xref: /openbmc/qemu/include/exec/memory.h (revision e7fa3377)
1 /*
2  * Physical memory management API
3  *
4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
5  *
6  * Authors:
7  *  Avi Kivity <avi@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #ifndef MEMORY_H
15 #define MEMORY_H
16 
17 #ifndef CONFIG_USER_ONLY
18 
19 #include "exec/cpu-common.h"
20 #include "exec/hwaddr.h"
21 #include "exec/memattrs.h"
22 #include "exec/memop.h"
23 #include "exec/ramlist.h"
24 #include "qemu/bswap.h"
25 #include "qemu/queue.h"
26 #include "qemu/int128.h"
27 #include "qemu/notify.h"
28 #include "qom/object.h"
29 #include "qemu/rcu.h"
30 
31 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
32 
33 #define MAX_PHYS_ADDR_SPACE_BITS 62
34 #define MAX_PHYS_ADDR            (((hwaddr)1 << MAX_PHYS_ADDR_SPACE_BITS) - 1)
35 
36 #define TYPE_MEMORY_REGION "memory-region"
37 DECLARE_INSTANCE_CHECKER(MemoryRegion, MEMORY_REGION,
38                          TYPE_MEMORY_REGION)
39 
40 #define TYPE_IOMMU_MEMORY_REGION "iommu-memory-region"
41 typedef struct IOMMUMemoryRegionClass IOMMUMemoryRegionClass;
42 DECLARE_OBJ_CHECKERS(IOMMUMemoryRegion, IOMMUMemoryRegionClass,
43                      IOMMU_MEMORY_REGION, TYPE_IOMMU_MEMORY_REGION)
44 
45 #define TYPE_RAM_DISCARD_MANAGER "qemu:ram-discard-manager"
46 typedef struct RamDiscardManagerClass RamDiscardManagerClass;
47 typedef struct RamDiscardManager RamDiscardManager;
48 DECLARE_OBJ_CHECKERS(RamDiscardManager, RamDiscardManagerClass,
49                      RAM_DISCARD_MANAGER, TYPE_RAM_DISCARD_MANAGER);
50 
51 #ifdef CONFIG_FUZZ
52 void fuzz_dma_read_cb(size_t addr,
53                       size_t len,
54                       MemoryRegion *mr);
55 #else
56 static inline void fuzz_dma_read_cb(size_t addr,
57                                     size_t len,
58                                     MemoryRegion *mr)
59 {
60     /* Do Nothing */
61 }
62 #endif
63 
64 /* Possible bits for global_dirty_log_{start|stop} */
65 
66 /* Dirty tracking enabled because migration is running */
67 #define GLOBAL_DIRTY_MIGRATION  (1U << 0)
68 
69 /* Dirty tracking enabled because measuring dirty rate */
70 #define GLOBAL_DIRTY_DIRTY_RATE (1U << 1)
71 
72 #define GLOBAL_DIRTY_MASK  (0x3)
73 
74 extern unsigned int global_dirty_tracking;
75 
76 typedef struct MemoryRegionOps MemoryRegionOps;
77 
78 struct ReservedRegion {
79     hwaddr low;
80     hwaddr high;
81     unsigned type;
82 };
83 
84 /**
85  * struct MemoryRegionSection: describes a fragment of a #MemoryRegion
86  *
87  * @mr: the region, or %NULL if empty
88  * @fv: the flat view of the address space the region is mapped in
89  * @offset_within_region: the beginning of the section, relative to @mr's start
90  * @size: the size of the section; will not exceed @mr's boundaries
91  * @offset_within_address_space: the address of the first byte of the section
92  *     relative to the region's address space
93  * @readonly: writes to this section are ignored
94  * @nonvolatile: this section is non-volatile
95  */
96 struct MemoryRegionSection {
97     Int128 size;
98     MemoryRegion *mr;
99     FlatView *fv;
100     hwaddr offset_within_region;
101     hwaddr offset_within_address_space;
102     bool readonly;
103     bool nonvolatile;
104 };
105 
106 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
107 
108 /* See address_space_translate: bit 0 is read, bit 1 is write.  */
109 typedef enum {
110     IOMMU_NONE = 0,
111     IOMMU_RO   = 1,
112     IOMMU_WO   = 2,
113     IOMMU_RW   = 3,
114 } IOMMUAccessFlags;
115 
116 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0))
117 
118 struct IOMMUTLBEntry {
119     AddressSpace    *target_as;
120     hwaddr           iova;
121     hwaddr           translated_addr;
122     hwaddr           addr_mask;  /* 0xfff = 4k translation */
123     IOMMUAccessFlags perm;
124 };
125 
126 /*
127  * Bitmap for different IOMMUNotifier capabilities. Each notifier can
128  * register with one or multiple IOMMU Notifier capability bit(s).
129  */
130 typedef enum {
131     IOMMU_NOTIFIER_NONE = 0,
132     /* Notify cache invalidations */
133     IOMMU_NOTIFIER_UNMAP = 0x1,
134     /* Notify entry changes (newly created entries) */
135     IOMMU_NOTIFIER_MAP = 0x2,
136     /* Notify changes on device IOTLB entries */
137     IOMMU_NOTIFIER_DEVIOTLB_UNMAP = 0x04,
138 } IOMMUNotifierFlag;
139 
140 #define IOMMU_NOTIFIER_IOTLB_EVENTS (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
141 #define IOMMU_NOTIFIER_DEVIOTLB_EVENTS IOMMU_NOTIFIER_DEVIOTLB_UNMAP
142 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_IOTLB_EVENTS | \
143                             IOMMU_NOTIFIER_DEVIOTLB_EVENTS)
144 
145 struct IOMMUNotifier;
146 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier,
147                             IOMMUTLBEntry *data);
148 
149 struct IOMMUNotifier {
150     IOMMUNotify notify;
151     IOMMUNotifierFlag notifier_flags;
152     /* Notify for address space range start <= addr <= end */
153     hwaddr start;
154     hwaddr end;
155     int iommu_idx;
156     QLIST_ENTRY(IOMMUNotifier) node;
157 };
158 typedef struct IOMMUNotifier IOMMUNotifier;
159 
160 typedef struct IOMMUTLBEvent {
161     IOMMUNotifierFlag type;
162     IOMMUTLBEntry entry;
163 } IOMMUTLBEvent;
164 
165 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
166 #define RAM_PREALLOC   (1 << 0)
167 
168 /* RAM is mmap-ed with MAP_SHARED */
169 #define RAM_SHARED     (1 << 1)
170 
171 /* Only a portion of RAM (used_length) is actually used, and migrated.
172  * Resizing RAM while migrating can result in the migration being canceled.
173  */
174 #define RAM_RESIZEABLE (1 << 2)
175 
176 /* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically
177  * zero the page and wake waiting processes.
178  * (Set during postcopy)
179  */
180 #define RAM_UF_ZEROPAGE (1 << 3)
181 
182 /* RAM can be migrated */
183 #define RAM_MIGRATABLE (1 << 4)
184 
185 /* RAM is a persistent kind memory */
186 #define RAM_PMEM (1 << 5)
187 
188 
189 /*
190  * UFFDIO_WRITEPROTECT is used on this RAMBlock to
191  * support 'write-tracking' migration type.
192  * Implies ram_state->ram_wt_enabled.
193  */
194 #define RAM_UF_WRITEPROTECT (1 << 6)
195 
196 /*
197  * RAM is mmap-ed with MAP_NORESERVE. When set, reserving swap space (or huge
198  * pages if applicable) is skipped: will bail out if not supported. When not
199  * set, the OS will do the reservation, if supported for the memory type.
200  */
201 #define RAM_NORESERVE (1 << 7)
202 
203 /* RAM that isn't accessible through normal means. */
204 #define RAM_PROTECTED (1 << 8)
205 
206 static inline void iommu_notifier_init(IOMMUNotifier *n, IOMMUNotify fn,
207                                        IOMMUNotifierFlag flags,
208                                        hwaddr start, hwaddr end,
209                                        int iommu_idx)
210 {
211     n->notify = fn;
212     n->notifier_flags = flags;
213     n->start = start;
214     n->end = end;
215     n->iommu_idx = iommu_idx;
216 }
217 
218 /*
219  * Memory region callbacks
220  */
221 struct MemoryRegionOps {
222     /* Read from the memory region. @addr is relative to @mr; @size is
223      * in bytes. */
224     uint64_t (*read)(void *opaque,
225                      hwaddr addr,
226                      unsigned size);
227     /* Write to the memory region. @addr is relative to @mr; @size is
228      * in bytes. */
229     void (*write)(void *opaque,
230                   hwaddr addr,
231                   uint64_t data,
232                   unsigned size);
233 
234     MemTxResult (*read_with_attrs)(void *opaque,
235                                    hwaddr addr,
236                                    uint64_t *data,
237                                    unsigned size,
238                                    MemTxAttrs attrs);
239     MemTxResult (*write_with_attrs)(void *opaque,
240                                     hwaddr addr,
241                                     uint64_t data,
242                                     unsigned size,
243                                     MemTxAttrs attrs);
244 
245     enum device_endian endianness;
246     /* Guest-visible constraints: */
247     struct {
248         /* If nonzero, specify bounds on access sizes beyond which a machine
249          * check is thrown.
250          */
251         unsigned min_access_size;
252         unsigned max_access_size;
253         /* If true, unaligned accesses are supported.  Otherwise unaligned
254          * accesses throw machine checks.
255          */
256          bool unaligned;
257         /*
258          * If present, and returns #false, the transaction is not accepted
259          * by the device (and results in machine dependent behaviour such
260          * as a machine check exception).
261          */
262         bool (*accepts)(void *opaque, hwaddr addr,
263                         unsigned size, bool is_write,
264                         MemTxAttrs attrs);
265     } valid;
266     /* Internal implementation constraints: */
267     struct {
268         /* If nonzero, specifies the minimum size implemented.  Smaller sizes
269          * will be rounded upwards and a partial result will be returned.
270          */
271         unsigned min_access_size;
272         /* If nonzero, specifies the maximum size implemented.  Larger sizes
273          * will be done as a series of accesses with smaller sizes.
274          */
275         unsigned max_access_size;
276         /* If true, unaligned accesses are supported.  Otherwise all accesses
277          * are converted to (possibly multiple) naturally aligned accesses.
278          */
279         bool unaligned;
280     } impl;
281 };
282 
283 typedef struct MemoryRegionClass {
284     /* private */
285     ObjectClass parent_class;
286 } MemoryRegionClass;
287 
288 
289 enum IOMMUMemoryRegionAttr {
290     IOMMU_ATTR_SPAPR_TCE_FD
291 };
292 
293 /*
294  * IOMMUMemoryRegionClass:
295  *
296  * All IOMMU implementations need to subclass TYPE_IOMMU_MEMORY_REGION
297  * and provide an implementation of at least the @translate method here
298  * to handle requests to the memory region. Other methods are optional.
299  *
300  * The IOMMU implementation must use the IOMMU notifier infrastructure
301  * to report whenever mappings are changed, by calling
302  * memory_region_notify_iommu() (or, if necessary, by calling
303  * memory_region_notify_iommu_one() for each registered notifier).
304  *
305  * Conceptually an IOMMU provides a mapping from input address
306  * to an output TLB entry. If the IOMMU is aware of memory transaction
307  * attributes and the output TLB entry depends on the transaction
308  * attributes, we represent this using IOMMU indexes. Each index
309  * selects a particular translation table that the IOMMU has:
310  *
311  *   @attrs_to_index returns the IOMMU index for a set of transaction attributes
312  *
313  *   @translate takes an input address and an IOMMU index
314  *
315  * and the mapping returned can only depend on the input address and the
316  * IOMMU index.
317  *
318  * Most IOMMUs don't care about the transaction attributes and support
319  * only a single IOMMU index. A more complex IOMMU might have one index
320  * for secure transactions and one for non-secure transactions.
321  */
322 struct IOMMUMemoryRegionClass {
323     /* private: */
324     MemoryRegionClass parent_class;
325 
326     /* public: */
327     /**
328      * @translate:
329      *
330      * Return a TLB entry that contains a given address.
331      *
332      * The IOMMUAccessFlags indicated via @flag are optional and may
333      * be specified as IOMMU_NONE to indicate that the caller needs
334      * the full translation information for both reads and writes. If
335      * the access flags are specified then the IOMMU implementation
336      * may use this as an optimization, to stop doing a page table
337      * walk as soon as it knows that the requested permissions are not
338      * allowed. If IOMMU_NONE is passed then the IOMMU must do the
339      * full page table walk and report the permissions in the returned
340      * IOMMUTLBEntry. (Note that this implies that an IOMMU may not
341      * return different mappings for reads and writes.)
342      *
343      * The returned information remains valid while the caller is
344      * holding the big QEMU lock or is inside an RCU critical section;
345      * if the caller wishes to cache the mapping beyond that it must
346      * register an IOMMU notifier so it can invalidate its cached
347      * information when the IOMMU mapping changes.
348      *
349      * @iommu: the IOMMUMemoryRegion
350      *
351      * @hwaddr: address to be translated within the memory region
352      *
353      * @flag: requested access permission
354      *
355      * @iommu_idx: IOMMU index for the translation
356      */
357     IOMMUTLBEntry (*translate)(IOMMUMemoryRegion *iommu, hwaddr addr,
358                                IOMMUAccessFlags flag, int iommu_idx);
359     /**
360      * @get_min_page_size:
361      *
362      * Returns minimum supported page size in bytes.
363      *
364      * If this method is not provided then the minimum is assumed to
365      * be TARGET_PAGE_SIZE.
366      *
367      * @iommu: the IOMMUMemoryRegion
368      */
369     uint64_t (*get_min_page_size)(IOMMUMemoryRegion *iommu);
370     /**
371      * @notify_flag_changed:
372      *
373      * Called when IOMMU Notifier flag changes (ie when the set of
374      * events which IOMMU users are requesting notification for changes).
375      * Optional method -- need not be provided if the IOMMU does not
376      * need to know exactly which events must be notified.
377      *
378      * @iommu: the IOMMUMemoryRegion
379      *
380      * @old_flags: events which previously needed to be notified
381      *
382      * @new_flags: events which now need to be notified
383      *
384      * Returns 0 on success, or a negative errno; in particular
385      * returns -EINVAL if the new flag bitmap is not supported by the
386      * IOMMU memory region. In case of failure, the error object
387      * must be created
388      */
389     int (*notify_flag_changed)(IOMMUMemoryRegion *iommu,
390                                IOMMUNotifierFlag old_flags,
391                                IOMMUNotifierFlag new_flags,
392                                Error **errp);
393     /**
394      * @replay:
395      *
396      * Called to handle memory_region_iommu_replay().
397      *
398      * The default implementation of memory_region_iommu_replay() is to
399      * call the IOMMU translate method for every page in the address space
400      * with flag == IOMMU_NONE and then call the notifier if translate
401      * returns a valid mapping. If this method is implemented then it
402      * overrides the default behaviour, and must provide the full semantics
403      * of memory_region_iommu_replay(), by calling @notifier for every
404      * translation present in the IOMMU.
405      *
406      * Optional method -- an IOMMU only needs to provide this method
407      * if the default is inefficient or produces undesirable side effects.
408      *
409      * Note: this is not related to record-and-replay functionality.
410      */
411     void (*replay)(IOMMUMemoryRegion *iommu, IOMMUNotifier *notifier);
412 
413     /**
414      * @get_attr:
415      *
416      * Get IOMMU misc attributes. This is an optional method that
417      * can be used to allow users of the IOMMU to get implementation-specific
418      * information. The IOMMU implements this method to handle calls
419      * by IOMMU users to memory_region_iommu_get_attr() by filling in
420      * the arbitrary data pointer for any IOMMUMemoryRegionAttr values that
421      * the IOMMU supports. If the method is unimplemented then
422      * memory_region_iommu_get_attr() will always return -EINVAL.
423      *
424      * @iommu: the IOMMUMemoryRegion
425      *
426      * @attr: attribute being queried
427      *
428      * @data: memory to fill in with the attribute data
429      *
430      * Returns 0 on success, or a negative errno; in particular
431      * returns -EINVAL for unrecognized or unimplemented attribute types.
432      */
433     int (*get_attr)(IOMMUMemoryRegion *iommu, enum IOMMUMemoryRegionAttr attr,
434                     void *data);
435 
436     /**
437      * @attrs_to_index:
438      *
439      * Return the IOMMU index to use for a given set of transaction attributes.
440      *
441      * Optional method: if an IOMMU only supports a single IOMMU index then
442      * the default implementation of memory_region_iommu_attrs_to_index()
443      * will return 0.
444      *
445      * The indexes supported by an IOMMU must be contiguous, starting at 0.
446      *
447      * @iommu: the IOMMUMemoryRegion
448      * @attrs: memory transaction attributes
449      */
450     int (*attrs_to_index)(IOMMUMemoryRegion *iommu, MemTxAttrs attrs);
451 
452     /**
453      * @num_indexes:
454      *
455      * Return the number of IOMMU indexes this IOMMU supports.
456      *
457      * Optional method: if this method is not provided, then
458      * memory_region_iommu_num_indexes() will return 1, indicating that
459      * only a single IOMMU index is supported.
460      *
461      * @iommu: the IOMMUMemoryRegion
462      */
463     int (*num_indexes)(IOMMUMemoryRegion *iommu);
464 
465     /**
466      * @iommu_set_page_size_mask:
467      *
468      * Restrict the page size mask that can be supported with a given IOMMU
469      * memory region. Used for example to propagate host physical IOMMU page
470      * size mask limitations to the virtual IOMMU.
471      *
472      * Optional method: if this method is not provided, then the default global
473      * page mask is used.
474      *
475      * @iommu: the IOMMUMemoryRegion
476      *
477      * @page_size_mask: a bitmask of supported page sizes. At least one bit,
478      * representing the smallest page size, must be set. Additional set bits
479      * represent supported block sizes. For example a host physical IOMMU that
480      * uses page tables with a page size of 4kB, and supports 2MB and 4GB
481      * blocks, will set mask 0x40201000. A granule of 4kB with indiscriminate
482      * block sizes is specified with mask 0xfffffffffffff000.
483      *
484      * Returns 0 on success, or a negative error. In case of failure, the error
485      * object must be created.
486      */
487      int (*iommu_set_page_size_mask)(IOMMUMemoryRegion *iommu,
488                                      uint64_t page_size_mask,
489                                      Error **errp);
490 };
491 
492 typedef struct RamDiscardListener RamDiscardListener;
493 typedef int (*NotifyRamPopulate)(RamDiscardListener *rdl,
494                                  MemoryRegionSection *section);
495 typedef void (*NotifyRamDiscard)(RamDiscardListener *rdl,
496                                  MemoryRegionSection *section);
497 
498 struct RamDiscardListener {
499     /*
500      * @notify_populate:
501      *
502      * Notification that previously discarded memory is about to get populated.
503      * Listeners are able to object. If any listener objects, already
504      * successfully notified listeners are notified about a discard again.
505      *
506      * @rdl: the #RamDiscardListener getting notified
507      * @section: the #MemoryRegionSection to get populated. The section
508      *           is aligned within the memory region to the minimum granularity
509      *           unless it would exceed the registered section.
510      *
511      * Returns 0 on success. If the notification is rejected by the listener,
512      * an error is returned.
513      */
514     NotifyRamPopulate notify_populate;
515 
516     /*
517      * @notify_discard:
518      *
519      * Notification that previously populated memory was discarded successfully
520      * and listeners should drop all references to such memory and prevent
521      * new population (e.g., unmap).
522      *
523      * @rdl: the #RamDiscardListener getting notified
524      * @section: the #MemoryRegionSection to get populated. The section
525      *           is aligned within the memory region to the minimum granularity
526      *           unless it would exceed the registered section.
527      */
528     NotifyRamDiscard notify_discard;
529 
530     /*
531      * @double_discard_supported:
532      *
533      * The listener suppors getting @notify_discard notifications that span
534      * already discarded parts.
535      */
536     bool double_discard_supported;
537 
538     MemoryRegionSection *section;
539     QLIST_ENTRY(RamDiscardListener) next;
540 };
541 
542 static inline void ram_discard_listener_init(RamDiscardListener *rdl,
543                                              NotifyRamPopulate populate_fn,
544                                              NotifyRamDiscard discard_fn,
545                                              bool double_discard_supported)
546 {
547     rdl->notify_populate = populate_fn;
548     rdl->notify_discard = discard_fn;
549     rdl->double_discard_supported = double_discard_supported;
550 }
551 
552 typedef int (*ReplayRamPopulate)(MemoryRegionSection *section, void *opaque);
553 typedef void (*ReplayRamDiscard)(MemoryRegionSection *section, void *opaque);
554 
555 /*
556  * RamDiscardManagerClass:
557  *
558  * A #RamDiscardManager coordinates which parts of specific RAM #MemoryRegion
559  * regions are currently populated to be used/accessed by the VM, notifying
560  * after parts were discarded (freeing up memory) and before parts will be
561  * populated (consuming memory), to be used/acessed by the VM.
562  *
563  * A #RamDiscardManager can only be set for a RAM #MemoryRegion while the
564  * #MemoryRegion isn't mapped yet; it cannot change while the #MemoryRegion is
565  * mapped.
566  *
567  * The #RamDiscardManager is intended to be used by technologies that are
568  * incompatible with discarding of RAM (e.g., VFIO, which may pin all
569  * memory inside a #MemoryRegion), and require proper coordination to only
570  * map the currently populated parts, to hinder parts that are expected to
571  * remain discarded from silently getting populated and consuming memory.
572  * Technologies that support discarding of RAM don't have to bother and can
573  * simply map the whole #MemoryRegion.
574  *
575  * An example #RamDiscardManager is virtio-mem, which logically (un)plugs
576  * memory within an assigned RAM #MemoryRegion, coordinated with the VM.
577  * Logically unplugging memory consists of discarding RAM. The VM agreed to not
578  * access unplugged (discarded) memory - especially via DMA. virtio-mem will
579  * properly coordinate with listeners before memory is plugged (populated),
580  * and after memory is unplugged (discarded).
581  *
582  * Listeners are called in multiples of the minimum granularity (unless it
583  * would exceed the registered range) and changes are aligned to the minimum
584  * granularity within the #MemoryRegion. Listeners have to prepare for memory
585  * becomming discarded in a different granularity than it was populated and the
586  * other way around.
587  */
588 struct RamDiscardManagerClass {
589     /* private */
590     InterfaceClass parent_class;
591 
592     /* public */
593 
594     /**
595      * @get_min_granularity:
596      *
597      * Get the minimum granularity in which listeners will get notified
598      * about changes within the #MemoryRegion via the #RamDiscardManager.
599      *
600      * @rdm: the #RamDiscardManager
601      * @mr: the #MemoryRegion
602      *
603      * Returns the minimum granularity.
604      */
605     uint64_t (*get_min_granularity)(const RamDiscardManager *rdm,
606                                     const MemoryRegion *mr);
607 
608     /**
609      * @is_populated:
610      *
611      * Check whether the given #MemoryRegionSection is completely populated
612      * (i.e., no parts are currently discarded) via the #RamDiscardManager.
613      * There are no alignment requirements.
614      *
615      * @rdm: the #RamDiscardManager
616      * @section: the #MemoryRegionSection
617      *
618      * Returns whether the given range is completely populated.
619      */
620     bool (*is_populated)(const RamDiscardManager *rdm,
621                          const MemoryRegionSection *section);
622 
623     /**
624      * @replay_populated:
625      *
626      * Call the #ReplayRamPopulate callback for all populated parts within the
627      * #MemoryRegionSection via the #RamDiscardManager.
628      *
629      * In case any call fails, no further calls are made.
630      *
631      * @rdm: the #RamDiscardManager
632      * @section: the #MemoryRegionSection
633      * @replay_fn: the #ReplayRamPopulate callback
634      * @opaque: pointer to forward to the callback
635      *
636      * Returns 0 on success, or a negative error if any notification failed.
637      */
638     int (*replay_populated)(const RamDiscardManager *rdm,
639                             MemoryRegionSection *section,
640                             ReplayRamPopulate replay_fn, void *opaque);
641 
642     /**
643      * @replay_discarded:
644      *
645      * Call the #ReplayRamDiscard callback for all discarded parts within the
646      * #MemoryRegionSection via the #RamDiscardManager.
647      *
648      * @rdm: the #RamDiscardManager
649      * @section: the #MemoryRegionSection
650      * @replay_fn: the #ReplayRamDiscard callback
651      * @opaque: pointer to forward to the callback
652      */
653     void (*replay_discarded)(const RamDiscardManager *rdm,
654                              MemoryRegionSection *section,
655                              ReplayRamDiscard replay_fn, void *opaque);
656 
657     /**
658      * @register_listener:
659      *
660      * Register a #RamDiscardListener for the given #MemoryRegionSection and
661      * immediately notify the #RamDiscardListener about all populated parts
662      * within the #MemoryRegionSection via the #RamDiscardManager.
663      *
664      * In case any notification fails, no further notifications are triggered
665      * and an error is logged.
666      *
667      * @rdm: the #RamDiscardManager
668      * @rdl: the #RamDiscardListener
669      * @section: the #MemoryRegionSection
670      */
671     void (*register_listener)(RamDiscardManager *rdm,
672                               RamDiscardListener *rdl,
673                               MemoryRegionSection *section);
674 
675     /**
676      * @unregister_listener:
677      *
678      * Unregister a previously registered #RamDiscardListener via the
679      * #RamDiscardManager after notifying the #RamDiscardListener about all
680      * populated parts becoming unpopulated within the registered
681      * #MemoryRegionSection.
682      *
683      * @rdm: the #RamDiscardManager
684      * @rdl: the #RamDiscardListener
685      */
686     void (*unregister_listener)(RamDiscardManager *rdm,
687                                 RamDiscardListener *rdl);
688 };
689 
690 uint64_t ram_discard_manager_get_min_granularity(const RamDiscardManager *rdm,
691                                                  const MemoryRegion *mr);
692 
693 bool ram_discard_manager_is_populated(const RamDiscardManager *rdm,
694                                       const MemoryRegionSection *section);
695 
696 int ram_discard_manager_replay_populated(const RamDiscardManager *rdm,
697                                          MemoryRegionSection *section,
698                                          ReplayRamPopulate replay_fn,
699                                          void *opaque);
700 
701 void ram_discard_manager_replay_discarded(const RamDiscardManager *rdm,
702                                           MemoryRegionSection *section,
703                                           ReplayRamDiscard replay_fn,
704                                           void *opaque);
705 
706 void ram_discard_manager_register_listener(RamDiscardManager *rdm,
707                                            RamDiscardListener *rdl,
708                                            MemoryRegionSection *section);
709 
710 void ram_discard_manager_unregister_listener(RamDiscardManager *rdm,
711                                              RamDiscardListener *rdl);
712 
713 typedef struct CoalescedMemoryRange CoalescedMemoryRange;
714 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd;
715 
716 /** MemoryRegion:
717  *
718  * A struct representing a memory region.
719  */
720 struct MemoryRegion {
721     Object parent_obj;
722 
723     /* private: */
724 
725     /* The following fields should fit in a cache line */
726     bool romd_mode;
727     bool ram;
728     bool subpage;
729     bool readonly; /* For RAM regions */
730     bool nonvolatile;
731     bool rom_device;
732     bool flush_coalesced_mmio;
733     uint8_t dirty_log_mask;
734     bool is_iommu;
735     RAMBlock *ram_block;
736     Object *owner;
737 
738     const MemoryRegionOps *ops;
739     void *opaque;
740     MemoryRegion *container;
741     Int128 size;
742     hwaddr addr;
743     void (*destructor)(MemoryRegion *mr);
744     uint64_t align;
745     bool terminates;
746     bool ram_device;
747     bool enabled;
748     bool warning_printed; /* For reservations */
749     uint8_t vga_logging_count;
750     MemoryRegion *alias;
751     hwaddr alias_offset;
752     int32_t priority;
753     QTAILQ_HEAD(, MemoryRegion) subregions;
754     QTAILQ_ENTRY(MemoryRegion) subregions_link;
755     QTAILQ_HEAD(, CoalescedMemoryRange) coalesced;
756     const char *name;
757     unsigned ioeventfd_nb;
758     MemoryRegionIoeventfd *ioeventfds;
759     RamDiscardManager *rdm; /* Only for RAM */
760 };
761 
762 struct IOMMUMemoryRegion {
763     MemoryRegion parent_obj;
764 
765     QLIST_HEAD(, IOMMUNotifier) iommu_notify;
766     IOMMUNotifierFlag iommu_notify_flags;
767 };
768 
769 #define IOMMU_NOTIFIER_FOREACH(n, mr) \
770     QLIST_FOREACH((n), &(mr)->iommu_notify, node)
771 
772 /**
773  * struct MemoryListener: callbacks structure for updates to the physical memory map
774  *
775  * Allows a component to adjust to changes in the guest-visible memory map.
776  * Use with memory_listener_register() and memory_listener_unregister().
777  */
778 struct MemoryListener {
779     /**
780      * @begin:
781      *
782      * Called at the beginning of an address space update transaction.
783      * Followed by calls to #MemoryListener.region_add(),
784      * #MemoryListener.region_del(), #MemoryListener.region_nop(),
785      * #MemoryListener.log_start() and #MemoryListener.log_stop() in
786      * increasing address order.
787      *
788      * @listener: The #MemoryListener.
789      */
790     void (*begin)(MemoryListener *listener);
791 
792     /**
793      * @commit:
794      *
795      * Called at the end of an address space update transaction,
796      * after the last call to #MemoryListener.region_add(),
797      * #MemoryListener.region_del() or #MemoryListener.region_nop(),
798      * #MemoryListener.log_start() and #MemoryListener.log_stop().
799      *
800      * @listener: The #MemoryListener.
801      */
802     void (*commit)(MemoryListener *listener);
803 
804     /**
805      * @region_add:
806      *
807      * Called during an address space update transaction,
808      * for a section of the address space that is new in this address space
809      * space since the last transaction.
810      *
811      * @listener: The #MemoryListener.
812      * @section: The new #MemoryRegionSection.
813      */
814     void (*region_add)(MemoryListener *listener, MemoryRegionSection *section);
815 
816     /**
817      * @region_del:
818      *
819      * Called during an address space update transaction,
820      * for a section of the address space that has disappeared in the address
821      * space since the last transaction.
822      *
823      * @listener: The #MemoryListener.
824      * @section: The old #MemoryRegionSection.
825      */
826     void (*region_del)(MemoryListener *listener, MemoryRegionSection *section);
827 
828     /**
829      * @region_nop:
830      *
831      * Called during an address space update transaction,
832      * for a section of the address space that is in the same place in the address
833      * space as in the last transaction.
834      *
835      * @listener: The #MemoryListener.
836      * @section: The #MemoryRegionSection.
837      */
838     void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section);
839 
840     /**
841      * @log_start:
842      *
843      * Called during an address space update transaction, after
844      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
845      * #MemoryListener.region_nop(), if dirty memory logging clients have
846      * become active since the last transaction.
847      *
848      * @listener: The #MemoryListener.
849      * @section: The #MemoryRegionSection.
850      * @old: A bitmap of dirty memory logging clients that were active in
851      * the previous transaction.
852      * @new: A bitmap of dirty memory logging clients that are active in
853      * the current transaction.
854      */
855     void (*log_start)(MemoryListener *listener, MemoryRegionSection *section,
856                       int old, int new);
857 
858     /**
859      * @log_stop:
860      *
861      * Called during an address space update transaction, after
862      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
863      * #MemoryListener.region_nop() and possibly after
864      * #MemoryListener.log_start(), if dirty memory logging clients have
865      * become inactive since the last transaction.
866      *
867      * @listener: The #MemoryListener.
868      * @section: The #MemoryRegionSection.
869      * @old: A bitmap of dirty memory logging clients that were active in
870      * the previous transaction.
871      * @new: A bitmap of dirty memory logging clients that are active in
872      * the current transaction.
873      */
874     void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section,
875                      int old, int new);
876 
877     /**
878      * @log_sync:
879      *
880      * Called by memory_region_snapshot_and_clear_dirty() and
881      * memory_global_dirty_log_sync(), before accessing QEMU's "official"
882      * copy of the dirty memory bitmap for a #MemoryRegionSection.
883      *
884      * @listener: The #MemoryListener.
885      * @section: The #MemoryRegionSection.
886      */
887     void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section);
888 
889     /**
890      * @log_sync_global:
891      *
892      * This is the global version of @log_sync when the listener does
893      * not have a way to synchronize the log with finer granularity.
894      * When the listener registers with @log_sync_global defined, then
895      * its @log_sync must be NULL.  Vice versa.
896      *
897      * @listener: The #MemoryListener.
898      */
899     void (*log_sync_global)(MemoryListener *listener);
900 
901     /**
902      * @log_clear:
903      *
904      * Called before reading the dirty memory bitmap for a
905      * #MemoryRegionSection.
906      *
907      * @listener: The #MemoryListener.
908      * @section: The #MemoryRegionSection.
909      */
910     void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section);
911 
912     /**
913      * @log_global_start:
914      *
915      * Called by memory_global_dirty_log_start(), which
916      * enables the %DIRTY_LOG_MIGRATION client on all memory regions in
917      * the address space.  #MemoryListener.log_global_start() is also
918      * called when a #MemoryListener is added, if global dirty logging is
919      * active at that time.
920      *
921      * @listener: The #MemoryListener.
922      */
923     void (*log_global_start)(MemoryListener *listener);
924 
925     /**
926      * @log_global_stop:
927      *
928      * Called by memory_global_dirty_log_stop(), which
929      * disables the %DIRTY_LOG_MIGRATION client on all memory regions in
930      * the address space.
931      *
932      * @listener: The #MemoryListener.
933      */
934     void (*log_global_stop)(MemoryListener *listener);
935 
936     /**
937      * @log_global_after_sync:
938      *
939      * Called after reading the dirty memory bitmap
940      * for any #MemoryRegionSection.
941      *
942      * @listener: The #MemoryListener.
943      */
944     void (*log_global_after_sync)(MemoryListener *listener);
945 
946     /**
947      * @eventfd_add:
948      *
949      * Called during an address space update transaction,
950      * for a section of the address space that has had a new ioeventfd
951      * registration since the last transaction.
952      *
953      * @listener: The #MemoryListener.
954      * @section: The new #MemoryRegionSection.
955      * @match_data: The @match_data parameter for the new ioeventfd.
956      * @data: The @data parameter for the new ioeventfd.
957      * @e: The #EventNotifier parameter for the new ioeventfd.
958      */
959     void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section,
960                         bool match_data, uint64_t data, EventNotifier *e);
961 
962     /**
963      * @eventfd_del:
964      *
965      * Called during an address space update transaction,
966      * for a section of the address space that has dropped an ioeventfd
967      * registration since the last transaction.
968      *
969      * @listener: The #MemoryListener.
970      * @section: The new #MemoryRegionSection.
971      * @match_data: The @match_data parameter for the dropped ioeventfd.
972      * @data: The @data parameter for the dropped ioeventfd.
973      * @e: The #EventNotifier parameter for the dropped ioeventfd.
974      */
975     void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section,
976                         bool match_data, uint64_t data, EventNotifier *e);
977 
978     /**
979      * @coalesced_io_add:
980      *
981      * Called during an address space update transaction,
982      * for a section of the address space that has had a new coalesced
983      * MMIO range registration since the last transaction.
984      *
985      * @listener: The #MemoryListener.
986      * @section: The new #MemoryRegionSection.
987      * @addr: The starting address for the coalesced MMIO range.
988      * @len: The length of the coalesced MMIO range.
989      */
990     void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section,
991                                hwaddr addr, hwaddr len);
992 
993     /**
994      * @coalesced_io_del:
995      *
996      * Called during an address space update transaction,
997      * for a section of the address space that has dropped a coalesced
998      * MMIO range since the last transaction.
999      *
1000      * @listener: The #MemoryListener.
1001      * @section: The new #MemoryRegionSection.
1002      * @addr: The starting address for the coalesced MMIO range.
1003      * @len: The length of the coalesced MMIO range.
1004      */
1005     void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section,
1006                                hwaddr addr, hwaddr len);
1007     /**
1008      * @priority:
1009      *
1010      * Govern the order in which memory listeners are invoked. Lower priorities
1011      * are invoked earlier for "add" or "start" callbacks, and later for "delete"
1012      * or "stop" callbacks.
1013      */
1014     unsigned priority;
1015 
1016     /**
1017      * @name:
1018      *
1019      * Name of the listener.  It can be used in contexts where we'd like to
1020      * identify one memory listener with the rest.
1021      */
1022     const char *name;
1023 
1024     /* private: */
1025     AddressSpace *address_space;
1026     QTAILQ_ENTRY(MemoryListener) link;
1027     QTAILQ_ENTRY(MemoryListener) link_as;
1028 };
1029 
1030 /**
1031  * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects
1032  */
1033 struct AddressSpace {
1034     /* private: */
1035     struct rcu_head rcu;
1036     char *name;
1037     MemoryRegion *root;
1038 
1039     /* Accessed via RCU.  */
1040     struct FlatView *current_map;
1041 
1042     int ioeventfd_nb;
1043     struct MemoryRegionIoeventfd *ioeventfds;
1044     QTAILQ_HEAD(, MemoryListener) listeners;
1045     QTAILQ_ENTRY(AddressSpace) address_spaces_link;
1046 };
1047 
1048 typedef struct AddressSpaceDispatch AddressSpaceDispatch;
1049 typedef struct FlatRange FlatRange;
1050 
1051 /* Flattened global view of current active memory hierarchy.  Kept in sorted
1052  * order.
1053  */
1054 struct FlatView {
1055     struct rcu_head rcu;
1056     unsigned ref;
1057     FlatRange *ranges;
1058     unsigned nr;
1059     unsigned nr_allocated;
1060     struct AddressSpaceDispatch *dispatch;
1061     MemoryRegion *root;
1062 };
1063 
1064 static inline FlatView *address_space_to_flatview(AddressSpace *as)
1065 {
1066     return qatomic_rcu_read(&as->current_map);
1067 }
1068 
1069 /**
1070  * typedef flatview_cb: callback for flatview_for_each_range()
1071  *
1072  * @start: start address of the range within the FlatView
1073  * @len: length of the range in bytes
1074  * @mr: MemoryRegion covering this range
1075  * @offset_in_region: offset of the first byte of the range within @mr
1076  * @opaque: data pointer passed to flatview_for_each_range()
1077  *
1078  * Returns: true to stop the iteration, false to keep going.
1079  */
1080 typedef bool (*flatview_cb)(Int128 start,
1081                             Int128 len,
1082                             const MemoryRegion *mr,
1083                             hwaddr offset_in_region,
1084                             void *opaque);
1085 
1086 /**
1087  * flatview_for_each_range: Iterate through a FlatView
1088  * @fv: the FlatView to iterate through
1089  * @cb: function to call for each range
1090  * @opaque: opaque data pointer to pass to @cb
1091  *
1092  * A FlatView is made up of a list of non-overlapping ranges, each of
1093  * which is a slice of a MemoryRegion. This function iterates through
1094  * each range in @fv, calling @cb. The callback function can terminate
1095  * iteration early by returning 'true'.
1096  */
1097 void flatview_for_each_range(FlatView *fv, flatview_cb cb, void *opaque);
1098 
1099 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a,
1100                                           MemoryRegionSection *b)
1101 {
1102     return a->mr == b->mr &&
1103            a->fv == b->fv &&
1104            a->offset_within_region == b->offset_within_region &&
1105            a->offset_within_address_space == b->offset_within_address_space &&
1106            int128_eq(a->size, b->size) &&
1107            a->readonly == b->readonly &&
1108            a->nonvolatile == b->nonvolatile;
1109 }
1110 
1111 /**
1112  * memory_region_section_new_copy: Copy a memory region section
1113  *
1114  * Allocate memory for a new copy, copy the memory region section, and
1115  * properly take a reference on all relevant members.
1116  *
1117  * @s: the #MemoryRegionSection to copy
1118  */
1119 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s);
1120 
1121 /**
1122  * memory_region_section_new_copy: Free a copied memory region section
1123  *
1124  * Free a copy of a memory section created via memory_region_section_new_copy().
1125  * properly dropping references on all relevant members.
1126  *
1127  * @s: the #MemoryRegionSection to copy
1128  */
1129 void memory_region_section_free_copy(MemoryRegionSection *s);
1130 
1131 /**
1132  * memory_region_init: Initialize a memory region
1133  *
1134  * The region typically acts as a container for other memory regions.  Use
1135  * memory_region_add_subregion() to add subregions.
1136  *
1137  * @mr: the #MemoryRegion to be initialized
1138  * @owner: the object that tracks the region's reference count
1139  * @name: used for debugging; not visible to the user or ABI
1140  * @size: size of the region; any subregions beyond this size will be clipped
1141  */
1142 void memory_region_init(MemoryRegion *mr,
1143                         Object *owner,
1144                         const char *name,
1145                         uint64_t size);
1146 
1147 /**
1148  * memory_region_ref: Add 1 to a memory region's reference count
1149  *
1150  * Whenever memory regions are accessed outside the BQL, they need to be
1151  * preserved against hot-unplug.  MemoryRegions actually do not have their
1152  * own reference count; they piggyback on a QOM object, their "owner".
1153  * This function adds a reference to the owner.
1154  *
1155  * All MemoryRegions must have an owner if they can disappear, even if the
1156  * device they belong to operates exclusively under the BQL.  This is because
1157  * the region could be returned at any time by memory_region_find, and this
1158  * is usually under guest control.
1159  *
1160  * @mr: the #MemoryRegion
1161  */
1162 void memory_region_ref(MemoryRegion *mr);
1163 
1164 /**
1165  * memory_region_unref: Remove 1 to a memory region's reference count
1166  *
1167  * Whenever memory regions are accessed outside the BQL, they need to be
1168  * preserved against hot-unplug.  MemoryRegions actually do not have their
1169  * own reference count; they piggyback on a QOM object, their "owner".
1170  * This function removes a reference to the owner and possibly destroys it.
1171  *
1172  * @mr: the #MemoryRegion
1173  */
1174 void memory_region_unref(MemoryRegion *mr);
1175 
1176 /**
1177  * memory_region_init_io: Initialize an I/O memory region.
1178  *
1179  * Accesses into the region will cause the callbacks in @ops to be called.
1180  * if @size is nonzero, subregions will be clipped to @size.
1181  *
1182  * @mr: the #MemoryRegion to be initialized.
1183  * @owner: the object that tracks the region's reference count
1184  * @ops: a structure containing read and write callbacks to be used when
1185  *       I/O is performed on the region.
1186  * @opaque: passed to the read and write callbacks of the @ops structure.
1187  * @name: used for debugging; not visible to the user or ABI
1188  * @size: size of the region.
1189  */
1190 void memory_region_init_io(MemoryRegion *mr,
1191                            Object *owner,
1192                            const MemoryRegionOps *ops,
1193                            void *opaque,
1194                            const char *name,
1195                            uint64_t size);
1196 
1197 /**
1198  * memory_region_init_ram_nomigrate:  Initialize RAM memory region.  Accesses
1199  *                                    into the region will modify memory
1200  *                                    directly.
1201  *
1202  * @mr: the #MemoryRegion to be initialized.
1203  * @owner: the object that tracks the region's reference count
1204  * @name: Region name, becomes part of RAMBlock name used in migration stream
1205  *        must be unique within any device
1206  * @size: size of the region.
1207  * @errp: pointer to Error*, to store an error if it happens.
1208  *
1209  * Note that this function does not do anything to cause the data in the
1210  * RAM memory region to be migrated; that is the responsibility of the caller.
1211  */
1212 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1213                                       Object *owner,
1214                                       const char *name,
1215                                       uint64_t size,
1216                                       Error **errp);
1217 
1218 /**
1219  * memory_region_init_ram_flags_nomigrate:  Initialize RAM memory region.
1220  *                                          Accesses into the region will
1221  *                                          modify memory directly.
1222  *
1223  * @mr: the #MemoryRegion to be initialized.
1224  * @owner: the object that tracks the region's reference count
1225  * @name: Region name, becomes part of RAMBlock name used in migration stream
1226  *        must be unique within any device
1227  * @size: size of the region.
1228  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_NORESERVE.
1229  * @errp: pointer to Error*, to store an error if it happens.
1230  *
1231  * Note that this function does not do anything to cause the data in the
1232  * RAM memory region to be migrated; that is the responsibility of the caller.
1233  */
1234 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1235                                             Object *owner,
1236                                             const char *name,
1237                                             uint64_t size,
1238                                             uint32_t ram_flags,
1239                                             Error **errp);
1240 
1241 /**
1242  * memory_region_init_resizeable_ram:  Initialize memory region with resizeable
1243  *                                     RAM.  Accesses into the region will
1244  *                                     modify memory directly.  Only an initial
1245  *                                     portion of this RAM is actually used.
1246  *                                     Changing the size while migrating
1247  *                                     can result in the migration being
1248  *                                     canceled.
1249  *
1250  * @mr: the #MemoryRegion to be initialized.
1251  * @owner: the object that tracks the region's reference count
1252  * @name: Region name, becomes part of RAMBlock name used in migration stream
1253  *        must be unique within any device
1254  * @size: used size of the region.
1255  * @max_size: max size of the region.
1256  * @resized: callback to notify owner about used size change.
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_resizeable_ram(MemoryRegion *mr,
1263                                        Object *owner,
1264                                        const char *name,
1265                                        uint64_t size,
1266                                        uint64_t max_size,
1267                                        void (*resized)(const char*,
1268                                                        uint64_t length,
1269                                                        void *host),
1270                                        Error **errp);
1271 #ifdef CONFIG_POSIX
1272 
1273 /**
1274  * memory_region_init_ram_from_file:  Initialize RAM memory region with a
1275  *                                    mmap-ed backend.
1276  *
1277  * @mr: the #MemoryRegion to be initialized.
1278  * @owner: the object that tracks the region's reference count
1279  * @name: Region name, becomes part of RAMBlock name used in migration stream
1280  *        must be unique within any device
1281  * @size: size of the region.
1282  * @align: alignment of the region base address; if 0, the default alignment
1283  *         (getpagesize()) will be used.
1284  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1285  *             RAM_NORESERVE,
1286  * @path: the path in which to allocate the RAM.
1287  * @readonly: true to open @path for reading, false for read/write.
1288  * @errp: pointer to Error*, to store an error if it happens.
1289  *
1290  * Note that this function does not do anything to cause the data in the
1291  * RAM memory region to be migrated; that is the responsibility of the caller.
1292  */
1293 void memory_region_init_ram_from_file(MemoryRegion *mr,
1294                                       Object *owner,
1295                                       const char *name,
1296                                       uint64_t size,
1297                                       uint64_t align,
1298                                       uint32_t ram_flags,
1299                                       const char *path,
1300                                       bool readonly,
1301                                       Error **errp);
1302 
1303 /**
1304  * memory_region_init_ram_from_fd:  Initialize RAM memory region with a
1305  *                                  mmap-ed backend.
1306  *
1307  * @mr: the #MemoryRegion to be initialized.
1308  * @owner: the object that tracks the region's reference count
1309  * @name: the name of the region.
1310  * @size: size of the region.
1311  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1312  *             RAM_NORESERVE, RAM_PROTECTED.
1313  * @fd: the fd to mmap.
1314  * @offset: offset within the file referenced by fd
1315  * @errp: pointer to Error*, to store an error if it happens.
1316  *
1317  * Note that this function does not do anything to cause the data in the
1318  * RAM memory region to be migrated; that is the responsibility of the caller.
1319  */
1320 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1321                                     Object *owner,
1322                                     const char *name,
1323                                     uint64_t size,
1324                                     uint32_t ram_flags,
1325                                     int fd,
1326                                     ram_addr_t offset,
1327                                     Error **errp);
1328 #endif
1329 
1330 /**
1331  * memory_region_init_ram_ptr:  Initialize RAM memory region from a
1332  *                              user-provided pointer.  Accesses into the
1333  *                              region will modify memory directly.
1334  *
1335  * @mr: the #MemoryRegion to be initialized.
1336  * @owner: the object that tracks the region's reference count
1337  * @name: Region name, becomes part of RAMBlock name used in migration stream
1338  *        must be unique within any device
1339  * @size: size of the region.
1340  * @ptr: memory to be mapped; must contain at least @size bytes.
1341  *
1342  * Note that this function does not do anything to cause the data in the
1343  * RAM memory region to be migrated; that is the responsibility of the caller.
1344  */
1345 void memory_region_init_ram_ptr(MemoryRegion *mr,
1346                                 Object *owner,
1347                                 const char *name,
1348                                 uint64_t size,
1349                                 void *ptr);
1350 
1351 /**
1352  * memory_region_init_ram_device_ptr:  Initialize RAM device memory region from
1353  *                                     a user-provided pointer.
1354  *
1355  * A RAM device represents a mapping to a physical device, such as to a PCI
1356  * MMIO BAR of an vfio-pci assigned device.  The memory region may be mapped
1357  * into the VM address space and access to the region will modify memory
1358  * directly.  However, the memory region should not be included in a memory
1359  * dump (device may not be enabled/mapped at the time of the dump), and
1360  * operations incompatible with manipulating MMIO should be avoided.  Replaces
1361  * skip_dump flag.
1362  *
1363  * @mr: the #MemoryRegion to be initialized.
1364  * @owner: the object that tracks the region's reference count
1365  * @name: the name of the region.
1366  * @size: size of the region.
1367  * @ptr: memory to be mapped; must contain at least @size bytes.
1368  *
1369  * Note that this function does not do anything to cause the data in the
1370  * RAM memory region to be migrated; that is the responsibility of the caller.
1371  * (For RAM device memory regions, migrating the contents rarely makes sense.)
1372  */
1373 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1374                                        Object *owner,
1375                                        const char *name,
1376                                        uint64_t size,
1377                                        void *ptr);
1378 
1379 /**
1380  * memory_region_init_alias: Initialize a memory region that aliases all or a
1381  *                           part of another memory region.
1382  *
1383  * @mr: the #MemoryRegion to be initialized.
1384  * @owner: the object that tracks the region's reference count
1385  * @name: used for debugging; not visible to the user or ABI
1386  * @orig: the region to be referenced; @mr will be equivalent to
1387  *        @orig between @offset and @offset + @size - 1.
1388  * @offset: start of the section in @orig to be referenced.
1389  * @size: size of the region.
1390  */
1391 void memory_region_init_alias(MemoryRegion *mr,
1392                               Object *owner,
1393                               const char *name,
1394                               MemoryRegion *orig,
1395                               hwaddr offset,
1396                               uint64_t size);
1397 
1398 /**
1399  * memory_region_init_rom_nomigrate: Initialize a ROM memory region.
1400  *
1401  * This has the same effect as calling memory_region_init_ram_nomigrate()
1402  * and then marking the resulting region read-only with
1403  * memory_region_set_readonly().
1404  *
1405  * Note that this function does not do anything to cause the data in the
1406  * RAM side of the memory region to be migrated; that is the responsibility
1407  * of the caller.
1408  *
1409  * @mr: the #MemoryRegion to be initialized.
1410  * @owner: the object that tracks the region's reference count
1411  * @name: Region name, becomes part of RAMBlock name used in migration stream
1412  *        must be unique within any device
1413  * @size: size of the region.
1414  * @errp: pointer to Error*, to store an error if it happens.
1415  */
1416 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1417                                       Object *owner,
1418                                       const char *name,
1419                                       uint64_t size,
1420                                       Error **errp);
1421 
1422 /**
1423  * memory_region_init_rom_device_nomigrate:  Initialize a ROM memory region.
1424  *                                 Writes are handled via callbacks.
1425  *
1426  * Note that this function does not do anything to cause the data in the
1427  * RAM side of the memory region to be migrated; that is the responsibility
1428  * of the caller.
1429  *
1430  * @mr: the #MemoryRegion to be initialized.
1431  * @owner: the object that tracks the region's reference count
1432  * @ops: callbacks for write access handling (must not be NULL).
1433  * @opaque: passed to the read and write callbacks of the @ops structure.
1434  * @name: Region name, becomes part of RAMBlock name used in migration stream
1435  *        must be unique within any device
1436  * @size: size of the region.
1437  * @errp: pointer to Error*, to store an error if it happens.
1438  */
1439 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1440                                              Object *owner,
1441                                              const MemoryRegionOps *ops,
1442                                              void *opaque,
1443                                              const char *name,
1444                                              uint64_t size,
1445                                              Error **errp);
1446 
1447 /**
1448  * memory_region_init_iommu: Initialize a memory region of a custom type
1449  * that translates addresses
1450  *
1451  * An IOMMU region translates addresses and forwards accesses to a target
1452  * memory region.
1453  *
1454  * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION.
1455  * @_iommu_mr should be a pointer to enough memory for an instance of
1456  * that subclass, @instance_size is the size of that subclass, and
1457  * @mrtypename is its name. This function will initialize @_iommu_mr as an
1458  * instance of the subclass, and its methods will then be called to handle
1459  * accesses to the memory region. See the documentation of
1460  * #IOMMUMemoryRegionClass for further details.
1461  *
1462  * @_iommu_mr: the #IOMMUMemoryRegion to be initialized
1463  * @instance_size: the IOMMUMemoryRegion subclass instance size
1464  * @mrtypename: the type name of the #IOMMUMemoryRegion
1465  * @owner: the object that tracks the region's reference count
1466  * @name: used for debugging; not visible to the user or ABI
1467  * @size: size of the region.
1468  */
1469 void memory_region_init_iommu(void *_iommu_mr,
1470                               size_t instance_size,
1471                               const char *mrtypename,
1472                               Object *owner,
1473                               const char *name,
1474                               uint64_t size);
1475 
1476 /**
1477  * memory_region_init_ram - Initialize RAM memory region.  Accesses into the
1478  *                          region will modify memory directly.
1479  *
1480  * @mr: the #MemoryRegion to be initialized
1481  * @owner: the object that tracks the region's reference count (must be
1482  *         TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL)
1483  * @name: name of the memory region
1484  * @size: size of the region in bytes
1485  * @errp: pointer to Error*, to store an error if it happens.
1486  *
1487  * This function allocates RAM for a board model or device, and
1488  * arranges for it to be migrated (by calling vmstate_register_ram()
1489  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1490  * @owner is NULL).
1491  *
1492  * TODO: Currently we restrict @owner to being either NULL (for
1493  * global RAM regions with no owner) or devices, so that we can
1494  * give the RAM block a unique name for migration purposes.
1495  * We should lift this restriction and allow arbitrary Objects.
1496  * If you pass a non-NULL non-device @owner then we will assert.
1497  */
1498 void memory_region_init_ram(MemoryRegion *mr,
1499                             Object *owner,
1500                             const char *name,
1501                             uint64_t size,
1502                             Error **errp);
1503 
1504 /**
1505  * memory_region_init_rom: Initialize a ROM memory region.
1506  *
1507  * This has the same effect as calling memory_region_init_ram()
1508  * and then marking the resulting region read-only with
1509  * memory_region_set_readonly(). This includes arranging for the
1510  * contents to be migrated.
1511  *
1512  * TODO: Currently we restrict @owner to being either NULL (for
1513  * global RAM regions with no owner) or devices, so that we can
1514  * give the RAM block a unique name for migration purposes.
1515  * We should lift this restriction and allow arbitrary Objects.
1516  * If you pass a non-NULL non-device @owner then we will assert.
1517  *
1518  * @mr: the #MemoryRegion to be initialized.
1519  * @owner: the object that tracks the region's reference count
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(MemoryRegion *mr,
1526                             Object *owner,
1527                             const char *name,
1528                             uint64_t size,
1529                             Error **errp);
1530 
1531 /**
1532  * memory_region_init_rom_device:  Initialize a ROM memory region.
1533  *                                 Writes are handled via callbacks.
1534  *
1535  * This function initializes a memory region backed by RAM for reads
1536  * and callbacks for writes, and arranges for the RAM backing to
1537  * be migrated (by calling vmstate_register_ram()
1538  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1539  * @owner is NULL).
1540  *
1541  * TODO: Currently we restrict @owner to being either NULL (for
1542  * global RAM regions with no owner) or devices, so that we can
1543  * give the RAM block a unique name for migration purposes.
1544  * We should lift this restriction and allow arbitrary Objects.
1545  * If you pass a non-NULL non-device @owner then we will assert.
1546  *
1547  * @mr: the #MemoryRegion to be initialized.
1548  * @owner: the object that tracks the region's reference count
1549  * @ops: callbacks for write access handling (must not be NULL).
1550  * @opaque: passed to the read and write callbacks of the @ops structure.
1551  * @name: Region name, becomes part of RAMBlock name used in migration stream
1552  *        must be unique within any device
1553  * @size: size of the region.
1554  * @errp: pointer to Error*, to store an error if it happens.
1555  */
1556 void memory_region_init_rom_device(MemoryRegion *mr,
1557                                    Object *owner,
1558                                    const MemoryRegionOps *ops,
1559                                    void *opaque,
1560                                    const char *name,
1561                                    uint64_t size,
1562                                    Error **errp);
1563 
1564 
1565 /**
1566  * memory_region_owner: get a memory region's owner.
1567  *
1568  * @mr: the memory region being queried.
1569  */
1570 Object *memory_region_owner(MemoryRegion *mr);
1571 
1572 /**
1573  * memory_region_size: get a memory region's size.
1574  *
1575  * @mr: the memory region being queried.
1576  */
1577 uint64_t memory_region_size(MemoryRegion *mr);
1578 
1579 /**
1580  * memory_region_is_ram: check whether a memory region is random access
1581  *
1582  * Returns %true if a memory region is random access.
1583  *
1584  * @mr: the memory region being queried
1585  */
1586 static inline bool memory_region_is_ram(MemoryRegion *mr)
1587 {
1588     return mr->ram;
1589 }
1590 
1591 /**
1592  * memory_region_is_ram_device: check whether a memory region is a ram device
1593  *
1594  * Returns %true if a memory region is a device backed ram region
1595  *
1596  * @mr: the memory region being queried
1597  */
1598 bool memory_region_is_ram_device(MemoryRegion *mr);
1599 
1600 /**
1601  * memory_region_is_romd: check whether a memory region is in ROMD mode
1602  *
1603  * Returns %true if a memory region is a ROM device and currently set to allow
1604  * direct reads.
1605  *
1606  * @mr: the memory region being queried
1607  */
1608 static inline bool memory_region_is_romd(MemoryRegion *mr)
1609 {
1610     return mr->rom_device && mr->romd_mode;
1611 }
1612 
1613 /**
1614  * memory_region_is_protected: check whether a memory region is protected
1615  *
1616  * Returns %true if a memory region is protected RAM and cannot be accessed
1617  * via standard mechanisms, e.g. DMA.
1618  *
1619  * @mr: the memory region being queried
1620  */
1621 bool memory_region_is_protected(MemoryRegion *mr);
1622 
1623 /**
1624  * memory_region_get_iommu: check whether a memory region is an iommu
1625  *
1626  * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu,
1627  * otherwise NULL.
1628  *
1629  * @mr: the memory region being queried
1630  */
1631 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr)
1632 {
1633     if (mr->alias) {
1634         return memory_region_get_iommu(mr->alias);
1635     }
1636     if (mr->is_iommu) {
1637         return (IOMMUMemoryRegion *) mr;
1638     }
1639     return NULL;
1640 }
1641 
1642 /**
1643  * memory_region_get_iommu_class_nocheck: returns iommu memory region class
1644  *   if an iommu or NULL if not
1645  *
1646  * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu,
1647  * otherwise NULL. This is fast path avoiding QOM checking, use with caution.
1648  *
1649  * @iommu_mr: the memory region being queried
1650  */
1651 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck(
1652         IOMMUMemoryRegion *iommu_mr)
1653 {
1654     return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class);
1655 }
1656 
1657 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL)
1658 
1659 /**
1660  * memory_region_iommu_get_min_page_size: get minimum supported page size
1661  * for an iommu
1662  *
1663  * Returns minimum supported page size for an iommu.
1664  *
1665  * @iommu_mr: the memory region being queried
1666  */
1667 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr);
1668 
1669 /**
1670  * memory_region_notify_iommu: notify a change in an IOMMU translation entry.
1671  *
1672  * Note: for any IOMMU implementation, an in-place mapping change
1673  * should be notified with an UNMAP followed by a MAP.
1674  *
1675  * @iommu_mr: the memory region that was changed
1676  * @iommu_idx: the IOMMU index for the translation table which has changed
1677  * @event: TLB event with the new entry in the IOMMU translation table.
1678  *         The entry replaces all old entries for the same virtual I/O address
1679  *         range.
1680  */
1681 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1682                                 int iommu_idx,
1683                                 IOMMUTLBEvent event);
1684 
1685 /**
1686  * memory_region_notify_iommu_one: notify a change in an IOMMU translation
1687  *                           entry to a single notifier
1688  *
1689  * This works just like memory_region_notify_iommu(), but it only
1690  * notifies a specific notifier, not all of them.
1691  *
1692  * @notifier: the notifier to be notified
1693  * @event: TLB event with the new entry in the IOMMU translation table.
1694  *         The entry replaces all old entries for the same virtual I/O address
1695  *         range.
1696  */
1697 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1698                                     IOMMUTLBEvent *event);
1699 
1700 /**
1701  * memory_region_register_iommu_notifier: register a notifier for changes to
1702  * IOMMU translation entries.
1703  *
1704  * Returns 0 on success, or a negative errno otherwise. In particular,
1705  * -EINVAL indicates that at least one of the attributes of the notifier
1706  * is not supported (flag/range) by the IOMMU memory region. In case of error
1707  * the error object must be created.
1708  *
1709  * @mr: the memory region to observe
1710  * @n: the IOMMUNotifier to be added; the notify callback receives a
1711  *     pointer to an #IOMMUTLBEntry as the opaque value; the pointer
1712  *     ceases to be valid on exit from the notifier.
1713  * @errp: pointer to Error*, to store an error if it happens.
1714  */
1715 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1716                                           IOMMUNotifier *n, Error **errp);
1717 
1718 /**
1719  * memory_region_iommu_replay: replay existing IOMMU translations to
1720  * a notifier with the minimum page granularity returned by
1721  * mr->iommu_ops->get_page_size().
1722  *
1723  * Note: this is not related to record-and-replay functionality.
1724  *
1725  * @iommu_mr: the memory region to observe
1726  * @n: the notifier to which to replay iommu mappings
1727  */
1728 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
1729 
1730 /**
1731  * memory_region_unregister_iommu_notifier: unregister a notifier for
1732  * changes to IOMMU translation entries.
1733  *
1734  * @mr: the memory region which was observed and for which notity_stopped()
1735  *      needs to be called
1736  * @n: the notifier to be removed.
1737  */
1738 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1739                                              IOMMUNotifier *n);
1740 
1741 /**
1742  * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is
1743  * defined on the IOMMU.
1744  *
1745  * Returns 0 on success, or a negative errno otherwise. In particular,
1746  * -EINVAL indicates that the IOMMU does not support the requested
1747  * attribute.
1748  *
1749  * @iommu_mr: the memory region
1750  * @attr: the requested attribute
1751  * @data: a pointer to the requested attribute data
1752  */
1753 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1754                                  enum IOMMUMemoryRegionAttr attr,
1755                                  void *data);
1756 
1757 /**
1758  * memory_region_iommu_attrs_to_index: return the IOMMU index to
1759  * use for translations with the given memory transaction attributes.
1760  *
1761  * @iommu_mr: the memory region
1762  * @attrs: the memory transaction attributes
1763  */
1764 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1765                                        MemTxAttrs attrs);
1766 
1767 /**
1768  * memory_region_iommu_num_indexes: return the total number of IOMMU
1769  * indexes that this IOMMU supports.
1770  *
1771  * @iommu_mr: the memory region
1772  */
1773 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr);
1774 
1775 /**
1776  * memory_region_iommu_set_page_size_mask: set the supported page
1777  * sizes for a given IOMMU memory region
1778  *
1779  * @iommu_mr: IOMMU memory region
1780  * @page_size_mask: supported page size mask
1781  * @errp: pointer to Error*, to store an error if it happens.
1782  */
1783 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1784                                            uint64_t page_size_mask,
1785                                            Error **errp);
1786 
1787 /**
1788  * memory_region_name: get a memory region's name
1789  *
1790  * Returns the string that was used to initialize the memory region.
1791  *
1792  * @mr: the memory region being queried
1793  */
1794 const char *memory_region_name(const MemoryRegion *mr);
1795 
1796 /**
1797  * memory_region_is_logging: return whether a memory region is logging writes
1798  *
1799  * Returns %true if the memory region is logging writes for the given client
1800  *
1801  * @mr: the memory region being queried
1802  * @client: the client being queried
1803  */
1804 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client);
1805 
1806 /**
1807  * memory_region_get_dirty_log_mask: return the clients for which a
1808  * memory region is logging writes.
1809  *
1810  * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants
1811  * are the bit indices.
1812  *
1813  * @mr: the memory region being queried
1814  */
1815 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr);
1816 
1817 /**
1818  * memory_region_is_rom: check whether a memory region is ROM
1819  *
1820  * Returns %true if a memory region is read-only memory.
1821  *
1822  * @mr: the memory region being queried
1823  */
1824 static inline bool memory_region_is_rom(MemoryRegion *mr)
1825 {
1826     return mr->ram && mr->readonly;
1827 }
1828 
1829 /**
1830  * memory_region_is_nonvolatile: check whether a memory region is non-volatile
1831  *
1832  * Returns %true is a memory region is non-volatile memory.
1833  *
1834  * @mr: the memory region being queried
1835  */
1836 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr)
1837 {
1838     return mr->nonvolatile;
1839 }
1840 
1841 /**
1842  * memory_region_get_fd: Get a file descriptor backing a RAM memory region.
1843  *
1844  * Returns a file descriptor backing a file-based RAM memory region,
1845  * or -1 if the region is not a file-based RAM memory region.
1846  *
1847  * @mr: the RAM or alias memory region being queried.
1848  */
1849 int memory_region_get_fd(MemoryRegion *mr);
1850 
1851 /**
1852  * memory_region_from_host: Convert a pointer into a RAM memory region
1853  * and an offset within it.
1854  *
1855  * Given a host pointer inside a RAM memory region (created with
1856  * memory_region_init_ram() or memory_region_init_ram_ptr()), return
1857  * the MemoryRegion and the offset within it.
1858  *
1859  * Use with care; by the time this function returns, the returned pointer is
1860  * not protected by RCU anymore.  If the caller is not within an RCU critical
1861  * section and does not hold the iothread lock, it must have other means of
1862  * protecting the pointer, such as a reference to the region that includes
1863  * the incoming ram_addr_t.
1864  *
1865  * @ptr: the host pointer to be converted
1866  * @offset: the offset within memory region
1867  */
1868 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset);
1869 
1870 /**
1871  * memory_region_get_ram_ptr: Get a pointer into a RAM memory region.
1872  *
1873  * Returns a host pointer to a RAM memory region (created with
1874  * memory_region_init_ram() or memory_region_init_ram_ptr()).
1875  *
1876  * Use with care; by the time this function returns, the returned pointer is
1877  * not protected by RCU anymore.  If the caller is not within an RCU critical
1878  * section and does not hold the iothread lock, it must have other means of
1879  * protecting the pointer, such as a reference to the region that includes
1880  * the incoming ram_addr_t.
1881  *
1882  * @mr: the memory region being queried.
1883  */
1884 void *memory_region_get_ram_ptr(MemoryRegion *mr);
1885 
1886 /* memory_region_ram_resize: Resize a RAM region.
1887  *
1888  * Resizing RAM while migrating can result in the migration being canceled.
1889  * Care has to be taken if the guest might have already detected the memory.
1890  *
1891  * @mr: a memory region created with @memory_region_init_resizeable_ram.
1892  * @newsize: the new size the region
1893  * @errp: pointer to Error*, to store an error if it happens.
1894  */
1895 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize,
1896                               Error **errp);
1897 
1898 /**
1899  * memory_region_msync: Synchronize selected address range of
1900  * a memory mapped region
1901  *
1902  * @mr: the memory region to be msync
1903  * @addr: the initial address of the range to be sync
1904  * @size: the size of the range to be sync
1905  */
1906 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size);
1907 
1908 /**
1909  * memory_region_writeback: Trigger cache writeback for
1910  * selected address range
1911  *
1912  * @mr: the memory region to be updated
1913  * @addr: the initial address of the range to be written back
1914  * @size: the size of the range to be written back
1915  */
1916 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size);
1917 
1918 /**
1919  * memory_region_set_log: Turn dirty logging on or off for a region.
1920  *
1921  * Turns dirty logging on or off for a specified client (display, migration).
1922  * Only meaningful for RAM regions.
1923  *
1924  * @mr: the memory region being updated.
1925  * @log: whether dirty logging is to be enabled or disabled.
1926  * @client: the user of the logging information; %DIRTY_MEMORY_VGA only.
1927  */
1928 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client);
1929 
1930 /**
1931  * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region.
1932  *
1933  * Marks a range of bytes as dirty, after it has been dirtied outside
1934  * guest code.
1935  *
1936  * @mr: the memory region being dirtied.
1937  * @addr: the address (relative to the start of the region) being dirtied.
1938  * @size: size of the range being dirtied.
1939  */
1940 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1941                              hwaddr size);
1942 
1943 /**
1944  * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range
1945  *
1946  * This function is called when the caller wants to clear the remote
1947  * dirty bitmap of a memory range within the memory region.  This can
1948  * be used by e.g. KVM to manually clear dirty log when
1949  * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host
1950  * kernel.
1951  *
1952  * @mr:     the memory region to clear the dirty log upon
1953  * @start:  start address offset within the memory region
1954  * @len:    length of the memory region to clear dirty bitmap
1955  */
1956 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
1957                                       hwaddr len);
1958 
1959 /**
1960  * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty
1961  *                                         bitmap and clear it.
1962  *
1963  * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and
1964  * returns the snapshot.  The snapshot can then be used to query dirty
1965  * status, using memory_region_snapshot_get_dirty.  Snapshotting allows
1966  * querying the same page multiple times, which is especially useful for
1967  * display updates where the scanlines often are not page aligned.
1968  *
1969  * The dirty bitmap region which gets copyed into the snapshot (and
1970  * cleared afterwards) can be larger than requested.  The boundaries
1971  * are rounded up/down so complete bitmap longs (covering 64 pages on
1972  * 64bit hosts) can be copied over into the bitmap snapshot.  Which
1973  * isn't a problem for display updates as the extra pages are outside
1974  * the visible area, and in case the visible area changes a full
1975  * display redraw is due anyway.  Should other use cases for this
1976  * function emerge we might have to revisit this implementation
1977  * detail.
1978  *
1979  * Use g_free to release DirtyBitmapSnapshot.
1980  *
1981  * @mr: the memory region being queried.
1982  * @addr: the address (relative to the start of the region) being queried.
1983  * @size: the size of the range being queried.
1984  * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA.
1985  */
1986 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
1987                                                             hwaddr addr,
1988                                                             hwaddr size,
1989                                                             unsigned client);
1990 
1991 /**
1992  * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty
1993  *                                   in the specified dirty bitmap snapshot.
1994  *
1995  * @mr: the memory region being queried.
1996  * @snap: the dirty bitmap snapshot
1997  * @addr: the address (relative to the start of the region) being queried.
1998  * @size: the size of the range being queried.
1999  */
2000 bool memory_region_snapshot_get_dirty(MemoryRegion *mr,
2001                                       DirtyBitmapSnapshot *snap,
2002                                       hwaddr addr, hwaddr size);
2003 
2004 /**
2005  * memory_region_reset_dirty: Mark a range of pages as clean, for a specified
2006  *                            client.
2007  *
2008  * Marks a range of pages as no longer dirty.
2009  *
2010  * @mr: the region being updated.
2011  * @addr: the start of the subrange being cleaned.
2012  * @size: the size of the subrange being cleaned.
2013  * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or
2014  *          %DIRTY_MEMORY_VGA.
2015  */
2016 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2017                                hwaddr size, unsigned client);
2018 
2019 /**
2020  * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate
2021  *                                 TBs (for self-modifying code).
2022  *
2023  * The MemoryRegionOps->write() callback of a ROM device must use this function
2024  * to mark byte ranges that have been modified internally, such as by directly
2025  * accessing the memory returned by memory_region_get_ram_ptr().
2026  *
2027  * This function marks the range dirty and invalidates TBs so that TCG can
2028  * detect self-modifying code.
2029  *
2030  * @mr: the region being flushed.
2031  * @addr: the start, relative to the start of the region, of the range being
2032  *        flushed.
2033  * @size: the size, in bytes, of the range being flushed.
2034  */
2035 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size);
2036 
2037 /**
2038  * memory_region_set_readonly: Turn a memory region read-only (or read-write)
2039  *
2040  * Allows a memory region to be marked as read-only (turning it into a ROM).
2041  * only useful on RAM regions.
2042  *
2043  * @mr: the region being updated.
2044  * @readonly: whether rhe region is to be ROM or RAM.
2045  */
2046 void memory_region_set_readonly(MemoryRegion *mr, bool readonly);
2047 
2048 /**
2049  * memory_region_set_nonvolatile: Turn a memory region non-volatile
2050  *
2051  * Allows a memory region to be marked as non-volatile.
2052  * only useful on RAM regions.
2053  *
2054  * @mr: the region being updated.
2055  * @nonvolatile: whether rhe region is to be non-volatile.
2056  */
2057 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile);
2058 
2059 /**
2060  * memory_region_rom_device_set_romd: enable/disable ROMD mode
2061  *
2062  * Allows a ROM device (initialized with memory_region_init_rom_device() to
2063  * set to ROMD mode (default) or MMIO mode.  When it is in ROMD mode, the
2064  * device is mapped to guest memory and satisfies read access directly.
2065  * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function.
2066  * Writes are always handled by the #MemoryRegion.write function.
2067  *
2068  * @mr: the memory region to be updated
2069  * @romd_mode: %true to put the region into ROMD mode
2070  */
2071 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode);
2072 
2073 /**
2074  * memory_region_set_coalescing: Enable memory coalescing for the region.
2075  *
2076  * Enabled writes to a region to be queued for later processing. MMIO ->write
2077  * callbacks may be delayed until a non-coalesced MMIO is issued.
2078  * Only useful for IO regions.  Roughly similar to write-combining hardware.
2079  *
2080  * @mr: the memory region to be write coalesced
2081  */
2082 void memory_region_set_coalescing(MemoryRegion *mr);
2083 
2084 /**
2085  * memory_region_add_coalescing: Enable memory coalescing for a sub-range of
2086  *                               a region.
2087  *
2088  * Like memory_region_set_coalescing(), but works on a sub-range of a region.
2089  * Multiple calls can be issued coalesced disjoint ranges.
2090  *
2091  * @mr: the memory region to be updated.
2092  * @offset: the start of the range within the region to be coalesced.
2093  * @size: the size of the subrange to be coalesced.
2094  */
2095 void memory_region_add_coalescing(MemoryRegion *mr,
2096                                   hwaddr offset,
2097                                   uint64_t size);
2098 
2099 /**
2100  * memory_region_clear_coalescing: Disable MMIO coalescing for the region.
2101  *
2102  * Disables any coalescing caused by memory_region_set_coalescing() or
2103  * memory_region_add_coalescing().  Roughly equivalent to uncacheble memory
2104  * hardware.
2105  *
2106  * @mr: the memory region to be updated.
2107  */
2108 void memory_region_clear_coalescing(MemoryRegion *mr);
2109 
2110 /**
2111  * memory_region_set_flush_coalesced: Enforce memory coalescing flush before
2112  *                                    accesses.
2113  *
2114  * Ensure that pending coalesced MMIO request are flushed before the memory
2115  * region is accessed. This property is automatically enabled for all regions
2116  * passed to memory_region_set_coalescing() and memory_region_add_coalescing().
2117  *
2118  * @mr: the memory region to be updated.
2119  */
2120 void memory_region_set_flush_coalesced(MemoryRegion *mr);
2121 
2122 /**
2123  * memory_region_clear_flush_coalesced: Disable memory coalescing flush before
2124  *                                      accesses.
2125  *
2126  * Clear the automatic coalesced MMIO flushing enabled via
2127  * memory_region_set_flush_coalesced. Note that this service has no effect on
2128  * memory regions that have MMIO coalescing enabled for themselves. For them,
2129  * automatic flushing will stop once coalescing is disabled.
2130  *
2131  * @mr: the memory region to be updated.
2132  */
2133 void memory_region_clear_flush_coalesced(MemoryRegion *mr);
2134 
2135 /**
2136  * memory_region_add_eventfd: Request an eventfd to be triggered when a word
2137  *                            is written to a location.
2138  *
2139  * Marks a word in an IO region (initialized with memory_region_init_io())
2140  * as a trigger for an eventfd event.  The I/O callback will not be called.
2141  * The caller must be prepared to handle failure (that is, take the required
2142  * action if the callback _is_ called).
2143  *
2144  * @mr: the memory region being updated.
2145  * @addr: the address within @mr that is to be monitored
2146  * @size: the size of the access to trigger the eventfd
2147  * @match_data: whether to match against @data, instead of just @addr
2148  * @data: the data to match against the guest write
2149  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2150  **/
2151 void memory_region_add_eventfd(MemoryRegion *mr,
2152                                hwaddr addr,
2153                                unsigned size,
2154                                bool match_data,
2155                                uint64_t data,
2156                                EventNotifier *e);
2157 
2158 /**
2159  * memory_region_del_eventfd: Cancel an eventfd.
2160  *
2161  * Cancels an eventfd trigger requested by a previous
2162  * memory_region_add_eventfd() call.
2163  *
2164  * @mr: the memory region being updated.
2165  * @addr: the address within @mr that is to be monitored
2166  * @size: the size of the access to trigger the eventfd
2167  * @match_data: whether to match against @data, instead of just @addr
2168  * @data: the data to match against the guest write
2169  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2170  */
2171 void memory_region_del_eventfd(MemoryRegion *mr,
2172                                hwaddr addr,
2173                                unsigned size,
2174                                bool match_data,
2175                                uint64_t data,
2176                                EventNotifier *e);
2177 
2178 /**
2179  * memory_region_add_subregion: Add a subregion to a container.
2180  *
2181  * Adds a subregion at @offset.  The subregion may not overlap with other
2182  * subregions (except for those explicitly marked as overlapping).  A region
2183  * may only be added once as a subregion (unless removed with
2184  * memory_region_del_subregion()); use memory_region_init_alias() if you
2185  * want a region to be a subregion in multiple locations.
2186  *
2187  * @mr: the region to contain the new subregion; must be a container
2188  *      initialized with memory_region_init().
2189  * @offset: the offset relative to @mr where @subregion is added.
2190  * @subregion: the subregion to be added.
2191  */
2192 void memory_region_add_subregion(MemoryRegion *mr,
2193                                  hwaddr offset,
2194                                  MemoryRegion *subregion);
2195 /**
2196  * memory_region_add_subregion_overlap: Add a subregion to a container
2197  *                                      with overlap.
2198  *
2199  * Adds a subregion at @offset.  The subregion may overlap with other
2200  * subregions.  Conflicts are resolved by having a higher @priority hide a
2201  * lower @priority. Subregions without priority are taken as @priority 0.
2202  * A region may only be added once as a subregion (unless removed with
2203  * memory_region_del_subregion()); use memory_region_init_alias() if you
2204  * want a region to be a subregion in multiple locations.
2205  *
2206  * @mr: the region to contain the new subregion; must be a container
2207  *      initialized with memory_region_init().
2208  * @offset: the offset relative to @mr where @subregion is added.
2209  * @subregion: the subregion to be added.
2210  * @priority: used for resolving overlaps; highest priority wins.
2211  */
2212 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2213                                          hwaddr offset,
2214                                          MemoryRegion *subregion,
2215                                          int priority);
2216 
2217 /**
2218  * memory_region_get_ram_addr: Get the ram address associated with a memory
2219  *                             region
2220  *
2221  * @mr: the region to be queried
2222  */
2223 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr);
2224 
2225 uint64_t memory_region_get_alignment(const MemoryRegion *mr);
2226 /**
2227  * memory_region_del_subregion: Remove a subregion.
2228  *
2229  * Removes a subregion from its container.
2230  *
2231  * @mr: the container to be updated.
2232  * @subregion: the region being removed; must be a current subregion of @mr.
2233  */
2234 void memory_region_del_subregion(MemoryRegion *mr,
2235                                  MemoryRegion *subregion);
2236 
2237 /*
2238  * memory_region_set_enabled: dynamically enable or disable a region
2239  *
2240  * Enables or disables a memory region.  A disabled memory region
2241  * ignores all accesses to itself and its subregions.  It does not
2242  * obscure sibling subregions with lower priority - it simply behaves as
2243  * if it was removed from the hierarchy.
2244  *
2245  * Regions default to being enabled.
2246  *
2247  * @mr: the region to be updated
2248  * @enabled: whether to enable or disable the region
2249  */
2250 void memory_region_set_enabled(MemoryRegion *mr, bool enabled);
2251 
2252 /*
2253  * memory_region_set_address: dynamically update the address of a region
2254  *
2255  * Dynamically updates the address of a region, relative to its container.
2256  * May be used on regions are currently part of a memory hierarchy.
2257  *
2258  * @mr: the region to be updated
2259  * @addr: new address, relative to container region
2260  */
2261 void memory_region_set_address(MemoryRegion *mr, hwaddr addr);
2262 
2263 /*
2264  * memory_region_set_size: dynamically update the size of a region.
2265  *
2266  * Dynamically updates the size of a region.
2267  *
2268  * @mr: the region to be updated
2269  * @size: used size of the region.
2270  */
2271 void memory_region_set_size(MemoryRegion *mr, uint64_t size);
2272 
2273 /*
2274  * memory_region_set_alias_offset: dynamically update a memory alias's offset
2275  *
2276  * Dynamically updates the offset into the target region that an alias points
2277  * to, as if the fourth argument to memory_region_init_alias() has changed.
2278  *
2279  * @mr: the #MemoryRegion to be updated; should be an alias.
2280  * @offset: the new offset into the target memory region
2281  */
2282 void memory_region_set_alias_offset(MemoryRegion *mr,
2283                                     hwaddr offset);
2284 
2285 /**
2286  * memory_region_present: checks if an address relative to a @container
2287  * translates into #MemoryRegion within @container
2288  *
2289  * Answer whether a #MemoryRegion within @container covers the address
2290  * @addr.
2291  *
2292  * @container: a #MemoryRegion within which @addr is a relative address
2293  * @addr: the area within @container to be searched
2294  */
2295 bool memory_region_present(MemoryRegion *container, hwaddr addr);
2296 
2297 /**
2298  * memory_region_is_mapped: returns true if #MemoryRegion is mapped
2299  * into any address space.
2300  *
2301  * @mr: a #MemoryRegion which should be checked if it's mapped
2302  */
2303 bool memory_region_is_mapped(MemoryRegion *mr);
2304 
2305 /**
2306  * memory_region_get_ram_discard_manager: get the #RamDiscardManager for a
2307  * #MemoryRegion
2308  *
2309  * The #RamDiscardManager cannot change while a memory region is mapped.
2310  *
2311  * @mr: the #MemoryRegion
2312  */
2313 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr);
2314 
2315 /**
2316  * memory_region_has_ram_discard_manager: check whether a #MemoryRegion has a
2317  * #RamDiscardManager assigned
2318  *
2319  * @mr: the #MemoryRegion
2320  */
2321 static inline bool memory_region_has_ram_discard_manager(MemoryRegion *mr)
2322 {
2323     return !!memory_region_get_ram_discard_manager(mr);
2324 }
2325 
2326 /**
2327  * memory_region_set_ram_discard_manager: set the #RamDiscardManager for a
2328  * #MemoryRegion
2329  *
2330  * This function must not be called for a mapped #MemoryRegion, a #MemoryRegion
2331  * that does not cover RAM, or a #MemoryRegion that already has a
2332  * #RamDiscardManager assigned.
2333  *
2334  * @mr: the #MemoryRegion
2335  * @rdm: #RamDiscardManager to set
2336  */
2337 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2338                                            RamDiscardManager *rdm);
2339 
2340 /**
2341  * memory_region_find: translate an address/size relative to a
2342  * MemoryRegion into a #MemoryRegionSection.
2343  *
2344  * Locates the first #MemoryRegion within @mr that overlaps the range
2345  * given by @addr and @size.
2346  *
2347  * Returns a #MemoryRegionSection that describes a contiguous overlap.
2348  * It will have the following characteristics:
2349  * - @size = 0 iff no overlap was found
2350  * - @mr is non-%NULL iff an overlap was found
2351  *
2352  * Remember that in the return value the @offset_within_region is
2353  * relative to the returned region (in the .@mr field), not to the
2354  * @mr argument.
2355  *
2356  * Similarly, the .@offset_within_address_space is relative to the
2357  * address space that contains both regions, the passed and the
2358  * returned one.  However, in the special case where the @mr argument
2359  * has no container (and thus is the root of the address space), the
2360  * following will hold:
2361  * - @offset_within_address_space >= @addr
2362  * - @offset_within_address_space + .@size <= @addr + @size
2363  *
2364  * @mr: a MemoryRegion within which @addr is a relative address
2365  * @addr: start of the area within @as to be searched
2366  * @size: size of the area to be searched
2367  */
2368 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2369                                        hwaddr addr, uint64_t size);
2370 
2371 /**
2372  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2373  *
2374  * Synchronizes the dirty page log for all address spaces.
2375  */
2376 void memory_global_dirty_log_sync(void);
2377 
2378 /**
2379  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2380  *
2381  * Synchronizes the vCPUs with a thread that is reading the dirty bitmap.
2382  * This function must be called after the dirty log bitmap is cleared, and
2383  * before dirty guest memory pages are read.  If you are using
2384  * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes
2385  * care of doing this.
2386  */
2387 void memory_global_after_dirty_log_sync(void);
2388 
2389 /**
2390  * memory_region_transaction_begin: Start a transaction.
2391  *
2392  * During a transaction, changes will be accumulated and made visible
2393  * only when the transaction ends (is committed).
2394  */
2395 void memory_region_transaction_begin(void);
2396 
2397 /**
2398  * memory_region_transaction_commit: Commit a transaction and make changes
2399  *                                   visible to the guest.
2400  */
2401 void memory_region_transaction_commit(void);
2402 
2403 /**
2404  * memory_listener_register: register callbacks to be called when memory
2405  *                           sections are mapped or unmapped into an address
2406  *                           space
2407  *
2408  * @listener: an object containing the callbacks to be called
2409  * @filter: if non-%NULL, only regions in this address space will be observed
2410  */
2411 void memory_listener_register(MemoryListener *listener, AddressSpace *filter);
2412 
2413 /**
2414  * memory_listener_unregister: undo the effect of memory_listener_register()
2415  *
2416  * @listener: an object containing the callbacks to be removed
2417  */
2418 void memory_listener_unregister(MemoryListener *listener);
2419 
2420 /**
2421  * memory_global_dirty_log_start: begin dirty logging for all regions
2422  *
2423  * @flags: purpose of starting dirty log, migration or dirty rate
2424  */
2425 void memory_global_dirty_log_start(unsigned int flags);
2426 
2427 /**
2428  * memory_global_dirty_log_stop: end dirty logging for all regions
2429  *
2430  * @flags: purpose of stopping dirty log, migration or dirty rate
2431  */
2432 void memory_global_dirty_log_stop(unsigned int flags);
2433 
2434 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled);
2435 
2436 /**
2437  * memory_region_dispatch_read: perform a read directly to the specified
2438  * MemoryRegion.
2439  *
2440  * @mr: #MemoryRegion to access
2441  * @addr: address within that region
2442  * @pval: pointer to uint64_t which the data is written to
2443  * @op: size, sign, and endianness of the memory operation
2444  * @attrs: memory transaction attributes to use for the access
2445  */
2446 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
2447                                         hwaddr addr,
2448                                         uint64_t *pval,
2449                                         MemOp op,
2450                                         MemTxAttrs attrs);
2451 /**
2452  * memory_region_dispatch_write: perform a write directly to the specified
2453  * MemoryRegion.
2454  *
2455  * @mr: #MemoryRegion to access
2456  * @addr: address within that region
2457  * @data: data to write
2458  * @op: size, sign, and endianness of the memory operation
2459  * @attrs: memory transaction attributes to use for the access
2460  */
2461 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
2462                                          hwaddr addr,
2463                                          uint64_t data,
2464                                          MemOp op,
2465                                          MemTxAttrs attrs);
2466 
2467 /**
2468  * address_space_init: initializes an address space
2469  *
2470  * @as: an uninitialized #AddressSpace
2471  * @root: a #MemoryRegion that routes addresses for the address space
2472  * @name: an address space name.  The name is only used for debugging
2473  *        output.
2474  */
2475 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name);
2476 
2477 /**
2478  * address_space_destroy: destroy an address space
2479  *
2480  * Releases all resources associated with an address space.  After an address space
2481  * is destroyed, its root memory region (given by address_space_init()) may be destroyed
2482  * as well.
2483  *
2484  * @as: address space to be destroyed
2485  */
2486 void address_space_destroy(AddressSpace *as);
2487 
2488 /**
2489  * address_space_remove_listeners: unregister all listeners of an address space
2490  *
2491  * Removes all callbacks previously registered with memory_listener_register()
2492  * for @as.
2493  *
2494  * @as: an initialized #AddressSpace
2495  */
2496 void address_space_remove_listeners(AddressSpace *as);
2497 
2498 /**
2499  * address_space_rw: read from or write to an address space.
2500  *
2501  * Return a MemTxResult indicating whether the operation succeeded
2502  * or failed (eg unassigned memory, device rejected the transaction,
2503  * IOMMU fault).
2504  *
2505  * @as: #AddressSpace to be accessed
2506  * @addr: address within that address space
2507  * @attrs: memory transaction attributes
2508  * @buf: buffer with the data transferred
2509  * @len: the number of bytes to read or write
2510  * @is_write: indicates the transfer direction
2511  */
2512 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
2513                              MemTxAttrs attrs, void *buf,
2514                              hwaddr len, bool is_write);
2515 
2516 /**
2517  * address_space_write: write to address space.
2518  *
2519  * Return a MemTxResult indicating whether the operation succeeded
2520  * or failed (eg unassigned memory, device rejected the transaction,
2521  * IOMMU fault).
2522  *
2523  * @as: #AddressSpace to be accessed
2524  * @addr: address within that address space
2525  * @attrs: memory transaction attributes
2526  * @buf: buffer with the data transferred
2527  * @len: the number of bytes to write
2528  */
2529 MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
2530                                 MemTxAttrs attrs,
2531                                 const void *buf, hwaddr len);
2532 
2533 /**
2534  * address_space_write_rom: write to address space, including ROM.
2535  *
2536  * This function writes to the specified address space, but will
2537  * write data to both ROM and RAM. This is used for non-guest
2538  * writes like writes from the gdb debug stub or initial loading
2539  * of ROM contents.
2540  *
2541  * Note that portions of the write which attempt to write data to
2542  * a device will be silently ignored -- only real RAM and ROM will
2543  * be written to.
2544  *
2545  * Return a MemTxResult indicating whether the operation succeeded
2546  * or failed (eg unassigned memory, device rejected the transaction,
2547  * IOMMU fault).
2548  *
2549  * @as: #AddressSpace to be accessed
2550  * @addr: address within that address space
2551  * @attrs: memory transaction attributes
2552  * @buf: buffer with the data transferred
2553  * @len: the number of bytes to write
2554  */
2555 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr,
2556                                     MemTxAttrs attrs,
2557                                     const void *buf, hwaddr len);
2558 
2559 /* address_space_ld*: load from an address space
2560  * address_space_st*: store to an address space
2561  *
2562  * These functions perform a load or store of the byte, word,
2563  * longword or quad to the specified address within the AddressSpace.
2564  * The _le suffixed functions treat the data as little endian;
2565  * _be indicates big endian; no suffix indicates "same endianness
2566  * as guest CPU".
2567  *
2568  * The "guest CPU endianness" accessors are deprecated for use outside
2569  * target-* code; devices should be CPU-agnostic and use either the LE
2570  * or the BE accessors.
2571  *
2572  * @as #AddressSpace to be accessed
2573  * @addr: address within that address space
2574  * @val: data value, for stores
2575  * @attrs: memory transaction attributes
2576  * @result: location to write the success/failure of the transaction;
2577  *   if NULL, this information is discarded
2578  */
2579 
2580 #define SUFFIX
2581 #define ARG1         as
2582 #define ARG1_DECL    AddressSpace *as
2583 #include "exec/memory_ldst.h.inc"
2584 
2585 #define SUFFIX
2586 #define ARG1         as
2587 #define ARG1_DECL    AddressSpace *as
2588 #include "exec/memory_ldst_phys.h.inc"
2589 
2590 struct MemoryRegionCache {
2591     void *ptr;
2592     hwaddr xlat;
2593     hwaddr len;
2594     FlatView *fv;
2595     MemoryRegionSection mrs;
2596     bool is_write;
2597 };
2598 
2599 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL })
2600 
2601 
2602 /* address_space_ld*_cached: load from a cached #MemoryRegion
2603  * address_space_st*_cached: store into a cached #MemoryRegion
2604  *
2605  * These functions perform a load or store of the byte, word,
2606  * longword or quad to the specified address.  The address is
2607  * a physical address in the AddressSpace, but it must lie within
2608  * a #MemoryRegion that was mapped with address_space_cache_init.
2609  *
2610  * The _le suffixed functions treat the data as little endian;
2611  * _be indicates big endian; no suffix indicates "same endianness
2612  * as guest CPU".
2613  *
2614  * The "guest CPU endianness" accessors are deprecated for use outside
2615  * target-* code; devices should be CPU-agnostic and use either the LE
2616  * or the BE accessors.
2617  *
2618  * @cache: previously initialized #MemoryRegionCache to be accessed
2619  * @addr: address within the address space
2620  * @val: data value, for stores
2621  * @attrs: memory transaction attributes
2622  * @result: location to write the success/failure of the transaction;
2623  *   if NULL, this information is discarded
2624  */
2625 
2626 #define SUFFIX       _cached_slow
2627 #define ARG1         cache
2628 #define ARG1_DECL    MemoryRegionCache *cache
2629 #include "exec/memory_ldst.h.inc"
2630 
2631 /* Inline fast path for direct RAM access.  */
2632 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache,
2633     hwaddr addr, MemTxAttrs attrs, MemTxResult *result)
2634 {
2635     assert(addr < cache->len);
2636     if (likely(cache->ptr)) {
2637         return ldub_p(cache->ptr + addr);
2638     } else {
2639         return address_space_ldub_cached_slow(cache, addr, attrs, result);
2640     }
2641 }
2642 
2643 static inline void address_space_stb_cached(MemoryRegionCache *cache,
2644     hwaddr addr, uint8_t val, MemTxAttrs attrs, MemTxResult *result)
2645 {
2646     assert(addr < cache->len);
2647     if (likely(cache->ptr)) {
2648         stb_p(cache->ptr + addr, val);
2649     } else {
2650         address_space_stb_cached_slow(cache, addr, val, attrs, result);
2651     }
2652 }
2653 
2654 #define ENDIANNESS   _le
2655 #include "exec/memory_ldst_cached.h.inc"
2656 
2657 #define ENDIANNESS   _be
2658 #include "exec/memory_ldst_cached.h.inc"
2659 
2660 #define SUFFIX       _cached
2661 #define ARG1         cache
2662 #define ARG1_DECL    MemoryRegionCache *cache
2663 #include "exec/memory_ldst_phys.h.inc"
2664 
2665 /* address_space_cache_init: prepare for repeated access to a physical
2666  * memory region
2667  *
2668  * @cache: #MemoryRegionCache to be filled
2669  * @as: #AddressSpace to be accessed
2670  * @addr: address within that address space
2671  * @len: length of buffer
2672  * @is_write: indicates the transfer direction
2673  *
2674  * Will only work with RAM, and may map a subset of the requested range by
2675  * returning a value that is less than @len.  On failure, return a negative
2676  * errno value.
2677  *
2678  * Because it only works with RAM, this function can be used for
2679  * read-modify-write operations.  In this case, is_write should be %true.
2680  *
2681  * Note that addresses passed to the address_space_*_cached functions
2682  * are relative to @addr.
2683  */
2684 int64_t address_space_cache_init(MemoryRegionCache *cache,
2685                                  AddressSpace *as,
2686                                  hwaddr addr,
2687                                  hwaddr len,
2688                                  bool is_write);
2689 
2690 /**
2691  * address_space_cache_invalidate: complete a write to a #MemoryRegionCache
2692  *
2693  * @cache: The #MemoryRegionCache to operate on.
2694  * @addr: The first physical address that was written, relative to the
2695  * address that was passed to @address_space_cache_init.
2696  * @access_len: The number of bytes that were written starting at @addr.
2697  */
2698 void address_space_cache_invalidate(MemoryRegionCache *cache,
2699                                     hwaddr addr,
2700                                     hwaddr access_len);
2701 
2702 /**
2703  * address_space_cache_destroy: free a #MemoryRegionCache
2704  *
2705  * @cache: The #MemoryRegionCache whose memory should be released.
2706  */
2707 void address_space_cache_destroy(MemoryRegionCache *cache);
2708 
2709 /* address_space_get_iotlb_entry: translate an address into an IOTLB
2710  * entry. Should be called from an RCU critical section.
2711  */
2712 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
2713                                             bool is_write, MemTxAttrs attrs);
2714 
2715 /* address_space_translate: translate an address range into an address space
2716  * into a MemoryRegion and an address range into that section.  Should be
2717  * called from an RCU critical section, to avoid that the last reference
2718  * to the returned region disappears after address_space_translate returns.
2719  *
2720  * @fv: #FlatView to be accessed
2721  * @addr: address within that address space
2722  * @xlat: pointer to address within the returned memory region section's
2723  * #MemoryRegion.
2724  * @len: pointer to length
2725  * @is_write: indicates the transfer direction
2726  * @attrs: memory attributes
2727  */
2728 MemoryRegion *flatview_translate(FlatView *fv,
2729                                  hwaddr addr, hwaddr *xlat,
2730                                  hwaddr *len, bool is_write,
2731                                  MemTxAttrs attrs);
2732 
2733 static inline MemoryRegion *address_space_translate(AddressSpace *as,
2734                                                     hwaddr addr, hwaddr *xlat,
2735                                                     hwaddr *len, bool is_write,
2736                                                     MemTxAttrs attrs)
2737 {
2738     return flatview_translate(address_space_to_flatview(as),
2739                               addr, xlat, len, is_write, attrs);
2740 }
2741 
2742 /* address_space_access_valid: check for validity of accessing an address
2743  * space range
2744  *
2745  * Check whether memory is assigned to the given address space range, and
2746  * access is permitted by any IOMMU regions that are active for the address
2747  * space.
2748  *
2749  * For now, addr and len should be aligned to a page size.  This limitation
2750  * will be lifted in the future.
2751  *
2752  * @as: #AddressSpace to be accessed
2753  * @addr: address within that address space
2754  * @len: length of the area to be checked
2755  * @is_write: indicates the transfer direction
2756  * @attrs: memory attributes
2757  */
2758 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len,
2759                                 bool is_write, MemTxAttrs attrs);
2760 
2761 /* address_space_map: map a physical memory region into a host virtual address
2762  *
2763  * May map a subset of the requested range, given by and returned in @plen.
2764  * May return %NULL and set *@plen to zero(0), if resources needed to perform
2765  * the mapping are exhausted.
2766  * Use only for reads OR writes - not for read-modify-write operations.
2767  * Use cpu_register_map_client() to know when retrying the map operation is
2768  * likely to succeed.
2769  *
2770  * @as: #AddressSpace to be accessed
2771  * @addr: address within that address space
2772  * @plen: pointer to length of buffer; updated on return
2773  * @is_write: indicates the transfer direction
2774  * @attrs: memory attributes
2775  */
2776 void *address_space_map(AddressSpace *as, hwaddr addr,
2777                         hwaddr *plen, bool is_write, MemTxAttrs attrs);
2778 
2779 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map()
2780  *
2781  * Will also mark the memory as dirty if @is_write == %true.  @access_len gives
2782  * the amount of memory that was actually read or written by the caller.
2783  *
2784  * @as: #AddressSpace used
2785  * @buffer: host pointer as returned by address_space_map()
2786  * @len: buffer length as returned by address_space_map()
2787  * @access_len: amount of data actually transferred
2788  * @is_write: indicates the transfer direction
2789  */
2790 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2791                          bool is_write, hwaddr access_len);
2792 
2793 
2794 /* Internal functions, part of the implementation of address_space_read.  */
2795 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
2796                                     MemTxAttrs attrs, void *buf, hwaddr len);
2797 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
2798                                    MemTxAttrs attrs, void *buf,
2799                                    hwaddr len, hwaddr addr1, hwaddr l,
2800                                    MemoryRegion *mr);
2801 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr);
2802 
2803 /* Internal functions, part of the implementation of address_space_read_cached
2804  * and address_space_write_cached.  */
2805 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache,
2806                                            hwaddr addr, void *buf, hwaddr len);
2807 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache,
2808                                             hwaddr addr, const void *buf,
2809                                             hwaddr len);
2810 
2811 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
2812 {
2813     if (is_write) {
2814         return memory_region_is_ram(mr) && !mr->readonly &&
2815                !mr->rom_device && !memory_region_is_ram_device(mr);
2816     } else {
2817         return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) ||
2818                memory_region_is_romd(mr);
2819     }
2820 }
2821 
2822 /**
2823  * address_space_read: read from an address space.
2824  *
2825  * Return a MemTxResult indicating whether the operation succeeded
2826  * or failed (eg unassigned memory, device rejected the transaction,
2827  * IOMMU fault).  Called within RCU critical section.
2828  *
2829  * @as: #AddressSpace to be accessed
2830  * @addr: address within that address space
2831  * @attrs: memory transaction attributes
2832  * @buf: buffer with the data transferred
2833  * @len: length of the data transferred
2834  */
2835 static inline __attribute__((__always_inline__))
2836 MemTxResult address_space_read(AddressSpace *as, hwaddr addr,
2837                                MemTxAttrs attrs, void *buf,
2838                                hwaddr len)
2839 {
2840     MemTxResult result = MEMTX_OK;
2841     hwaddr l, addr1;
2842     void *ptr;
2843     MemoryRegion *mr;
2844     FlatView *fv;
2845 
2846     if (__builtin_constant_p(len)) {
2847         if (len) {
2848             RCU_READ_LOCK_GUARD();
2849             fv = address_space_to_flatview(as);
2850             l = len;
2851             mr = flatview_translate(fv, addr, &addr1, &l, false, attrs);
2852             if (len == l && memory_access_is_direct(mr, false)) {
2853                 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
2854                 memcpy(buf, ptr, len);
2855             } else {
2856                 result = flatview_read_continue(fv, addr, attrs, buf, len,
2857                                                 addr1, l, mr);
2858             }
2859         }
2860     } else {
2861         result = address_space_read_full(as, addr, attrs, buf, len);
2862     }
2863     return result;
2864 }
2865 
2866 /**
2867  * address_space_read_cached: read from a cached RAM region
2868  *
2869  * @cache: Cached region to be addressed
2870  * @addr: address relative to the base of the RAM region
2871  * @buf: buffer with the data transferred
2872  * @len: length of the data transferred
2873  */
2874 static inline MemTxResult
2875 address_space_read_cached(MemoryRegionCache *cache, hwaddr addr,
2876                           void *buf, hwaddr len)
2877 {
2878     assert(addr < cache->len && len <= cache->len - addr);
2879     fuzz_dma_read_cb(cache->xlat + addr, len, cache->mrs.mr);
2880     if (likely(cache->ptr)) {
2881         memcpy(buf, cache->ptr + addr, len);
2882         return MEMTX_OK;
2883     } else {
2884         return address_space_read_cached_slow(cache, addr, buf, len);
2885     }
2886 }
2887 
2888 /**
2889  * address_space_write_cached: write to a cached RAM region
2890  *
2891  * @cache: Cached region to be addressed
2892  * @addr: address relative to the base of the RAM region
2893  * @buf: buffer with the data transferred
2894  * @len: length of the data transferred
2895  */
2896 static inline MemTxResult
2897 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr,
2898                            const void *buf, hwaddr len)
2899 {
2900     assert(addr < cache->len && len <= cache->len - addr);
2901     if (likely(cache->ptr)) {
2902         memcpy(cache->ptr + addr, buf, len);
2903         return MEMTX_OK;
2904     } else {
2905         return address_space_write_cached_slow(cache, addr, buf, len);
2906     }
2907 }
2908 
2909 #ifdef NEED_CPU_H
2910 /* enum device_endian to MemOp.  */
2911 static inline MemOp devend_memop(enum device_endian end)
2912 {
2913     QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN &&
2914                       DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN);
2915 
2916 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
2917     /* Swap if non-host endianness or native (target) endianness */
2918     return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP;
2919 #else
2920     const int non_host_endianness =
2921         DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN;
2922 
2923     /* In this case, native (target) endianness needs no swap.  */
2924     return (end == non_host_endianness) ? MO_BSWAP : 0;
2925 #endif
2926 }
2927 #endif
2928 
2929 /*
2930  * Inhibit technologies that require discarding of pages in RAM blocks, e.g.,
2931  * to manage the actual amount of memory consumed by the VM (then, the memory
2932  * provided by RAM blocks might be bigger than the desired memory consumption).
2933  * This *must* be set if:
2934  * - Discarding parts of a RAM blocks does not result in the change being
2935  *   reflected in the VM and the pages getting freed.
2936  * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous
2937  *   discards blindly.
2938  * - Discarding parts of a RAM blocks will result in integrity issues (e.g.,
2939  *   encrypted VMs).
2940  * Technologies that only temporarily pin the current working set of a
2941  * driver are fine, because we don't expect such pages to be discarded
2942  * (esp. based on guest action like balloon inflation).
2943  *
2944  * This is *not* to be used to protect from concurrent discards (esp.,
2945  * postcopy).
2946  *
2947  * Returns 0 if successful. Returns -EBUSY if a technology that relies on
2948  * discards to work reliably is active.
2949  */
2950 int ram_block_discard_disable(bool state);
2951 
2952 /*
2953  * See ram_block_discard_disable(): only disable uncoordinated discards,
2954  * keeping coordinated discards (via the RamDiscardManager) enabled.
2955  */
2956 int ram_block_uncoordinated_discard_disable(bool state);
2957 
2958 /*
2959  * Inhibit technologies that disable discarding of pages in RAM blocks.
2960  *
2961  * Returns 0 if successful. Returns -EBUSY if discards are already set to
2962  * broken.
2963  */
2964 int ram_block_discard_require(bool state);
2965 
2966 /*
2967  * See ram_block_discard_require(): only inhibit technologies that disable
2968  * uncoordinated discarding of pages in RAM blocks, allowing co-existance with
2969  * technologies that only inhibit uncoordinated discards (via the
2970  * RamDiscardManager).
2971  */
2972 int ram_block_coordinated_discard_require(bool state);
2973 
2974 /*
2975  * Test if any discarding of memory in ram blocks is disabled.
2976  */
2977 bool ram_block_discard_is_disabled(void);
2978 
2979 /*
2980  * Test if any discarding of memory in ram blocks is required to work reliably.
2981  */
2982 bool ram_block_discard_is_required(void);
2983 
2984 #endif
2985 
2986 #endif
2987