xref: /qemu/include/exec/memory.h (revision b83a80e8)
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     int mapped_via_alias; /* Mapped via an alias, container might be NULL */
742     Int128 size;
743     hwaddr addr;
744     void (*destructor)(MemoryRegion *mr);
745     uint64_t align;
746     bool terminates;
747     bool ram_device;
748     bool enabled;
749     bool warning_printed; /* For reservations */
750     uint8_t vga_logging_count;
751     MemoryRegion *alias;
752     hwaddr alias_offset;
753     int32_t priority;
754     QTAILQ_HEAD(, MemoryRegion) subregions;
755     QTAILQ_ENTRY(MemoryRegion) subregions_link;
756     QTAILQ_HEAD(, CoalescedMemoryRange) coalesced;
757     const char *name;
758     unsigned ioeventfd_nb;
759     MemoryRegionIoeventfd *ioeventfds;
760     RamDiscardManager *rdm; /* Only for RAM */
761 };
762 
763 struct IOMMUMemoryRegion {
764     MemoryRegion parent_obj;
765 
766     QLIST_HEAD(, IOMMUNotifier) iommu_notify;
767     IOMMUNotifierFlag iommu_notify_flags;
768 };
769 
770 #define IOMMU_NOTIFIER_FOREACH(n, mr) \
771     QLIST_FOREACH((n), &(mr)->iommu_notify, node)
772 
773 /**
774  * struct MemoryListener: callbacks structure for updates to the physical memory map
775  *
776  * Allows a component to adjust to changes in the guest-visible memory map.
777  * Use with memory_listener_register() and memory_listener_unregister().
778  */
779 struct MemoryListener {
780     /**
781      * @begin:
782      *
783      * Called at the beginning of an address space update transaction.
784      * Followed by calls to #MemoryListener.region_add(),
785      * #MemoryListener.region_del(), #MemoryListener.region_nop(),
786      * #MemoryListener.log_start() and #MemoryListener.log_stop() in
787      * increasing address order.
788      *
789      * @listener: The #MemoryListener.
790      */
791     void (*begin)(MemoryListener *listener);
792 
793     /**
794      * @commit:
795      *
796      * Called at the end of an address space update transaction,
797      * after the last call to #MemoryListener.region_add(),
798      * #MemoryListener.region_del() or #MemoryListener.region_nop(),
799      * #MemoryListener.log_start() and #MemoryListener.log_stop().
800      *
801      * @listener: The #MemoryListener.
802      */
803     void (*commit)(MemoryListener *listener);
804 
805     /**
806      * @region_add:
807      *
808      * Called during an address space update transaction,
809      * for a section of the address space that is new in this address space
810      * space since the last transaction.
811      *
812      * @listener: The #MemoryListener.
813      * @section: The new #MemoryRegionSection.
814      */
815     void (*region_add)(MemoryListener *listener, MemoryRegionSection *section);
816 
817     /**
818      * @region_del:
819      *
820      * Called during an address space update transaction,
821      * for a section of the address space that has disappeared in the address
822      * space since the last transaction.
823      *
824      * @listener: The #MemoryListener.
825      * @section: The old #MemoryRegionSection.
826      */
827     void (*region_del)(MemoryListener *listener, MemoryRegionSection *section);
828 
829     /**
830      * @region_nop:
831      *
832      * Called during an address space update transaction,
833      * for a section of the address space that is in the same place in the address
834      * space as in the last transaction.
835      *
836      * @listener: The #MemoryListener.
837      * @section: The #MemoryRegionSection.
838      */
839     void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section);
840 
841     /**
842      * @log_start:
843      *
844      * Called during an address space update transaction, after
845      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
846      * #MemoryListener.region_nop(), if dirty memory logging clients have
847      * become active since the last transaction.
848      *
849      * @listener: The #MemoryListener.
850      * @section: The #MemoryRegionSection.
851      * @old: A bitmap of dirty memory logging clients that were active in
852      * the previous transaction.
853      * @new: A bitmap of dirty memory logging clients that are active in
854      * the current transaction.
855      */
856     void (*log_start)(MemoryListener *listener, MemoryRegionSection *section,
857                       int old, int new);
858 
859     /**
860      * @log_stop:
861      *
862      * Called during an address space update transaction, after
863      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
864      * #MemoryListener.region_nop() and possibly after
865      * #MemoryListener.log_start(), if dirty memory logging clients have
866      * become inactive since the last transaction.
867      *
868      * @listener: The #MemoryListener.
869      * @section: The #MemoryRegionSection.
870      * @old: A bitmap of dirty memory logging clients that were active in
871      * the previous transaction.
872      * @new: A bitmap of dirty memory logging clients that are active in
873      * the current transaction.
874      */
875     void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section,
876                      int old, int new);
877 
878     /**
879      * @log_sync:
880      *
881      * Called by memory_region_snapshot_and_clear_dirty() and
882      * memory_global_dirty_log_sync(), before accessing QEMU's "official"
883      * copy of the dirty memory bitmap for a #MemoryRegionSection.
884      *
885      * @listener: The #MemoryListener.
886      * @section: The #MemoryRegionSection.
887      */
888     void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section);
889 
890     /**
891      * @log_sync_global:
892      *
893      * This is the global version of @log_sync when the listener does
894      * not have a way to synchronize the log with finer granularity.
895      * When the listener registers with @log_sync_global defined, then
896      * its @log_sync must be NULL.  Vice versa.
897      *
898      * @listener: The #MemoryListener.
899      */
900     void (*log_sync_global)(MemoryListener *listener);
901 
902     /**
903      * @log_clear:
904      *
905      * Called before reading the dirty memory bitmap for a
906      * #MemoryRegionSection.
907      *
908      * @listener: The #MemoryListener.
909      * @section: The #MemoryRegionSection.
910      */
911     void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section);
912 
913     /**
914      * @log_global_start:
915      *
916      * Called by memory_global_dirty_log_start(), which
917      * enables the %DIRTY_LOG_MIGRATION client on all memory regions in
918      * the address space.  #MemoryListener.log_global_start() is also
919      * called when a #MemoryListener is added, if global dirty logging is
920      * active at that time.
921      *
922      * @listener: The #MemoryListener.
923      */
924     void (*log_global_start)(MemoryListener *listener);
925 
926     /**
927      * @log_global_stop:
928      *
929      * Called by memory_global_dirty_log_stop(), which
930      * disables the %DIRTY_LOG_MIGRATION client on all memory regions in
931      * the address space.
932      *
933      * @listener: The #MemoryListener.
934      */
935     void (*log_global_stop)(MemoryListener *listener);
936 
937     /**
938      * @log_global_after_sync:
939      *
940      * Called after reading the dirty memory bitmap
941      * for any #MemoryRegionSection.
942      *
943      * @listener: The #MemoryListener.
944      */
945     void (*log_global_after_sync)(MemoryListener *listener);
946 
947     /**
948      * @eventfd_add:
949      *
950      * Called during an address space update transaction,
951      * for a section of the address space that has had a new ioeventfd
952      * registration since the last transaction.
953      *
954      * @listener: The #MemoryListener.
955      * @section: The new #MemoryRegionSection.
956      * @match_data: The @match_data parameter for the new ioeventfd.
957      * @data: The @data parameter for the new ioeventfd.
958      * @e: The #EventNotifier parameter for the new ioeventfd.
959      */
960     void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section,
961                         bool match_data, uint64_t data, EventNotifier *e);
962 
963     /**
964      * @eventfd_del:
965      *
966      * Called during an address space update transaction,
967      * for a section of the address space that has dropped an ioeventfd
968      * registration since the last transaction.
969      *
970      * @listener: The #MemoryListener.
971      * @section: The new #MemoryRegionSection.
972      * @match_data: The @match_data parameter for the dropped ioeventfd.
973      * @data: The @data parameter for the dropped ioeventfd.
974      * @e: The #EventNotifier parameter for the dropped ioeventfd.
975      */
976     void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section,
977                         bool match_data, uint64_t data, EventNotifier *e);
978 
979     /**
980      * @coalesced_io_add:
981      *
982      * Called during an address space update transaction,
983      * for a section of the address space that has had a new coalesced
984      * MMIO range registration since the last transaction.
985      *
986      * @listener: The #MemoryListener.
987      * @section: The new #MemoryRegionSection.
988      * @addr: The starting address for the coalesced MMIO range.
989      * @len: The length of the coalesced MMIO range.
990      */
991     void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section,
992                                hwaddr addr, hwaddr len);
993 
994     /**
995      * @coalesced_io_del:
996      *
997      * Called during an address space update transaction,
998      * for a section of the address space that has dropped a coalesced
999      * MMIO range since the last transaction.
1000      *
1001      * @listener: The #MemoryListener.
1002      * @section: The new #MemoryRegionSection.
1003      * @addr: The starting address for the coalesced MMIO range.
1004      * @len: The length of the coalesced MMIO range.
1005      */
1006     void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section,
1007                                hwaddr addr, hwaddr len);
1008     /**
1009      * @priority:
1010      *
1011      * Govern the order in which memory listeners are invoked. Lower priorities
1012      * are invoked earlier for "add" or "start" callbacks, and later for "delete"
1013      * or "stop" callbacks.
1014      */
1015     unsigned priority;
1016 
1017     /**
1018      * @name:
1019      *
1020      * Name of the listener.  It can be used in contexts where we'd like to
1021      * identify one memory listener with the rest.
1022      */
1023     const char *name;
1024 
1025     /* private: */
1026     AddressSpace *address_space;
1027     QTAILQ_ENTRY(MemoryListener) link;
1028     QTAILQ_ENTRY(MemoryListener) link_as;
1029 };
1030 
1031 /**
1032  * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects
1033  */
1034 struct AddressSpace {
1035     /* private: */
1036     struct rcu_head rcu;
1037     char *name;
1038     MemoryRegion *root;
1039 
1040     /* Accessed via RCU.  */
1041     struct FlatView *current_map;
1042 
1043     int ioeventfd_nb;
1044     struct MemoryRegionIoeventfd *ioeventfds;
1045     QTAILQ_HEAD(, MemoryListener) listeners;
1046     QTAILQ_ENTRY(AddressSpace) address_spaces_link;
1047 };
1048 
1049 typedef struct AddressSpaceDispatch AddressSpaceDispatch;
1050 typedef struct FlatRange FlatRange;
1051 
1052 /* Flattened global view of current active memory hierarchy.  Kept in sorted
1053  * order.
1054  */
1055 struct FlatView {
1056     struct rcu_head rcu;
1057     unsigned ref;
1058     FlatRange *ranges;
1059     unsigned nr;
1060     unsigned nr_allocated;
1061     struct AddressSpaceDispatch *dispatch;
1062     MemoryRegion *root;
1063 };
1064 
1065 static inline FlatView *address_space_to_flatview(AddressSpace *as)
1066 {
1067     return qatomic_rcu_read(&as->current_map);
1068 }
1069 
1070 /**
1071  * typedef flatview_cb: callback for flatview_for_each_range()
1072  *
1073  * @start: start address of the range within the FlatView
1074  * @len: length of the range in bytes
1075  * @mr: MemoryRegion covering this range
1076  * @offset_in_region: offset of the first byte of the range within @mr
1077  * @opaque: data pointer passed to flatview_for_each_range()
1078  *
1079  * Returns: true to stop the iteration, false to keep going.
1080  */
1081 typedef bool (*flatview_cb)(Int128 start,
1082                             Int128 len,
1083                             const MemoryRegion *mr,
1084                             hwaddr offset_in_region,
1085                             void *opaque);
1086 
1087 /**
1088  * flatview_for_each_range: Iterate through a FlatView
1089  * @fv: the FlatView to iterate through
1090  * @cb: function to call for each range
1091  * @opaque: opaque data pointer to pass to @cb
1092  *
1093  * A FlatView is made up of a list of non-overlapping ranges, each of
1094  * which is a slice of a MemoryRegion. This function iterates through
1095  * each range in @fv, calling @cb. The callback function can terminate
1096  * iteration early by returning 'true'.
1097  */
1098 void flatview_for_each_range(FlatView *fv, flatview_cb cb, void *opaque);
1099 
1100 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a,
1101                                           MemoryRegionSection *b)
1102 {
1103     return a->mr == b->mr &&
1104            a->fv == b->fv &&
1105            a->offset_within_region == b->offset_within_region &&
1106            a->offset_within_address_space == b->offset_within_address_space &&
1107            int128_eq(a->size, b->size) &&
1108            a->readonly == b->readonly &&
1109            a->nonvolatile == b->nonvolatile;
1110 }
1111 
1112 /**
1113  * memory_region_section_new_copy: Copy a memory region section
1114  *
1115  * Allocate memory for a new copy, copy the memory region section, and
1116  * properly take a reference on all relevant members.
1117  *
1118  * @s: the #MemoryRegionSection to copy
1119  */
1120 MemoryRegionSection *memory_region_section_new_copy(MemoryRegionSection *s);
1121 
1122 /**
1123  * memory_region_section_new_copy: Free a copied memory region section
1124  *
1125  * Free a copy of a memory section created via memory_region_section_new_copy().
1126  * properly dropping references on all relevant members.
1127  *
1128  * @s: the #MemoryRegionSection to copy
1129  */
1130 void memory_region_section_free_copy(MemoryRegionSection *s);
1131 
1132 /**
1133  * memory_region_init: Initialize a memory region
1134  *
1135  * The region typically acts as a container for other memory regions.  Use
1136  * memory_region_add_subregion() to add subregions.
1137  *
1138  * @mr: the #MemoryRegion to be initialized
1139  * @owner: the object that tracks the region's reference count
1140  * @name: used for debugging; not visible to the user or ABI
1141  * @size: size of the region; any subregions beyond this size will be clipped
1142  */
1143 void memory_region_init(MemoryRegion *mr,
1144                         Object *owner,
1145                         const char *name,
1146                         uint64_t size);
1147 
1148 /**
1149  * memory_region_ref: Add 1 to a memory region's reference count
1150  *
1151  * Whenever memory regions are accessed outside the BQL, they need to be
1152  * preserved against hot-unplug.  MemoryRegions actually do not have their
1153  * own reference count; they piggyback on a QOM object, their "owner".
1154  * This function adds a reference to the owner.
1155  *
1156  * All MemoryRegions must have an owner if they can disappear, even if the
1157  * device they belong to operates exclusively under the BQL.  This is because
1158  * the region could be returned at any time by memory_region_find, and this
1159  * is usually under guest control.
1160  *
1161  * @mr: the #MemoryRegion
1162  */
1163 void memory_region_ref(MemoryRegion *mr);
1164 
1165 /**
1166  * memory_region_unref: Remove 1 to a memory region's reference count
1167  *
1168  * Whenever memory regions are accessed outside the BQL, they need to be
1169  * preserved against hot-unplug.  MemoryRegions actually do not have their
1170  * own reference count; they piggyback on a QOM object, their "owner".
1171  * This function removes a reference to the owner and possibly destroys it.
1172  *
1173  * @mr: the #MemoryRegion
1174  */
1175 void memory_region_unref(MemoryRegion *mr);
1176 
1177 /**
1178  * memory_region_init_io: Initialize an I/O memory region.
1179  *
1180  * Accesses into the region will cause the callbacks in @ops to be called.
1181  * if @size is nonzero, subregions will be clipped to @size.
1182  *
1183  * @mr: the #MemoryRegion to be initialized.
1184  * @owner: the object that tracks the region's reference count
1185  * @ops: a structure containing read and write callbacks to be used when
1186  *       I/O is performed on the region.
1187  * @opaque: passed to the read and write callbacks of the @ops structure.
1188  * @name: used for debugging; not visible to the user or ABI
1189  * @size: size of the region.
1190  */
1191 void memory_region_init_io(MemoryRegion *mr,
1192                            Object *owner,
1193                            const MemoryRegionOps *ops,
1194                            void *opaque,
1195                            const char *name,
1196                            uint64_t size);
1197 
1198 /**
1199  * memory_region_init_ram_nomigrate:  Initialize RAM memory region.  Accesses
1200  *                                    into the region will modify memory
1201  *                                    directly.
1202  *
1203  * @mr: the #MemoryRegion to be initialized.
1204  * @owner: the object that tracks the region's reference count
1205  * @name: Region name, becomes part of RAMBlock name used in migration stream
1206  *        must be unique within any device
1207  * @size: size of the region.
1208  * @errp: pointer to Error*, to store an error if it happens.
1209  *
1210  * Note that this function does not do anything to cause the data in the
1211  * RAM memory region to be migrated; that is the responsibility of the caller.
1212  */
1213 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
1214                                       Object *owner,
1215                                       const char *name,
1216                                       uint64_t size,
1217                                       Error **errp);
1218 
1219 /**
1220  * memory_region_init_ram_flags_nomigrate:  Initialize RAM memory region.
1221  *                                          Accesses into the region will
1222  *                                          modify memory directly.
1223  *
1224  * @mr: the #MemoryRegion to be initialized.
1225  * @owner: the object that tracks the region's reference count
1226  * @name: Region name, becomes part of RAMBlock name used in migration stream
1227  *        must be unique within any device
1228  * @size: size of the region.
1229  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_NORESERVE.
1230  * @errp: pointer to Error*, to store an error if it happens.
1231  *
1232  * Note that this function does not do anything to cause the data in the
1233  * RAM memory region to be migrated; that is the responsibility of the caller.
1234  */
1235 void memory_region_init_ram_flags_nomigrate(MemoryRegion *mr,
1236                                             Object *owner,
1237                                             const char *name,
1238                                             uint64_t size,
1239                                             uint32_t ram_flags,
1240                                             Error **errp);
1241 
1242 /**
1243  * memory_region_init_resizeable_ram:  Initialize memory region with resizeable
1244  *                                     RAM.  Accesses into the region will
1245  *                                     modify memory directly.  Only an initial
1246  *                                     portion of this RAM is actually used.
1247  *                                     Changing the size while migrating
1248  *                                     can result in the migration being
1249  *                                     canceled.
1250  *
1251  * @mr: the #MemoryRegion to be initialized.
1252  * @owner: the object that tracks the region's reference count
1253  * @name: Region name, becomes part of RAMBlock name used in migration stream
1254  *        must be unique within any device
1255  * @size: used size of the region.
1256  * @max_size: max size of the region.
1257  * @resized: callback to notify owner about used size change.
1258  * @errp: pointer to Error*, to store an error if it happens.
1259  *
1260  * Note that this function does not do anything to cause the data in the
1261  * RAM memory region to be migrated; that is the responsibility of the caller.
1262  */
1263 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1264                                        Object *owner,
1265                                        const char *name,
1266                                        uint64_t size,
1267                                        uint64_t max_size,
1268                                        void (*resized)(const char*,
1269                                                        uint64_t length,
1270                                                        void *host),
1271                                        Error **errp);
1272 #ifdef CONFIG_POSIX
1273 
1274 /**
1275  * memory_region_init_ram_from_file:  Initialize RAM memory region with a
1276  *                                    mmap-ed backend.
1277  *
1278  * @mr: the #MemoryRegion to be initialized.
1279  * @owner: the object that tracks the region's reference count
1280  * @name: Region name, becomes part of RAMBlock name used in migration stream
1281  *        must be unique within any device
1282  * @size: size of the region.
1283  * @align: alignment of the region base address; if 0, the default alignment
1284  *         (getpagesize()) will be used.
1285  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1286  *             RAM_NORESERVE,
1287  * @path: the path in which to allocate the RAM.
1288  * @readonly: true to open @path for reading, false for read/write.
1289  * @errp: pointer to Error*, to store an error if it happens.
1290  *
1291  * Note that this function does not do anything to cause the data in the
1292  * RAM memory region to be migrated; that is the responsibility of the caller.
1293  */
1294 void memory_region_init_ram_from_file(MemoryRegion *mr,
1295                                       Object *owner,
1296                                       const char *name,
1297                                       uint64_t size,
1298                                       uint64_t align,
1299                                       uint32_t ram_flags,
1300                                       const char *path,
1301                                       bool readonly,
1302                                       Error **errp);
1303 
1304 /**
1305  * memory_region_init_ram_from_fd:  Initialize RAM memory region with a
1306  *                                  mmap-ed backend.
1307  *
1308  * @mr: the #MemoryRegion to be initialized.
1309  * @owner: the object that tracks the region's reference count
1310  * @name: the name of the region.
1311  * @size: size of the region.
1312  * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM,
1313  *             RAM_NORESERVE, RAM_PROTECTED.
1314  * @fd: the fd to mmap.
1315  * @offset: offset within the file referenced by fd
1316  * @errp: pointer to Error*, to store an error if it happens.
1317  *
1318  * Note that this function does not do anything to cause the data in the
1319  * RAM memory region to be migrated; that is the responsibility of the caller.
1320  */
1321 void memory_region_init_ram_from_fd(MemoryRegion *mr,
1322                                     Object *owner,
1323                                     const char *name,
1324                                     uint64_t size,
1325                                     uint32_t ram_flags,
1326                                     int fd,
1327                                     ram_addr_t offset,
1328                                     Error **errp);
1329 #endif
1330 
1331 /**
1332  * memory_region_init_ram_ptr:  Initialize RAM memory region from a
1333  *                              user-provided pointer.  Accesses into the
1334  *                              region will modify memory directly.
1335  *
1336  * @mr: the #MemoryRegion to be initialized.
1337  * @owner: the object that tracks the region's reference count
1338  * @name: Region name, becomes part of RAMBlock name used in migration stream
1339  *        must be unique within any device
1340  * @size: size of the region.
1341  * @ptr: memory to be mapped; must contain at least @size bytes.
1342  *
1343  * Note that this function does not do anything to cause the data in the
1344  * RAM memory region to be migrated; that is the responsibility of the caller.
1345  */
1346 void memory_region_init_ram_ptr(MemoryRegion *mr,
1347                                 Object *owner,
1348                                 const char *name,
1349                                 uint64_t size,
1350                                 void *ptr);
1351 
1352 /**
1353  * memory_region_init_ram_device_ptr:  Initialize RAM device memory region from
1354  *                                     a user-provided pointer.
1355  *
1356  * A RAM device represents a mapping to a physical device, such as to a PCI
1357  * MMIO BAR of an vfio-pci assigned device.  The memory region may be mapped
1358  * into the VM address space and access to the region will modify memory
1359  * directly.  However, the memory region should not be included in a memory
1360  * dump (device may not be enabled/mapped at the time of the dump), and
1361  * operations incompatible with manipulating MMIO should be avoided.  Replaces
1362  * skip_dump flag.
1363  *
1364  * @mr: the #MemoryRegion to be initialized.
1365  * @owner: the object that tracks the region's reference count
1366  * @name: the name of the region.
1367  * @size: size of the region.
1368  * @ptr: memory to be mapped; must contain at least @size bytes.
1369  *
1370  * Note that this function does not do anything to cause the data in the
1371  * RAM memory region to be migrated; that is the responsibility of the caller.
1372  * (For RAM device memory regions, migrating the contents rarely makes sense.)
1373  */
1374 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
1375                                        Object *owner,
1376                                        const char *name,
1377                                        uint64_t size,
1378                                        void *ptr);
1379 
1380 /**
1381  * memory_region_init_alias: Initialize a memory region that aliases all or a
1382  *                           part of another memory region.
1383  *
1384  * @mr: the #MemoryRegion to be initialized.
1385  * @owner: the object that tracks the region's reference count
1386  * @name: used for debugging; not visible to the user or ABI
1387  * @orig: the region to be referenced; @mr will be equivalent to
1388  *        @orig between @offset and @offset + @size - 1.
1389  * @offset: start of the section in @orig to be referenced.
1390  * @size: size of the region.
1391  */
1392 void memory_region_init_alias(MemoryRegion *mr,
1393                               Object *owner,
1394                               const char *name,
1395                               MemoryRegion *orig,
1396                               hwaddr offset,
1397                               uint64_t size);
1398 
1399 /**
1400  * memory_region_init_rom_nomigrate: Initialize a ROM memory region.
1401  *
1402  * This has the same effect as calling memory_region_init_ram_nomigrate()
1403  * and then marking the resulting region read-only with
1404  * memory_region_set_readonly().
1405  *
1406  * Note that this function does not do anything to cause the data in the
1407  * RAM side of the memory region to be migrated; that is the responsibility
1408  * of the caller.
1409  *
1410  * @mr: the #MemoryRegion to be initialized.
1411  * @owner: the object that tracks the region's reference count
1412  * @name: Region name, becomes part of RAMBlock name used in migration stream
1413  *        must be unique within any device
1414  * @size: size of the region.
1415  * @errp: pointer to Error*, to store an error if it happens.
1416  */
1417 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1418                                       Object *owner,
1419                                       const char *name,
1420                                       uint64_t size,
1421                                       Error **errp);
1422 
1423 /**
1424  * memory_region_init_rom_device_nomigrate:  Initialize a ROM memory region.
1425  *                                 Writes are handled via callbacks.
1426  *
1427  * Note that this function does not do anything to cause the data in the
1428  * RAM side of the memory region to be migrated; that is the responsibility
1429  * of the caller.
1430  *
1431  * @mr: the #MemoryRegion to be initialized.
1432  * @owner: the object that tracks the region's reference count
1433  * @ops: callbacks for write access handling (must not be NULL).
1434  * @opaque: passed to the read and write callbacks of the @ops structure.
1435  * @name: Region name, becomes part of RAMBlock name used in migration stream
1436  *        must be unique within any device
1437  * @size: size of the region.
1438  * @errp: pointer to Error*, to store an error if it happens.
1439  */
1440 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1441                                              Object *owner,
1442                                              const MemoryRegionOps *ops,
1443                                              void *opaque,
1444                                              const char *name,
1445                                              uint64_t size,
1446                                              Error **errp);
1447 
1448 /**
1449  * memory_region_init_iommu: Initialize a memory region of a custom type
1450  * that translates addresses
1451  *
1452  * An IOMMU region translates addresses and forwards accesses to a target
1453  * memory region.
1454  *
1455  * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION.
1456  * @_iommu_mr should be a pointer to enough memory for an instance of
1457  * that subclass, @instance_size is the size of that subclass, and
1458  * @mrtypename is its name. This function will initialize @_iommu_mr as an
1459  * instance of the subclass, and its methods will then be called to handle
1460  * accesses to the memory region. See the documentation of
1461  * #IOMMUMemoryRegionClass for further details.
1462  *
1463  * @_iommu_mr: the #IOMMUMemoryRegion to be initialized
1464  * @instance_size: the IOMMUMemoryRegion subclass instance size
1465  * @mrtypename: the type name of the #IOMMUMemoryRegion
1466  * @owner: the object that tracks the region's reference count
1467  * @name: used for debugging; not visible to the user or ABI
1468  * @size: size of the region.
1469  */
1470 void memory_region_init_iommu(void *_iommu_mr,
1471                               size_t instance_size,
1472                               const char *mrtypename,
1473                               Object *owner,
1474                               const char *name,
1475                               uint64_t size);
1476 
1477 /**
1478  * memory_region_init_ram - Initialize RAM memory region.  Accesses into the
1479  *                          region will modify memory directly.
1480  *
1481  * @mr: the #MemoryRegion to be initialized
1482  * @owner: the object that tracks the region's reference count (must be
1483  *         TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL)
1484  * @name: name of the memory region
1485  * @size: size of the region in bytes
1486  * @errp: pointer to Error*, to store an error if it happens.
1487  *
1488  * This function allocates RAM for a board model or device, and
1489  * arranges for it to be migrated (by calling vmstate_register_ram()
1490  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1491  * @owner is NULL).
1492  *
1493  * TODO: Currently we restrict @owner to being either NULL (for
1494  * global RAM regions with no owner) or devices, so that we can
1495  * give the RAM block a unique name for migration purposes.
1496  * We should lift this restriction and allow arbitrary Objects.
1497  * If you pass a non-NULL non-device @owner then we will assert.
1498  */
1499 void memory_region_init_ram(MemoryRegion *mr,
1500                             Object *owner,
1501                             const char *name,
1502                             uint64_t size,
1503                             Error **errp);
1504 
1505 /**
1506  * memory_region_init_rom: Initialize a ROM memory region.
1507  *
1508  * This has the same effect as calling memory_region_init_ram()
1509  * and then marking the resulting region read-only with
1510  * memory_region_set_readonly(). This includes arranging for the
1511  * contents to be migrated.
1512  *
1513  * TODO: Currently we restrict @owner to being either NULL (for
1514  * global RAM regions with no owner) or devices, so that we can
1515  * give the RAM block a unique name for migration purposes.
1516  * We should lift this restriction and allow arbitrary Objects.
1517  * If you pass a non-NULL non-device @owner then we will assert.
1518  *
1519  * @mr: the #MemoryRegion to be initialized.
1520  * @owner: the object that tracks the region's reference count
1521  * @name: Region name, becomes part of RAMBlock name used in migration stream
1522  *        must be unique within any device
1523  * @size: size of the region.
1524  * @errp: pointer to Error*, to store an error if it happens.
1525  */
1526 void memory_region_init_rom(MemoryRegion *mr,
1527                             Object *owner,
1528                             const char *name,
1529                             uint64_t size,
1530                             Error **errp);
1531 
1532 /**
1533  * memory_region_init_rom_device:  Initialize a ROM memory region.
1534  *                                 Writes are handled via callbacks.
1535  *
1536  * This function initializes a memory region backed by RAM for reads
1537  * and callbacks for writes, and arranges for the RAM backing to
1538  * be migrated (by calling vmstate_register_ram()
1539  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1540  * @owner is NULL).
1541  *
1542  * TODO: Currently we restrict @owner to being either NULL (for
1543  * global RAM regions with no owner) or devices, so that we can
1544  * give the RAM block a unique name for migration purposes.
1545  * We should lift this restriction and allow arbitrary Objects.
1546  * If you pass a non-NULL non-device @owner then we will assert.
1547  *
1548  * @mr: the #MemoryRegion to be initialized.
1549  * @owner: the object that tracks the region's reference count
1550  * @ops: callbacks for write access handling (must not be NULL).
1551  * @opaque: passed to the read and write callbacks of the @ops structure.
1552  * @name: Region name, becomes part of RAMBlock name used in migration stream
1553  *        must be unique within any device
1554  * @size: size of the region.
1555  * @errp: pointer to Error*, to store an error if it happens.
1556  */
1557 void memory_region_init_rom_device(MemoryRegion *mr,
1558                                    Object *owner,
1559                                    const MemoryRegionOps *ops,
1560                                    void *opaque,
1561                                    const char *name,
1562                                    uint64_t size,
1563                                    Error **errp);
1564 
1565 
1566 /**
1567  * memory_region_owner: get a memory region's owner.
1568  *
1569  * @mr: the memory region being queried.
1570  */
1571 Object *memory_region_owner(MemoryRegion *mr);
1572 
1573 /**
1574  * memory_region_size: get a memory region's size.
1575  *
1576  * @mr: the memory region being queried.
1577  */
1578 uint64_t memory_region_size(MemoryRegion *mr);
1579 
1580 /**
1581  * memory_region_is_ram: check whether a memory region is random access
1582  *
1583  * Returns %true if a memory region is random access.
1584  *
1585  * @mr: the memory region being queried
1586  */
1587 static inline bool memory_region_is_ram(MemoryRegion *mr)
1588 {
1589     return mr->ram;
1590 }
1591 
1592 /**
1593  * memory_region_is_ram_device: check whether a memory region is a ram device
1594  *
1595  * Returns %true if a memory region is a device backed ram region
1596  *
1597  * @mr: the memory region being queried
1598  */
1599 bool memory_region_is_ram_device(MemoryRegion *mr);
1600 
1601 /**
1602  * memory_region_is_romd: check whether a memory region is in ROMD mode
1603  *
1604  * Returns %true if a memory region is a ROM device and currently set to allow
1605  * direct reads.
1606  *
1607  * @mr: the memory region being queried
1608  */
1609 static inline bool memory_region_is_romd(MemoryRegion *mr)
1610 {
1611     return mr->rom_device && mr->romd_mode;
1612 }
1613 
1614 /**
1615  * memory_region_is_protected: check whether a memory region is protected
1616  *
1617  * Returns %true if a memory region is protected RAM and cannot be accessed
1618  * via standard mechanisms, e.g. DMA.
1619  *
1620  * @mr: the memory region being queried
1621  */
1622 bool memory_region_is_protected(MemoryRegion *mr);
1623 
1624 /**
1625  * memory_region_get_iommu: check whether a memory region is an iommu
1626  *
1627  * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu,
1628  * otherwise NULL.
1629  *
1630  * @mr: the memory region being queried
1631  */
1632 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr)
1633 {
1634     if (mr->alias) {
1635         return memory_region_get_iommu(mr->alias);
1636     }
1637     if (mr->is_iommu) {
1638         return (IOMMUMemoryRegion *) mr;
1639     }
1640     return NULL;
1641 }
1642 
1643 /**
1644  * memory_region_get_iommu_class_nocheck: returns iommu memory region class
1645  *   if an iommu or NULL if not
1646  *
1647  * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu,
1648  * otherwise NULL. This is fast path avoiding QOM checking, use with caution.
1649  *
1650  * @iommu_mr: the memory region being queried
1651  */
1652 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck(
1653         IOMMUMemoryRegion *iommu_mr)
1654 {
1655     return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class);
1656 }
1657 
1658 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL)
1659 
1660 /**
1661  * memory_region_iommu_get_min_page_size: get minimum supported page size
1662  * for an iommu
1663  *
1664  * Returns minimum supported page size for an iommu.
1665  *
1666  * @iommu_mr: the memory region being queried
1667  */
1668 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr);
1669 
1670 /**
1671  * memory_region_notify_iommu: notify a change in an IOMMU translation entry.
1672  *
1673  * Note: for any IOMMU implementation, an in-place mapping change
1674  * should be notified with an UNMAP followed by a MAP.
1675  *
1676  * @iommu_mr: the memory region that was changed
1677  * @iommu_idx: the IOMMU index for the translation table which has changed
1678  * @event: TLB event with the new entry in the IOMMU translation table.
1679  *         The entry replaces all old entries for the same virtual I/O address
1680  *         range.
1681  */
1682 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1683                                 int iommu_idx,
1684                                 IOMMUTLBEvent event);
1685 
1686 /**
1687  * memory_region_notify_iommu_one: notify a change in an IOMMU translation
1688  *                           entry to a single notifier
1689  *
1690  * This works just like memory_region_notify_iommu(), but it only
1691  * notifies a specific notifier, not all of them.
1692  *
1693  * @notifier: the notifier to be notified
1694  * @event: TLB event with the new entry in the IOMMU translation table.
1695  *         The entry replaces all old entries for the same virtual I/O address
1696  *         range.
1697  */
1698 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
1699                                     IOMMUTLBEvent *event);
1700 
1701 /**
1702  * memory_region_register_iommu_notifier: register a notifier for changes to
1703  * IOMMU translation entries.
1704  *
1705  * Returns 0 on success, or a negative errno otherwise. In particular,
1706  * -EINVAL indicates that at least one of the attributes of the notifier
1707  * is not supported (flag/range) by the IOMMU memory region. In case of error
1708  * the error object must be created.
1709  *
1710  * @mr: the memory region to observe
1711  * @n: the IOMMUNotifier to be added; the notify callback receives a
1712  *     pointer to an #IOMMUTLBEntry as the opaque value; the pointer
1713  *     ceases to be valid on exit from the notifier.
1714  * @errp: pointer to Error*, to store an error if it happens.
1715  */
1716 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1717                                           IOMMUNotifier *n, Error **errp);
1718 
1719 /**
1720  * memory_region_iommu_replay: replay existing IOMMU translations to
1721  * a notifier with the minimum page granularity returned by
1722  * mr->iommu_ops->get_page_size().
1723  *
1724  * Note: this is not related to record-and-replay functionality.
1725  *
1726  * @iommu_mr: the memory region to observe
1727  * @n: the notifier to which to replay iommu mappings
1728  */
1729 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
1730 
1731 /**
1732  * memory_region_unregister_iommu_notifier: unregister a notifier for
1733  * changes to IOMMU translation entries.
1734  *
1735  * @mr: the memory region which was observed and for which notity_stopped()
1736  *      needs to be called
1737  * @n: the notifier to be removed.
1738  */
1739 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1740                                              IOMMUNotifier *n);
1741 
1742 /**
1743  * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is
1744  * defined on the IOMMU.
1745  *
1746  * Returns 0 on success, or a negative errno otherwise. In particular,
1747  * -EINVAL indicates that the IOMMU does not support the requested
1748  * attribute.
1749  *
1750  * @iommu_mr: the memory region
1751  * @attr: the requested attribute
1752  * @data: a pointer to the requested attribute data
1753  */
1754 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1755                                  enum IOMMUMemoryRegionAttr attr,
1756                                  void *data);
1757 
1758 /**
1759  * memory_region_iommu_attrs_to_index: return the IOMMU index to
1760  * use for translations with the given memory transaction attributes.
1761  *
1762  * @iommu_mr: the memory region
1763  * @attrs: the memory transaction attributes
1764  */
1765 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1766                                        MemTxAttrs attrs);
1767 
1768 /**
1769  * memory_region_iommu_num_indexes: return the total number of IOMMU
1770  * indexes that this IOMMU supports.
1771  *
1772  * @iommu_mr: the memory region
1773  */
1774 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr);
1775 
1776 /**
1777  * memory_region_iommu_set_page_size_mask: set the supported page
1778  * sizes for a given IOMMU memory region
1779  *
1780  * @iommu_mr: IOMMU memory region
1781  * @page_size_mask: supported page size mask
1782  * @errp: pointer to Error*, to store an error if it happens.
1783  */
1784 int memory_region_iommu_set_page_size_mask(IOMMUMemoryRegion *iommu_mr,
1785                                            uint64_t page_size_mask,
1786                                            Error **errp);
1787 
1788 /**
1789  * memory_region_name: get a memory region's name
1790  *
1791  * Returns the string that was used to initialize the memory region.
1792  *
1793  * @mr: the memory region being queried
1794  */
1795 const char *memory_region_name(const MemoryRegion *mr);
1796 
1797 /**
1798  * memory_region_is_logging: return whether a memory region is logging writes
1799  *
1800  * Returns %true if the memory region is logging writes for the given client
1801  *
1802  * @mr: the memory region being queried
1803  * @client: the client being queried
1804  */
1805 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client);
1806 
1807 /**
1808  * memory_region_get_dirty_log_mask: return the clients for which a
1809  * memory region is logging writes.
1810  *
1811  * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants
1812  * are the bit indices.
1813  *
1814  * @mr: the memory region being queried
1815  */
1816 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr);
1817 
1818 /**
1819  * memory_region_is_rom: check whether a memory region is ROM
1820  *
1821  * Returns %true if a memory region is read-only memory.
1822  *
1823  * @mr: the memory region being queried
1824  */
1825 static inline bool memory_region_is_rom(MemoryRegion *mr)
1826 {
1827     return mr->ram && mr->readonly;
1828 }
1829 
1830 /**
1831  * memory_region_is_nonvolatile: check whether a memory region is non-volatile
1832  *
1833  * Returns %true is a memory region is non-volatile memory.
1834  *
1835  * @mr: the memory region being queried
1836  */
1837 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr)
1838 {
1839     return mr->nonvolatile;
1840 }
1841 
1842 /**
1843  * memory_region_get_fd: Get a file descriptor backing a RAM memory region.
1844  *
1845  * Returns a file descriptor backing a file-based RAM memory region,
1846  * or -1 if the region is not a file-based RAM memory region.
1847  *
1848  * @mr: the RAM or alias memory region being queried.
1849  */
1850 int memory_region_get_fd(MemoryRegion *mr);
1851 
1852 /**
1853  * memory_region_from_host: Convert a pointer into a RAM memory region
1854  * and an offset within it.
1855  *
1856  * Given a host pointer inside a RAM memory region (created with
1857  * memory_region_init_ram() or memory_region_init_ram_ptr()), return
1858  * the MemoryRegion and the offset within it.
1859  *
1860  * Use with care; by the time this function returns, the returned pointer is
1861  * not protected by RCU anymore.  If the caller is not within an RCU critical
1862  * section and does not hold the iothread lock, it must have other means of
1863  * protecting the pointer, such as a reference to the region that includes
1864  * the incoming ram_addr_t.
1865  *
1866  * @ptr: the host pointer to be converted
1867  * @offset: the offset within memory region
1868  */
1869 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset);
1870 
1871 /**
1872  * memory_region_get_ram_ptr: Get a pointer into a RAM memory region.
1873  *
1874  * Returns a host pointer to a RAM memory region (created with
1875  * memory_region_init_ram() or memory_region_init_ram_ptr()).
1876  *
1877  * Use with care; by the time this function returns, the returned pointer is
1878  * not protected by RCU anymore.  If the caller is not within an RCU critical
1879  * section and does not hold the iothread lock, it must have other means of
1880  * protecting the pointer, such as a reference to the region that includes
1881  * the incoming ram_addr_t.
1882  *
1883  * @mr: the memory region being queried.
1884  */
1885 void *memory_region_get_ram_ptr(MemoryRegion *mr);
1886 
1887 /* memory_region_ram_resize: Resize a RAM region.
1888  *
1889  * Resizing RAM while migrating can result in the migration being canceled.
1890  * Care has to be taken if the guest might have already detected the memory.
1891  *
1892  * @mr: a memory region created with @memory_region_init_resizeable_ram.
1893  * @newsize: the new size the region
1894  * @errp: pointer to Error*, to store an error if it happens.
1895  */
1896 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize,
1897                               Error **errp);
1898 
1899 /**
1900  * memory_region_msync: Synchronize selected address range of
1901  * a memory mapped region
1902  *
1903  * @mr: the memory region to be msync
1904  * @addr: the initial address of the range to be sync
1905  * @size: the size of the range to be sync
1906  */
1907 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size);
1908 
1909 /**
1910  * memory_region_writeback: Trigger cache writeback for
1911  * selected address range
1912  *
1913  * @mr: the memory region to be updated
1914  * @addr: the initial address of the range to be written back
1915  * @size: the size of the range to be written back
1916  */
1917 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size);
1918 
1919 /**
1920  * memory_region_set_log: Turn dirty logging on or off for a region.
1921  *
1922  * Turns dirty logging on or off for a specified client (display, migration).
1923  * Only meaningful for RAM regions.
1924  *
1925  * @mr: the memory region being updated.
1926  * @log: whether dirty logging is to be enabled or disabled.
1927  * @client: the user of the logging information; %DIRTY_MEMORY_VGA only.
1928  */
1929 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client);
1930 
1931 /**
1932  * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region.
1933  *
1934  * Marks a range of bytes as dirty, after it has been dirtied outside
1935  * guest code.
1936  *
1937  * @mr: the memory region being dirtied.
1938  * @addr: the address (relative to the start of the region) being dirtied.
1939  * @size: size of the range being dirtied.
1940  */
1941 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1942                              hwaddr size);
1943 
1944 /**
1945  * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range
1946  *
1947  * This function is called when the caller wants to clear the remote
1948  * dirty bitmap of a memory range within the memory region.  This can
1949  * be used by e.g. KVM to manually clear dirty log when
1950  * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host
1951  * kernel.
1952  *
1953  * @mr:     the memory region to clear the dirty log upon
1954  * @start:  start address offset within the memory region
1955  * @len:    length of the memory region to clear dirty bitmap
1956  */
1957 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
1958                                       hwaddr len);
1959 
1960 /**
1961  * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty
1962  *                                         bitmap and clear it.
1963  *
1964  * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and
1965  * returns the snapshot.  The snapshot can then be used to query dirty
1966  * status, using memory_region_snapshot_get_dirty.  Snapshotting allows
1967  * querying the same page multiple times, which is especially useful for
1968  * display updates where the scanlines often are not page aligned.
1969  *
1970  * The dirty bitmap region which gets copyed into the snapshot (and
1971  * cleared afterwards) can be larger than requested.  The boundaries
1972  * are rounded up/down so complete bitmap longs (covering 64 pages on
1973  * 64bit hosts) can be copied over into the bitmap snapshot.  Which
1974  * isn't a problem for display updates as the extra pages are outside
1975  * the visible area, and in case the visible area changes a full
1976  * display redraw is due anyway.  Should other use cases for this
1977  * function emerge we might have to revisit this implementation
1978  * detail.
1979  *
1980  * Use g_free to release DirtyBitmapSnapshot.
1981  *
1982  * @mr: the memory region being queried.
1983  * @addr: the address (relative to the start of the region) being queried.
1984  * @size: the size of the range being queried.
1985  * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA.
1986  */
1987 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
1988                                                             hwaddr addr,
1989                                                             hwaddr size,
1990                                                             unsigned client);
1991 
1992 /**
1993  * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty
1994  *                                   in the specified dirty bitmap snapshot.
1995  *
1996  * @mr: the memory region being queried.
1997  * @snap: the dirty bitmap snapshot
1998  * @addr: the address (relative to the start of the region) being queried.
1999  * @size: the size of the range being queried.
2000  */
2001 bool memory_region_snapshot_get_dirty(MemoryRegion *mr,
2002                                       DirtyBitmapSnapshot *snap,
2003                                       hwaddr addr, hwaddr size);
2004 
2005 /**
2006  * memory_region_reset_dirty: Mark a range of pages as clean, for a specified
2007  *                            client.
2008  *
2009  * Marks a range of pages as no longer dirty.
2010  *
2011  * @mr: the region being updated.
2012  * @addr: the start of the subrange being cleaned.
2013  * @size: the size of the subrange being cleaned.
2014  * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or
2015  *          %DIRTY_MEMORY_VGA.
2016  */
2017 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
2018                                hwaddr size, unsigned client);
2019 
2020 /**
2021  * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate
2022  *                                 TBs (for self-modifying code).
2023  *
2024  * The MemoryRegionOps->write() callback of a ROM device must use this function
2025  * to mark byte ranges that have been modified internally, such as by directly
2026  * accessing the memory returned by memory_region_get_ram_ptr().
2027  *
2028  * This function marks the range dirty and invalidates TBs so that TCG can
2029  * detect self-modifying code.
2030  *
2031  * @mr: the region being flushed.
2032  * @addr: the start, relative to the start of the region, of the range being
2033  *        flushed.
2034  * @size: the size, in bytes, of the range being flushed.
2035  */
2036 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size);
2037 
2038 /**
2039  * memory_region_set_readonly: Turn a memory region read-only (or read-write)
2040  *
2041  * Allows a memory region to be marked as read-only (turning it into a ROM).
2042  * only useful on RAM regions.
2043  *
2044  * @mr: the region being updated.
2045  * @readonly: whether rhe region is to be ROM or RAM.
2046  */
2047 void memory_region_set_readonly(MemoryRegion *mr, bool readonly);
2048 
2049 /**
2050  * memory_region_set_nonvolatile: Turn a memory region non-volatile
2051  *
2052  * Allows a memory region to be marked as non-volatile.
2053  * only useful on RAM regions.
2054  *
2055  * @mr: the region being updated.
2056  * @nonvolatile: whether rhe region is to be non-volatile.
2057  */
2058 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile);
2059 
2060 /**
2061  * memory_region_rom_device_set_romd: enable/disable ROMD mode
2062  *
2063  * Allows a ROM device (initialized with memory_region_init_rom_device() to
2064  * set to ROMD mode (default) or MMIO mode.  When it is in ROMD mode, the
2065  * device is mapped to guest memory and satisfies read access directly.
2066  * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function.
2067  * Writes are always handled by the #MemoryRegion.write function.
2068  *
2069  * @mr: the memory region to be updated
2070  * @romd_mode: %true to put the region into ROMD mode
2071  */
2072 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode);
2073 
2074 /**
2075  * memory_region_set_coalescing: Enable memory coalescing for the region.
2076  *
2077  * Enabled writes to a region to be queued for later processing. MMIO ->write
2078  * callbacks may be delayed until a non-coalesced MMIO is issued.
2079  * Only useful for IO regions.  Roughly similar to write-combining hardware.
2080  *
2081  * @mr: the memory region to be write coalesced
2082  */
2083 void memory_region_set_coalescing(MemoryRegion *mr);
2084 
2085 /**
2086  * memory_region_add_coalescing: Enable memory coalescing for a sub-range of
2087  *                               a region.
2088  *
2089  * Like memory_region_set_coalescing(), but works on a sub-range of a region.
2090  * Multiple calls can be issued coalesced disjoint ranges.
2091  *
2092  * @mr: the memory region to be updated.
2093  * @offset: the start of the range within the region to be coalesced.
2094  * @size: the size of the subrange to be coalesced.
2095  */
2096 void memory_region_add_coalescing(MemoryRegion *mr,
2097                                   hwaddr offset,
2098                                   uint64_t size);
2099 
2100 /**
2101  * memory_region_clear_coalescing: Disable MMIO coalescing for the region.
2102  *
2103  * Disables any coalescing caused by memory_region_set_coalescing() or
2104  * memory_region_add_coalescing().  Roughly equivalent to uncacheble memory
2105  * hardware.
2106  *
2107  * @mr: the memory region to be updated.
2108  */
2109 void memory_region_clear_coalescing(MemoryRegion *mr);
2110 
2111 /**
2112  * memory_region_set_flush_coalesced: Enforce memory coalescing flush before
2113  *                                    accesses.
2114  *
2115  * Ensure that pending coalesced MMIO request are flushed before the memory
2116  * region is accessed. This property is automatically enabled for all regions
2117  * passed to memory_region_set_coalescing() and memory_region_add_coalescing().
2118  *
2119  * @mr: the memory region to be updated.
2120  */
2121 void memory_region_set_flush_coalesced(MemoryRegion *mr);
2122 
2123 /**
2124  * memory_region_clear_flush_coalesced: Disable memory coalescing flush before
2125  *                                      accesses.
2126  *
2127  * Clear the automatic coalesced MMIO flushing enabled via
2128  * memory_region_set_flush_coalesced. Note that this service has no effect on
2129  * memory regions that have MMIO coalescing enabled for themselves. For them,
2130  * automatic flushing will stop once coalescing is disabled.
2131  *
2132  * @mr: the memory region to be updated.
2133  */
2134 void memory_region_clear_flush_coalesced(MemoryRegion *mr);
2135 
2136 /**
2137  * memory_region_add_eventfd: Request an eventfd to be triggered when a word
2138  *                            is written to a location.
2139  *
2140  * Marks a word in an IO region (initialized with memory_region_init_io())
2141  * as a trigger for an eventfd event.  The I/O callback will not be called.
2142  * The caller must be prepared to handle failure (that is, take the required
2143  * action if the callback _is_ called).
2144  *
2145  * @mr: the memory region being updated.
2146  * @addr: the address within @mr that is to be monitored
2147  * @size: the size of the access to trigger the eventfd
2148  * @match_data: whether to match against @data, instead of just @addr
2149  * @data: the data to match against the guest write
2150  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2151  **/
2152 void memory_region_add_eventfd(MemoryRegion *mr,
2153                                hwaddr addr,
2154                                unsigned size,
2155                                bool match_data,
2156                                uint64_t data,
2157                                EventNotifier *e);
2158 
2159 /**
2160  * memory_region_del_eventfd: Cancel an eventfd.
2161  *
2162  * Cancels an eventfd trigger requested by a previous
2163  * memory_region_add_eventfd() call.
2164  *
2165  * @mr: the memory region being updated.
2166  * @addr: the address within @mr that is to be monitored
2167  * @size: the size of the access to trigger the eventfd
2168  * @match_data: whether to match against @data, instead of just @addr
2169  * @data: the data to match against the guest write
2170  * @e: event notifier to be triggered when @addr, @size, and @data all match.
2171  */
2172 void memory_region_del_eventfd(MemoryRegion *mr,
2173                                hwaddr addr,
2174                                unsigned size,
2175                                bool match_data,
2176                                uint64_t data,
2177                                EventNotifier *e);
2178 
2179 /**
2180  * memory_region_add_subregion: Add a subregion to a container.
2181  *
2182  * Adds a subregion at @offset.  The subregion may not overlap with other
2183  * subregions (except for those explicitly marked as overlapping).  A region
2184  * may only be added once as a subregion (unless removed with
2185  * memory_region_del_subregion()); use memory_region_init_alias() if you
2186  * want a region to be a subregion in multiple locations.
2187  *
2188  * @mr: the region to contain the new subregion; must be a container
2189  *      initialized with memory_region_init().
2190  * @offset: the offset relative to @mr where @subregion is added.
2191  * @subregion: the subregion to be added.
2192  */
2193 void memory_region_add_subregion(MemoryRegion *mr,
2194                                  hwaddr offset,
2195                                  MemoryRegion *subregion);
2196 /**
2197  * memory_region_add_subregion_overlap: Add a subregion to a container
2198  *                                      with overlap.
2199  *
2200  * Adds a subregion at @offset.  The subregion may overlap with other
2201  * subregions.  Conflicts are resolved by having a higher @priority hide a
2202  * lower @priority. Subregions without priority are taken as @priority 0.
2203  * A region may only be added once as a subregion (unless removed with
2204  * memory_region_del_subregion()); use memory_region_init_alias() if you
2205  * want a region to be a subregion in multiple locations.
2206  *
2207  * @mr: the region to contain the new subregion; must be a container
2208  *      initialized with memory_region_init().
2209  * @offset: the offset relative to @mr where @subregion is added.
2210  * @subregion: the subregion to be added.
2211  * @priority: used for resolving overlaps; highest priority wins.
2212  */
2213 void memory_region_add_subregion_overlap(MemoryRegion *mr,
2214                                          hwaddr offset,
2215                                          MemoryRegion *subregion,
2216                                          int priority);
2217 
2218 /**
2219  * memory_region_get_ram_addr: Get the ram address associated with a memory
2220  *                             region
2221  *
2222  * @mr: the region to be queried
2223  */
2224 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr);
2225 
2226 uint64_t memory_region_get_alignment(const MemoryRegion *mr);
2227 /**
2228  * memory_region_del_subregion: Remove a subregion.
2229  *
2230  * Removes a subregion from its container.
2231  *
2232  * @mr: the container to be updated.
2233  * @subregion: the region being removed; must be a current subregion of @mr.
2234  */
2235 void memory_region_del_subregion(MemoryRegion *mr,
2236                                  MemoryRegion *subregion);
2237 
2238 /*
2239  * memory_region_set_enabled: dynamically enable or disable a region
2240  *
2241  * Enables or disables a memory region.  A disabled memory region
2242  * ignores all accesses to itself and its subregions.  It does not
2243  * obscure sibling subregions with lower priority - it simply behaves as
2244  * if it was removed from the hierarchy.
2245  *
2246  * Regions default to being enabled.
2247  *
2248  * @mr: the region to be updated
2249  * @enabled: whether to enable or disable the region
2250  */
2251 void memory_region_set_enabled(MemoryRegion *mr, bool enabled);
2252 
2253 /*
2254  * memory_region_set_address: dynamically update the address of a region
2255  *
2256  * Dynamically updates the address of a region, relative to its container.
2257  * May be used on regions are currently part of a memory hierarchy.
2258  *
2259  * @mr: the region to be updated
2260  * @addr: new address, relative to container region
2261  */
2262 void memory_region_set_address(MemoryRegion *mr, hwaddr addr);
2263 
2264 /*
2265  * memory_region_set_size: dynamically update the size of a region.
2266  *
2267  * Dynamically updates the size of a region.
2268  *
2269  * @mr: the region to be updated
2270  * @size: used size of the region.
2271  */
2272 void memory_region_set_size(MemoryRegion *mr, uint64_t size);
2273 
2274 /*
2275  * memory_region_set_alias_offset: dynamically update a memory alias's offset
2276  *
2277  * Dynamically updates the offset into the target region that an alias points
2278  * to, as if the fourth argument to memory_region_init_alias() has changed.
2279  *
2280  * @mr: the #MemoryRegion to be updated; should be an alias.
2281  * @offset: the new offset into the target memory region
2282  */
2283 void memory_region_set_alias_offset(MemoryRegion *mr,
2284                                     hwaddr offset);
2285 
2286 /**
2287  * memory_region_present: checks if an address relative to a @container
2288  * translates into #MemoryRegion within @container
2289  *
2290  * Answer whether a #MemoryRegion within @container covers the address
2291  * @addr.
2292  *
2293  * @container: a #MemoryRegion within which @addr is a relative address
2294  * @addr: the area within @container to be searched
2295  */
2296 bool memory_region_present(MemoryRegion *container, hwaddr addr);
2297 
2298 /**
2299  * memory_region_is_mapped: returns true if #MemoryRegion is mapped
2300  * into another memory region, which does not necessarily imply that it is
2301  * mapped into an address space.
2302  *
2303  * @mr: a #MemoryRegion which should be checked if it's mapped
2304  */
2305 bool memory_region_is_mapped(MemoryRegion *mr);
2306 
2307 /**
2308  * memory_region_get_ram_discard_manager: get the #RamDiscardManager for a
2309  * #MemoryRegion
2310  *
2311  * The #RamDiscardManager cannot change while a memory region is mapped.
2312  *
2313  * @mr: the #MemoryRegion
2314  */
2315 RamDiscardManager *memory_region_get_ram_discard_manager(MemoryRegion *mr);
2316 
2317 /**
2318  * memory_region_has_ram_discard_manager: check whether a #MemoryRegion has a
2319  * #RamDiscardManager assigned
2320  *
2321  * @mr: the #MemoryRegion
2322  */
2323 static inline bool memory_region_has_ram_discard_manager(MemoryRegion *mr)
2324 {
2325     return !!memory_region_get_ram_discard_manager(mr);
2326 }
2327 
2328 /**
2329  * memory_region_set_ram_discard_manager: set the #RamDiscardManager for a
2330  * #MemoryRegion
2331  *
2332  * This function must not be called for a mapped #MemoryRegion, a #MemoryRegion
2333  * that does not cover RAM, or a #MemoryRegion that already has a
2334  * #RamDiscardManager assigned.
2335  *
2336  * @mr: the #MemoryRegion
2337  * @rdm: #RamDiscardManager to set
2338  */
2339 void memory_region_set_ram_discard_manager(MemoryRegion *mr,
2340                                            RamDiscardManager *rdm);
2341 
2342 /**
2343  * memory_region_find: translate an address/size relative to a
2344  * MemoryRegion into a #MemoryRegionSection.
2345  *
2346  * Locates the first #MemoryRegion within @mr that overlaps the range
2347  * given by @addr and @size.
2348  *
2349  * Returns a #MemoryRegionSection that describes a contiguous overlap.
2350  * It will have the following characteristics:
2351  * - @size = 0 iff no overlap was found
2352  * - @mr is non-%NULL iff an overlap was found
2353  *
2354  * Remember that in the return value the @offset_within_region is
2355  * relative to the returned region (in the .@mr field), not to the
2356  * @mr argument.
2357  *
2358  * Similarly, the .@offset_within_address_space is relative to the
2359  * address space that contains both regions, the passed and the
2360  * returned one.  However, in the special case where the @mr argument
2361  * has no container (and thus is the root of the address space), the
2362  * following will hold:
2363  * - @offset_within_address_space >= @addr
2364  * - @offset_within_address_space + .@size <= @addr + @size
2365  *
2366  * @mr: a MemoryRegion within which @addr is a relative address
2367  * @addr: start of the area within @as to be searched
2368  * @size: size of the area to be searched
2369  */
2370 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2371                                        hwaddr addr, uint64_t size);
2372 
2373 /**
2374  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2375  *
2376  * Synchronizes the dirty page log for all address spaces.
2377  */
2378 void memory_global_dirty_log_sync(void);
2379 
2380 /**
2381  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
2382  *
2383  * Synchronizes the vCPUs with a thread that is reading the dirty bitmap.
2384  * This function must be called after the dirty log bitmap is cleared, and
2385  * before dirty guest memory pages are read.  If you are using
2386  * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes
2387  * care of doing this.
2388  */
2389 void memory_global_after_dirty_log_sync(void);
2390 
2391 /**
2392  * memory_region_transaction_begin: Start a transaction.
2393  *
2394  * During a transaction, changes will be accumulated and made visible
2395  * only when the transaction ends (is committed).
2396  */
2397 void memory_region_transaction_begin(void);
2398 
2399 /**
2400  * memory_region_transaction_commit: Commit a transaction and make changes
2401  *                                   visible to the guest.
2402  */
2403 void memory_region_transaction_commit(void);
2404 
2405 /**
2406  * memory_listener_register: register callbacks to be called when memory
2407  *                           sections are mapped or unmapped into an address
2408  *                           space
2409  *
2410  * @listener: an object containing the callbacks to be called
2411  * @filter: if non-%NULL, only regions in this address space will be observed
2412  */
2413 void memory_listener_register(MemoryListener *listener, AddressSpace *filter);
2414 
2415 /**
2416  * memory_listener_unregister: undo the effect of memory_listener_register()
2417  *
2418  * @listener: an object containing the callbacks to be removed
2419  */
2420 void memory_listener_unregister(MemoryListener *listener);
2421 
2422 /**
2423  * memory_global_dirty_log_start: begin dirty logging for all regions
2424  *
2425  * @flags: purpose of starting dirty log, migration or dirty rate
2426  */
2427 void memory_global_dirty_log_start(unsigned int flags);
2428 
2429 /**
2430  * memory_global_dirty_log_stop: end dirty logging for all regions
2431  *
2432  * @flags: purpose of stopping dirty log, migration or dirty rate
2433  */
2434 void memory_global_dirty_log_stop(unsigned int flags);
2435 
2436 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled);
2437 
2438 /**
2439  * memory_region_dispatch_read: perform a read directly to the specified
2440  * MemoryRegion.
2441  *
2442  * @mr: #MemoryRegion to access
2443  * @addr: address within that region
2444  * @pval: pointer to uint64_t which the data is written to
2445  * @op: size, sign, and endianness of the memory operation
2446  * @attrs: memory transaction attributes to use for the access
2447  */
2448 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
2449                                         hwaddr addr,
2450                                         uint64_t *pval,
2451                                         MemOp op,
2452                                         MemTxAttrs attrs);
2453 /**
2454  * memory_region_dispatch_write: perform a write directly to the specified
2455  * MemoryRegion.
2456  *
2457  * @mr: #MemoryRegion to access
2458  * @addr: address within that region
2459  * @data: data to write
2460  * @op: size, sign, and endianness of the memory operation
2461  * @attrs: memory transaction attributes to use for the access
2462  */
2463 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
2464                                          hwaddr addr,
2465                                          uint64_t data,
2466                                          MemOp op,
2467                                          MemTxAttrs attrs);
2468 
2469 /**
2470  * address_space_init: initializes an address space
2471  *
2472  * @as: an uninitialized #AddressSpace
2473  * @root: a #MemoryRegion that routes addresses for the address space
2474  * @name: an address space name.  The name is only used for debugging
2475  *        output.
2476  */
2477 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name);
2478 
2479 /**
2480  * address_space_destroy: destroy an address space
2481  *
2482  * Releases all resources associated with an address space.  After an address space
2483  * is destroyed, its root memory region (given by address_space_init()) may be destroyed
2484  * as well.
2485  *
2486  * @as: address space to be destroyed
2487  */
2488 void address_space_destroy(AddressSpace *as);
2489 
2490 /**
2491  * address_space_remove_listeners: unregister all listeners of an address space
2492  *
2493  * Removes all callbacks previously registered with memory_listener_register()
2494  * for @as.
2495  *
2496  * @as: an initialized #AddressSpace
2497  */
2498 void address_space_remove_listeners(AddressSpace *as);
2499 
2500 /**
2501  * address_space_rw: read from or write to an address space.
2502  *
2503  * Return a MemTxResult indicating whether the operation succeeded
2504  * or failed (eg unassigned memory, device rejected the transaction,
2505  * IOMMU fault).
2506  *
2507  * @as: #AddressSpace to be accessed
2508  * @addr: address within that address space
2509  * @attrs: memory transaction attributes
2510  * @buf: buffer with the data transferred
2511  * @len: the number of bytes to read or write
2512  * @is_write: indicates the transfer direction
2513  */
2514 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
2515                              MemTxAttrs attrs, void *buf,
2516                              hwaddr len, bool is_write);
2517 
2518 /**
2519  * address_space_write: write to address space.
2520  *
2521  * Return a MemTxResult indicating whether the operation succeeded
2522  * or failed (eg unassigned memory, device rejected the transaction,
2523  * IOMMU fault).
2524  *
2525  * @as: #AddressSpace to be accessed
2526  * @addr: address within that address space
2527  * @attrs: memory transaction attributes
2528  * @buf: buffer with the data transferred
2529  * @len: the number of bytes to write
2530  */
2531 MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
2532                                 MemTxAttrs attrs,
2533                                 const void *buf, hwaddr len);
2534 
2535 /**
2536  * address_space_write_rom: write to address space, including ROM.
2537  *
2538  * This function writes to the specified address space, but will
2539  * write data to both ROM and RAM. This is used for non-guest
2540  * writes like writes from the gdb debug stub or initial loading
2541  * of ROM contents.
2542  *
2543  * Note that portions of the write which attempt to write data to
2544  * a device will be silently ignored -- only real RAM and ROM will
2545  * be written to.
2546  *
2547  * Return a MemTxResult indicating whether the operation succeeded
2548  * or failed (eg unassigned memory, device rejected the transaction,
2549  * IOMMU fault).
2550  *
2551  * @as: #AddressSpace to be accessed
2552  * @addr: address within that address space
2553  * @attrs: memory transaction attributes
2554  * @buf: buffer with the data transferred
2555  * @len: the number of bytes to write
2556  */
2557 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr,
2558                                     MemTxAttrs attrs,
2559                                     const void *buf, hwaddr len);
2560 
2561 /* address_space_ld*: load from an address space
2562  * address_space_st*: store to an address space
2563  *
2564  * These functions perform a load or store of the byte, word,
2565  * longword or quad to the specified address within the AddressSpace.
2566  * The _le suffixed functions treat the data as little endian;
2567  * _be indicates big endian; no suffix indicates "same endianness
2568  * as guest CPU".
2569  *
2570  * The "guest CPU endianness" accessors are deprecated for use outside
2571  * target-* code; devices should be CPU-agnostic and use either the LE
2572  * or the BE accessors.
2573  *
2574  * @as #AddressSpace to be accessed
2575  * @addr: address within that address space
2576  * @val: data value, for stores
2577  * @attrs: memory transaction attributes
2578  * @result: location to write the success/failure of the transaction;
2579  *   if NULL, this information is discarded
2580  */
2581 
2582 #define SUFFIX
2583 #define ARG1         as
2584 #define ARG1_DECL    AddressSpace *as
2585 #include "exec/memory_ldst.h.inc"
2586 
2587 #define SUFFIX
2588 #define ARG1         as
2589 #define ARG1_DECL    AddressSpace *as
2590 #include "exec/memory_ldst_phys.h.inc"
2591 
2592 struct MemoryRegionCache {
2593     void *ptr;
2594     hwaddr xlat;
2595     hwaddr len;
2596     FlatView *fv;
2597     MemoryRegionSection mrs;
2598     bool is_write;
2599 };
2600 
2601 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL })
2602 
2603 
2604 /* address_space_ld*_cached: load from a cached #MemoryRegion
2605  * address_space_st*_cached: store into a cached #MemoryRegion
2606  *
2607  * These functions perform a load or store of the byte, word,
2608  * longword or quad to the specified address.  The address is
2609  * a physical address in the AddressSpace, but it must lie within
2610  * a #MemoryRegion that was mapped with address_space_cache_init.
2611  *
2612  * The _le suffixed functions treat the data as little endian;
2613  * _be indicates big endian; no suffix indicates "same endianness
2614  * as guest CPU".
2615  *
2616  * The "guest CPU endianness" accessors are deprecated for use outside
2617  * target-* code; devices should be CPU-agnostic and use either the LE
2618  * or the BE accessors.
2619  *
2620  * @cache: previously initialized #MemoryRegionCache to be accessed
2621  * @addr: address within the address space
2622  * @val: data value, for stores
2623  * @attrs: memory transaction attributes
2624  * @result: location to write the success/failure of the transaction;
2625  *   if NULL, this information is discarded
2626  */
2627 
2628 #define SUFFIX       _cached_slow
2629 #define ARG1         cache
2630 #define ARG1_DECL    MemoryRegionCache *cache
2631 #include "exec/memory_ldst.h.inc"
2632 
2633 /* Inline fast path for direct RAM access.  */
2634 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache,
2635     hwaddr addr, MemTxAttrs attrs, MemTxResult *result)
2636 {
2637     assert(addr < cache->len);
2638     if (likely(cache->ptr)) {
2639         return ldub_p(cache->ptr + addr);
2640     } else {
2641         return address_space_ldub_cached_slow(cache, addr, attrs, result);
2642     }
2643 }
2644 
2645 static inline void address_space_stb_cached(MemoryRegionCache *cache,
2646     hwaddr addr, uint8_t val, MemTxAttrs attrs, MemTxResult *result)
2647 {
2648     assert(addr < cache->len);
2649     if (likely(cache->ptr)) {
2650         stb_p(cache->ptr + addr, val);
2651     } else {
2652         address_space_stb_cached_slow(cache, addr, val, attrs, result);
2653     }
2654 }
2655 
2656 #define ENDIANNESS   _le
2657 #include "exec/memory_ldst_cached.h.inc"
2658 
2659 #define ENDIANNESS   _be
2660 #include "exec/memory_ldst_cached.h.inc"
2661 
2662 #define SUFFIX       _cached
2663 #define ARG1         cache
2664 #define ARG1_DECL    MemoryRegionCache *cache
2665 #include "exec/memory_ldst_phys.h.inc"
2666 
2667 /* address_space_cache_init: prepare for repeated access to a physical
2668  * memory region
2669  *
2670  * @cache: #MemoryRegionCache to be filled
2671  * @as: #AddressSpace to be accessed
2672  * @addr: address within that address space
2673  * @len: length of buffer
2674  * @is_write: indicates the transfer direction
2675  *
2676  * Will only work with RAM, and may map a subset of the requested range by
2677  * returning a value that is less than @len.  On failure, return a negative
2678  * errno value.
2679  *
2680  * Because it only works with RAM, this function can be used for
2681  * read-modify-write operations.  In this case, is_write should be %true.
2682  *
2683  * Note that addresses passed to the address_space_*_cached functions
2684  * are relative to @addr.
2685  */
2686 int64_t address_space_cache_init(MemoryRegionCache *cache,
2687                                  AddressSpace *as,
2688                                  hwaddr addr,
2689                                  hwaddr len,
2690                                  bool is_write);
2691 
2692 /**
2693  * address_space_cache_invalidate: complete a write to a #MemoryRegionCache
2694  *
2695  * @cache: The #MemoryRegionCache to operate on.
2696  * @addr: The first physical address that was written, relative to the
2697  * address that was passed to @address_space_cache_init.
2698  * @access_len: The number of bytes that were written starting at @addr.
2699  */
2700 void address_space_cache_invalidate(MemoryRegionCache *cache,
2701                                     hwaddr addr,
2702                                     hwaddr access_len);
2703 
2704 /**
2705  * address_space_cache_destroy: free a #MemoryRegionCache
2706  *
2707  * @cache: The #MemoryRegionCache whose memory should be released.
2708  */
2709 void address_space_cache_destroy(MemoryRegionCache *cache);
2710 
2711 /* address_space_get_iotlb_entry: translate an address into an IOTLB
2712  * entry. Should be called from an RCU critical section.
2713  */
2714 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
2715                                             bool is_write, MemTxAttrs attrs);
2716 
2717 /* address_space_translate: translate an address range into an address space
2718  * into a MemoryRegion and an address range into that section.  Should be
2719  * called from an RCU critical section, to avoid that the last reference
2720  * to the returned region disappears after address_space_translate returns.
2721  *
2722  * @fv: #FlatView to be accessed
2723  * @addr: address within that address space
2724  * @xlat: pointer to address within the returned memory region section's
2725  * #MemoryRegion.
2726  * @len: pointer to length
2727  * @is_write: indicates the transfer direction
2728  * @attrs: memory attributes
2729  */
2730 MemoryRegion *flatview_translate(FlatView *fv,
2731                                  hwaddr addr, hwaddr *xlat,
2732                                  hwaddr *len, bool is_write,
2733                                  MemTxAttrs attrs);
2734 
2735 static inline MemoryRegion *address_space_translate(AddressSpace *as,
2736                                                     hwaddr addr, hwaddr *xlat,
2737                                                     hwaddr *len, bool is_write,
2738                                                     MemTxAttrs attrs)
2739 {
2740     return flatview_translate(address_space_to_flatview(as),
2741                               addr, xlat, len, is_write, attrs);
2742 }
2743 
2744 /* address_space_access_valid: check for validity of accessing an address
2745  * space range
2746  *
2747  * Check whether memory is assigned to the given address space range, and
2748  * access is permitted by any IOMMU regions that are active for the address
2749  * space.
2750  *
2751  * For now, addr and len should be aligned to a page size.  This limitation
2752  * will be lifted in the future.
2753  *
2754  * @as: #AddressSpace to be accessed
2755  * @addr: address within that address space
2756  * @len: length of the area to be checked
2757  * @is_write: indicates the transfer direction
2758  * @attrs: memory attributes
2759  */
2760 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len,
2761                                 bool is_write, MemTxAttrs attrs);
2762 
2763 /* address_space_map: map a physical memory region into a host virtual address
2764  *
2765  * May map a subset of the requested range, given by and returned in @plen.
2766  * May return %NULL and set *@plen to zero(0), if resources needed to perform
2767  * the mapping are exhausted.
2768  * Use only for reads OR writes - not for read-modify-write operations.
2769  * Use cpu_register_map_client() to know when retrying the map operation is
2770  * likely to succeed.
2771  *
2772  * @as: #AddressSpace to be accessed
2773  * @addr: address within that address space
2774  * @plen: pointer to length of buffer; updated on return
2775  * @is_write: indicates the transfer direction
2776  * @attrs: memory attributes
2777  */
2778 void *address_space_map(AddressSpace *as, hwaddr addr,
2779                         hwaddr *plen, bool is_write, MemTxAttrs attrs);
2780 
2781 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map()
2782  *
2783  * Will also mark the memory as dirty if @is_write == %true.  @access_len gives
2784  * the amount of memory that was actually read or written by the caller.
2785  *
2786  * @as: #AddressSpace used
2787  * @buffer: host pointer as returned by address_space_map()
2788  * @len: buffer length as returned by address_space_map()
2789  * @access_len: amount of data actually transferred
2790  * @is_write: indicates the transfer direction
2791  */
2792 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2793                          bool is_write, hwaddr access_len);
2794 
2795 
2796 /* Internal functions, part of the implementation of address_space_read.  */
2797 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
2798                                     MemTxAttrs attrs, void *buf, hwaddr len);
2799 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
2800                                    MemTxAttrs attrs, void *buf,
2801                                    hwaddr len, hwaddr addr1, hwaddr l,
2802                                    MemoryRegion *mr);
2803 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr);
2804 
2805 /* Internal functions, part of the implementation of address_space_read_cached
2806  * and address_space_write_cached.  */
2807 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache,
2808                                            hwaddr addr, void *buf, hwaddr len);
2809 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache,
2810                                             hwaddr addr, const void *buf,
2811                                             hwaddr len);
2812 
2813 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
2814 {
2815     if (is_write) {
2816         return memory_region_is_ram(mr) && !mr->readonly &&
2817                !mr->rom_device && !memory_region_is_ram_device(mr);
2818     } else {
2819         return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) ||
2820                memory_region_is_romd(mr);
2821     }
2822 }
2823 
2824 /**
2825  * address_space_read: read from an address space.
2826  *
2827  * Return a MemTxResult indicating whether the operation succeeded
2828  * or failed (eg unassigned memory, device rejected the transaction,
2829  * IOMMU fault).  Called within RCU critical section.
2830  *
2831  * @as: #AddressSpace to be accessed
2832  * @addr: address within that address space
2833  * @attrs: memory transaction attributes
2834  * @buf: buffer with the data transferred
2835  * @len: length of the data transferred
2836  */
2837 static inline __attribute__((__always_inline__))
2838 MemTxResult address_space_read(AddressSpace *as, hwaddr addr,
2839                                MemTxAttrs attrs, void *buf,
2840                                hwaddr len)
2841 {
2842     MemTxResult result = MEMTX_OK;
2843     hwaddr l, addr1;
2844     void *ptr;
2845     MemoryRegion *mr;
2846     FlatView *fv;
2847 
2848     if (__builtin_constant_p(len)) {
2849         if (len) {
2850             RCU_READ_LOCK_GUARD();
2851             fv = address_space_to_flatview(as);
2852             l = len;
2853             mr = flatview_translate(fv, addr, &addr1, &l, false, attrs);
2854             if (len == l && memory_access_is_direct(mr, false)) {
2855                 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
2856                 memcpy(buf, ptr, len);
2857             } else {
2858                 result = flatview_read_continue(fv, addr, attrs, buf, len,
2859                                                 addr1, l, mr);
2860             }
2861         }
2862     } else {
2863         result = address_space_read_full(as, addr, attrs, buf, len);
2864     }
2865     return result;
2866 }
2867 
2868 /**
2869  * address_space_read_cached: read from a cached RAM region
2870  *
2871  * @cache: Cached region to be addressed
2872  * @addr: address relative to the base of the RAM region
2873  * @buf: buffer with the data transferred
2874  * @len: length of the data transferred
2875  */
2876 static inline MemTxResult
2877 address_space_read_cached(MemoryRegionCache *cache, hwaddr addr,
2878                           void *buf, hwaddr len)
2879 {
2880     assert(addr < cache->len && len <= cache->len - addr);
2881     fuzz_dma_read_cb(cache->xlat + addr, len, cache->mrs.mr);
2882     if (likely(cache->ptr)) {
2883         memcpy(buf, cache->ptr + addr, len);
2884         return MEMTX_OK;
2885     } else {
2886         return address_space_read_cached_slow(cache, addr, buf, len);
2887     }
2888 }
2889 
2890 /**
2891  * address_space_write_cached: write to a cached RAM region
2892  *
2893  * @cache: Cached region to be addressed
2894  * @addr: address relative to the base of the RAM region
2895  * @buf: buffer with the data transferred
2896  * @len: length of the data transferred
2897  */
2898 static inline MemTxResult
2899 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr,
2900                            const void *buf, hwaddr len)
2901 {
2902     assert(addr < cache->len && len <= cache->len - addr);
2903     if (likely(cache->ptr)) {
2904         memcpy(cache->ptr + addr, buf, len);
2905         return MEMTX_OK;
2906     } else {
2907         return address_space_write_cached_slow(cache, addr, buf, len);
2908     }
2909 }
2910 
2911 /**
2912  * address_space_set: Fill address space with a constant byte.
2913  *
2914  * Return a MemTxResult indicating whether the operation succeeded
2915  * or failed (eg unassigned memory, device rejected the transaction,
2916  * IOMMU fault).
2917  *
2918  * @as: #AddressSpace to be accessed
2919  * @addr: address within that address space
2920  * @c: constant byte to fill the memory
2921  * @len: the number of bytes to fill with the constant byte
2922  * @attrs: memory transaction attributes
2923  */
2924 MemTxResult address_space_set(AddressSpace *as, hwaddr addr,
2925                               uint8_t c, hwaddr len, MemTxAttrs attrs);
2926 
2927 #ifdef NEED_CPU_H
2928 /* enum device_endian to MemOp.  */
2929 static inline MemOp devend_memop(enum device_endian end)
2930 {
2931     QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN &&
2932                       DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN);
2933 
2934 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
2935     /* Swap if non-host endianness or native (target) endianness */
2936     return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP;
2937 #else
2938     const int non_host_endianness =
2939         DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN;
2940 
2941     /* In this case, native (target) endianness needs no swap.  */
2942     return (end == non_host_endianness) ? MO_BSWAP : 0;
2943 #endif
2944 }
2945 #endif
2946 
2947 /*
2948  * Inhibit technologies that require discarding of pages in RAM blocks, e.g.,
2949  * to manage the actual amount of memory consumed by the VM (then, the memory
2950  * provided by RAM blocks might be bigger than the desired memory consumption).
2951  * This *must* be set if:
2952  * - Discarding parts of a RAM blocks does not result in the change being
2953  *   reflected in the VM and the pages getting freed.
2954  * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous
2955  *   discards blindly.
2956  * - Discarding parts of a RAM blocks will result in integrity issues (e.g.,
2957  *   encrypted VMs).
2958  * Technologies that only temporarily pin the current working set of a
2959  * driver are fine, because we don't expect such pages to be discarded
2960  * (esp. based on guest action like balloon inflation).
2961  *
2962  * This is *not* to be used to protect from concurrent discards (esp.,
2963  * postcopy).
2964  *
2965  * Returns 0 if successful. Returns -EBUSY if a technology that relies on
2966  * discards to work reliably is active.
2967  */
2968 int ram_block_discard_disable(bool state);
2969 
2970 /*
2971  * See ram_block_discard_disable(): only disable uncoordinated discards,
2972  * keeping coordinated discards (via the RamDiscardManager) enabled.
2973  */
2974 int ram_block_uncoordinated_discard_disable(bool state);
2975 
2976 /*
2977  * Inhibit technologies that disable discarding of pages in RAM blocks.
2978  *
2979  * Returns 0 if successful. Returns -EBUSY if discards are already set to
2980  * broken.
2981  */
2982 int ram_block_discard_require(bool state);
2983 
2984 /*
2985  * See ram_block_discard_require(): only inhibit technologies that disable
2986  * uncoordinated discarding of pages in RAM blocks, allowing co-existance with
2987  * technologies that only inhibit uncoordinated discards (via the
2988  * RamDiscardManager).
2989  */
2990 int ram_block_coordinated_discard_require(bool state);
2991 
2992 /*
2993  * Test if any discarding of memory in ram blocks is disabled.
2994  */
2995 bool ram_block_discard_is_disabled(void);
2996 
2997 /*
2998  * Test if any discarding of memory in ram blocks is required to work reliably.
2999  */
3000 bool ram_block_discard_is_required(void);
3001 
3002 #endif
3003 
3004 #endif
3005