xref: /qemu/include/exec/memory.h (revision bbc35fc2)
1 /*
2  * Physical memory management API
3  *
4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
5  *
6  * Authors:
7  *  Avi Kivity <avi@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #ifndef MEMORY_H
15 #define MEMORY_H
16 
17 #ifndef CONFIG_USER_ONLY
18 
19 #include "exec/cpu-common.h"
20 #include "exec/hwaddr.h"
21 #include "exec/memattrs.h"
22 #include "exec/memop.h"
23 #include "exec/ramlist.h"
24 #include "qemu/bswap.h"
25 #include "qemu/queue.h"
26 #include "qemu/int128.h"
27 #include "qemu/notify.h"
28 #include "qom/object.h"
29 #include "qemu/rcu.h"
30 
31 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
32 
33 #define MAX_PHYS_ADDR_SPACE_BITS 62
34 #define MAX_PHYS_ADDR            (((hwaddr)1 << MAX_PHYS_ADDR_SPACE_BITS) - 1)
35 
36 #define TYPE_MEMORY_REGION "qemu:memory-region"
37 DECLARE_INSTANCE_CHECKER(MemoryRegion, MEMORY_REGION,
38                          TYPE_MEMORY_REGION)
39 
40 #define TYPE_IOMMU_MEMORY_REGION "qemu:iommu-memory-region"
41 typedef struct IOMMUMemoryRegionClass IOMMUMemoryRegionClass;
42 DECLARE_OBJ_CHECKERS(IOMMUMemoryRegion, IOMMUMemoryRegionClass,
43                      IOMMU_MEMORY_REGION, TYPE_IOMMU_MEMORY_REGION)
44 
45 extern bool global_dirty_log;
46 
47 typedef struct MemoryRegionOps MemoryRegionOps;
48 
49 struct ReservedRegion {
50     hwaddr low;
51     hwaddr high;
52     unsigned type;
53 };
54 
55 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
56 
57 /* See address_space_translate: bit 0 is read, bit 1 is write.  */
58 typedef enum {
59     IOMMU_NONE = 0,
60     IOMMU_RO   = 1,
61     IOMMU_WO   = 2,
62     IOMMU_RW   = 3,
63 } IOMMUAccessFlags;
64 
65 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0))
66 
67 struct IOMMUTLBEntry {
68     AddressSpace    *target_as;
69     hwaddr           iova;
70     hwaddr           translated_addr;
71     hwaddr           addr_mask;  /* 0xfff = 4k translation */
72     IOMMUAccessFlags perm;
73 };
74 
75 /*
76  * Bitmap for different IOMMUNotifier capabilities. Each notifier can
77  * register with one or multiple IOMMU Notifier capability bit(s).
78  */
79 typedef enum {
80     IOMMU_NOTIFIER_NONE = 0,
81     /* Notify cache invalidations */
82     IOMMU_NOTIFIER_UNMAP = 0x1,
83     /* Notify entry changes (newly created entries) */
84     IOMMU_NOTIFIER_MAP = 0x2,
85 } IOMMUNotifierFlag;
86 
87 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
88 
89 struct IOMMUNotifier;
90 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier,
91                             IOMMUTLBEntry *data);
92 
93 struct IOMMUNotifier {
94     IOMMUNotify notify;
95     IOMMUNotifierFlag notifier_flags;
96     /* Notify for address space range start <= addr <= end */
97     hwaddr start;
98     hwaddr end;
99     int iommu_idx;
100     QLIST_ENTRY(IOMMUNotifier) node;
101 };
102 typedef struct IOMMUNotifier IOMMUNotifier;
103 
104 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
105 #define RAM_PREALLOC   (1 << 0)
106 
107 /* RAM is mmap-ed with MAP_SHARED */
108 #define RAM_SHARED     (1 << 1)
109 
110 /* Only a portion of RAM (used_length) is actually used, and migrated.
111  * This used_length size can change across reboots.
112  */
113 #define RAM_RESIZEABLE (1 << 2)
114 
115 /* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically
116  * zero the page and wake waiting processes.
117  * (Set during postcopy)
118  */
119 #define RAM_UF_ZEROPAGE (1 << 3)
120 
121 /* RAM can be migrated */
122 #define RAM_MIGRATABLE (1 << 4)
123 
124 /* RAM is a persistent kind memory */
125 #define RAM_PMEM (1 << 5)
126 
127 static inline void iommu_notifier_init(IOMMUNotifier *n, IOMMUNotify fn,
128                                        IOMMUNotifierFlag flags,
129                                        hwaddr start, hwaddr end,
130                                        int iommu_idx)
131 {
132     n->notify = fn;
133     n->notifier_flags = flags;
134     n->start = start;
135     n->end = end;
136     n->iommu_idx = iommu_idx;
137 }
138 
139 /*
140  * Memory region callbacks
141  */
142 struct MemoryRegionOps {
143     /* Read from the memory region. @addr is relative to @mr; @size is
144      * in bytes. */
145     uint64_t (*read)(void *opaque,
146                      hwaddr addr,
147                      unsigned size);
148     /* Write to the memory region. @addr is relative to @mr; @size is
149      * in bytes. */
150     void (*write)(void *opaque,
151                   hwaddr addr,
152                   uint64_t data,
153                   unsigned size);
154 
155     MemTxResult (*read_with_attrs)(void *opaque,
156                                    hwaddr addr,
157                                    uint64_t *data,
158                                    unsigned size,
159                                    MemTxAttrs attrs);
160     MemTxResult (*write_with_attrs)(void *opaque,
161                                     hwaddr addr,
162                                     uint64_t data,
163                                     unsigned size,
164                                     MemTxAttrs attrs);
165 
166     enum device_endian endianness;
167     /* Guest-visible constraints: */
168     struct {
169         /* If nonzero, specify bounds on access sizes beyond which a machine
170          * check is thrown.
171          */
172         unsigned min_access_size;
173         unsigned max_access_size;
174         /* If true, unaligned accesses are supported.  Otherwise unaligned
175          * accesses throw machine checks.
176          */
177          bool unaligned;
178         /*
179          * If present, and returns #false, the transaction is not accepted
180          * by the device (and results in machine dependent behaviour such
181          * as a machine check exception).
182          */
183         bool (*accepts)(void *opaque, hwaddr addr,
184                         unsigned size, bool is_write,
185                         MemTxAttrs attrs);
186     } valid;
187     /* Internal implementation constraints: */
188     struct {
189         /* If nonzero, specifies the minimum size implemented.  Smaller sizes
190          * will be rounded upwards and a partial result will be returned.
191          */
192         unsigned min_access_size;
193         /* If nonzero, specifies the maximum size implemented.  Larger sizes
194          * will be done as a series of accesses with smaller sizes.
195          */
196         unsigned max_access_size;
197         /* If true, unaligned accesses are supported.  Otherwise all accesses
198          * are converted to (possibly multiple) naturally aligned accesses.
199          */
200         bool unaligned;
201     } impl;
202 };
203 
204 typedef struct MemoryRegionClass {
205     /* private */
206     ObjectClass parent_class;
207 } MemoryRegionClass;
208 
209 
210 enum IOMMUMemoryRegionAttr {
211     IOMMU_ATTR_SPAPR_TCE_FD
212 };
213 
214 /*
215  * IOMMUMemoryRegionClass:
216  *
217  * All IOMMU implementations need to subclass TYPE_IOMMU_MEMORY_REGION
218  * and provide an implementation of at least the @translate method here
219  * to handle requests to the memory region. Other methods are optional.
220  *
221  * The IOMMU implementation must use the IOMMU notifier infrastructure
222  * to report whenever mappings are changed, by calling
223  * memory_region_notify_iommu() (or, if necessary, by calling
224  * memory_region_notify_one() for each registered notifier).
225  *
226  * Conceptually an IOMMU provides a mapping from input address
227  * to an output TLB entry. If the IOMMU is aware of memory transaction
228  * attributes and the output TLB entry depends on the transaction
229  * attributes, we represent this using IOMMU indexes. Each index
230  * selects a particular translation table that the IOMMU has:
231  *
232  *   @attrs_to_index returns the IOMMU index for a set of transaction attributes
233  *
234  *   @translate takes an input address and an IOMMU index
235  *
236  * and the mapping returned can only depend on the input address and the
237  * IOMMU index.
238  *
239  * Most IOMMUs don't care about the transaction attributes and support
240  * only a single IOMMU index. A more complex IOMMU might have one index
241  * for secure transactions and one for non-secure transactions.
242  */
243 struct IOMMUMemoryRegionClass {
244     /* private: */
245     MemoryRegionClass parent_class;
246 
247     /* public: */
248     /**
249      * @translate:
250      *
251      * Return a TLB entry that contains a given address.
252      *
253      * The IOMMUAccessFlags indicated via @flag are optional and may
254      * be specified as IOMMU_NONE to indicate that the caller needs
255      * the full translation information for both reads and writes. If
256      * the access flags are specified then the IOMMU implementation
257      * may use this as an optimization, to stop doing a page table
258      * walk as soon as it knows that the requested permissions are not
259      * allowed. If IOMMU_NONE is passed then the IOMMU must do the
260      * full page table walk and report the permissions in the returned
261      * IOMMUTLBEntry. (Note that this implies that an IOMMU may not
262      * return different mappings for reads and writes.)
263      *
264      * The returned information remains valid while the caller is
265      * holding the big QEMU lock or is inside an RCU critical section;
266      * if the caller wishes to cache the mapping beyond that it must
267      * register an IOMMU notifier so it can invalidate its cached
268      * information when the IOMMU mapping changes.
269      *
270      * @iommu: the IOMMUMemoryRegion
271      *
272      * @hwaddr: address to be translated within the memory region
273      *
274      * @flag: requested access permission
275      *
276      * @iommu_idx: IOMMU index for the translation
277      */
278     IOMMUTLBEntry (*translate)(IOMMUMemoryRegion *iommu, hwaddr addr,
279                                IOMMUAccessFlags flag, int iommu_idx);
280     /**
281      * @get_min_page_size:
282      *
283      * Returns minimum supported page size in bytes.
284      *
285      * If this method is not provided then the minimum is assumed to
286      * be TARGET_PAGE_SIZE.
287      *
288      * @iommu: the IOMMUMemoryRegion
289      */
290     uint64_t (*get_min_page_size)(IOMMUMemoryRegion *iommu);
291     /**
292      * @notify_flag_changed:
293      *
294      * Called when IOMMU Notifier flag changes (ie when the set of
295      * events which IOMMU users are requesting notification for changes).
296      * Optional method -- need not be provided if the IOMMU does not
297      * need to know exactly which events must be notified.
298      *
299      * @iommu: the IOMMUMemoryRegion
300      *
301      * @old_flags: events which previously needed to be notified
302      *
303      * @new_flags: events which now need to be notified
304      *
305      * Returns 0 on success, or a negative errno; in particular
306      * returns -EINVAL if the new flag bitmap is not supported by the
307      * IOMMU memory region. In case of failure, the error object
308      * must be created
309      */
310     int (*notify_flag_changed)(IOMMUMemoryRegion *iommu,
311                                IOMMUNotifierFlag old_flags,
312                                IOMMUNotifierFlag new_flags,
313                                Error **errp);
314     /**
315      * @replay:
316      *
317      * Called to handle memory_region_iommu_replay().
318      *
319      * The default implementation of memory_region_iommu_replay() is to
320      * call the IOMMU translate method for every page in the address space
321      * with flag == IOMMU_NONE and then call the notifier if translate
322      * returns a valid mapping. If this method is implemented then it
323      * overrides the default behaviour, and must provide the full semantics
324      * of memory_region_iommu_replay(), by calling @notifier for every
325      * translation present in the IOMMU.
326      *
327      * Optional method -- an IOMMU only needs to provide this method
328      * if the default is inefficient or produces undesirable side effects.
329      *
330      * Note: this is not related to record-and-replay functionality.
331      */
332     void (*replay)(IOMMUMemoryRegion *iommu, IOMMUNotifier *notifier);
333 
334     /**
335      * @get_attr:
336      *
337      * Get IOMMU misc attributes. This is an optional method that
338      * can be used to allow users of the IOMMU to get implementation-specific
339      * information. The IOMMU implements this method to handle calls
340      * by IOMMU users to memory_region_iommu_get_attr() by filling in
341      * the arbitrary data pointer for any IOMMUMemoryRegionAttr values that
342      * the IOMMU supports. If the method is unimplemented then
343      * memory_region_iommu_get_attr() will always return -EINVAL.
344      *
345      * @iommu: the IOMMUMemoryRegion
346      *
347      * @attr: attribute being queried
348      *
349      * @data: memory to fill in with the attribute data
350      *
351      * Returns 0 on success, or a negative errno; in particular
352      * returns -EINVAL for unrecognized or unimplemented attribute types.
353      */
354     int (*get_attr)(IOMMUMemoryRegion *iommu, enum IOMMUMemoryRegionAttr attr,
355                     void *data);
356 
357     /**
358      * @attrs_to_index:
359      *
360      * Return the IOMMU index to use for a given set of transaction attributes.
361      *
362      * Optional method: if an IOMMU only supports a single IOMMU index then
363      * the default implementation of memory_region_iommu_attrs_to_index()
364      * will return 0.
365      *
366      * The indexes supported by an IOMMU must be contiguous, starting at 0.
367      *
368      * @iommu: the IOMMUMemoryRegion
369      * @attrs: memory transaction attributes
370      */
371     int (*attrs_to_index)(IOMMUMemoryRegion *iommu, MemTxAttrs attrs);
372 
373     /**
374      * @num_indexes:
375      *
376      * Return the number of IOMMU indexes this IOMMU supports.
377      *
378      * Optional method: if this method is not provided, then
379      * memory_region_iommu_num_indexes() will return 1, indicating that
380      * only a single IOMMU index is supported.
381      *
382      * @iommu: the IOMMUMemoryRegion
383      */
384     int (*num_indexes)(IOMMUMemoryRegion *iommu);
385 };
386 
387 typedef struct CoalescedMemoryRange CoalescedMemoryRange;
388 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd;
389 
390 /** MemoryRegion:
391  *
392  * A struct representing a memory region.
393  */
394 struct MemoryRegion {
395     Object parent_obj;
396 
397     /* private: */
398 
399     /* The following fields should fit in a cache line */
400     bool romd_mode;
401     bool ram;
402     bool subpage;
403     bool readonly; /* For RAM regions */
404     bool nonvolatile;
405     bool rom_device;
406     bool flush_coalesced_mmio;
407     uint8_t dirty_log_mask;
408     bool is_iommu;
409     RAMBlock *ram_block;
410     Object *owner;
411 
412     const MemoryRegionOps *ops;
413     void *opaque;
414     MemoryRegion *container;
415     Int128 size;
416     hwaddr addr;
417     void (*destructor)(MemoryRegion *mr);
418     uint64_t align;
419     bool terminates;
420     bool ram_device;
421     bool enabled;
422     bool warning_printed; /* For reservations */
423     uint8_t vga_logging_count;
424     MemoryRegion *alias;
425     hwaddr alias_offset;
426     int32_t priority;
427     QTAILQ_HEAD(, MemoryRegion) subregions;
428     QTAILQ_ENTRY(MemoryRegion) subregions_link;
429     QTAILQ_HEAD(, CoalescedMemoryRange) coalesced;
430     const char *name;
431     unsigned ioeventfd_nb;
432     MemoryRegionIoeventfd *ioeventfds;
433 };
434 
435 struct IOMMUMemoryRegion {
436     MemoryRegion parent_obj;
437 
438     QLIST_HEAD(, IOMMUNotifier) iommu_notify;
439     IOMMUNotifierFlag iommu_notify_flags;
440 };
441 
442 #define IOMMU_NOTIFIER_FOREACH(n, mr) \
443     QLIST_FOREACH((n), &(mr)->iommu_notify, node)
444 
445 /**
446  * struct MemoryListener: callbacks structure for updates to the physical memory map
447  *
448  * Allows a component to adjust to changes in the guest-visible memory map.
449  * Use with memory_listener_register() and memory_listener_unregister().
450  */
451 struct MemoryListener {
452     /**
453      * @begin:
454      *
455      * Called at the beginning of an address space update transaction.
456      * Followed by calls to #MemoryListener.region_add(),
457      * #MemoryListener.region_del(), #MemoryListener.region_nop(),
458      * #MemoryListener.log_start() and #MemoryListener.log_stop() in
459      * increasing address order.
460      *
461      * @listener: The #MemoryListener.
462      */
463     void (*begin)(MemoryListener *listener);
464 
465     /**
466      * @commit:
467      *
468      * Called at the end of an address space update transaction,
469      * after the last call to #MemoryListener.region_add(),
470      * #MemoryListener.region_del() or #MemoryListener.region_nop(),
471      * #MemoryListener.log_start() and #MemoryListener.log_stop().
472      *
473      * @listener: The #MemoryListener.
474      */
475     void (*commit)(MemoryListener *listener);
476 
477     /**
478      * @region_add:
479      *
480      * Called during an address space update transaction,
481      * for a section of the address space that is new in this address space
482      * space since the last transaction.
483      *
484      * @listener: The #MemoryListener.
485      * @section: The new #MemoryRegionSection.
486      */
487     void (*region_add)(MemoryListener *listener, MemoryRegionSection *section);
488 
489     /**
490      * @region_del:
491      *
492      * Called during an address space update transaction,
493      * for a section of the address space that has disappeared in the address
494      * space since the last transaction.
495      *
496      * @listener: The #MemoryListener.
497      * @section: The old #MemoryRegionSection.
498      */
499     void (*region_del)(MemoryListener *listener, MemoryRegionSection *section);
500 
501     /**
502      * @region_nop:
503      *
504      * Called during an address space update transaction,
505      * for a section of the address space that is in the same place in the address
506      * space as in the last transaction.
507      *
508      * @listener: The #MemoryListener.
509      * @section: The #MemoryRegionSection.
510      */
511     void (*region_nop)(MemoryListener *listener, MemoryRegionSection *section);
512 
513     /**
514      * @log_start:
515      *
516      * Called during an address space update transaction, after
517      * one of #MemoryListener.region_add(),#MemoryListener.region_del() or
518      * #MemoryListener.region_nop(), if dirty memory logging clients have
519      * become active since the last transaction.
520      *
521      * @listener: The #MemoryListener.
522      * @section: The #MemoryRegionSection.
523      * @old: A bitmap of dirty memory logging clients that were active in
524      * the previous transaction.
525      * @new: A bitmap of dirty memory logging clients that are active in
526      * the current transaction.
527      */
528     void (*log_start)(MemoryListener *listener, MemoryRegionSection *section,
529                       int old, int new);
530 
531     /**
532      * @log_stop:
533      *
534      * Called during an address space update transaction, after
535      * one of #MemoryListener.region_add(), #MemoryListener.region_del() or
536      * #MemoryListener.region_nop() and possibly after
537      * #MemoryListener.log_start(), if dirty memory logging clients have
538      * become inactive since the last transaction.
539      *
540      * @listener: The #MemoryListener.
541      * @section: The #MemoryRegionSection.
542      * @old: A bitmap of dirty memory logging clients that were active in
543      * the previous transaction.
544      * @new: A bitmap of dirty memory logging clients that are active in
545      * the current transaction.
546      */
547     void (*log_stop)(MemoryListener *listener, MemoryRegionSection *section,
548                      int old, int new);
549 
550     /**
551      * @log_sync:
552      *
553      * Called by memory_region_snapshot_and_clear_dirty() and
554      * memory_global_dirty_log_sync(), before accessing QEMU's "official"
555      * copy of the dirty memory bitmap for a #MemoryRegionSection.
556      *
557      * @listener: The #MemoryListener.
558      * @section: The #MemoryRegionSection.
559      */
560     void (*log_sync)(MemoryListener *listener, MemoryRegionSection *section);
561 
562     /**
563      * @log_clear:
564      *
565      * Called before reading the dirty memory bitmap for a
566      * #MemoryRegionSection.
567      *
568      * @listener: The #MemoryListener.
569      * @section: The #MemoryRegionSection.
570      */
571     void (*log_clear)(MemoryListener *listener, MemoryRegionSection *section);
572 
573     /**
574      * @log_global_start:
575      *
576      * Called by memory_global_dirty_log_start(), which
577      * enables the %DIRTY_LOG_MIGRATION client on all memory regions in
578      * the address space.  #MemoryListener.log_global_start() is also
579      * called when a #MemoryListener is added, if global dirty logging is
580      * active at that time.
581      *
582      * @listener: The #MemoryListener.
583      */
584     void (*log_global_start)(MemoryListener *listener);
585 
586     /**
587      * @log_global_stop:
588      *
589      * Called by memory_global_dirty_log_stop(), which
590      * disables the %DIRTY_LOG_MIGRATION client on all memory regions in
591      * the address space.
592      *
593      * @listener: The #MemoryListener.
594      */
595     void (*log_global_stop)(MemoryListener *listener);
596 
597     /**
598      * @log_global_after_sync:
599      *
600      * Called after reading the dirty memory bitmap
601      * for any #MemoryRegionSection.
602      *
603      * @listener: The #MemoryListener.
604      */
605     void (*log_global_after_sync)(MemoryListener *listener);
606 
607     /**
608      * @eventfd_add:
609      *
610      * Called during an address space update transaction,
611      * for a section of the address space that has had a new ioeventfd
612      * registration since the last transaction.
613      *
614      * @listener: The #MemoryListener.
615      * @section: The new #MemoryRegionSection.
616      * @match_data: The @match_data parameter for the new ioeventfd.
617      * @data: The @data parameter for the new ioeventfd.
618      * @e: The #EventNotifier parameter for the new ioeventfd.
619      */
620     void (*eventfd_add)(MemoryListener *listener, MemoryRegionSection *section,
621                         bool match_data, uint64_t data, EventNotifier *e);
622 
623     /**
624      * @eventfd_del:
625      *
626      * Called during an address space update transaction,
627      * for a section of the address space that has dropped an ioeventfd
628      * registration since the last transaction.
629      *
630      * @listener: The #MemoryListener.
631      * @section: The new #MemoryRegionSection.
632      * @match_data: The @match_data parameter for the dropped ioeventfd.
633      * @data: The @data parameter for the dropped ioeventfd.
634      * @e: The #EventNotifier parameter for the dropped ioeventfd.
635      */
636     void (*eventfd_del)(MemoryListener *listener, MemoryRegionSection *section,
637                         bool match_data, uint64_t data, EventNotifier *e);
638 
639     /**
640      * @coalesced_io_add:
641      *
642      * Called during an address space update transaction,
643      * for a section of the address space that has had a new coalesced
644      * MMIO range registration since the last transaction.
645      *
646      * @listener: The #MemoryListener.
647      * @section: The new #MemoryRegionSection.
648      * @addr: The starting address for the coalesced MMIO range.
649      * @len: The length of the coalesced MMIO range.
650      */
651     void (*coalesced_io_add)(MemoryListener *listener, MemoryRegionSection *section,
652                                hwaddr addr, hwaddr len);
653 
654     /**
655      * @coalesced_io_del:
656      *
657      * Called during an address space update transaction,
658      * for a section of the address space that has dropped a coalesced
659      * MMIO range since the last transaction.
660      *
661      * @listener: The #MemoryListener.
662      * @section: The new #MemoryRegionSection.
663      * @addr: The starting address for the coalesced MMIO range.
664      * @len: The length of the coalesced MMIO range.
665      */
666     void (*coalesced_io_del)(MemoryListener *listener, MemoryRegionSection *section,
667                                hwaddr addr, hwaddr len);
668     /**
669      * @priority:
670      *
671      * Govern the order in which memory listeners are invoked. Lower priorities
672      * are invoked earlier for "add" or "start" callbacks, and later for "delete"
673      * or "stop" callbacks.
674      */
675     unsigned priority;
676 
677     /* private: */
678     AddressSpace *address_space;
679     QTAILQ_ENTRY(MemoryListener) link;
680     QTAILQ_ENTRY(MemoryListener) link_as;
681 };
682 
683 /**
684  * struct AddressSpace: describes a mapping of addresses to #MemoryRegion objects
685  */
686 struct AddressSpace {
687     /* private: */
688     struct rcu_head rcu;
689     char *name;
690     MemoryRegion *root;
691 
692     /* Accessed via RCU.  */
693     struct FlatView *current_map;
694 
695     int ioeventfd_nb;
696     struct MemoryRegionIoeventfd *ioeventfds;
697     QTAILQ_HEAD(, MemoryListener) listeners;
698     QTAILQ_ENTRY(AddressSpace) address_spaces_link;
699 };
700 
701 typedef struct AddressSpaceDispatch AddressSpaceDispatch;
702 typedef struct FlatRange FlatRange;
703 
704 /* Flattened global view of current active memory hierarchy.  Kept in sorted
705  * order.
706  */
707 struct FlatView {
708     struct rcu_head rcu;
709     unsigned ref;
710     FlatRange *ranges;
711     unsigned nr;
712     unsigned nr_allocated;
713     struct AddressSpaceDispatch *dispatch;
714     MemoryRegion *root;
715 };
716 
717 static inline FlatView *address_space_to_flatview(AddressSpace *as)
718 {
719     return qatomic_rcu_read(&as->current_map);
720 }
721 
722 
723 /**
724  * struct MemoryRegionSection: describes a fragment of a #MemoryRegion
725  *
726  * @mr: the region, or %NULL if empty
727  * @fv: the flat view of the address space the region is mapped in
728  * @offset_within_region: the beginning of the section, relative to @mr's start
729  * @size: the size of the section; will not exceed @mr's boundaries
730  * @offset_within_address_space: the address of the first byte of the section
731  *     relative to the region's address space
732  * @readonly: writes to this section are ignored
733  * @nonvolatile: this section is non-volatile
734  */
735 struct MemoryRegionSection {
736     Int128 size;
737     MemoryRegion *mr;
738     FlatView *fv;
739     hwaddr offset_within_region;
740     hwaddr offset_within_address_space;
741     bool readonly;
742     bool nonvolatile;
743 };
744 
745 static inline bool MemoryRegionSection_eq(MemoryRegionSection *a,
746                                           MemoryRegionSection *b)
747 {
748     return a->mr == b->mr &&
749            a->fv == b->fv &&
750            a->offset_within_region == b->offset_within_region &&
751            a->offset_within_address_space == b->offset_within_address_space &&
752            int128_eq(a->size, b->size) &&
753            a->readonly == b->readonly &&
754            a->nonvolatile == b->nonvolatile;
755 }
756 
757 /**
758  * memory_region_init: Initialize a memory region
759  *
760  * The region typically acts as a container for other memory regions.  Use
761  * memory_region_add_subregion() to add subregions.
762  *
763  * @mr: the #MemoryRegion to be initialized
764  * @owner: the object that tracks the region's reference count
765  * @name: used for debugging; not visible to the user or ABI
766  * @size: size of the region; any subregions beyond this size will be clipped
767  */
768 void memory_region_init(MemoryRegion *mr,
769                         struct Object *owner,
770                         const char *name,
771                         uint64_t size);
772 
773 /**
774  * memory_region_ref: Add 1 to a memory region's reference count
775  *
776  * Whenever memory regions are accessed outside the BQL, they need to be
777  * preserved against hot-unplug.  MemoryRegions actually do not have their
778  * own reference count; they piggyback on a QOM object, their "owner".
779  * This function adds a reference to the owner.
780  *
781  * All MemoryRegions must have an owner if they can disappear, even if the
782  * device they belong to operates exclusively under the BQL.  This is because
783  * the region could be returned at any time by memory_region_find, and this
784  * is usually under guest control.
785  *
786  * @mr: the #MemoryRegion
787  */
788 void memory_region_ref(MemoryRegion *mr);
789 
790 /**
791  * memory_region_unref: Remove 1 to a memory region's reference count
792  *
793  * Whenever memory regions are accessed outside the BQL, they need to be
794  * preserved against hot-unplug.  MemoryRegions actually do not have their
795  * own reference count; they piggyback on a QOM object, their "owner".
796  * This function removes a reference to the owner and possibly destroys it.
797  *
798  * @mr: the #MemoryRegion
799  */
800 void memory_region_unref(MemoryRegion *mr);
801 
802 /**
803  * memory_region_init_io: Initialize an I/O memory region.
804  *
805  * Accesses into the region will cause the callbacks in @ops to be called.
806  * if @size is nonzero, subregions will be clipped to @size.
807  *
808  * @mr: the #MemoryRegion to be initialized.
809  * @owner: the object that tracks the region's reference count
810  * @ops: a structure containing read and write callbacks to be used when
811  *       I/O is performed on the region.
812  * @opaque: passed to the read and write callbacks of the @ops structure.
813  * @name: used for debugging; not visible to the user or ABI
814  * @size: size of the region.
815  */
816 void memory_region_init_io(MemoryRegion *mr,
817                            struct Object *owner,
818                            const MemoryRegionOps *ops,
819                            void *opaque,
820                            const char *name,
821                            uint64_t size);
822 
823 /**
824  * memory_region_init_ram_nomigrate:  Initialize RAM memory region.  Accesses
825  *                                    into the region will modify memory
826  *                                    directly.
827  *
828  * @mr: the #MemoryRegion to be initialized.
829  * @owner: the object that tracks the region's reference count
830  * @name: Region name, becomes part of RAMBlock name used in migration stream
831  *        must be unique within any device
832  * @size: size of the region.
833  * @errp: pointer to Error*, to store an error if it happens.
834  *
835  * Note that this function does not do anything to cause the data in the
836  * RAM memory region to be migrated; that is the responsibility of the caller.
837  */
838 void memory_region_init_ram_nomigrate(MemoryRegion *mr,
839                                       struct Object *owner,
840                                       const char *name,
841                                       uint64_t size,
842                                       Error **errp);
843 
844 /**
845  * memory_region_init_ram_shared_nomigrate:  Initialize RAM memory region.
846  *                                           Accesses into the region will
847  *                                           modify memory directly.
848  *
849  * @mr: the #MemoryRegion to be initialized.
850  * @owner: the object that tracks the region's reference count
851  * @name: Region name, becomes part of RAMBlock name used in migration stream
852  *        must be unique within any device
853  * @size: size of the region.
854  * @share: allow remapping RAM to different addresses
855  * @errp: pointer to Error*, to store an error if it happens.
856  *
857  * Note that this function is similar to memory_region_init_ram_nomigrate.
858  * The only difference is part of the RAM region can be remapped.
859  */
860 void memory_region_init_ram_shared_nomigrate(MemoryRegion *mr,
861                                              struct Object *owner,
862                                              const char *name,
863                                              uint64_t size,
864                                              bool share,
865                                              Error **errp);
866 
867 /**
868  * memory_region_init_resizeable_ram:  Initialize memory region with resizeable
869  *                                     RAM.  Accesses into the region will
870  *                                     modify memory directly.  Only an initial
871  *                                     portion of this RAM is actually used.
872  *                                     The used size can change across reboots.
873  *
874  * @mr: the #MemoryRegion to be initialized.
875  * @owner: the object that tracks the region's reference count
876  * @name: Region name, becomes part of RAMBlock name used in migration stream
877  *        must be unique within any device
878  * @size: used size of the region.
879  * @max_size: max size of the region.
880  * @resized: callback to notify owner about used size change.
881  * @errp: pointer to Error*, to store an error if it happens.
882  *
883  * Note that this function does not do anything to cause the data in the
884  * RAM memory region to be migrated; that is the responsibility of the caller.
885  */
886 void memory_region_init_resizeable_ram(MemoryRegion *mr,
887                                        struct Object *owner,
888                                        const char *name,
889                                        uint64_t size,
890                                        uint64_t max_size,
891                                        void (*resized)(const char*,
892                                                        uint64_t length,
893                                                        void *host),
894                                        Error **errp);
895 #ifdef CONFIG_POSIX
896 
897 /**
898  * memory_region_init_ram_from_file:  Initialize RAM memory region with a
899  *                                    mmap-ed backend.
900  *
901  * @mr: the #MemoryRegion to be initialized.
902  * @owner: the object that tracks the region's reference count
903  * @name: Region name, becomes part of RAMBlock name used in migration stream
904  *        must be unique within any device
905  * @size: size of the region.
906  * @align: alignment of the region base address; if 0, the default alignment
907  *         (getpagesize()) will be used.
908  * @ram_flags: Memory region features:
909  *             - RAM_SHARED: memory must be mmaped with the MAP_SHARED flag
910  *             - RAM_PMEM: the memory is persistent memory
911  *             Other bits are ignored now.
912  * @path: the path in which to allocate the RAM.
913  * @errp: pointer to Error*, to store an error if it happens.
914  *
915  * Note that this function does not do anything to cause the data in the
916  * RAM memory region to be migrated; that is the responsibility of the caller.
917  */
918 void memory_region_init_ram_from_file(MemoryRegion *mr,
919                                       struct Object *owner,
920                                       const char *name,
921                                       uint64_t size,
922                                       uint64_t align,
923                                       uint32_t ram_flags,
924                                       const char *path,
925                                       Error **errp);
926 
927 /**
928  * memory_region_init_ram_from_fd:  Initialize RAM memory region with a
929  *                                  mmap-ed backend.
930  *
931  * @mr: the #MemoryRegion to be initialized.
932  * @owner: the object that tracks the region's reference count
933  * @name: the name of the region.
934  * @size: size of the region.
935  * @share: %true if memory must be mmaped with the MAP_SHARED flag
936  * @fd: the fd to mmap.
937  * @errp: pointer to Error*, to store an error if it happens.
938  *
939  * Note that this function does not do anything to cause the data in the
940  * RAM memory region to be migrated; that is the responsibility of the caller.
941  */
942 void memory_region_init_ram_from_fd(MemoryRegion *mr,
943                                     struct Object *owner,
944                                     const char *name,
945                                     uint64_t size,
946                                     bool share,
947                                     int fd,
948                                     Error **errp);
949 #endif
950 
951 /**
952  * memory_region_init_ram_ptr:  Initialize RAM memory region from a
953  *                              user-provided pointer.  Accesses into the
954  *                              region will modify memory directly.
955  *
956  * @mr: the #MemoryRegion to be initialized.
957  * @owner: the object that tracks the region's reference count
958  * @name: Region name, becomes part of RAMBlock name used in migration stream
959  *        must be unique within any device
960  * @size: size of the region.
961  * @ptr: memory to be mapped; must contain at least @size bytes.
962  *
963  * Note that this function does not do anything to cause the data in the
964  * RAM memory region to be migrated; that is the responsibility of the caller.
965  */
966 void memory_region_init_ram_ptr(MemoryRegion *mr,
967                                 struct Object *owner,
968                                 const char *name,
969                                 uint64_t size,
970                                 void *ptr);
971 
972 /**
973  * memory_region_init_ram_device_ptr:  Initialize RAM device memory region from
974  *                                     a user-provided pointer.
975  *
976  * A RAM device represents a mapping to a physical device, such as to a PCI
977  * MMIO BAR of an vfio-pci assigned device.  The memory region may be mapped
978  * into the VM address space and access to the region will modify memory
979  * directly.  However, the memory region should not be included in a memory
980  * dump (device may not be enabled/mapped at the time of the dump), and
981  * operations incompatible with manipulating MMIO should be avoided.  Replaces
982  * skip_dump flag.
983  *
984  * @mr: the #MemoryRegion to be initialized.
985  * @owner: the object that tracks the region's reference count
986  * @name: the name of the region.
987  * @size: size of the region.
988  * @ptr: memory to be mapped; must contain at least @size bytes.
989  *
990  * Note that this function does not do anything to cause the data in the
991  * RAM memory region to be migrated; that is the responsibility of the caller.
992  * (For RAM device memory regions, migrating the contents rarely makes sense.)
993  */
994 void memory_region_init_ram_device_ptr(MemoryRegion *mr,
995                                        struct Object *owner,
996                                        const char *name,
997                                        uint64_t size,
998                                        void *ptr);
999 
1000 /**
1001  * memory_region_init_alias: Initialize a memory region that aliases all or a
1002  *                           part of another memory region.
1003  *
1004  * @mr: the #MemoryRegion to be initialized.
1005  * @owner: the object that tracks the region's reference count
1006  * @name: used for debugging; not visible to the user or ABI
1007  * @orig: the region to be referenced; @mr will be equivalent to
1008  *        @orig between @offset and @offset + @size - 1.
1009  * @offset: start of the section in @orig to be referenced.
1010  * @size: size of the region.
1011  */
1012 void memory_region_init_alias(MemoryRegion *mr,
1013                               struct Object *owner,
1014                               const char *name,
1015                               MemoryRegion *orig,
1016                               hwaddr offset,
1017                               uint64_t size);
1018 
1019 /**
1020  * memory_region_init_rom_nomigrate: Initialize a ROM memory region.
1021  *
1022  * This has the same effect as calling memory_region_init_ram_nomigrate()
1023  * and then marking the resulting region read-only with
1024  * memory_region_set_readonly().
1025  *
1026  * Note that this function does not do anything to cause the data in the
1027  * RAM side of the memory region to be migrated; that is the responsibility
1028  * of the caller.
1029  *
1030  * @mr: the #MemoryRegion to be initialized.
1031  * @owner: the object that tracks the region's reference count
1032  * @name: Region name, becomes part of RAMBlock name used in migration stream
1033  *        must be unique within any device
1034  * @size: size of the region.
1035  * @errp: pointer to Error*, to store an error if it happens.
1036  */
1037 void memory_region_init_rom_nomigrate(MemoryRegion *mr,
1038                                       struct Object *owner,
1039                                       const char *name,
1040                                       uint64_t size,
1041                                       Error **errp);
1042 
1043 /**
1044  * memory_region_init_rom_device_nomigrate:  Initialize a ROM memory region.
1045  *                                 Writes are handled via callbacks.
1046  *
1047  * Note that this function does not do anything to cause the data in the
1048  * RAM side of the memory region to be migrated; that is the responsibility
1049  * of the caller.
1050  *
1051  * @mr: the #MemoryRegion to be initialized.
1052  * @owner: the object that tracks the region's reference count
1053  * @ops: callbacks for write access handling (must not be NULL).
1054  * @opaque: passed to the read and write callbacks of the @ops structure.
1055  * @name: Region name, becomes part of RAMBlock name used in migration stream
1056  *        must be unique within any device
1057  * @size: size of the region.
1058  * @errp: pointer to Error*, to store an error if it happens.
1059  */
1060 void memory_region_init_rom_device_nomigrate(MemoryRegion *mr,
1061                                              struct Object *owner,
1062                                              const MemoryRegionOps *ops,
1063                                              void *opaque,
1064                                              const char *name,
1065                                              uint64_t size,
1066                                              Error **errp);
1067 
1068 /**
1069  * memory_region_init_iommu: Initialize a memory region of a custom type
1070  * that translates addresses
1071  *
1072  * An IOMMU region translates addresses and forwards accesses to a target
1073  * memory region.
1074  *
1075  * The IOMMU implementation must define a subclass of TYPE_IOMMU_MEMORY_REGION.
1076  * @_iommu_mr should be a pointer to enough memory for an instance of
1077  * that subclass, @instance_size is the size of that subclass, and
1078  * @mrtypename is its name. This function will initialize @_iommu_mr as an
1079  * instance of the subclass, and its methods will then be called to handle
1080  * accesses to the memory region. See the documentation of
1081  * #IOMMUMemoryRegionClass for further details.
1082  *
1083  * @_iommu_mr: the #IOMMUMemoryRegion to be initialized
1084  * @instance_size: the IOMMUMemoryRegion subclass instance size
1085  * @mrtypename: the type name of the #IOMMUMemoryRegion
1086  * @owner: the object that tracks the region's reference count
1087  * @name: used for debugging; not visible to the user or ABI
1088  * @size: size of the region.
1089  */
1090 void memory_region_init_iommu(void *_iommu_mr,
1091                               size_t instance_size,
1092                               const char *mrtypename,
1093                               Object *owner,
1094                               const char *name,
1095                               uint64_t size);
1096 
1097 /**
1098  * memory_region_init_ram - Initialize RAM memory region.  Accesses into the
1099  *                          region will modify memory directly.
1100  *
1101  * @mr: the #MemoryRegion to be initialized
1102  * @owner: the object that tracks the region's reference count (must be
1103  *         TYPE_DEVICE or a subclass of TYPE_DEVICE, or NULL)
1104  * @name: name of the memory region
1105  * @size: size of the region in bytes
1106  * @errp: pointer to Error*, to store an error if it happens.
1107  *
1108  * This function allocates RAM for a board model or device, and
1109  * arranges for it to be migrated (by calling vmstate_register_ram()
1110  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1111  * @owner is NULL).
1112  *
1113  * TODO: Currently we restrict @owner to being either NULL (for
1114  * global RAM regions with no owner) or devices, so that we can
1115  * give the RAM block a unique name for migration purposes.
1116  * We should lift this restriction and allow arbitrary Objects.
1117  * If you pass a non-NULL non-device @owner then we will assert.
1118  */
1119 void memory_region_init_ram(MemoryRegion *mr,
1120                             struct Object *owner,
1121                             const char *name,
1122                             uint64_t size,
1123                             Error **errp);
1124 
1125 /**
1126  * memory_region_init_rom: Initialize a ROM memory region.
1127  *
1128  * This has the same effect as calling memory_region_init_ram()
1129  * and then marking the resulting region read-only with
1130  * memory_region_set_readonly(). This includes arranging for the
1131  * contents to be migrated.
1132  *
1133  * TODO: Currently we restrict @owner to being either NULL (for
1134  * global RAM regions with no owner) or devices, so that we can
1135  * give the RAM block a unique name for migration purposes.
1136  * We should lift this restriction and allow arbitrary Objects.
1137  * If you pass a non-NULL non-device @owner then we will assert.
1138  *
1139  * @mr: the #MemoryRegion to be initialized.
1140  * @owner: the object that tracks the region's reference count
1141  * @name: Region name, becomes part of RAMBlock name used in migration stream
1142  *        must be unique within any device
1143  * @size: size of the region.
1144  * @errp: pointer to Error*, to store an error if it happens.
1145  */
1146 void memory_region_init_rom(MemoryRegion *mr,
1147                             struct Object *owner,
1148                             const char *name,
1149                             uint64_t size,
1150                             Error **errp);
1151 
1152 /**
1153  * memory_region_init_rom_device:  Initialize a ROM memory region.
1154  *                                 Writes are handled via callbacks.
1155  *
1156  * This function initializes a memory region backed by RAM for reads
1157  * and callbacks for writes, and arranges for the RAM backing to
1158  * be migrated (by calling vmstate_register_ram()
1159  * if @owner is a DeviceState, or vmstate_register_ram_global() if
1160  * @owner is NULL).
1161  *
1162  * TODO: Currently we restrict @owner to being either NULL (for
1163  * global RAM regions with no owner) or devices, so that we can
1164  * give the RAM block a unique name for migration purposes.
1165  * We should lift this restriction and allow arbitrary Objects.
1166  * If you pass a non-NULL non-device @owner then we will assert.
1167  *
1168  * @mr: the #MemoryRegion to be initialized.
1169  * @owner: the object that tracks the region's reference count
1170  * @ops: callbacks for write access handling (must not be NULL).
1171  * @opaque: passed to the read and write callbacks of the @ops structure.
1172  * @name: Region name, becomes part of RAMBlock name used in migration stream
1173  *        must be unique within any device
1174  * @size: size of the region.
1175  * @errp: pointer to Error*, to store an error if it happens.
1176  */
1177 void memory_region_init_rom_device(MemoryRegion *mr,
1178                                    struct Object *owner,
1179                                    const MemoryRegionOps *ops,
1180                                    void *opaque,
1181                                    const char *name,
1182                                    uint64_t size,
1183                                    Error **errp);
1184 
1185 
1186 /**
1187  * memory_region_owner: get a memory region's owner.
1188  *
1189  * @mr: the memory region being queried.
1190  */
1191 struct Object *memory_region_owner(MemoryRegion *mr);
1192 
1193 /**
1194  * memory_region_size: get a memory region's size.
1195  *
1196  * @mr: the memory region being queried.
1197  */
1198 uint64_t memory_region_size(MemoryRegion *mr);
1199 
1200 /**
1201  * memory_region_is_ram: check whether a memory region is random access
1202  *
1203  * Returns %true if a memory region is random access.
1204  *
1205  * @mr: the memory region being queried
1206  */
1207 static inline bool memory_region_is_ram(MemoryRegion *mr)
1208 {
1209     return mr->ram;
1210 }
1211 
1212 /**
1213  * memory_region_is_ram_device: check whether a memory region is a ram device
1214  *
1215  * Returns %true if a memory region is a device backed ram region
1216  *
1217  * @mr: the memory region being queried
1218  */
1219 bool memory_region_is_ram_device(MemoryRegion *mr);
1220 
1221 /**
1222  * memory_region_is_romd: check whether a memory region is in ROMD mode
1223  *
1224  * Returns %true if a memory region is a ROM device and currently set to allow
1225  * direct reads.
1226  *
1227  * @mr: the memory region being queried
1228  */
1229 static inline bool memory_region_is_romd(MemoryRegion *mr)
1230 {
1231     return mr->rom_device && mr->romd_mode;
1232 }
1233 
1234 /**
1235  * memory_region_get_iommu: check whether a memory region is an iommu
1236  *
1237  * Returns pointer to IOMMUMemoryRegion if a memory region is an iommu,
1238  * otherwise NULL.
1239  *
1240  * @mr: the memory region being queried
1241  */
1242 static inline IOMMUMemoryRegion *memory_region_get_iommu(MemoryRegion *mr)
1243 {
1244     if (mr->alias) {
1245         return memory_region_get_iommu(mr->alias);
1246     }
1247     if (mr->is_iommu) {
1248         return (IOMMUMemoryRegion *) mr;
1249     }
1250     return NULL;
1251 }
1252 
1253 /**
1254  * memory_region_get_iommu_class_nocheck: returns iommu memory region class
1255  *   if an iommu or NULL if not
1256  *
1257  * Returns pointer to IOMMUMemoryRegionClass if a memory region is an iommu,
1258  * otherwise NULL. This is fast path avoiding QOM checking, use with caution.
1259  *
1260  * @iommu_mr: the memory region being queried
1261  */
1262 static inline IOMMUMemoryRegionClass *memory_region_get_iommu_class_nocheck(
1263         IOMMUMemoryRegion *iommu_mr)
1264 {
1265     return (IOMMUMemoryRegionClass *) (((Object *)iommu_mr)->class);
1266 }
1267 
1268 #define memory_region_is_iommu(mr) (memory_region_get_iommu(mr) != NULL)
1269 
1270 /**
1271  * memory_region_iommu_get_min_page_size: get minimum supported page size
1272  * for an iommu
1273  *
1274  * Returns minimum supported page size for an iommu.
1275  *
1276  * @iommu_mr: the memory region being queried
1277  */
1278 uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr);
1279 
1280 /**
1281  * memory_region_notify_iommu: notify a change in an IOMMU translation entry.
1282  *
1283  * The notification type will be decided by entry.perm bits:
1284  *
1285  * - For UNMAP (cache invalidation) notifies: set entry.perm to IOMMU_NONE.
1286  * - For MAP (newly added entry) notifies: set entry.perm to the
1287  *   permission of the page (which is definitely !IOMMU_NONE).
1288  *
1289  * Note: for any IOMMU implementation, an in-place mapping change
1290  * should be notified with an UNMAP followed by a MAP.
1291  *
1292  * @iommu_mr: the memory region that was changed
1293  * @iommu_idx: the IOMMU index for the translation table which has changed
1294  * @entry: the new entry in the IOMMU translation table.  The entry
1295  *         replaces all old entries for the same virtual I/O address range.
1296  *         Deleted entries have .@perm == 0.
1297  */
1298 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
1299                                 int iommu_idx,
1300                                 IOMMUTLBEntry entry);
1301 
1302 /**
1303  * memory_region_notify_one: notify a change in an IOMMU translation
1304  *                           entry to a single notifier
1305  *
1306  * This works just like memory_region_notify_iommu(), but it only
1307  * notifies a specific notifier, not all of them.
1308  *
1309  * @notifier: the notifier to be notified
1310  * @entry: the new entry in the IOMMU translation table.  The entry
1311  *         replaces all old entries for the same virtual I/O address range.
1312  *         Deleted entries have .@perm == 0.
1313  */
1314 void memory_region_notify_one(IOMMUNotifier *notifier,
1315                               IOMMUTLBEntry *entry);
1316 
1317 /**
1318  * memory_region_register_iommu_notifier: register a notifier for changes to
1319  * IOMMU translation entries.
1320  *
1321  * Returns 0 on success, or a negative errno otherwise. In particular,
1322  * -EINVAL indicates that at least one of the attributes of the notifier
1323  * is not supported (flag/range) by the IOMMU memory region. In case of error
1324  * the error object must be created.
1325  *
1326  * @mr: the memory region to observe
1327  * @n: the IOMMUNotifier to be added; the notify callback receives a
1328  *     pointer to an #IOMMUTLBEntry as the opaque value; the pointer
1329  *     ceases to be valid on exit from the notifier.
1330  * @errp: pointer to Error*, to store an error if it happens.
1331  */
1332 int memory_region_register_iommu_notifier(MemoryRegion *mr,
1333                                           IOMMUNotifier *n, Error **errp);
1334 
1335 /**
1336  * memory_region_iommu_replay: replay existing IOMMU translations to
1337  * a notifier with the minimum page granularity returned by
1338  * mr->iommu_ops->get_page_size().
1339  *
1340  * Note: this is not related to record-and-replay functionality.
1341  *
1342  * @iommu_mr: the memory region to observe
1343  * @n: the notifier to which to replay iommu mappings
1344  */
1345 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
1346 
1347 /**
1348  * memory_region_unregister_iommu_notifier: unregister a notifier for
1349  * changes to IOMMU translation entries.
1350  *
1351  * @mr: the memory region which was observed and for which notity_stopped()
1352  *      needs to be called
1353  * @n: the notifier to be removed.
1354  */
1355 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
1356                                              IOMMUNotifier *n);
1357 
1358 /**
1359  * memory_region_iommu_get_attr: return an IOMMU attr if get_attr() is
1360  * defined on the IOMMU.
1361  *
1362  * Returns 0 on success, or a negative errno otherwise. In particular,
1363  * -EINVAL indicates that the IOMMU does not support the requested
1364  * attribute.
1365  *
1366  * @iommu_mr: the memory region
1367  * @attr: the requested attribute
1368  * @data: a pointer to the requested attribute data
1369  */
1370 int memory_region_iommu_get_attr(IOMMUMemoryRegion *iommu_mr,
1371                                  enum IOMMUMemoryRegionAttr attr,
1372                                  void *data);
1373 
1374 /**
1375  * memory_region_iommu_attrs_to_index: return the IOMMU index to
1376  * use for translations with the given memory transaction attributes.
1377  *
1378  * @iommu_mr: the memory region
1379  * @attrs: the memory transaction attributes
1380  */
1381 int memory_region_iommu_attrs_to_index(IOMMUMemoryRegion *iommu_mr,
1382                                        MemTxAttrs attrs);
1383 
1384 /**
1385  * memory_region_iommu_num_indexes: return the total number of IOMMU
1386  * indexes that this IOMMU supports.
1387  *
1388  * @iommu_mr: the memory region
1389  */
1390 int memory_region_iommu_num_indexes(IOMMUMemoryRegion *iommu_mr);
1391 
1392 /**
1393  * memory_region_name: get a memory region's name
1394  *
1395  * Returns the string that was used to initialize the memory region.
1396  *
1397  * @mr: the memory region being queried
1398  */
1399 const char *memory_region_name(const MemoryRegion *mr);
1400 
1401 /**
1402  * memory_region_is_logging: return whether a memory region is logging writes
1403  *
1404  * Returns %true if the memory region is logging writes for the given client
1405  *
1406  * @mr: the memory region being queried
1407  * @client: the client being queried
1408  */
1409 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client);
1410 
1411 /**
1412  * memory_region_get_dirty_log_mask: return the clients for which a
1413  * memory region is logging writes.
1414  *
1415  * Returns a bitmap of clients, in which the DIRTY_MEMORY_* constants
1416  * are the bit indices.
1417  *
1418  * @mr: the memory region being queried
1419  */
1420 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr);
1421 
1422 /**
1423  * memory_region_is_rom: check whether a memory region is ROM
1424  *
1425  * Returns %true if a memory region is read-only memory.
1426  *
1427  * @mr: the memory region being queried
1428  */
1429 static inline bool memory_region_is_rom(MemoryRegion *mr)
1430 {
1431     return mr->ram && mr->readonly;
1432 }
1433 
1434 /**
1435  * memory_region_is_nonvolatile: check whether a memory region is non-volatile
1436  *
1437  * Returns %true is a memory region is non-volatile memory.
1438  *
1439  * @mr: the memory region being queried
1440  */
1441 static inline bool memory_region_is_nonvolatile(MemoryRegion *mr)
1442 {
1443     return mr->nonvolatile;
1444 }
1445 
1446 /**
1447  * memory_region_get_fd: Get a file descriptor backing a RAM memory region.
1448  *
1449  * Returns a file descriptor backing a file-based RAM memory region,
1450  * or -1 if the region is not a file-based RAM memory region.
1451  *
1452  * @mr: the RAM or alias memory region being queried.
1453  */
1454 int memory_region_get_fd(MemoryRegion *mr);
1455 
1456 /**
1457  * memory_region_from_host: Convert a pointer into a RAM memory region
1458  * and an offset within it.
1459  *
1460  * Given a host pointer inside a RAM memory region (created with
1461  * memory_region_init_ram() or memory_region_init_ram_ptr()), return
1462  * the MemoryRegion and the offset within it.
1463  *
1464  * Use with care; by the time this function returns, the returned pointer is
1465  * not protected by RCU anymore.  If the caller is not within an RCU critical
1466  * section and does not hold the iothread lock, it must have other means of
1467  * protecting the pointer, such as a reference to the region that includes
1468  * the incoming ram_addr_t.
1469  *
1470  * @ptr: the host pointer to be converted
1471  * @offset: the offset within memory region
1472  */
1473 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset);
1474 
1475 /**
1476  * memory_region_get_ram_ptr: Get a pointer into a RAM memory region.
1477  *
1478  * Returns a host pointer to a RAM memory region (created with
1479  * memory_region_init_ram() or memory_region_init_ram_ptr()).
1480  *
1481  * Use with care; by the time this function returns, the returned pointer is
1482  * not protected by RCU anymore.  If the caller is not within an RCU critical
1483  * section and does not hold the iothread lock, it must have other means of
1484  * protecting the pointer, such as a reference to the region that includes
1485  * the incoming ram_addr_t.
1486  *
1487  * @mr: the memory region being queried.
1488  */
1489 void *memory_region_get_ram_ptr(MemoryRegion *mr);
1490 
1491 /* memory_region_ram_resize: Resize a RAM region.
1492  *
1493  * Only legal before guest might have detected the memory size: e.g. on
1494  * incoming migration, or right after reset.
1495  *
1496  * @mr: a memory region created with @memory_region_init_resizeable_ram.
1497  * @newsize: the new size the region
1498  * @errp: pointer to Error*, to store an error if it happens.
1499  */
1500 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize,
1501                               Error **errp);
1502 
1503 /**
1504  * memory_region_msync: Synchronize selected address range of
1505  * a memory mapped region
1506  *
1507  * @mr: the memory region to be msync
1508  * @addr: the initial address of the range to be sync
1509  * @size: the size of the range to be sync
1510  */
1511 void memory_region_msync(MemoryRegion *mr, hwaddr addr, hwaddr size);
1512 
1513 /**
1514  * memory_region_writeback: Trigger cache writeback for
1515  * selected address range
1516  *
1517  * @mr: the memory region to be updated
1518  * @addr: the initial address of the range to be written back
1519  * @size: the size of the range to be written back
1520  */
1521 void memory_region_writeback(MemoryRegion *mr, hwaddr addr, hwaddr size);
1522 
1523 /**
1524  * memory_region_set_log: Turn dirty logging on or off for a region.
1525  *
1526  * Turns dirty logging on or off for a specified client (display, migration).
1527  * Only meaningful for RAM regions.
1528  *
1529  * @mr: the memory region being updated.
1530  * @log: whether dirty logging is to be enabled or disabled.
1531  * @client: the user of the logging information; %DIRTY_MEMORY_VGA only.
1532  */
1533 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client);
1534 
1535 /**
1536  * memory_region_set_dirty: Mark a range of bytes as dirty in a memory region.
1537  *
1538  * Marks a range of bytes as dirty, after it has been dirtied outside
1539  * guest code.
1540  *
1541  * @mr: the memory region being dirtied.
1542  * @addr: the address (relative to the start of the region) being dirtied.
1543  * @size: size of the range being dirtied.
1544  */
1545 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1546                              hwaddr size);
1547 
1548 /**
1549  * memory_region_clear_dirty_bitmap - clear dirty bitmap for memory range
1550  *
1551  * This function is called when the caller wants to clear the remote
1552  * dirty bitmap of a memory range within the memory region.  This can
1553  * be used by e.g. KVM to manually clear dirty log when
1554  * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is declared support by the host
1555  * kernel.
1556  *
1557  * @mr:     the memory region to clear the dirty log upon
1558  * @start:  start address offset within the memory region
1559  * @len:    length of the memory region to clear dirty bitmap
1560  */
1561 void memory_region_clear_dirty_bitmap(MemoryRegion *mr, hwaddr start,
1562                                       hwaddr len);
1563 
1564 /**
1565  * memory_region_snapshot_and_clear_dirty: Get a snapshot of the dirty
1566  *                                         bitmap and clear it.
1567  *
1568  * Creates a snapshot of the dirty bitmap, clears the dirty bitmap and
1569  * returns the snapshot.  The snapshot can then be used to query dirty
1570  * status, using memory_region_snapshot_get_dirty.  Snapshotting allows
1571  * querying the same page multiple times, which is especially useful for
1572  * display updates where the scanlines often are not page aligned.
1573  *
1574  * The dirty bitmap region which gets copyed into the snapshot (and
1575  * cleared afterwards) can be larger than requested.  The boundaries
1576  * are rounded up/down so complete bitmap longs (covering 64 pages on
1577  * 64bit hosts) can be copied over into the bitmap snapshot.  Which
1578  * isn't a problem for display updates as the extra pages are outside
1579  * the visible area, and in case the visible area changes a full
1580  * display redraw is due anyway.  Should other use cases for this
1581  * function emerge we might have to revisit this implementation
1582  * detail.
1583  *
1584  * Use g_free to release DirtyBitmapSnapshot.
1585  *
1586  * @mr: the memory region being queried.
1587  * @addr: the address (relative to the start of the region) being queried.
1588  * @size: the size of the range being queried.
1589  * @client: the user of the logging information; typically %DIRTY_MEMORY_VGA.
1590  */
1591 DirtyBitmapSnapshot *memory_region_snapshot_and_clear_dirty(MemoryRegion *mr,
1592                                                             hwaddr addr,
1593                                                             hwaddr size,
1594                                                             unsigned client);
1595 
1596 /**
1597  * memory_region_snapshot_get_dirty: Check whether a range of bytes is dirty
1598  *                                   in the specified dirty bitmap snapshot.
1599  *
1600  * @mr: the memory region being queried.
1601  * @snap: the dirty bitmap snapshot
1602  * @addr: the address (relative to the start of the region) being queried.
1603  * @size: the size of the range being queried.
1604  */
1605 bool memory_region_snapshot_get_dirty(MemoryRegion *mr,
1606                                       DirtyBitmapSnapshot *snap,
1607                                       hwaddr addr, hwaddr size);
1608 
1609 /**
1610  * memory_region_reset_dirty: Mark a range of pages as clean, for a specified
1611  *                            client.
1612  *
1613  * Marks a range of pages as no longer dirty.
1614  *
1615  * @mr: the region being updated.
1616  * @addr: the start of the subrange being cleaned.
1617  * @size: the size of the subrange being cleaned.
1618  * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or
1619  *          %DIRTY_MEMORY_VGA.
1620  */
1621 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1622                                hwaddr size, unsigned client);
1623 
1624 /**
1625  * memory_region_flush_rom_device: Mark a range of pages dirty and invalidate
1626  *                                 TBs (for self-modifying code).
1627  *
1628  * The MemoryRegionOps->write() callback of a ROM device must use this function
1629  * to mark byte ranges that have been modified internally, such as by directly
1630  * accessing the memory returned by memory_region_get_ram_ptr().
1631  *
1632  * This function marks the range dirty and invalidates TBs so that TCG can
1633  * detect self-modifying code.
1634  *
1635  * @mr: the region being flushed.
1636  * @addr: the start, relative to the start of the region, of the range being
1637  *        flushed.
1638  * @size: the size, in bytes, of the range being flushed.
1639  */
1640 void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size);
1641 
1642 /**
1643  * memory_region_set_readonly: Turn a memory region read-only (or read-write)
1644  *
1645  * Allows a memory region to be marked as read-only (turning it into a ROM).
1646  * only useful on RAM regions.
1647  *
1648  * @mr: the region being updated.
1649  * @readonly: whether rhe region is to be ROM or RAM.
1650  */
1651 void memory_region_set_readonly(MemoryRegion *mr, bool readonly);
1652 
1653 /**
1654  * memory_region_set_nonvolatile: Turn a memory region non-volatile
1655  *
1656  * Allows a memory region to be marked as non-volatile.
1657  * only useful on RAM regions.
1658  *
1659  * @mr: the region being updated.
1660  * @nonvolatile: whether rhe region is to be non-volatile.
1661  */
1662 void memory_region_set_nonvolatile(MemoryRegion *mr, bool nonvolatile);
1663 
1664 /**
1665  * memory_region_rom_device_set_romd: enable/disable ROMD mode
1666  *
1667  * Allows a ROM device (initialized with memory_region_init_rom_device() to
1668  * set to ROMD mode (default) or MMIO mode.  When it is in ROMD mode, the
1669  * device is mapped to guest memory and satisfies read access directly.
1670  * When in MMIO mode, reads are forwarded to the #MemoryRegion.read function.
1671  * Writes are always handled by the #MemoryRegion.write function.
1672  *
1673  * @mr: the memory region to be updated
1674  * @romd_mode: %true to put the region into ROMD mode
1675  */
1676 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode);
1677 
1678 /**
1679  * memory_region_set_coalescing: Enable memory coalescing for the region.
1680  *
1681  * Enabled writes to a region to be queued for later processing. MMIO ->write
1682  * callbacks may be delayed until a non-coalesced MMIO is issued.
1683  * Only useful for IO regions.  Roughly similar to write-combining hardware.
1684  *
1685  * @mr: the memory region to be write coalesced
1686  */
1687 void memory_region_set_coalescing(MemoryRegion *mr);
1688 
1689 /**
1690  * memory_region_add_coalescing: Enable memory coalescing for a sub-range of
1691  *                               a region.
1692  *
1693  * Like memory_region_set_coalescing(), but works on a sub-range of a region.
1694  * Multiple calls can be issued coalesced disjoint ranges.
1695  *
1696  * @mr: the memory region to be updated.
1697  * @offset: the start of the range within the region to be coalesced.
1698  * @size: the size of the subrange to be coalesced.
1699  */
1700 void memory_region_add_coalescing(MemoryRegion *mr,
1701                                   hwaddr offset,
1702                                   uint64_t size);
1703 
1704 /**
1705  * memory_region_clear_coalescing: Disable MMIO coalescing for the region.
1706  *
1707  * Disables any coalescing caused by memory_region_set_coalescing() or
1708  * memory_region_add_coalescing().  Roughly equivalent to uncacheble memory
1709  * hardware.
1710  *
1711  * @mr: the memory region to be updated.
1712  */
1713 void memory_region_clear_coalescing(MemoryRegion *mr);
1714 
1715 /**
1716  * memory_region_set_flush_coalesced: Enforce memory coalescing flush before
1717  *                                    accesses.
1718  *
1719  * Ensure that pending coalesced MMIO request are flushed before the memory
1720  * region is accessed. This property is automatically enabled for all regions
1721  * passed to memory_region_set_coalescing() and memory_region_add_coalescing().
1722  *
1723  * @mr: the memory region to be updated.
1724  */
1725 void memory_region_set_flush_coalesced(MemoryRegion *mr);
1726 
1727 /**
1728  * memory_region_clear_flush_coalesced: Disable memory coalescing flush before
1729  *                                      accesses.
1730  *
1731  * Clear the automatic coalesced MMIO flushing enabled via
1732  * memory_region_set_flush_coalesced. Note that this service has no effect on
1733  * memory regions that have MMIO coalescing enabled for themselves. For them,
1734  * automatic flushing will stop once coalescing is disabled.
1735  *
1736  * @mr: the memory region to be updated.
1737  */
1738 void memory_region_clear_flush_coalesced(MemoryRegion *mr);
1739 
1740 /**
1741  * memory_region_add_eventfd: Request an eventfd to be triggered when a word
1742  *                            is written to a location.
1743  *
1744  * Marks a word in an IO region (initialized with memory_region_init_io())
1745  * as a trigger for an eventfd event.  The I/O callback will not be called.
1746  * The caller must be prepared to handle failure (that is, take the required
1747  * action if the callback _is_ called).
1748  *
1749  * @mr: the memory region being updated.
1750  * @addr: the address within @mr that is to be monitored
1751  * @size: the size of the access to trigger the eventfd
1752  * @match_data: whether to match against @data, instead of just @addr
1753  * @data: the data to match against the guest write
1754  * @e: event notifier to be triggered when @addr, @size, and @data all match.
1755  **/
1756 void memory_region_add_eventfd(MemoryRegion *mr,
1757                                hwaddr addr,
1758                                unsigned size,
1759                                bool match_data,
1760                                uint64_t data,
1761                                EventNotifier *e);
1762 
1763 /**
1764  * memory_region_del_eventfd: Cancel an eventfd.
1765  *
1766  * Cancels an eventfd trigger requested by a previous
1767  * memory_region_add_eventfd() call.
1768  *
1769  * @mr: the memory region being updated.
1770  * @addr: the address within @mr that is to be monitored
1771  * @size: the size of the access to trigger the eventfd
1772  * @match_data: whether to match against @data, instead of just @addr
1773  * @data: the data to match against the guest write
1774  * @e: event notifier to be triggered when @addr, @size, and @data all match.
1775  */
1776 void memory_region_del_eventfd(MemoryRegion *mr,
1777                                hwaddr addr,
1778                                unsigned size,
1779                                bool match_data,
1780                                uint64_t data,
1781                                EventNotifier *e);
1782 
1783 /**
1784  * memory_region_add_subregion: Add a subregion to a container.
1785  *
1786  * Adds a subregion at @offset.  The subregion may not overlap with other
1787  * subregions (except for those explicitly marked as overlapping).  A region
1788  * may only be added once as a subregion (unless removed with
1789  * memory_region_del_subregion()); use memory_region_init_alias() if you
1790  * want a region to be a subregion in multiple locations.
1791  *
1792  * @mr: the region to contain the new subregion; must be a container
1793  *      initialized with memory_region_init().
1794  * @offset: the offset relative to @mr where @subregion is added.
1795  * @subregion: the subregion to be added.
1796  */
1797 void memory_region_add_subregion(MemoryRegion *mr,
1798                                  hwaddr offset,
1799                                  MemoryRegion *subregion);
1800 /**
1801  * memory_region_add_subregion_overlap: Add a subregion to a container
1802  *                                      with overlap.
1803  *
1804  * Adds a subregion at @offset.  The subregion may overlap with other
1805  * subregions.  Conflicts are resolved by having a higher @priority hide a
1806  * lower @priority. Subregions without priority are taken as @priority 0.
1807  * A region may only be added once as a subregion (unless removed with
1808  * memory_region_del_subregion()); use memory_region_init_alias() if you
1809  * want a region to be a subregion in multiple locations.
1810  *
1811  * @mr: the region to contain the new subregion; must be a container
1812  *      initialized with memory_region_init().
1813  * @offset: the offset relative to @mr where @subregion is added.
1814  * @subregion: the subregion to be added.
1815  * @priority: used for resolving overlaps; highest priority wins.
1816  */
1817 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1818                                          hwaddr offset,
1819                                          MemoryRegion *subregion,
1820                                          int priority);
1821 
1822 /**
1823  * memory_region_get_ram_addr: Get the ram address associated with a memory
1824  *                             region
1825  *
1826  * @mr: the region to be queried
1827  */
1828 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr);
1829 
1830 uint64_t memory_region_get_alignment(const MemoryRegion *mr);
1831 /**
1832  * memory_region_del_subregion: Remove a subregion.
1833  *
1834  * Removes a subregion from its container.
1835  *
1836  * @mr: the container to be updated.
1837  * @subregion: the region being removed; must be a current subregion of @mr.
1838  */
1839 void memory_region_del_subregion(MemoryRegion *mr,
1840                                  MemoryRegion *subregion);
1841 
1842 /*
1843  * memory_region_set_enabled: dynamically enable or disable a region
1844  *
1845  * Enables or disables a memory region.  A disabled memory region
1846  * ignores all accesses to itself and its subregions.  It does not
1847  * obscure sibling subregions with lower priority - it simply behaves as
1848  * if it was removed from the hierarchy.
1849  *
1850  * Regions default to being enabled.
1851  *
1852  * @mr: the region to be updated
1853  * @enabled: whether to enable or disable the region
1854  */
1855 void memory_region_set_enabled(MemoryRegion *mr, bool enabled);
1856 
1857 /*
1858  * memory_region_set_address: dynamically update the address of a region
1859  *
1860  * Dynamically updates the address of a region, relative to its container.
1861  * May be used on regions are currently part of a memory hierarchy.
1862  *
1863  * @mr: the region to be updated
1864  * @addr: new address, relative to container region
1865  */
1866 void memory_region_set_address(MemoryRegion *mr, hwaddr addr);
1867 
1868 /*
1869  * memory_region_set_size: dynamically update the size of a region.
1870  *
1871  * Dynamically updates the size of a region.
1872  *
1873  * @mr: the region to be updated
1874  * @size: used size of the region.
1875  */
1876 void memory_region_set_size(MemoryRegion *mr, uint64_t size);
1877 
1878 /*
1879  * memory_region_set_alias_offset: dynamically update a memory alias's offset
1880  *
1881  * Dynamically updates the offset into the target region that an alias points
1882  * to, as if the fourth argument to memory_region_init_alias() has changed.
1883  *
1884  * @mr: the #MemoryRegion to be updated; should be an alias.
1885  * @offset: the new offset into the target memory region
1886  */
1887 void memory_region_set_alias_offset(MemoryRegion *mr,
1888                                     hwaddr offset);
1889 
1890 /**
1891  * memory_region_present: checks if an address relative to a @container
1892  * translates into #MemoryRegion within @container
1893  *
1894  * Answer whether a #MemoryRegion within @container covers the address
1895  * @addr.
1896  *
1897  * @container: a #MemoryRegion within which @addr is a relative address
1898  * @addr: the area within @container to be searched
1899  */
1900 bool memory_region_present(MemoryRegion *container, hwaddr addr);
1901 
1902 /**
1903  * memory_region_is_mapped: returns true if #MemoryRegion is mapped
1904  * into any address space.
1905  *
1906  * @mr: a #MemoryRegion which should be checked if it's mapped
1907  */
1908 bool memory_region_is_mapped(MemoryRegion *mr);
1909 
1910 /**
1911  * memory_region_find: translate an address/size relative to a
1912  * MemoryRegion into a #MemoryRegionSection.
1913  *
1914  * Locates the first #MemoryRegion within @mr that overlaps the range
1915  * given by @addr and @size.
1916  *
1917  * Returns a #MemoryRegionSection that describes a contiguous overlap.
1918  * It will have the following characteristics:
1919  * - @size = 0 iff no overlap was found
1920  * - @mr is non-%NULL iff an overlap was found
1921  *
1922  * Remember that in the return value the @offset_within_region is
1923  * relative to the returned region (in the .@mr field), not to the
1924  * @mr argument.
1925  *
1926  * Similarly, the .@offset_within_address_space is relative to the
1927  * address space that contains both regions, the passed and the
1928  * returned one.  However, in the special case where the @mr argument
1929  * has no container (and thus is the root of the address space), the
1930  * following will hold:
1931  * - @offset_within_address_space >= @addr
1932  * - @offset_within_address_space + .@size <= @addr + @size
1933  *
1934  * @mr: a MemoryRegion within which @addr is a relative address
1935  * @addr: start of the area within @as to be searched
1936  * @size: size of the area to be searched
1937  */
1938 MemoryRegionSection memory_region_find(MemoryRegion *mr,
1939                                        hwaddr addr, uint64_t size);
1940 
1941 /**
1942  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
1943  *
1944  * Synchronizes the dirty page log for all address spaces.
1945  */
1946 void memory_global_dirty_log_sync(void);
1947 
1948 /**
1949  * memory_global_dirty_log_sync: synchronize the dirty log for all memory
1950  *
1951  * Synchronizes the vCPUs with a thread that is reading the dirty bitmap.
1952  * This function must be called after the dirty log bitmap is cleared, and
1953  * before dirty guest memory pages are read.  If you are using
1954  * #DirtyBitmapSnapshot, memory_region_snapshot_and_clear_dirty() takes
1955  * care of doing this.
1956  */
1957 void memory_global_after_dirty_log_sync(void);
1958 
1959 /**
1960  * memory_region_transaction_begin: Start a transaction.
1961  *
1962  * During a transaction, changes will be accumulated and made visible
1963  * only when the transaction ends (is committed).
1964  */
1965 void memory_region_transaction_begin(void);
1966 
1967 /**
1968  * memory_region_transaction_commit: Commit a transaction and make changes
1969  *                                   visible to the guest.
1970  */
1971 void memory_region_transaction_commit(void);
1972 
1973 /**
1974  * memory_listener_register: register callbacks to be called when memory
1975  *                           sections are mapped or unmapped into an address
1976  *                           space
1977  *
1978  * @listener: an object containing the callbacks to be called
1979  * @filter: if non-%NULL, only regions in this address space will be observed
1980  */
1981 void memory_listener_register(MemoryListener *listener, AddressSpace *filter);
1982 
1983 /**
1984  * memory_listener_unregister: undo the effect of memory_listener_register()
1985  *
1986  * @listener: an object containing the callbacks to be removed
1987  */
1988 void memory_listener_unregister(MemoryListener *listener);
1989 
1990 /**
1991  * memory_global_dirty_log_start: begin dirty logging for all regions
1992  */
1993 void memory_global_dirty_log_start(void);
1994 
1995 /**
1996  * memory_global_dirty_log_stop: end dirty logging for all regions
1997  */
1998 void memory_global_dirty_log_stop(void);
1999 
2000 void mtree_info(bool flatview, bool dispatch_tree, bool owner, bool disabled);
2001 
2002 /**
2003  * memory_region_dispatch_read: perform a read directly to the specified
2004  * MemoryRegion.
2005  *
2006  * @mr: #MemoryRegion to access
2007  * @addr: address within that region
2008  * @pval: pointer to uint64_t which the data is written to
2009  * @op: size, sign, and endianness of the memory operation
2010  * @attrs: memory transaction attributes to use for the access
2011  */
2012 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
2013                                         hwaddr addr,
2014                                         uint64_t *pval,
2015                                         MemOp op,
2016                                         MemTxAttrs attrs);
2017 /**
2018  * memory_region_dispatch_write: perform a write directly to the specified
2019  * MemoryRegion.
2020  *
2021  * @mr: #MemoryRegion to access
2022  * @addr: address within that region
2023  * @data: data to write
2024  * @op: size, sign, and endianness of the memory operation
2025  * @attrs: memory transaction attributes to use for the access
2026  */
2027 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
2028                                          hwaddr addr,
2029                                          uint64_t data,
2030                                          MemOp op,
2031                                          MemTxAttrs attrs);
2032 
2033 /**
2034  * address_space_init: initializes an address space
2035  *
2036  * @as: an uninitialized #AddressSpace
2037  * @root: a #MemoryRegion that routes addresses for the address space
2038  * @name: an address space name.  The name is only used for debugging
2039  *        output.
2040  */
2041 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name);
2042 
2043 /**
2044  * address_space_destroy: destroy an address space
2045  *
2046  * Releases all resources associated with an address space.  After an address space
2047  * is destroyed, its root memory region (given by address_space_init()) may be destroyed
2048  * as well.
2049  *
2050  * @as: address space to be destroyed
2051  */
2052 void address_space_destroy(AddressSpace *as);
2053 
2054 /**
2055  * address_space_remove_listeners: unregister all listeners of an address space
2056  *
2057  * Removes all callbacks previously registered with memory_listener_register()
2058  * for @as.
2059  *
2060  * @as: an initialized #AddressSpace
2061  */
2062 void address_space_remove_listeners(AddressSpace *as);
2063 
2064 /**
2065  * address_space_rw: read from or write to an address space.
2066  *
2067  * Return a MemTxResult indicating whether the operation succeeded
2068  * or failed (eg unassigned memory, device rejected the transaction,
2069  * IOMMU fault).
2070  *
2071  * @as: #AddressSpace to be accessed
2072  * @addr: address within that address space
2073  * @attrs: memory transaction attributes
2074  * @buf: buffer with the data transferred
2075  * @len: the number of bytes to read or write
2076  * @is_write: indicates the transfer direction
2077  */
2078 MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
2079                              MemTxAttrs attrs, void *buf,
2080                              hwaddr len, bool is_write);
2081 
2082 /**
2083  * address_space_write: write to address space.
2084  *
2085  * Return a MemTxResult indicating whether the operation succeeded
2086  * or failed (eg unassigned memory, device rejected the transaction,
2087  * IOMMU fault).
2088  *
2089  * @as: #AddressSpace to be accessed
2090  * @addr: address within that address space
2091  * @attrs: memory transaction attributes
2092  * @buf: buffer with the data transferred
2093  * @len: the number of bytes to write
2094  */
2095 MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
2096                                 MemTxAttrs attrs,
2097                                 const void *buf, hwaddr len);
2098 
2099 /**
2100  * address_space_write_rom: write to address space, including ROM.
2101  *
2102  * This function writes to the specified address space, but will
2103  * write data to both ROM and RAM. This is used for non-guest
2104  * writes like writes from the gdb debug stub or initial loading
2105  * of ROM contents.
2106  *
2107  * Note that portions of the write which attempt to write data to
2108  * a device will be silently ignored -- only real RAM and ROM will
2109  * be written to.
2110  *
2111  * Return a MemTxResult indicating whether the operation succeeded
2112  * or failed (eg unassigned memory, device rejected the transaction,
2113  * IOMMU fault).
2114  *
2115  * @as: #AddressSpace to be accessed
2116  * @addr: address within that address space
2117  * @attrs: memory transaction attributes
2118  * @buf: buffer with the data transferred
2119  * @len: the number of bytes to write
2120  */
2121 MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr,
2122                                     MemTxAttrs attrs,
2123                                     const void *buf, hwaddr len);
2124 
2125 /* address_space_ld*: load from an address space
2126  * address_space_st*: store to an address space
2127  *
2128  * These functions perform a load or store of the byte, word,
2129  * longword or quad to the specified address within the AddressSpace.
2130  * The _le suffixed functions treat the data as little endian;
2131  * _be indicates big endian; no suffix indicates "same endianness
2132  * as guest CPU".
2133  *
2134  * The "guest CPU endianness" accessors are deprecated for use outside
2135  * target-* code; devices should be CPU-agnostic and use either the LE
2136  * or the BE accessors.
2137  *
2138  * @as #AddressSpace to be accessed
2139  * @addr: address within that address space
2140  * @val: data value, for stores
2141  * @attrs: memory transaction attributes
2142  * @result: location to write the success/failure of the transaction;
2143  *   if NULL, this information is discarded
2144  */
2145 
2146 #define SUFFIX
2147 #define ARG1         as
2148 #define ARG1_DECL    AddressSpace *as
2149 #include "exec/memory_ldst.h.inc"
2150 
2151 #define SUFFIX
2152 #define ARG1         as
2153 #define ARG1_DECL    AddressSpace *as
2154 #include "exec/memory_ldst_phys.h.inc"
2155 
2156 struct MemoryRegionCache {
2157     void *ptr;
2158     hwaddr xlat;
2159     hwaddr len;
2160     FlatView *fv;
2161     MemoryRegionSection mrs;
2162     bool is_write;
2163 };
2164 
2165 #define MEMORY_REGION_CACHE_INVALID ((MemoryRegionCache) { .mrs.mr = NULL })
2166 
2167 
2168 /* address_space_ld*_cached: load from a cached #MemoryRegion
2169  * address_space_st*_cached: store into a cached #MemoryRegion
2170  *
2171  * These functions perform a load or store of the byte, word,
2172  * longword or quad to the specified address.  The address is
2173  * a physical address in the AddressSpace, but it must lie within
2174  * a #MemoryRegion that was mapped with address_space_cache_init.
2175  *
2176  * The _le suffixed functions treat the data as little endian;
2177  * _be indicates big endian; no suffix indicates "same endianness
2178  * as guest CPU".
2179  *
2180  * The "guest CPU endianness" accessors are deprecated for use outside
2181  * target-* code; devices should be CPU-agnostic and use either the LE
2182  * or the BE accessors.
2183  *
2184  * @cache: previously initialized #MemoryRegionCache to be accessed
2185  * @addr: address within the address space
2186  * @val: data value, for stores
2187  * @attrs: memory transaction attributes
2188  * @result: location to write the success/failure of the transaction;
2189  *   if NULL, this information is discarded
2190  */
2191 
2192 #define SUFFIX       _cached_slow
2193 #define ARG1         cache
2194 #define ARG1_DECL    MemoryRegionCache *cache
2195 #include "exec/memory_ldst.h.inc"
2196 
2197 /* Inline fast path for direct RAM access.  */
2198 static inline uint8_t address_space_ldub_cached(MemoryRegionCache *cache,
2199     hwaddr addr, MemTxAttrs attrs, MemTxResult *result)
2200 {
2201     assert(addr < cache->len);
2202     if (likely(cache->ptr)) {
2203         return ldub_p(cache->ptr + addr);
2204     } else {
2205         return address_space_ldub_cached_slow(cache, addr, attrs, result);
2206     }
2207 }
2208 
2209 static inline void address_space_stb_cached(MemoryRegionCache *cache,
2210     hwaddr addr, uint32_t val, MemTxAttrs attrs, MemTxResult *result)
2211 {
2212     assert(addr < cache->len);
2213     if (likely(cache->ptr)) {
2214         stb_p(cache->ptr + addr, val);
2215     } else {
2216         address_space_stb_cached_slow(cache, addr, val, attrs, result);
2217     }
2218 }
2219 
2220 #define ENDIANNESS   _le
2221 #include "exec/memory_ldst_cached.h.inc"
2222 
2223 #define ENDIANNESS   _be
2224 #include "exec/memory_ldst_cached.h.inc"
2225 
2226 #define SUFFIX       _cached
2227 #define ARG1         cache
2228 #define ARG1_DECL    MemoryRegionCache *cache
2229 #include "exec/memory_ldst_phys.h.inc"
2230 
2231 /* address_space_cache_init: prepare for repeated access to a physical
2232  * memory region
2233  *
2234  * @cache: #MemoryRegionCache to be filled
2235  * @as: #AddressSpace to be accessed
2236  * @addr: address within that address space
2237  * @len: length of buffer
2238  * @is_write: indicates the transfer direction
2239  *
2240  * Will only work with RAM, and may map a subset of the requested range by
2241  * returning a value that is less than @len.  On failure, return a negative
2242  * errno value.
2243  *
2244  * Because it only works with RAM, this function can be used for
2245  * read-modify-write operations.  In this case, is_write should be %true.
2246  *
2247  * Note that addresses passed to the address_space_*_cached functions
2248  * are relative to @addr.
2249  */
2250 int64_t address_space_cache_init(MemoryRegionCache *cache,
2251                                  AddressSpace *as,
2252                                  hwaddr addr,
2253                                  hwaddr len,
2254                                  bool is_write);
2255 
2256 /**
2257  * address_space_cache_invalidate: complete a write to a #MemoryRegionCache
2258  *
2259  * @cache: The #MemoryRegionCache to operate on.
2260  * @addr: The first physical address that was written, relative to the
2261  * address that was passed to @address_space_cache_init.
2262  * @access_len: The number of bytes that were written starting at @addr.
2263  */
2264 void address_space_cache_invalidate(MemoryRegionCache *cache,
2265                                     hwaddr addr,
2266                                     hwaddr access_len);
2267 
2268 /**
2269  * address_space_cache_destroy: free a #MemoryRegionCache
2270  *
2271  * @cache: The #MemoryRegionCache whose memory should be released.
2272  */
2273 void address_space_cache_destroy(MemoryRegionCache *cache);
2274 
2275 /* address_space_get_iotlb_entry: translate an address into an IOTLB
2276  * entry. Should be called from an RCU critical section.
2277  */
2278 IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
2279                                             bool is_write, MemTxAttrs attrs);
2280 
2281 /* address_space_translate: translate an address range into an address space
2282  * into a MemoryRegion and an address range into that section.  Should be
2283  * called from an RCU critical section, to avoid that the last reference
2284  * to the returned region disappears after address_space_translate returns.
2285  *
2286  * @fv: #FlatView to be accessed
2287  * @addr: address within that address space
2288  * @xlat: pointer to address within the returned memory region section's
2289  * #MemoryRegion.
2290  * @len: pointer to length
2291  * @is_write: indicates the transfer direction
2292  * @attrs: memory attributes
2293  */
2294 MemoryRegion *flatview_translate(FlatView *fv,
2295                                  hwaddr addr, hwaddr *xlat,
2296                                  hwaddr *len, bool is_write,
2297                                  MemTxAttrs attrs);
2298 
2299 static inline MemoryRegion *address_space_translate(AddressSpace *as,
2300                                                     hwaddr addr, hwaddr *xlat,
2301                                                     hwaddr *len, bool is_write,
2302                                                     MemTxAttrs attrs)
2303 {
2304     return flatview_translate(address_space_to_flatview(as),
2305                               addr, xlat, len, is_write, attrs);
2306 }
2307 
2308 /* address_space_access_valid: check for validity of accessing an address
2309  * space range
2310  *
2311  * Check whether memory is assigned to the given address space range, and
2312  * access is permitted by any IOMMU regions that are active for the address
2313  * space.
2314  *
2315  * For now, addr and len should be aligned to a page size.  This limitation
2316  * will be lifted in the future.
2317  *
2318  * @as: #AddressSpace to be accessed
2319  * @addr: address within that address space
2320  * @len: length of the area to be checked
2321  * @is_write: indicates the transfer direction
2322  * @attrs: memory attributes
2323  */
2324 bool address_space_access_valid(AddressSpace *as, hwaddr addr, hwaddr len,
2325                                 bool is_write, MemTxAttrs attrs);
2326 
2327 /* address_space_map: map a physical memory region into a host virtual address
2328  *
2329  * May map a subset of the requested range, given by and returned in @plen.
2330  * May return %NULL and set *@plen to zero(0), if resources needed to perform
2331  * the mapping are exhausted.
2332  * Use only for reads OR writes - not for read-modify-write operations.
2333  * Use cpu_register_map_client() to know when retrying the map operation is
2334  * likely to succeed.
2335  *
2336  * @as: #AddressSpace to be accessed
2337  * @addr: address within that address space
2338  * @plen: pointer to length of buffer; updated on return
2339  * @is_write: indicates the transfer direction
2340  * @attrs: memory attributes
2341  */
2342 void *address_space_map(AddressSpace *as, hwaddr addr,
2343                         hwaddr *plen, bool is_write, MemTxAttrs attrs);
2344 
2345 /* address_space_unmap: Unmaps a memory region previously mapped by address_space_map()
2346  *
2347  * Will also mark the memory as dirty if @is_write == %true.  @access_len gives
2348  * the amount of memory that was actually read or written by the caller.
2349  *
2350  * @as: #AddressSpace used
2351  * @buffer: host pointer as returned by address_space_map()
2352  * @len: buffer length as returned by address_space_map()
2353  * @access_len: amount of data actually transferred
2354  * @is_write: indicates the transfer direction
2355  */
2356 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2357                          bool is_write, hwaddr access_len);
2358 
2359 
2360 /* Internal functions, part of the implementation of address_space_read.  */
2361 MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
2362                                     MemTxAttrs attrs, void *buf, hwaddr len);
2363 MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
2364                                    MemTxAttrs attrs, void *buf,
2365                                    hwaddr len, hwaddr addr1, hwaddr l,
2366                                    MemoryRegion *mr);
2367 void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr);
2368 
2369 /* Internal functions, part of the implementation of address_space_read_cached
2370  * and address_space_write_cached.  */
2371 MemTxResult address_space_read_cached_slow(MemoryRegionCache *cache,
2372                                            hwaddr addr, void *buf, hwaddr len);
2373 MemTxResult address_space_write_cached_slow(MemoryRegionCache *cache,
2374                                             hwaddr addr, const void *buf,
2375                                             hwaddr len);
2376 
2377 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
2378 {
2379     if (is_write) {
2380         return memory_region_is_ram(mr) && !mr->readonly &&
2381                !mr->rom_device && !memory_region_is_ram_device(mr);
2382     } else {
2383         return (memory_region_is_ram(mr) && !memory_region_is_ram_device(mr)) ||
2384                memory_region_is_romd(mr);
2385     }
2386 }
2387 
2388 /**
2389  * address_space_read: read from an address space.
2390  *
2391  * Return a MemTxResult indicating whether the operation succeeded
2392  * or failed (eg unassigned memory, device rejected the transaction,
2393  * IOMMU fault).  Called within RCU critical section.
2394  *
2395  * @as: #AddressSpace to be accessed
2396  * @addr: address within that address space
2397  * @attrs: memory transaction attributes
2398  * @buf: buffer with the data transferred
2399  * @len: length of the data transferred
2400  */
2401 static inline __attribute__((__always_inline__))
2402 MemTxResult address_space_read(AddressSpace *as, hwaddr addr,
2403                                MemTxAttrs attrs, void *buf,
2404                                hwaddr len)
2405 {
2406     MemTxResult result = MEMTX_OK;
2407     hwaddr l, addr1;
2408     void *ptr;
2409     MemoryRegion *mr;
2410     FlatView *fv;
2411 
2412     if (__builtin_constant_p(len)) {
2413         if (len) {
2414             RCU_READ_LOCK_GUARD();
2415             fv = address_space_to_flatview(as);
2416             l = len;
2417             mr = flatview_translate(fv, addr, &addr1, &l, false, attrs);
2418             if (len == l && memory_access_is_direct(mr, false)) {
2419                 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
2420                 memcpy(buf, ptr, len);
2421             } else {
2422                 result = flatview_read_continue(fv, addr, attrs, buf, len,
2423                                                 addr1, l, mr);
2424             }
2425         }
2426     } else {
2427         result = address_space_read_full(as, addr, attrs, buf, len);
2428     }
2429     return result;
2430 }
2431 
2432 /**
2433  * address_space_read_cached: read from a cached RAM region
2434  *
2435  * @cache: Cached region to be addressed
2436  * @addr: address relative to the base of the RAM region
2437  * @buf: buffer with the data transferred
2438  * @len: length of the data transferred
2439  */
2440 static inline MemTxResult
2441 address_space_read_cached(MemoryRegionCache *cache, hwaddr addr,
2442                           void *buf, hwaddr len)
2443 {
2444     assert(addr < cache->len && len <= cache->len - addr);
2445     if (likely(cache->ptr)) {
2446         memcpy(buf, cache->ptr + addr, len);
2447         return MEMTX_OK;
2448     } else {
2449         return address_space_read_cached_slow(cache, addr, buf, len);
2450     }
2451 }
2452 
2453 /**
2454  * address_space_write_cached: write to a cached RAM region
2455  *
2456  * @cache: Cached region to be addressed
2457  * @addr: address relative to the base of the RAM region
2458  * @buf: buffer with the data transferred
2459  * @len: length of the data transferred
2460  */
2461 static inline MemTxResult
2462 address_space_write_cached(MemoryRegionCache *cache, hwaddr addr,
2463                            const void *buf, hwaddr len)
2464 {
2465     assert(addr < cache->len && len <= cache->len - addr);
2466     if (likely(cache->ptr)) {
2467         memcpy(cache->ptr + addr, buf, len);
2468         return MEMTX_OK;
2469     } else {
2470         return address_space_write_cached_slow(cache, addr, buf, len);
2471     }
2472 }
2473 
2474 #ifdef NEED_CPU_H
2475 /* enum device_endian to MemOp.  */
2476 static inline MemOp devend_memop(enum device_endian end)
2477 {
2478     QEMU_BUILD_BUG_ON(DEVICE_HOST_ENDIAN != DEVICE_LITTLE_ENDIAN &&
2479                       DEVICE_HOST_ENDIAN != DEVICE_BIG_ENDIAN);
2480 
2481 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
2482     /* Swap if non-host endianness or native (target) endianness */
2483     return (end == DEVICE_HOST_ENDIAN) ? 0 : MO_BSWAP;
2484 #else
2485     const int non_host_endianness =
2486         DEVICE_LITTLE_ENDIAN ^ DEVICE_BIG_ENDIAN ^ DEVICE_HOST_ENDIAN;
2487 
2488     /* In this case, native (target) endianness needs no swap.  */
2489     return (end == non_host_endianness) ? MO_BSWAP : 0;
2490 #endif
2491 }
2492 #endif
2493 
2494 /*
2495  * Inhibit technologies that require discarding of pages in RAM blocks, e.g.,
2496  * to manage the actual amount of memory consumed by the VM (then, the memory
2497  * provided by RAM blocks might be bigger than the desired memory consumption).
2498  * This *must* be set if:
2499  * - Discarding parts of a RAM blocks does not result in the change being
2500  *   reflected in the VM and the pages getting freed.
2501  * - All memory in RAM blocks is pinned or duplicated, invaldiating any previous
2502  *   discards blindly.
2503  * - Discarding parts of a RAM blocks will result in integrity issues (e.g.,
2504  *   encrypted VMs).
2505  * Technologies that only temporarily pin the current working set of a
2506  * driver are fine, because we don't expect such pages to be discarded
2507  * (esp. based on guest action like balloon inflation).
2508  *
2509  * This is *not* to be used to protect from concurrent discards (esp.,
2510  * postcopy).
2511  *
2512  * Returns 0 if successful. Returns -EBUSY if a technology that relies on
2513  * discards to work reliably is active.
2514  */
2515 int ram_block_discard_disable(bool state);
2516 
2517 /*
2518  * Inhibit technologies that disable discarding of pages in RAM blocks.
2519  *
2520  * Returns 0 if successful. Returns -EBUSY if discards are already set to
2521  * broken.
2522  */
2523 int ram_block_discard_require(bool state);
2524 
2525 /*
2526  * Test if discarding of memory in ram blocks is disabled.
2527  */
2528 bool ram_block_discard_is_disabled(void);
2529 
2530 /*
2531  * Test if discarding of memory in ram blocks is required to work reliably.
2532  */
2533 bool ram_block_discard_is_required(void);
2534 
2535 #endif
2536 
2537 #endif
2538