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