass for further details.
1430  *
1431  * @_iommu_mr: the #IOMMUMemoryRegion to be initialized
1432  * @instance_size: the IOMMUMemoryRegion subclass instance size
1433  * @mrtypename: the type name of the #IOMMUMemoryRegion
1434  * @owner: the object that tracks the region's reference count
1435  * @name: used for debugging; not visible to the user or ABI
1436  * @size: size of the region.
1437  */
1438 void memory_region_init_iommu(void *_iommu_mr,
1439                               size_t instance_size,
1440                               const char *mrtypename,
1441                               Object *owner,
1442                               const char *name,
1443                               uint64_t size);
1444 
1445 /**
1446  * memory_region_init_ram - Initialize RAM memory region.  Accesses into the
1447  *                          region will modify memory directly.
1448  *
1449  * @mr: the #MemoryRegion to be initialized
1450  * @owner: the object that tracks the region's reference count (must be
1451  *         TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL)
1452  * @name: name of the memory region
1453  * @size: size of the region in bytes
1454  * @errp: pointer to Error*, to store an error if it happens.
1455  *
1456  * This function allocates RAM for a board model or device, and
1457  * arranges for it to be migrated (by calling vmstate_register_ram()
1458  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1459  * @owner is NULL).
1460  *
1461  * TODO: Currently we restrict @owner to being either NULL (for
1462  * global RAM regions with no owner) or devices, so that we can
1463  * give the RAM block a unique name for migration purposes.
1464  * We should lift this restriction and allow arbitrary Objects.
1465  * If you pass a non-NULL non-device @owner then we will assert.
1466  */
1467 void memory_region_init_ram(MemoryRegion *mr,
1468                             Object *owner,
1469                             const char *name,
1470                             uint64_t size,
1471                             Error **errp);
1472 
1473 /**
1474  * memory_region_init_rom: Initialize a ROM memory region.
1475  *
1476  * This has the same effect as calling memory_region_init_ram()
1477  * and then marking the resulting region read-only with
1478  * memory_region_set_readonly(). This includes arranging for the
1479  * contents to be migrated.
1480  *
1481  * TODO: Currently we restrict @owner to being either NULL (for
1482  * global RAM regions with no owner) or devices, so that we can
1483  * give the RAM block a unique name for migration purposes.
1484  * We should lift this restriction and allow arbitrary Objects.
1485  * If you pass a non-NULL non-device @owner then we will assert.
1486  *
1487  * @mr: the #MemoryRegion to be initialized.
1488  * @owner: the object that tracks the region's reference count
1489  * @name: Region name, becomes part of RAMBlock name used in migration stream
1490  *        must be unique within any device
1491  * @size: size of the region.
1492  * @errp: pointer to Error*, to store an error if it happens.
1493  */
1494 void memory_region_init_rom(MemoryRegion *mr,
1495                             Object *owner,
1496                             const char *name,
1497                             uint64_t size,
1498                             Error **errp);
1499 
1500 /**
1501  * memory_region_init_rom_device:  Initialize a ROM memory region.
1502  *                                 Writes are handled via callbacks.
1503  *
1504  * This function initializes a memory region backed by RAM for reads
1505  * and callbacks for writes, and arranges for the RAM backing to
1506  * be migrated (by calling vmstate_register_ram()
1507  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1508  * @owner is NULL).
1509  *
1510  * TODO: Currently we restrict @owner to being either NULL (for
1511  * global RAM regions with no owner) or devices, so that we can
1512  * give the RAM block a unique name for migration purposes.
1513  * We should lift this restriction and allow arbitrary Objects.
1514  * If you pass a non-NULL non-device @owner then we will assert.
1515  *
1516  * @mr: the #MemoryRegion to be initialized.
1517  * @owner: the object that tracks the region's reference count
1518  * @ops: callbacks for write access handling (must not be NULL).
1519  * @opaque: passed to the read and write callbacks of the @ops structure.
1520  * @name: Region name, becomes part of RAMBlock name used in migration stream
1521  *        must be unique within any device
1522  * @size: size of the region.
1523  * @errp: pointer to Error*, to store an error if it happens.
1524  */
1525 void memory_region_init_rom_device(MemoryRegion *mr,
1526                                    Object *owner,
1527                                    const MemoryRegionOps *ops,
1528                                    void *opaque,
1529                                    const char *name,
1530                                    uint64_t size,
1531                                    Error **errp);
1532 
1533 
1534 /**
1535  * memory_region_owner: get a memory region's owner.
1536  *
1537  * @mr: the memory region being queried.
1538  */
1539 Object *memory_region_owner(MemoryRegion *mr);
1540 
1541 /**
1542  * memory_region_size: get a memory region's size.
1543  *
1544  * @mr: the memory region being queried.
1545  */
1546 uint64_t memory_region_size(MemoryRegion *mr);
1547 
1548 /**
1549  * memory_region_is_ram: check whether a memory region is random access
1550  *
1551  * Returns %true if a memory region is random access.
1552  *
1553  * @mr: the memory region being queried
1554  */
1555 static inline bool memory_region_is_ram(MemoryRegion *mr)
1556 {
1557     return mr->ram;
1558 }
1559 
1560 /**
1561  * memory_region_is_ram_device: check whether a memory region is a ram device
1562  *
1563  * Returns %true if a memory region is a device backed ram region
1564  *
1565  * @mr: the memory region being queried
1566  */
1567 bool memory_region_is_ram_device(MemoryRegion *mr);
1568 
1569 /**
1570  * memory_region_is_romd: check whether a memory region is in ROMD mode
1571  *
1572  * Returns %true if a memory region is a ROM device and currently set to allow
1573  * direct reads.
1574  *
1575  * @mr: the memory region being queried
1576  */
1577 static inline bool memory_region_is_romd(MemoryRegion *mr)
1578 {
1579     return mr->rom_device && mr->romd_mode;
1580 }
1581 
1582 /**
1583  * memory_region_is_protected: check whether a memory region is protected
1584  *
1585  * Returns %true if a memory region is protected RAM and cannot be accessed
1586  * via standard mechanisms, e.g. DMA.
1587  *
1588  * @mr: the memory region being queried
1589  */
1590 bool memory_region_is_protected(MemoryRegion *mr);
1591 
1592 /**
1593  * memory_region_get_iommu: check whether a memory region is an iommu
1594  *
1595  * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu,
1596  * otherwise NULL.
1597  *
1598  * @mr: the memory region being queried
1599  */
1600 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr)
1601 {
1602     if (mr->alias) {
1603         return memory_region_get_iommu(mr->alias);
1604     }
1605     if (mr->is_iommu) {
1606         return (IOMMUMemoryRegion *) mr;
1607     }
1608     return NULL;
1609 }
1610 
1611 /**
1612  * memory_region_get_iommu_class_nocheck: returns iommu memory region class
1613  *   if an iommu or NULL if not
1614  *
1615  * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu,
1616  * otherwise NULL. This is fast path avoiding QOM checking, use with caution.
1617  *
1618  * @iommu_mr: the memory region being queried
1619  */
1620 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck(
1621         IOMMUMemoryRegion *iommu_mr)
1622 {
1623     return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class);
1624 }
1625 
1626 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL)
1627 
1628 /**
1629  * memory_region_iommu_get_min_page_size: get minimum supported page size
1630  * for an iommu
1631  *
1632  * Returns minimum supported page size for an iommu.
1633  *
1634  * @iommu_mr: the memory region being queried
1635  */
1636 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr);
1637 
1638 /**
1639  * memory_region_notify_iommu: notify a change in an IOMMU translation entry.
1640  *
1641  * Note: for any IOMMU implementation, an in-place mapping change
1642  * should be notified with an UNMAP followed by a MAP.
1643  *
1644  * @iommu_mr: the memory region that was changed
1645  * @iommu_idx: the IOMMU index for the translation table which has changed
1646  * @event: TLB event with the new entry in the IOMMU translation table.
1647  *         The entry replaces all old entries for the same virtual I/O address
1648  *         range.
1649  */
1650 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1651                                 int iommu_idx,
1652                                 IOMMUTLBEvent event);
1653 
1654 /**
1655  * memory_region_notify_iommu_one: notify a change in an IOMMU translation
1656  *                           entry to a single notifier
1657  *
1658  * This works just like memory_region_notify_iommu(), but it only
1659  * notifies a specific notifier, not all of them.
1660  *
1661  * @notifier: the notifier to be notified
1662  * @event: TLB event with the new entry in the IOMMU translation table.
1663  *         The entry replaces all old entries for the same virtual I/O address
1664  *         range.
1665  */
1666 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1667                                     IOMMUTLBEvent *event);
1668 
1669 /**
1670  * memory_region_register_iommu_notifier: register a notifier for changes to
1671  * IOMMU translation entries.
1672  *
1673  * Returns 0 on success, or a negative errno otherwise. In particular,
1674  * -EINVAL indicates that at least one of the attributes of the notifier
1675  * is not supported (flag/range) by the IOMMU memory region. In case of error
1676  * the error object must be created.
1677  *
1678  * @mr: the memory region to observe
1679  * @n: the IOMMUNotifier to be added; the notify callback receives a
1680  *     pointer to an #IOMMUTLBEntry as the opaque value; the pointer
1681  *     ceases to be valid on exit from the notifier.
1682  * @errp: pointer to Error*, to store an error if it happens.
1683  */
1684 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1685                                           IOMMUNotifier *n, Error **errp);
1686 
1687 /**
1688  * memory_region_iommu_replay: replay existing IOMMU translations to
1689  * a notifier with the minimum page granularity returned by
1690  * mr->iommu_ops->get_page_size().
1691  *
1692  * Note: this is not related to record-and-replay functionality.
1693  *
1694  * @iommu_mr: the memory region to observe
1695  * @n: the notifier to which to replay iommu mappings
1696  */
1697 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
1698 
1699 /**
1700  * memory_region_unregister_iommu_notifier: unregister a notifier for
1701  * changes to IOMMU translation entries.
1702  *
1703  * @mr: the memory region which was observed and for which notity_stopped()
1704  *      needs to be called
1705  * @n: the notifier to be removed.
1706  */
1707 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1708                                              IOMMUNotifier *n);
1709 
1710 /**
1711  * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is
1712  * defined on the IOMMU.
1713  *
1714  * Returns 0 on success, or a negative errno otherwise. In particular,
1715  * -EINVAL indicates that the IOMMU does not support the requested
1716  * attribute.
1717  *
1718  * @iommu_mr: the memory region
1719  * @attr: the requested attribute
1720  * @data: a pointer to the requested attribute data
1721  */
1722 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1723                                  enum IOMMUMemoryRegionAttr attr,
1724                                  void *data);
1725 
1726 /**
1727  * memory_region_iommu_attrs_to_index: return the IOMMU index to
1728  * use for translations with the given memory transaction attributes.
1729  *
1730  * @iommu_mr: the memory region
1731  * @attrs: the memory transaction attributes
1732  */
1733 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1734                                        MemTxAttrs attrs);
1735 
1736 /**
1737  * memory_region_iommu_num_indexes: return the total number of IOMMU
1738  * indexes that this IOMMU supports.
1739  *
1740  * @iommu_mr: the memory region
1741  */
1742 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr);
1743 
1744 /**
1745  * memory_region_iommu_set_page_size_mask: set the supported page
1746  * sizes for a given IOMMU memory region
1747  *
1748  * @iommu_mr: IOMMU memory region
1749  * @page_size_mask: supported page size mask
1750  * @errp: pointer to Error*, to store an error if it happens.
1751  */
1752 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1753                                            uint64_t page_size_mask,
1754                                            Error **errp);
1755 
1756 /**
1757  * memory_region_name: get a memory region's name
1758  *
1759  * Returns the string that was used to initialize the memory region.
1760  *
1761  * @mr: the memory region being queried
1762  */
1763 const char *memory_region_name(const MemoryRegion *mr);
1764 
1765 /**
1766  * memory_region_is_logging: return whether a memory region is logging writes
1767  *
1768  * Returns %true if the memory region is logging writes for the given client
1769  *
1770  * @mr: the memory region being queried
1771  * @client: the client being queried
1772  */
1773 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client);
1774 
1775 /**
1776  * memory_region_get_dirty_log_mask: return the clients for which a
1777  * memory region is logging writes.
1778  *
1779  * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants
1780  * are the bit indices.
1781  *
1782  * @mr: the memory region being queried
1783  */
1784 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr);
1785 
1786 /**
1787  * memory_region_is_rom: check whether a memory region is ROM
1788  *
1789  * Returns %true if a memory region is read-only memory.
1790  *
1791  * @mr: the memory region being queried
1792  */
1793 static inline bool memory_region_is_rom(MemoryRegion *mr)
1794 {
1795     return mr->ram && mr->readonly;
1796 }
1797 
1798 /**
1799  * memory_region_is_nonvolatile: check whether a memory region is non-volatile
1800  *
1801  * Returns %true is a memory region is non-volatile memory.
1802  *
1803  * @mr: the memory region being queried
1804  */
1805 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr)
1806 {
1807     return mr->nonvolatile;
1808 }
1809 
1810 /**
1811  * memory_region_get_fd: Get a file descriptor backing a RAM memory region.
1812  *
1813  * Returns a file descriptor backing a file-based RAM memory region,
1814  * or -1 if the region is not a file-based RAM memory region.
1815  *
1816  * @mr: the RAM or alias memory region being queried.
1817  */
1818 int memory_region_get_fd(MemoryRegion *mr);
1819 
1820 /**
1821  * memory_region_from_host: Convert a pointer into a RAM memory region
1822  * and an offset within it.
1823  *
1824  * Given a host pointer inside a RAM memory region (created with
1825  * memory_region_init_ram() or memory_region_init_ram_ptr()), return
1826  * the MemoryRegion and the offset within it.
1827  *
1828  * Use with care; by the time this function returns, the returned pointer is
1829  * not protected by RCU anymore.  If the caller is not within an RCU critical
1830  * section and does not hold the iothread lock, it must have other means of
1831  * protecting the pointer, such as a reference to the region that includes
1832  * the incoming ram_addr_t.
1833  *
1834  * @ptr: the host pointer to be converted
1835  * @offset: the offset within memory region
1836  */
1837 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset);
1838 
1839 /**
1840  * memory_region_get_ram_ptr: Get a pointer into a RAM memory region.
1841  *
1842  * Returns a host pointer to a RAM memory region (created with
1843  * memory_region_init_ram() or memory_region_init_ram_ptr()).
1844  *
1845  * Use with care; by the time this function returns, the returned pointer is
1846  * not protected by RCU anymore.  If the caller is not within an RCU critical
1847  * section and does not hold the iothread lock, it must have other means of
1848  * protecting the pointer, such as a reference to the region that includes
1849  * the incoming ram_addr_t.
1850  *
1851  * @mr: the memory region being queried.
1852  */
1853 void *memory_region_get_ram_ptr(MemoryRegion *mr);
1854 
1855 /* memory_region_ram_resize: Resize a RAM region.
1856  *
1857  * Resizing RAM while migrating can result in the migration being canceled.
1858  * Care has to be taken if the guest might have already detected the memory.
1859  *
1860  * @mr: a memory region created with @memory_region_init_resizeable_ram.
1861  * @newsize: the new size the region
1862  * @errp: pointer to Error*, to store an error if it happens.
1863  */
1864 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize,
1865                               Error **errp);
1866 
1867 /**
1868  * memory_region_msync: Synchronize selected address range of
1869  * a memory mapped region
1870  *
1871  * @mr: the memory region to be msync
1872  * @addr: the initial address of the range to be sync
1873  * @size: the size of the range to be sync
1874  */
1875 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size);
1876 
1877 /**
1878  * memory_region_writeback: Trigger cache writeback for
1879  * selected address range
1880  *
1881  * @mr: the memory region to be updated
1882  * @addr: the initial address of the range to be written back
1883  * @size: the size of the range to be written back
1884  */
1885 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size);
1886 
1887 /**
1888  * memory_region_set_log: Turn dirty logging on or off for a region.
1889  *
1890  * Turns dirty logging on or off for a specified client (display, migration).
1891  * Only meaningful for RAM regions.
1892  *
1893  * @mr: the memory region being updated.
1894  * @log: whether dirty logging is to be enabled or disabled.
1895  * @client: the user of the logging information; %DIRTY_MEMORY_VGA only.
1896  */
1897 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client);
1898 
1899 /**
1900  * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region.
1901  *
1902  * Marks a range of bytes as dirty, after it has been dirtied outside
1903  * guest code.
1904  *
1905  * @mr: the memory region being dirtied.
1906  * @addr: the address (relative to the start of the region) being dirtied.
1907  * @size: size of the range being dirtied.
1908  */
1909 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1910                              hwaddr size);
1911 
1912 /**
1913  * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range
1914  *
1915  * This function is called when the caller wants to clear the remote
1916  * dirty bitmap of a memory range within the memory region.  This can
1917  * be used by e.g. KVM to manually clear dirty log when
1918  * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host
1919  * kernel.
1920  *
1921  * @mr:     the memory region to clear the dirty log upon
1922  * @start:  start address offset within the memory region
1923  * @len:    length of the memory region to clear dirty bitmap
1924  */
1925 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
1926                                       hwaddr len);
1927 
1928 /**
1929  * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty
1930  *                                         bitmap and clear it.
1931  *
1932  * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and
1933  * returns the snapshot.  The snapshot can then be used to query dirty
1934  * status, using memory_region_snapshot_get_dirty.  Snapshotting allows
1935  * querying the same page multiple times, which is especially useful for
1936  * display updates where the scanlines often are not page aligned.
1937  *
1938  * The dirty bitmap region which gets copyed into the snapshot (and
1939  * cleared afterwards) can be larger than requested.  The boundaries
1940  * are rounded up/down so complete bitmap longs (covering 64 pages on
1941  * 64bit hosts) can be copied over into the bitmap snapshot.  Which
1942  * isn't a problem for display updates as the extra pages are outside
1943  * the visible area, and in case the visible area changes a full
1944  * display redraw is due anyway.  Should other use cases for this
1945  * function emerge we might have to revisit this implementation
1946  * detail.
1947  *
1948  * Use g_free to release DirtyBitmapSnapshot.
1949  *
1950  * @mr: the memory region being queried.
1951  * @addr: the address (relative to the start of the region) being queried.
1952  * @size: the size of the range being queried.
1953  * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA.
1954  */
1955 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
1956                                                             hwaddr addr,
1957                                                             hwaddr size,
1958                                                             unsigned client);
1959 
1960 /**
1961  * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty
1962  *                                   in the specified dirty bitmap snapshot.
1963  *
1964  * @mr: the memory region being queried.
1965  * @snap: the dirty bitmap snapshot
1966  * @addr: the address (relative to the start of the region) being queried.
1967  * @size: the size of the range being queried.
1968  */
1969 bool memory_region_snapshot_get_dirty(MemoryRegion *mr,
1970                                       DirtyBitmapSnapshot *snap,
1971                                       hwaddr addr, hwaddr size);
1972 
1973 /**
1974  * memory_region_reset_dirty: Mark a range of pages as clean, for a specified
1975  *                            client.
1976  *
1977  * Marks a range of pages as no longer dirty.
1978  *
1979  * @mr: the region being updated.
1980  * @addr: the start of the subrange being cleaned.
1981  * @size: the size of the subrange being cleaned.
1982  * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or
1983  *          %DIRTY_MEMORY_VGA.
1984  */
1985 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1986                                hwaddr size, unsigned client);
1987 
1988 /**
1989  * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate
1990  *                                 TBs (for self-modifying code).
1991  *
1992  * The MemoryRegionOps->write() callback of a ROM device must use this function
1993  * to mark byte ranges that have been modified internally, such as by directly
1994  * accessing the memory returned by memory_region_get_ram_ptr().
1995  *
1996  * This function marks the range dirty and invalidates TBs so that TCG can
1997  * detect self-modifying code.
1998  *
1999  * @mr: the region being flushed.
2000  * @addr: the start, relative to the start of the region, of the range being
2001  *        flushed.
2002  * @size: the size, in bytes, of the range being flushed.
2003  */
2004 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size);
2005 
2006 /**
2007  * memory_region_set_readonly: Turn a memory region read-only (or read-write)
2008  *
2009  * Allows a memory region to be marked as read-only (turning it into a ROM).
2010  * only useful on RAM regions.
2011  *
2012  * @mr: the region being updated.
2013  * @readonly: whether rhe region is to be ROM or RAM.
2014  */
2015 void memory_region_set_readonly(MemoryRegion *mr, bool readonly);
2016 
2017 /**
2018  * memory_region_set_nonvolatile: Turn a memory region non-volatile
2019  *
2020  * Allows a memory region to be marked as non-volatile.
2021  * only useful on RAM regions.
2022  *
2023  * @mr: the region being updated.
2024  * @nonvolatile: whether rhe region is to be non-volatile.
2025  */
2026 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile);
2027 
2028 /**
2029  * memory_region_rom_device_set_romd: enable/disable ROMD mode
2030  *
2031  * Allows a ROM device (initialized with memory_region_init_rom_device() to
2032  * set to ROMD mode (default) or MMIO mode.  When it is in ROMD mode, the
2033  * device is mapped to guest memory and satisfies read access directly.
2034  * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function.
2035  * Writes are always handled by the #MemoryRegion.write function.
2036  *
2037  * @mr: the memory region to be updated
2038  * @romd_mode: %true to put the region into ROMD mode
2039  */
2040 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode);
2041 
2042 /**
2043  * memory_region_set_coalescing: Enable memory coalescing for the region.
2044  *
2045  * Enabled writes to a region to be queued for later processing. MMIO ->write
2046  * callbacks may be delayed until a non-coalesced MMIO is issued.
2047  * Only useful for IO regions.  Roughly similar to write-combining hardware.
2048  *
2049  * @mr: the memory region to be write coalesced
2050  */
2051 void memory_region_set_coalescing(MemoryRegion *mr);
2052 
2053 /**
2054  * memory_region_add_coalescing: Enable memory coalescing for a sub-range of
2055  *                               a region.
2056  *
2057  * Like memory_region_set_coalescing(), but works on a sub-range of a region.
2058  * Multiple calls can be issued coalesced disjoint ranges.
2059  *
2060  * @mr: the memory region to be updated.
2061  * @offset: the start of the range within the region to be coalesced.
2062  * @size: the size of the subrange to be coalesced.
2063  */
2064 void memory_region_add_coalescing(MemoryRegion *mr,
2065                                   hwaddr offset,
2066                                   uint64_t size);
2067 
2068 /**
2069  * memory_region_clear_coalescing: Disable MMIO coalescing for the region.
2070  *
2071  * Disables any coalescing caused by memory_region_set_coalescing() or
2072  * memory_region_add_coalescing().  Roughly equivalent to uncacheble memory
2073  * hardware.
2074  *
2075  * @mr: the memory region to be updated.
2076  */
2077 void memory_region_clear_coalescing(MemoryRegion *mr);
2078 
2079 /**
2080  * memory_region_set_flush_coalesced: Enforce memory coalescing flush before
2081  *                                    accesses.
2082  *
2083  * Ensure that pending coalesced MMIO request are flushed before the memory
2084  * region is accessed. This property is automatically enabled for all regions
2085  * passed to memory_region_set_coalescing() and memory_region_add_coalescing().
2086  *
2087  * @mr: the memory region to be updated.
2088  */
2089 void memory_region_set_flush_coalesced(MemoryRegion *mr);
2090 
2091 /**
2092  * memory_region_clear_flush_coalesced: Disable memory coalescing flush before
2093  *                                      accesses.
2094  *
2095  * Clear the automatic coalesced MMIO flushing enabled via
2096  * memory_region_set_flush_coalesced. Note that this service has no effect on
2097  * memory regions that have MMIO coalescing enabled for themselves. For them,
2098  * automatic flushing will stop once coalescing is disabled.
2099  *
2100  * @mr: the memory region to be updated.
2101  */
2102 void memory_region_clear_flush_coalesced(MemoryRegion *mr);
2103 
2104 /**
2105  * memory_region_add_eventfd: Request an eventfd to be triggered when a word
2106  *                            is written to a location.
2107  *
2108  * Marks a word in an IO region (initialized with memory_region_init_io())
2109  * as a trigger for an eventfd event.  The I/O callback will not be called.
2110  * The caller must be prepared to handle failure (that is, take the required
2111  * action if the callback _is_ called).
2112  *
2113  * @mr: the memory region being updated.
2114  * @addr: the address within @mr that is to be monitored
2115  * @size: the size of the access to trigger the eventfd
2116  * @match_data: whether to match against @data, instead of just @addr
2117  * @data: the data to match against the guest write
2118  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2119  **/
2120 void memory_region_add_eventfd(MemoryRegion *mr,
2121                                hwaddr addr,
2122                                unsigned size,
2123                                bool match_data,
2124                                uint64_t data,
2125                                EventNotifier *e);
2126 
2127 /**
2128  * memory_region_del_eventfd: Cancel an eventfd.
2129  *
2130  * Cancels an eventfd trigger requested by a previous
2131  * memory_region_add_eventfd() call.
2132  *
2133  * @mr: the memory region being updated.
2134  * @addr: the address within @mr that is to be monitored
2135  * @size: the size of the access to trigger the eventfd
2136  * @match_data: whether to match against @data, instead of just @addr
2137  * @data: the data to match against the guest write
2138  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2139  */
2140 void memory_region_del_eventfd(MemoryRegion *mr,
2141                                hwaddr addr,
2142                                unsigned size,
2143                                bool match_data,
2144                                uint64_t data,
2145                                EventNotifier *e);
2146 
2147 /**
2148  * memory_region_add_subregion: Add a subregion to a container.
2149  *
2150  * Adds a subregion at @offset.  The subregion may not overlap with other
2151  * subregions (except for those explicitly marked as overlapping).  A region
2152  * may only be added once as a subregion (unless removed with
2153  * memory_region_del_subregion()); use memory_region_init_alias() if you
2154  * want a region to be a subregion in multiple locations.
2155  *
2156  * @mr: the region to contain the new subregion; must be a container
2157  *      initialized with memory_region_init().
2158  * @offset: the offset relative to @mr where @subregion is added.
2159  * @subregion: the subregion to be added.
2160  */
2161 void memory_region_add_subregion(MemoryRegion *mr,
2162                                  hwaddr offset,
2163                                  MemoryRegion *subregion);
2164 /**
2165  * memory_region_add_subregion_overlap: Add a subregion to a container
2166  *                                      with overlap.
2167  *
2168  * Adds a subregion at @offset.  The subregion may overlap with other
2169  * subregions.  Conflicts are resolved by having a higher @priority hide a
2170  * lower @priority. Subregions without priority are taken as @priority 0.
2171  * A region may only be added once as a subregion (unless removed with
2172  * memory_region_del_subregion()); use memory_region_init_alias() if you
2173  * want a region to be a subregion in multiple locations.
2174  *
2175  * @mr: the region to contain the new subregion; must be a container
2176  *      initialized with memory_region_init().
2177  * @offset: the offset relative to @mr where @subregion is added.
2178  * @subregion: the subregion to be added.
2179  * @priority: used for resolving overlaps; highest priority wins.
2180  */
2181 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2182                                          hwaddr offset,
2183                                          MemoryRegion *subregion,
2184                                          int priority);
2185 
2186 /**
2187  * memory_region_get_ram_addr: Get the ram address associated with a memory
2188  *                             region
2189  *
2190  * @mr: the region to be queried
2191  */
2192 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr);
2193 
2194 uint64_t memory_region_get_alignment(const MemoryRegion *mr);
2195 /**
2196  * memory_region_del_subregion: Remove a subregion.
2197  *
2198  * Removes a subregion from its container.
2199  *
2200  * @mr: the container to be updated.
2201  * @subregion: the region being removed; must be a current subregion of @mr.
2202  */
2203 void memory_region_del_subregion(MemoryRegion *mr,
2204                                  MemoryRegion *subregion);
2205 
2206 /*
2207  * memory_region_set_enabled: dynamically enable or disable a region
2208  *
2209  * Enables or disables a memory region.  A disabled memory region
2210  * ignores all accesses to itself and its subregions.  It does not
2211  * obscure sibling subregions with lower priority - it simply behaves as
2212  * if it was removed from the hierarchy.
2213  *
2214  * Regions default to being enabled.
2215  *
2216  * @mr: the region to be updated
2217  * @enabled: whether to enable or disable the region
2218  */
2219 void memory_region_set_enabled(MemoryRegion *mr, bool enabled);
2220 
2221 /*
2222  * memory_region_set_address: dynamically update the address of a region
2223  *
2224  * Dynamically updates the address of a region, relative to its container.
2225  * May be used on regions are currently part of a memory hierarchy.
2226  *
2227  * @mr: the region to be updated
2228  * @addr: new address, relative to container region
2229  */
2230 void memory_region_set_address(MemoryRegion *mr, hwaddr addr);
2231 
2232 /*
2233  * memory_region_set_size: dynamically update the size of a region.
2234  *
2235  * Dynamically updates the size of a region.
2236  *
2237  * @mr: the region to be updated
2238  * @size: used size of the region.
2239  */
2240 void memory_region_set_size(MemoryRegion *mr, uint64_t size);
2241 
2242 /*
2243  * memory_region_set_alias_offset: dynamically update a memory alias's offset
2244  *
2245  * Dynamically updates the offset into the target region that an alias points
2246  * to, as if the fourth argument to memory_region_init_alias() has changed.
2247  *
2248  * @mr: the #MemoryRegion to be updated; should be an alias.
2249  * @offset: the new offset into the target memory region
2250  */
2251 void memory_region_set_alias_offset(MemoryRegion *mr,
2252                                     hwaddr offset);
2253 
2254 /**
2255  * memory_region_present: checks if an address relative to a @container
2256  * translates into #MemoryRegion within @container
2257  *
2258  * Answer whether a #MemoryRegion within @container covers the address
2259  * @addr.
2260  *
2261  * @container: a #MemoryRegion within which @addr is a relative address
2262  * @addr: the area within @container to be searched
2263  */
2264 bool memory_region_present(MemoryRegion *container, hwaddr addr);
2265 
2266 /**
2267  * memory_region_is_mapped: returns true if #MemoryRegion is mapped
2268  * into any address space.
2269  *
2270  * @mr: a #MemoryRegion which should be checked if it's mapped
2271  */
2272 bool memory_region_is_mapped(MemoryRegion *mr);
2273 
2274 /**
2275  * memory_region_get_ram_discard_manager: get the #RamDiscardManager for a
2276  * #MemoryRegion
2277  *
2278  * The #RamDiscardManager cannot change while a memory region is mapped.
2279  *
2280  * @mr: the #MemoryRegion
2281  */
2282 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr);
2283 
2284 /**
2285  * memory_region_has_ram_discard_manager: check whether a #MemoryRegion has a
2286  * #RamDiscardManager assigned
2287  *
2288  * @mr: the #MemoryRegion
2289  */
2290 static inline bool memory_region_has_ram_discard_manager(MemoryRegion *mr)
2291 {
2292     return !!memory_region_get_ram_discard_manager(mr);
2293 }
2294 
2295 /**
2296  * memory_region_set_ram_discard_manager: set the #RamDiscardManager for a
2297  * #MemoryRegion
2298  *
2299  * This function must not be called for a mapped #MemoryRegion, a #MemoryRegion
2300  * that does not cover RAM, or a #MemoryRegion that already has a
2301  * #RamDiscardManager assigned.
2302  *
2303  * @mr: the #MemoryRegion
2304  * @rdm: #RamDiscardManager to set
2305  */
2306 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2307                                            RamDiscardManager *rdm);
2308 
2309 /**
2310  * memory_region_find: translate an address/size relative to a
2311  * MemoryRegion into a #MemoryRegionSection.
2312  *
2313  * Locates the first #MemoryRegion within @mr that overlaps the range
2314  * given by @addr and @size.
2315  *
2316  * Returns a #MemoryRegionSection that describes a contiguous overlap.
2317  * It will have the following characteristics:
2318  * - @size = 0 iff no overlap was found
2319  * - @mr is non-%NULL iff an overlap was found
2320  *
2321  * Remember that in the return value the @offset_within_region is
2322  * relative to the returned region (in the .@mr field), not to the
2323  * @mr argument.
2324  *
2325  * Similarly, the .@offset_within_address_space is relative to the
2326  * address space that contains both regions, the passed and the
2327  * returned one.  However, in the special case where the @mr argument
2328  * has no container (and thus is the root of the address space), the
2329  * following will hold:
2330  * - @offset_within_address_space >= @addr
2331  * - @offset_within_address_space + .@size <= @addr + @size
2332  *
2333  * @mr: a MemoryRegion within which @addr is a relative address
2334  * @addr: start of the area within @as to be searched
2335  * @size: size of the area to be searched
2336  */
2337 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2338                                        hwaddr addr, uint64_t size);
2339 
2340 /**
2341  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2342  *
2343  * Synchronizes the dirty page log for all address spaces.
2344  */
2345 void memory_global_dirty_log_sync(void);
2346 
2347 /**
2348  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2349  *
2350  * Synchronizes the vCPUs with a thread that is reading the dirty bitmap.
2351  * This function must be called after the dirty log bitmap is cleared, and
2352  * before dirty guest memory pages are read.  If you are using
2353  * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes
2354  * care of doing this.
2355  */
2356 void memory_global_after_dirty_log_sync(void);
2357 
2358 /**
2359  * memory_region_transaction_begin: Start a transaction.
2360  *
2361  * During a transaction, changes will be accumulated and made visible
2362  * only when the transaction ends (is committed).
2363  */
2364 void memory_region_transaction_begin(void);
2365 
2366 /**
2367  * memory_region_transaction_commit: Commit a transaction and make changes
2368  *                                   visible to the guest.
2369  */
2370 void memory_region_transaction_commit(void);
2371 
2372 /**
2373  * memory_listener_register: register callbacks to be called when memory
2374  *                           sections are mapped or unmapped into an address
2375  *                           space
2376  *
2377  * @listener: an object containing the callbacks to be called
2378  * @filter: if non-%NULL, only regions in this address space will be observed
2379  */
2380 void memory_listener_register(MemoryListener *listener, AddressSpace *filter);
2381 
2382 /**
2383  * memory_listener_unregister: undo the effect of memory_listener_register()
2384  *
2385  * @listener: an object containing the callbacks to be removed
2386  */
2387 void memory_listener_unregister(MemoryListener *listener);
2388 
2389 /**
2390  * memory_global_dirty_log_start: begin dirty logging for all regions
2391  */
2392 void memory_global_dirty_log_start(void);
2393 
2394 /**
2395  * memory_global_dirty_log_stop: end dirty logging for all regions
2396  */
2397 void memory_global_dirty_log_stop(void);
2398 
2399 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled);
2400 
2401 /**
2402  * memory_region_dispatch_read: perform a read directly to the specified
2403  * MemoryRegion.
2404  *
2405  * @mr: #MemoryRegion to access
2406  * @addr: address within that region
2407  * @pval: pointer to uint64_t which the data is written to
2408  * @op: size, sign, and endianness of the memory operation
2409  * @attrs: memory transaction attributes to use for the access
2410  */
2411 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
2412                                         hwaddr addr,
2413                                         uint64_t *pval,
2414                                         MemOp op,
2415                                         MemTxAttrs attrs);
2416 /**
2417  * memory_region_dispatch_write: perform a write directly to the specified
2418  * MemoryRegion.
2419  *
2420  * @mr: #MemoryRegion to access
2421  * @addr: address within that region
2422  * @data: data to write
2423  * @op: size, sign, and endianness of the memory operation
2424  * @attrs: memory transaction attributes to use for the access
2425  */
2426 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
2427                                          hwaddr addr,
2428                                          uint64_t data,
2429                                          MemOp op,
2430                                          MemTxAttrs attrs);
2431 
2432 /**
2433  * address_space_init: initializes an address space
2434  *
2435  * @as: an uninitialized #AddressSpace
2436  * @root: a #MemoryRegion that routes addresses for the address space
2437  * @name: an address space name.  The name is only used for debugging
2438  *        output.
2439  */
2440 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name);
2441 
2442 /**
2443  * address_space_destroy: destroy an address space
2444  *
2445  * Releases all resources associated with an address space.  After an address space
2446  * is destroyed, its root memory region (given by address_space_init()) may be destroyed
2447  * as well.
2448  *
2449  * @as: address space to be destroyed
2450  */
2451 void address_space_destroy(AddressSpace *as);
2452 
2453 /**
2454  * address_space_remove_listeners: unregister all listeners of an address space
2455  *
2456  * Removes all callbacks previously registered with memory_listener_register()
2457  * for @as.
2458  *
2459  * @as: an initialized #AddressSpace
2460  */
2461 void address_space_remove_listeners(AddressSpace *as);
2462 
2463 /**
2464  * address_space_rw: read from or write to an address space.
2465  *
2466  * Return a MemTxResult indicating whether the operation succeeded
2467  * or failed (eg unassigned memory, device rejected the transaction,
2468  * IOMMU fault).
2469  *
2470  * @as: #AddressSpace to be accessed
2471  * @addr: address within that address space
2472  * @attrs: memory transaction attributes
2473  * @buf: buffer with the data transferred
2474  * @len: the number of bytes to read or write
2475  * @is_write: indicates the transfer direction
2476  */
2477 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
2478                              MemTxAttrs attrs, void *buf,
2479                              hwaddr len, bool is_write);
2480 
2481 /**
2482  * address_space_write: write to address space.
2483  *
2484  * Return a MemTxResult indicating whether the operation succeeded
2485  * or failed (eg unassigned memory, device rejected the transaction,
2486  * IOMMU fault).
2487  *
2488  * @as: #AddressSpace to be accessed
2489  * @addr: address within that address space
2490  * @attrs: memory transaction attributes
2491  * @buf: buffer with the data transferred
2492  * @len: the number of bytes to write
2493  */
2494 MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
2495                                 MemTxAttrs attrs,
2496                                 const void *buf, hwaddr len);
2497 
2498 /**
2499  * address_space_write_rom: write to address space, including ROM.
2500  *
2501  * This function writes to the specified address space, but will
2502  * write data to both ROM and RAM. This is used for non-guest
2503  * writes like writes from the gdb debug stub or initial loading
2504  * of ROM contents.
2505  *
2506  * Note that portions of the write which attempt to write data to
2507  * a device will be silently ignored -- only real RAM and ROM will
2508  * be written to.
2509  *
2510  * Return a MemTxResult indicating whether the operation succeeded
2511  * or failed (eg unassigned memory, device rejected the transaction,
2512  * IOMMU fault).
2513  *
2514  * @as: #AddressSpace to be accessed
2515  * @addr: address within that address space
2516  * @attrs: memory transaction attributes
2517  * @buf: buffer with the data transferred
2518  * @len: the number of bytes to write
2519  */
2520 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr,
2521                                     MemTxAttrs attrs,
2522                                     const void *buf, hwaddr len);
2523 
2524 /* address_space_ld*: load from an address space
2525  * address_space_st*: store to an address space
2526  *
2527  * These functions perform a load or store of the byte, word,
2528  * longword or quad to the specified address within the AddressSpace.
2529  * The _le suffixed functions treat the data as little endian;
2530  * _be indicates big endian; no suffix indicates "same endianness
2531  * as guest CPU".
2532  *
2533  * The "guest CPU endianness" accessors are deprecated for use outside
2534  * target-* code; devices should be CPU-agnostic and use either the LE
2535  * or the BE accessors.
2536  *
2537  * @as #AddressSpace to be accessed
2538  * @addr: address within that address space
2539  * @val: data value, for stores
2540  * @attrs: memory transaction attributes
2541  * @result: location to write the success/failure of the transaction;
2542  *   if NULL, this information is discarded
2543  */
2544 
2545 #define SUFFIX
2546 #define ARG1         as
2547 #define ARG1_DECL    AddressSpace *as
2548 #include "exec/memory_ldst.h.inc"
2549 
2550 #define SUFFIX
2551 #define ARG1         as
2552 #define ARG1_DECL    AddressSpace *as
2553 #include "exec/memory_ldst_phys.h.inc"
2554 
2555 struct MemoryRegionCache {
2556     void *ptr;
2557     hwaddr xlat;
2558     hwaddr len;
2559     FlatView *fv;
2560     MemoryRegionSection mrs;
2561     bool is_write;
2562 };
2563 
2564 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL })
2565 
2566 
2567 /* address_space_ld*_cached: load from a cached #MemoryRegion
2568  * address_space_st*_cached: store into a cached #MemoryRegion
2569  *
2570  * These functions perform a load or store of the byte, word,
2571  * longword or quad to the specified address.  The address is
2572  * a physical address in the AddressSpace, but it must lie within
2573  * a #MemoryRegion that was mapped with address_space_cache_init.
2574  *
2575  * The _le suffixed functions treat the data as little endian;
2576  * _be indicates big endian; no suffix indicates "same endianness
2577  * as guest CPU".
2578  *
2579  * The "guest CPU endianness" accessors are deprecated for use outside
2580  * target-* code; devices should be CPU-agnostic and use either the LE
2581  * or the BE accessors.
2582  *
2583  * @cache: previously initialized #MemoryRegionCache to be accessed
2584  * @addr: address within the address space
2585  * @val: data value, for stores
2586  * @attrs: memory transaction attributes
2587  * @result: location to write the success/failure of the transaction;
2588  *   if NULL, this information is discarded
2589  */
2590 
2591 #define SUFFIX       _cached_slow
2592 #define ARG1         cache
2593 #define ARG1_DECL    MemoryRegionCache *cache
2594 #include "exec/memory_ldst.h.inc"
2595 
2596 /* Inline fast path for direct RAM access.  */
2597 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache,
2598     hwaddr addr, MemTxAttrs attrs, MemTxResult *result)
2599 {
2600     assert(addr < cache->len);
2601     if (likely(cache->ptr)) {
2602         return ldub_p(cache->ptr + addr);
2603     } else {
2604         return address_space_ldub_cached_slow(cache, addr, attrs, result);
2605     }
2606 }
2607 
2608 static inline void address_space_stb_cached(MemoryRegionCache *cache,
2609     hwaddr addr, uint8_t val, MemTxAttrs attrs, MemTxResult *result)
2610 {
2611     assert(addr < cache->len);
2612     if (likely(cache->ptr)) {
2613         stb_p(cache->ptr + addr, val);
2614     } else {
2615         address_space_stb_cached_slow(cache, addr, val, attrs, result);
2616     }
2617 }
2618 
2619 #define ENDIANNESS   _le
2620 #include "exec/memory_ldst_cached.h.inc"
2621 
2622 #define ENDIANNESS   _be
2623 #include "exec/memory_ldst_cached.h.inc"
2624 
2625 #define SUFFIX       _cached
2626 #define ARG1         cache
2627 #define ARG1_DECL    MemoryRegionCache *cache
2628 #include "exec/memory_ldst_phys.h.inc"
2629 
2630 /* address_space_cache_init: prepare for repeated access to a physical
2631  * memory region
2632  *
2633  * @cache: #MemoryRegionCache to be filled
2634  * @as: #AddressSpace to be accessed
2635  * @addr: address within that address space
2636  * @len: length of buffer
2637  * @is_write: indicates the transfer direction
2638  *
2639  * Will only work with RAM, and may map a subset of the requested range by
2640  * returning a value that is less than @len.  On failure, return a negative
2641  * errno value.
2642  *
2643  * Because it only works with RAM, this function can be used for
2644  * read-modify-write operations.  In this case, is_write should be %true.
2645  *
2646  * Note that addresses passed to the address_space_*_cached functions
2647  * are relative to @addr.
2648  */
2649 int64_t address_space_cache_init(MemoryRegionCache *cache,
2650                                  AddressSpace *as,
2651                                  hwaddr addr,
2652                                  hwaddr len,
2653                                  bool is_write);
2654 
2655 /**
2656  * address_space_cache_invalidate: complete a write to a #MemoryRegionCache
2657  *
2658  * @cache: The #MemoryRegionCache to operate on.
2659  * @addr: The first physical address that was written, relative to the
2660  * address that was passed to @address_space_cache_init.
2661  * @access_len: The number of bytes that were written starting at @addr.
2662  */
2663 void address_space_cache_invalidate(MemoryRegionCache *cache,
2664                                     hwaddr addr,
2665                                     hwaddr access_len);
2666 
2667 /**
2668  * address_space_cache_destroy: free a #MemoryRegionCache
2669  *
2670  * @cache: The #MemoryRegionCache whose memory should be released.
2671  */
2672 void address_space_cache_destroy(MemoryRegionCache *cache);
2673 
2674 /* address_space_get_iotlb_entry: translate an address into an IOTLB
2675  * entry. Should be called from an RCU critical section.
2676  */
2677 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
2678                                             bool is_write, MemTxAttrs attrs);
2679 
2680 /* address_space_translate: translate an address range into an address space
2681  * into a MemoryRegion and an address range into that section.  Should be
2682  * called from an RCU critical section, to avoid that the last reference
2683  * to the returned region disappears after address_space_translate returns.
2684  *
2685  * @fv: #FlatView to be accessed
2686  * @addr: address within that address space
2687  * @xlat: pointer to address within the returned memory region section's
2688  * #MemoryRegion.
2689  * @len: pointer to length
2690  * @is_write: indicates the transfer direction
2691  * @attrs: memory attributes
2692  */
2693 MemoryRegion *flatview_translate(FlatView *fv,
2694                                  hwaddr addr, hwaddr *xlat,
2695                                  hwaddr *len, bool is_write,
2696                                  MemTxAttrs attrs);
2697 
2698 static inline MemoryRegion *address_space_translate(AddressSpace *as,
2699                                                     hwaddr addr, hwaddr *xlat,
2700                                                     hwaddr *len, bool is_write,
2701                                                     MemTxAttrs attrs)
2702 {
2703     return flatview_translate(address_space_to_flatview(as),
2704                               addr, xlat, len, is_write, attrs);
2705 }
2706 
2707 /* address_space_access_valid: check for validity of accessing an address
2708  * space range
2709  *
2710  * Check whether memory is assigned to the given address space range, and
2711  * access is permitted by any IOMMU regions that are active for the address
2712  * space.
2713  *
2714  * For now, addr and len should be aligned to a page size.  This limitation
2715  * will be lifted in the future.
2716  *
2717  * @as: #AddressSpace to be accessed
2718  * @addr: address within that address space
2719  * @len: length of the area to be checked
2720  * @is_write: indicates the transfer direction
2721  * @attrs: memory attributes
2722  */
2723 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len,
2724                                 bool is_write, MemTxAttrs attrs);
2725 
2726 /* address_space_map: map a physical memory region into a host virtual address
2727  *
2728  * May map a subset of the requested range, given by and returned in @plen.
2729  * May return %NULL and set *@plen to zero(0), if resources needed to perform
2730  * the mapping are exhausted.
2731  * Use only for reads OR writes - not for read-modify-write operations.
2732  * Use cpu_register_map_client() to know when retrying the map operation is
2733  * likely to succeed.
2734  *
2735  * @as: #AddressSpace to be accessed
2736  * @addr: address within that address space
2737  * @plen: pointer to length of buffer; updated on return
2738  * @is_write: indicates the transfer direction
2739  * @attrs: memory attributes
2740  */
2741 void *address_space_map(AddressSpace *as, hwaddr addr,
2742                         hwaddr *plen, bool is_write, MemTxAttrs attrs);
2743 
2744 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map()
2745  *
2746  * Will also mark the memory as dirty if @is_write == %true.  @access_len gives
2747  * the amount of memory that was actually read or written by the caller.
2748  *
2749  * @as: #AddressSpace used
2750  * @buffer: host pointer as returned by address_space_map()
2751  * @len: buffer length as returned by address_space_map()
2752  * @access_len: amount of data actually transferred
2753  * @is_write: indicates the transfer direction
2754  */
2755 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2756                          bool is_write, hwaddr access_len);
2757 
2758 
2759 /* Internal functions, part of the implementation of address_space_read.  */
2760 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
2761                                     MemTxAttrs attrs, void *buf, hwaddr len);
2762 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
2763                                    MemTxAttrs attrs, void *buf,
2764                                    hwaddr len, hwaddr addr1, hwaddr l,
2765                                    MemoryRegion *mr);
2766 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr);
2767 
2768 /* Internal functions, part of the implementation of address_space_read_cached
2769  * and address_space_write_cached.  */
2770 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache,
2771                                            hwaddr addr, void *buf, hwaddr len);
2772 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache,
2773                                             hwaddr addr, const void *buf,
2774                                             hwaddr len);
2775 
2776 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
2777 {
2778     if (is_write) {
2779         return memory_region_is_ram(mr) && !mr->readonly &&
2780                !mr->rom_device && !memory_region_is_ram_device(mr);
2781     } else {
2782         return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) ||
2783                memory_region_is_romd(mr);
2784     }
2785 }
2786 
2787 /**
2788  * address_space_read: read from an address space.
2789  *
2790  * Return a MemTxResult indicating whether the operation succeeded
2791  * or failed (eg unassigned memory, device rejected the transaction,
2792  * IOMMU fault).  Called within RCU critical section.
2793  *
2794  * @as: #AddressSpace to be accessed
2795  * @addr: address within that address space
2796  * @attrs: memory transaction attributes
2797  * @buf: buffer with the data transferred
2798  * @len: length of the data transferred
2799  */
2800 static inline __attribute__((__always_inline__))
2801 MemTxResult address_space_read(AddressSpace *as, hwaddr addr,
2802                                MemTxAttrs attrs, void *buf,
2803                                hwaddr len)
2804 {
2805     MemTxResult result = MEMTX_OK;
2806     hwaddr l, addr1;
2807     void *ptr;
2808     MemoryRegion *mr;
2809     FlatView *fv;
2810 
2811     if (__builtin_constant_p(len)) {
2812         if (len) {
2813             RCU_READ_LOCK_GUARD();
2814             fv = address_space_to_flatview(as);
2815             l = len;
2816             mr = flatview_translate(fv, addr, &addr1, &l, false, attrs);
2817             if (len == l && memory_access_is_direct(mr, false)) {
2818                 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
2819                 memcpy(buf, ptr, len);
2820             } else {
2821                 result = flatview_read_continue(fv, addr, attrs, buf, len,
2822                                                 addr1, l, mr);
2823             }
2824         }
2825     } else {
2826         result = address_space_read_full(as, addr, attrs, buf, len);
2827     }
2828     return result;
2829 }
2830 
2831 /**
2832  * address_space_read_cached: read from a cached RAM region
2833  *
2834  * @cache: Cached region to be addressed
2835  * @addr: address relative to the base of the RAM region
2836  * @buf: buffer with the data transferred
2837  * @len: length of the data transferred
2838  */
2839 static inline MemTxResult
2840 address_space_read_cached(MemoryRegionCache *cache, hwaddr addr,
2841                           void *buf, hwaddr len)
2842 {
2843     assert(addr < cache->len && len <= cache->len - addr);
2844     fuzz_dma_read_cb(cache->xlat + addr, len, cache->mrs.mr);
2845     if (likely(cache->ptr)) {
2846         memcpy(buf, cache->ptr + addr, len);
2847         return MEMTX_OK;
2848     } else {
2849         return address_space_read_cached_slow(cache, addr, buf, len);
2850     }
2851 }
2852 
2853 /**
2854  * address_space_write_cached: write to a cached RAM region
2855  *
2856  * @cache: Cached region to be addressed
2857  * @addr: address relative to the base of the RAM region
2858  * @buf: buffer with the data transferred
2859  * @len: length of the data transferred
2860  */
2861 static inline MemTxResult
2862 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr,
2863                            const void *buf, hwaddr len)
2864 {
2865     assert(addr < cache->len && len <= cache->len - addr);
2866     if (likely(cache->ptr)) {
2867         memcpy(cache->ptr + addr, buf, len);
2868         return MEMTX_OK;
2869     } else {
2870         return address_space_write_cached_slow(cache, addr, buf, len);
2871     }
2872 }
2873 
2874 #ifdef NEED_CPU_H
2875 /* enum device_endian to MemOp.  */
2876 static inline MemOp devend_memop(enum device_endian end)
2877 {
2878     QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN &&
2879                       DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN);
2880 
2881 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
2882     /* Swap if non-host endianness or native (target) endianness */
2883     return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP;
2884 #else
2885     const int non_host_endianness =
2886         DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN;
2887 
2888     /* In this case, native (target) endianness needs no swap.  */
2889     return (end == non_host_endianness) ? MO_BSWAP : 0;
2890 #endif
2891 }
2892 #endif
2893 
2894 /*
2895  * Inhibit technologies that require discarding of pages in RAM blocks, e.g.,
2896  * to manage the actual amount of memory consumed by the VM (then, the memory
2897  * provided by RAM blocks might be bigger than the desired memory consumption).
2898  * This *must* be set if:
2899  * - Discarding parts of a RAM blocks does not result in the change being
2900  *   reflected in the VM and the pages getting freed.
2901  * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous
2902  *   discards blindly.
2903  * - Discarding parts of a RAM blocks will result in integrity issues (e.g.,
2904  *   encrypted VMs).
2905  * Technologies that only temporarily pin the current working set of a
2906  * driver are fine, because we don't expect such pages to be discarded
2907  * (esp. based on guest action like balloon inflation).
2908  *
2909  * This is *not* to be used to protect from concurrent discards (esp.,
2910  * postcopy).
2911  *
2912  * Returns 0 if successful. Returns -EBUSY if a technology that relies on
2913  * discards to work reliably is active.
2914  */
2915 int ram_block_discard_disable(bool state);
2916 
2917 /*
2918  * See ram_block_discard_disable(): only disable uncoordinated discards,
2919  * keeping coordinated discards (via the RamDiscardManager) enabled.
2920  */
2921 int ram_block_uncoordinated_discard_disable(bool state);
2922 
2923 /*
2924  * Inhibit technologies that disable discarding of pages in RAM blocks.
2925  *
2926  * Returns 0 if successful. Returns -EBUSY if discards are already set to
2927  * broken.
2928  */
2929 int ram_block_discard_require(bool state);
2930 
2931 /*
2932  * See ram_block_discard_require(): only inhibit technologies that disable
2933  * uncoordinated discarding of pages in RAM blocks, allowing co-existance with
2934  * technologies that only inhibit uncoordinated discards (via the
2935  * RamDiscardManager).
2936  */
2937 int ram_block_coordinated_discard_require(bool state);
2938 
2939 /*
2940  * Test if any discarding of memory in ram blocks is disabled.
2941  */
2942 bool ram_block_discard_is_disabled(void);
2943 
2944 /*
2945  * Test if any discarding of memory in ram blocks is required to work reliably.
2946  */
2947 bool ram_block_discard_is_required(void);
2948 
2949 #endif
2950 
2951 #endif
2952