?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Bump version to 1.0-rc0
[qemu.git] / memory.c
blobdc5e35d667cc85adc42e1a397b4777c8ec2d316e
1 /*
2 * Physical memory management
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 #include "memory.h"
15 #include "exec-memory.h"
16 #include "ioport.h"
17 #include "bitops.h"
18 #include "kvm.h"
19 #include <assert.h>
20
21 unsigned memory_region_transaction_depth = 0;
22
23 typedef struct AddrRange AddrRange;
24
25 /*
26 * Note using signed integers limits us to physical addresses at most
27 * 63 bits wide. They are needed for negative offsetting in aliases
28 * (large MemoryRegion::alias_offset).
29 */
30 struct AddrRange {
31 int64_t start;
32 int64_t size;
33 };
34
35 static AddrRange addrrange_make(int64_t start, int64_t size)
36 {
37 return (AddrRange) { start, size };
38 }
39
40 static bool addrrange_equal(AddrRange r1, AddrRange r2)
41 {
42 return r1.start == r2.start && r1.size == r2.size;
43 }
44
45 static int64_t addrrange_end(AddrRange r)
46 {
47 return r.start + r.size;
48 }
49
50 static AddrRange addrrange_shift(AddrRange range, int64_t delta)
51 {
52 range.start += delta;
53 return range;
54 }
55
56 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
57 {
58 return (r1.start >= r2.start && (r1.start - r2.start) < r2.size)
59 || (r2.start >= r1.start && (r2.start - r1.start) < r1.size);
60 }
61
62 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
63 {
64 int64_t start = MAX(r1.start, r2.start);
65 /* off-by-one arithmetic to prevent overflow */
66 int64_t end = MIN(addrrange_end(r1) - 1, addrrange_end(r2) - 1);
67 return addrrange_make(start, end - start + 1);
68 }
69
70 struct CoalescedMemoryRange {
71 AddrRange addr;
72 QTAILQ_ENTRY(CoalescedMemoryRange) link;
73 };
74
75 struct MemoryRegionIoeventfd {
76 AddrRange addr;
77 bool match_data;
78 uint64_t data;
79 int fd;
80 };
81
82 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
83 MemoryRegionIoeventfd b)
84 {
85 if (a.addr.start < b.addr.start) {
86 return true;
87 } else if (a.addr.start > b.addr.start) {
88 return false;
89 } else if (a.addr.size < b.addr.size) {
90 return true;
91 } else if (a.addr.size > b.addr.size) {
92 return false;
93 } else if (a.match_data < b.match_data) {
94 return true;
95 } else if (a.match_data > b.match_data) {
96 return false;
97 } else if (a.match_data) {
98 if (a.data < b.data) {
99 return true;
100 } else if (a.data > b.data) {
101 return false;
102 }
103 }
104 if (a.fd < b.fd) {
105 return true;
106 } else if (a.fd > b.fd) {
107 return false;
108 }
109 return false;
110 }
111
112 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
113 MemoryRegionIoeventfd b)
114 {
115 return !memory_region_ioeventfd_before(a, b)
116 && !memory_region_ioeventfd_before(b, a);
117 }
118
119 typedef struct FlatRange FlatRange;
120 typedef struct FlatView FlatView;
121
122 /* Range of memory in the global map. Addresses are absolute. */
123 struct FlatRange {
124 MemoryRegion *mr;
125 target_phys_addr_t offset_in_region;
126 AddrRange addr;
127 uint8_t dirty_log_mask;
128 bool readable;
129 bool readonly;
130 };
131
132 /* Flattened global view of current active memory hierarchy. Kept in sorted
133 * order.
134 */
135 struct FlatView {
136 FlatRange *ranges;
137 unsigned nr;
138 unsigned nr_allocated;
139 };
140
141 typedef struct AddressSpace AddressSpace;
142 typedef struct AddressSpaceOps AddressSpaceOps;
143
144 /* A system address space - I/O, memory, etc. */
145 struct AddressSpace {
146 const AddressSpaceOps *ops;
147 MemoryRegion *root;
148 FlatView current_map;
149 int ioeventfd_nb;
150 MemoryRegionIoeventfd *ioeventfds;
151 };
152
153 struct AddressSpaceOps {
154 void (*range_add)(AddressSpace *as, FlatRange *fr);
155 void (*range_del)(AddressSpace *as, FlatRange *fr);
156 void (*log_start)(AddressSpace *as, FlatRange *fr);
157 void (*log_stop)(AddressSpace *as, FlatRange *fr);
158 void (*ioeventfd_add)(AddressSpace *as, MemoryRegionIoeventfd *fd);
159 void (*ioeventfd_del)(AddressSpace *as, MemoryRegionIoeventfd *fd);
160 };
161
162 #define FOR_EACH_FLAT_RANGE(var, view) \
163 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
164
165 static bool flatrange_equal(FlatRange *a, FlatRange *b)
166 {
167 return a->mr == b->mr
168 && addrrange_equal(a->addr, b->addr)
169 && a->offset_in_region == b->offset_in_region
170 && a->readable == b->readable
171 && a->readonly == b->readonly;
172 }
173
174 static void flatview_init(FlatView *view)
175 {
176 view->ranges = NULL;
177 view->nr = 0;
178 view->nr_allocated = 0;
179 }
180
181 /* Insert a range into a given position. Caller is responsible for maintaining
182 * sorting order.
183 */
184 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
185 {
186 if (view->nr == view->nr_allocated) {
187 view->nr_allocated = MAX(2 * view->nr, 10);
188 view->ranges = g_realloc(view->ranges,
189 view->nr_allocated * sizeof(*view->ranges));
190 }
191 memmove(view->ranges + pos + 1, view->ranges + pos,
192 (view->nr - pos) * sizeof(FlatRange));
193 view->ranges[pos] = *range;
194 ++view->nr;
195 }
196
197 static void flatview_destroy(FlatView *view)
198 {
199 g_free(view->ranges);
200 }
201
202 static bool can_merge(FlatRange *r1, FlatRange *r2)
203 {
204 return addrrange_end(r1->addr) == r2->addr.start
205 && r1->mr == r2->mr
206 && r1->offset_in_region + r1->addr.size == r2->offset_in_region
207 && r1->dirty_log_mask == r2->dirty_log_mask
208 && r1->readable == r2->readable
209 && r1->readonly == r2->readonly;
210 }
211
212 /* Attempt to simplify a view by merging ajacent ranges */
213 static void flatview_simplify(FlatView *view)
214 {
215 unsigned i, j;
216
217 i = 0;
218 while (i < view->nr) {
219 j = i + 1;
220 while (j < view->nr
221 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
222 view->ranges[i].addr.size += view->ranges[j].addr.size;
223 ++j;
224 }
225 ++i;
226 memmove(&view->ranges[i], &view->ranges[j],
227 (view->nr - j) * sizeof(view->ranges[j]));
228 view->nr -= j - i;
229 }
230 }
231
232 static void memory_region_read_accessor(void *opaque,
233 target_phys_addr_t addr,
234 uint64_t *value,
235 unsigned size,
236 unsigned shift,
237 uint64_t mask)
238 {
239 MemoryRegion *mr = opaque;
240 uint64_t tmp;
241
242 tmp = mr->ops->read(mr->opaque, addr, size);
243 *value |= (tmp & mask) << shift;
244 }
245
246 static void memory_region_write_accessor(void *opaque,
247 target_phys_addr_t addr,
248 uint64_t *value,
249 unsigned size,
250 unsigned shift,
251 uint64_t mask)
252 {
253 MemoryRegion *mr = opaque;
254 uint64_t tmp;
255
256 tmp = (*value >> shift) & mask;
257 mr->ops->write(mr->opaque, addr, tmp, size);
258 }
259
260 static void access_with_adjusted_size(target_phys_addr_t addr,
261 uint64_t *value,
262 unsigned size,
263 unsigned access_size_min,
264 unsigned access_size_max,
265 void (*access)(void *opaque,
266 target_phys_addr_t addr,
267 uint64_t *value,
268 unsigned size,
269 unsigned shift,
270 uint64_t mask),
271 void *opaque)
272 {
273 uint64_t access_mask;
274 unsigned access_size;
275 unsigned i;
276
277 if (!access_size_min) {
278 access_size_min = 1;
279 }
280 if (!access_size_max) {
281 access_size_max = 4;
282 }
283 access_size = MAX(MIN(size, access_size_max), access_size_min);
284 access_mask = -1ULL >> (64 - access_size * 8);
285 for (i = 0; i < size; i += access_size) {
286 /* FIXME: big-endian support */
287 access(opaque, addr + i, value, access_size, i * 8, access_mask);
288 }
289 }
290
291 static void memory_region_prepare_ram_addr(MemoryRegion *mr);
292
293 static void as_memory_range_add(AddressSpace *as, FlatRange *fr)
294 {
295 ram_addr_t phys_offset, region_offset;
296
297 memory_region_prepare_ram_addr(fr->mr);
298
299 phys_offset = fr->mr->ram_addr;
300 region_offset = fr->offset_in_region;
301 /* cpu_register_physical_memory_log() wants region_offset for
302 * mmio, but prefers offseting phys_offset for RAM. Humour it.
303 */
304 if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
305 phys_offset += region_offset;
306 region_offset = 0;
307 }
308
309 if (!fr->readable) {
310 phys_offset &= ~TARGET_PAGE_MASK & ~IO_MEM_ROMD;
311 }
312
313 if (fr->readonly) {
314 phys_offset |= IO_MEM_ROM;
315 }
316
317 cpu_register_physical_memory_log(fr->addr.start,
318 fr->addr.size,
319 phys_offset,
320 region_offset,
321 fr->dirty_log_mask);
322 }
323
324 static void as_memory_range_del(AddressSpace *as, FlatRange *fr)
325 {
326 if (fr->dirty_log_mask) {
327 cpu_physical_sync_dirty_bitmap(fr->addr.start,
328 fr->addr.start + fr->addr.size);
329 }
330 cpu_register_physical_memory(fr->addr.start, fr->addr.size,
331 IO_MEM_UNASSIGNED);
332 }
333
334 static void as_memory_log_start(AddressSpace *as, FlatRange *fr)
335 {
336 cpu_physical_log_start(fr->addr.start, fr->addr.size);
337 }
338
339 static void as_memory_log_stop(AddressSpace *as, FlatRange *fr)
340 {
341 cpu_physical_log_stop(fr->addr.start, fr->addr.size);
342 }
343
344 static void as_memory_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
345 {
346 int r;
347
348 assert(fd->match_data && fd->addr.size == 4);
349
350 r = kvm_set_ioeventfd_mmio_long(fd->fd, fd->addr.start, fd->data, true);
351 if (r < 0) {
352 abort();
353 }
354 }
355
356 static void as_memory_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
357 {
358 int r;
359
360 r = kvm_set_ioeventfd_mmio_long(fd->fd, fd->addr.start, fd->data, false);
361 if (r < 0) {
362 abort();
363 }
364 }
365
366 static const AddressSpaceOps address_space_ops_memory = {
367 .range_add = as_memory_range_add,
368 .range_del = as_memory_range_del,
369 .log_start = as_memory_log_start,
370 .log_stop = as_memory_log_stop,
371 .ioeventfd_add = as_memory_ioeventfd_add,
372 .ioeventfd_del = as_memory_ioeventfd_del,
373 };
374
375 static AddressSpace address_space_memory = {
376 .ops = &address_space_ops_memory,
377 };
378
379 static const MemoryRegionPortio *find_portio(MemoryRegion *mr, uint64_t offset,
380 unsigned width, bool write)
381 {
382 const MemoryRegionPortio *mrp;
383
384 for (mrp = mr->ops->old_portio; mrp->size; ++mrp) {
385 if (offset >= mrp->offset && offset < mrp->offset + mrp->len
386 && width == mrp->size
387 && (write ? (bool)mrp->write : (bool)mrp->read)) {
388 return mrp;
389 }
390 }
391 return NULL;
392 }
393
394 static void memory_region_iorange_read(IORange *iorange,
395 uint64_t offset,
396 unsigned width,
397 uint64_t *data)
398 {
399 MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
400
401 if (mr->ops->old_portio) {
402 const MemoryRegionPortio *mrp = find_portio(mr, offset, width, false);
403
404 *data = ((uint64_t)1 << (width * 8)) - 1;
405 if (mrp) {
406 *data = mrp->read(mr->opaque, offset + mr->offset);
407 } else if (width == 2) {
408 mrp = find_portio(mr, offset, 1, false);
409 assert(mrp);
410 *data = mrp->read(mr->opaque, offset + mr->offset) |
411 (mrp->read(mr->opaque, offset + mr->offset + 1) << 8);
412 }
413 return;
414 }
415 *data = 0;
416 access_with_adjusted_size(offset + mr->offset, data, width,
417 mr->ops->impl.min_access_size,
418 mr->ops->impl.max_access_size,
419 memory_region_read_accessor, mr);
420 }
421
422 static void memory_region_iorange_write(IORange *iorange,
423 uint64_t offset,
424 unsigned width,
425 uint64_t data)
426 {
427 MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
428
429 if (mr->ops->old_portio) {
430 const MemoryRegionPortio *mrp = find_portio(mr, offset, width, true);
431
432 if (mrp) {
433 mrp->write(mr->opaque, offset + mr->offset, data);
434 } else if (width == 2) {
435 mrp = find_portio(mr, offset, 1, false);
436 assert(mrp);
437 mrp->write(mr->opaque, offset + mr->offset, data & 0xff);
438 mrp->write(mr->opaque, offset + mr->offset + 1, data >> 8);
439 }
440 return;
441 }
442 access_with_adjusted_size(offset + mr->offset, &data, width,
443 mr->ops->impl.min_access_size,
444 mr->ops->impl.max_access_size,
445 memory_region_write_accessor, mr);
446 }
447
448 static const IORangeOps memory_region_iorange_ops = {
449 .read = memory_region_iorange_read,
450 .write = memory_region_iorange_write,
451 };
452
453 static void as_io_range_add(AddressSpace *as, FlatRange *fr)
454 {
455 iorange_init(&fr->mr->iorange, &memory_region_iorange_ops,
456 fr->addr.start,fr->addr.size);
457 ioport_register(&fr->mr->iorange);
458 }
459
460 static void as_io_range_del(AddressSpace *as, FlatRange *fr)
461 {
462 isa_unassign_ioport(fr->addr.start, fr->addr.size);
463 }
464
465 static void as_io_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
466 {
467 int r;
468
469 assert(fd->match_data && fd->addr.size == 2);
470
471 r = kvm_set_ioeventfd_pio_word(fd->fd, fd->addr.start, fd->data, true);
472 if (r < 0) {
473 abort();
474 }
475 }
476
477 static void as_io_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
478 {
479 int r;
480
481 r = kvm_set_ioeventfd_pio_word(fd->fd, fd->addr.start, fd->data, false);
482 if (r < 0) {
483 abort();
484 }
485 }
486
487 static const AddressSpaceOps address_space_ops_io = {
488 .range_add = as_io_range_add,
489 .range_del = as_io_range_del,
490 .ioeventfd_add = as_io_ioeventfd_add,
491 .ioeventfd_del = as_io_ioeventfd_del,
492 };
493
494 static AddressSpace address_space_io = {
495 .ops = &address_space_ops_io,
496 };
497
498 /* Render a memory region into the global view. Ranges in @view obscure
499 * ranges in @mr.
500 */
501 static void render_memory_region(FlatView *view,
502 MemoryRegion *mr,
503 target_phys_addr_t base,
504 AddrRange clip,
505 bool readonly)
506 {
507 MemoryRegion *subregion;
508 unsigned i;
509 target_phys_addr_t offset_in_region;
510 int64_t remain;
511 int64_t now;
512 FlatRange fr;
513 AddrRange tmp;
514
515 base += mr->addr;
516 readonly |= mr->readonly;
517
518 tmp = addrrange_make(base, mr->size);
519
520 if (!addrrange_intersects(tmp, clip)) {
521 return;
522 }
523
524 clip = addrrange_intersection(tmp, clip);
525
526 if (mr->alias) {
527 base -= mr->alias->addr;
528 base -= mr->alias_offset;
529 render_memory_region(view, mr->alias, base, clip, readonly);
530 return;
531 }
532
533 /* Render subregions in priority order. */
534 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
535 render_memory_region(view, subregion, base, clip, readonly);
536 }
537
538 if (!mr->terminates) {
539 return;
540 }
541
542 offset_in_region = clip.start - base;
543 base = clip.start;
544 remain = clip.size;
545
546 /* Render the region itself into any gaps left by the current view. */
547 for (i = 0; i < view->nr && remain; ++i) {
548 if (base >= addrrange_end(view->ranges[i].addr)) {
549 continue;
550 }
551 if (base < view->ranges[i].addr.start) {
552 now = MIN(remain, view->ranges[i].addr.start - base);
553 fr.mr = mr;
554 fr.offset_in_region = offset_in_region;
555 fr.addr = addrrange_make(base, now);
556 fr.dirty_log_mask = mr->dirty_log_mask;
557 fr.readable = mr->readable;
558 fr.readonly = readonly;
559 flatview_insert(view, i, &fr);
560 ++i;
561 base += now;
562 offset_in_region += now;
563 remain -= now;
564 }
565 if (base == view->ranges[i].addr.start) {
566 now = MIN(remain, view->ranges[i].addr.size);
567 base += now;
568 offset_in_region += now;
569 remain -= now;
570 }
571 }
572 if (remain) {
573 fr.mr = mr;
574 fr.offset_in_region = offset_in_region;
575 fr.addr = addrrange_make(base, remain);
576 fr.dirty_log_mask = mr->dirty_log_mask;
577 fr.readable = mr->readable;
578 fr.readonly = readonly;
579 flatview_insert(view, i, &fr);
580 }
581 }
582
583 /* Render a memory topology into a list of disjoint absolute ranges. */
584 static FlatView generate_memory_topology(MemoryRegion *mr)
585 {
586 FlatView view;
587
588 flatview_init(&view);
589
590 render_memory_region(&view, mr, 0, addrrange_make(0, INT64_MAX), false);
591 flatview_simplify(&view);
592
593 return view;
594 }
595
596 static void address_space_add_del_ioeventfds(AddressSpace *as,
597 MemoryRegionIoeventfd *fds_new,
598 unsigned fds_new_nb,
599 MemoryRegionIoeventfd *fds_old,
600 unsigned fds_old_nb)
601 {
602 unsigned iold, inew;
603
604 /* Generate a symmetric difference of the old and new fd sets, adding
605 * and deleting as necessary.
606 */
607
608 iold = inew = 0;
609 while (iold < fds_old_nb || inew < fds_new_nb) {
610 if (iold < fds_old_nb
611 && (inew == fds_new_nb
612 || memory_region_ioeventfd_before(fds_old[iold],
613 fds_new[inew]))) {
614 as->ops->ioeventfd_del(as, &fds_old[iold]);
615 ++iold;
616 } else if (inew < fds_new_nb
617 && (iold == fds_old_nb
618 || memory_region_ioeventfd_before(fds_new[inew],
619 fds_old[iold]))) {
620 as->ops->ioeventfd_add(as, &fds_new[inew]);
621 ++inew;
622 } else {
623 ++iold;
624 ++inew;
625 }
626 }
627 }
628
629 static void address_space_update_ioeventfds(AddressSpace *as)
630 {
631 FlatRange *fr;
632 unsigned ioeventfd_nb = 0;
633 MemoryRegionIoeventfd *ioeventfds = NULL;
634 AddrRange tmp;
635 unsigned i;
636
637 FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
638 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
639 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
640 fr->addr.start - fr->offset_in_region);
641 if (addrrange_intersects(fr->addr, tmp)) {
642 ++ioeventfd_nb;
643 ioeventfds = g_realloc(ioeventfds,
644 ioeventfd_nb * sizeof(*ioeventfds));
645 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
646 ioeventfds[ioeventfd_nb-1].addr = tmp;
647 }
648 }
649 }
650
651 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
652 as->ioeventfds, as->ioeventfd_nb);
653
654 g_free(as->ioeventfds);
655 as->ioeventfds = ioeventfds;
656 as->ioeventfd_nb = ioeventfd_nb;
657 }
658
659 static void address_space_update_topology_pass(AddressSpace *as,
660 FlatView old_view,
661 FlatView new_view,
662 bool adding)
663 {
664 unsigned iold, inew;
665 FlatRange *frold, *frnew;
666
667 /* Generate a symmetric difference of the old and new memory maps.
668 * Kill ranges in the old map, and instantiate ranges in the new map.
669 */
670 iold = inew = 0;
671 while (iold < old_view.nr || inew < new_view.nr) {
672 if (iold < old_view.nr) {
673 frold = &old_view.ranges[iold];
674 } else {
675 frold = NULL;
676 }
677 if (inew < new_view.nr) {
678 frnew = &new_view.ranges[inew];
679 } else {
680 frnew = NULL;
681 }
682
683 if (frold
684 && (!frnew
685 || frold->addr.start < frnew->addr.start
686 || (frold->addr.start == frnew->addr.start
687 && !flatrange_equal(frold, frnew)))) {
688 /* In old, but (not in new, or in new but attributes changed). */
689
690 if (!adding) {
691 as->ops->range_del(as, frold);
692 }
693
694 ++iold;
695 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
696 /* In both (logging may have changed) */
697
698 if (adding) {
699 if (frold->dirty_log_mask && !frnew->dirty_log_mask) {
700 as->ops->log_stop(as, frnew);
701 } else if (frnew->dirty_log_mask && !frold->dirty_log_mask) {
702 as->ops->log_start(as, frnew);
703 }
704 }
705
706 ++iold;
707 ++inew;
708 } else {
709 /* In new */
710
711 if (adding) {
712 as->ops->range_add(as, frnew);
713 }
714
715 ++inew;
716 }
717 }
718 }
719
720
721 static void address_space_update_topology(AddressSpace *as)
722 {
723 FlatView old_view = as->current_map;
724 FlatView new_view = generate_memory_topology(as->root);
725
726 address_space_update_topology_pass(as, old_view, new_view, false);
727 address_space_update_topology_pass(as, old_view, new_view, true);
728
729 as->current_map = new_view;
730 flatview_destroy(&old_view);
731 address_space_update_ioeventfds(as);
732 }
733
734 static void memory_region_update_topology(void)
735 {
736 if (memory_region_transaction_depth) {
737 return;
738 }
739
740 if (address_space_memory.root) {
741 address_space_update_topology(&address_space_memory);
742 }
743 if (address_space_io.root) {
744 address_space_update_topology(&address_space_io);
745 }
746 }
747
748 void memory_region_transaction_begin(void)
749 {
750 ++memory_region_transaction_depth;
751 }
752
753 void memory_region_transaction_commit(void)
754 {
755 assert(memory_region_transaction_depth);
756 --memory_region_transaction_depth;
757 memory_region_update_topology();
758 }
759
760 static void memory_region_destructor_none(MemoryRegion *mr)
761 {
762 }
763
764 static void memory_region_destructor_ram(MemoryRegion *mr)
765 {
766 qemu_ram_free(mr->ram_addr);
767 }
768
769 static void memory_region_destructor_ram_from_ptr(MemoryRegion *mr)
770 {
771 qemu_ram_free_from_ptr(mr->ram_addr);
772 }
773
774 static void memory_region_destructor_iomem(MemoryRegion *mr)
775 {
776 cpu_unregister_io_memory(mr->ram_addr);
777 }
778
779 static void memory_region_destructor_rom_device(MemoryRegion *mr)
780 {
781 qemu_ram_free(mr->ram_addr & TARGET_PAGE_MASK);
782 cpu_unregister_io_memory(mr->ram_addr & ~(TARGET_PAGE_MASK | IO_MEM_ROMD));
783 }
784
785 void memory_region_init(MemoryRegion *mr,
786 const char *name,
787 uint64_t size)
788 {
789 mr->ops = NULL;
790 mr->parent = NULL;
791 mr->size = size;
792 mr->addr = 0;
793 mr->offset = 0;
794 mr->terminates = false;
795 mr->readable = true;
796 mr->readonly = false;
797 mr->destructor = memory_region_destructor_none;
798 mr->priority = 0;
799 mr->may_overlap = false;
800 mr->alias = NULL;
801 QTAILQ_INIT(&mr->subregions);
802 memset(&mr->subregions_link, 0, sizeof mr->subregions_link);
803 QTAILQ_INIT(&mr->coalesced);
804 mr->name = g_strdup(name);
805 mr->dirty_log_mask = 0;
806 mr->ioeventfd_nb = 0;
807 mr->ioeventfds = NULL;
808 }
809
810 static bool memory_region_access_valid(MemoryRegion *mr,
811 target_phys_addr_t addr,
812 unsigned size)
813 {
814 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
815 return false;
816 }
817
818 /* Treat zero as compatibility all valid */
819 if (!mr->ops->valid.max_access_size) {
820 return true;
821 }
822
823 if (size > mr->ops->valid.max_access_size
824 || size < mr->ops->valid.min_access_size) {
825 return false;
826 }
827 return true;
828 }
829
830 static uint32_t memory_region_read_thunk_n(void *_mr,
831 target_phys_addr_t addr,
832 unsigned size)
833 {
834 MemoryRegion *mr = _mr;
835 uint64_t data = 0;
836
837 if (!memory_region_access_valid(mr, addr, size)) {
838 return -1U; /* FIXME: better signalling */
839 }
840
841 if (!mr->ops->read) {
842 return mr->ops->old_mmio.read[bitops_ffsl(size)](mr->opaque, addr);
843 }
844
845 /* FIXME: support unaligned access */
846 access_with_adjusted_size(addr + mr->offset, &data, size,
847 mr->ops->impl.min_access_size,
848 mr->ops->impl.max_access_size,
849 memory_region_read_accessor, mr);
850
851 return data;
852 }
853
854 static void memory_region_write_thunk_n(void *_mr,
855 target_phys_addr_t addr,
856 unsigned size,
857 uint64_t data)
858 {
859 MemoryRegion *mr = _mr;
860
861 if (!memory_region_access_valid(mr, addr, size)) {
862 return; /* FIXME: better signalling */
863 }
864
865 if (!mr->ops->write) {
866 mr->ops->old_mmio.write[bitops_ffsl(size)](mr->opaque, addr, data);
867 return;
868 }
869
870 /* FIXME: support unaligned access */
871 access_with_adjusted_size(addr + mr->offset, &data, size,
872 mr->ops->impl.min_access_size,
873 mr->ops->impl.max_access_size,
874 memory_region_write_accessor, mr);
875 }
876
877 static uint32_t memory_region_read_thunk_b(void *mr, target_phys_addr_t addr)
878 {
879 return memory_region_read_thunk_n(mr, addr, 1);
880 }
881
882 static uint32_t memory_region_read_thunk_w(void *mr, target_phys_addr_t addr)
883 {
884 return memory_region_read_thunk_n(mr, addr, 2);
885 }
886
887 static uint32_t memory_region_read_thunk_l(void *mr, target_phys_addr_t addr)
888 {
889 return memory_region_read_thunk_n(mr, addr, 4);
890 }
891
892 static void memory_region_write_thunk_b(void *mr, target_phys_addr_t addr,
893 uint32_t data)
894 {
895 memory_region_write_thunk_n(mr, addr, 1, data);
896 }
897
898 static void memory_region_write_thunk_w(void *mr, target_phys_addr_t addr,
899 uint32_t data)
900 {
901 memory_region_write_thunk_n(mr, addr, 2, data);
902 }
903
904 static void memory_region_write_thunk_l(void *mr, target_phys_addr_t addr,
905 uint32_t data)
906 {
907 memory_region_write_thunk_n(mr, addr, 4, data);
908 }
909
910 static CPUReadMemoryFunc * const memory_region_read_thunk[] = {
911 memory_region_read_thunk_b,
912 memory_region_read_thunk_w,
913 memory_region_read_thunk_l,
914 };
915
916 static CPUWriteMemoryFunc * const memory_region_write_thunk[] = {
917 memory_region_write_thunk_b,
918 memory_region_write_thunk_w,
919 memory_region_write_thunk_l,
920 };
921
922 static void memory_region_prepare_ram_addr(MemoryRegion *mr)
923 {
924 if (mr->backend_registered) {
925 return;
926 }
927
928 mr->destructor = memory_region_destructor_iomem;
929 mr->ram_addr = cpu_register_io_memory(memory_region_read_thunk,
930 memory_region_write_thunk,
931 mr,
932 mr->ops->endianness);
933 mr->backend_registered = true;
934 }
935
936 void memory_region_init_io(MemoryRegion *mr,
937 const MemoryRegionOps *ops,
938 void *opaque,
939 const char *name,
940 uint64_t size)
941 {
942 memory_region_init(mr, name, size);
943 mr->ops = ops;
944 mr->opaque = opaque;
945 mr->terminates = true;
946 mr->backend_registered = false;
947 }
948
949 void memory_region_init_ram(MemoryRegion *mr,
950 DeviceState *dev,
951 const char *name,
952 uint64_t size)
953 {
954 memory_region_init(mr, name, size);
955 mr->terminates = true;
956 mr->destructor = memory_region_destructor_ram;
957 mr->ram_addr = qemu_ram_alloc(dev, name, size);
958 mr->backend_registered = true;
959 }
960
961 void memory_region_init_ram_ptr(MemoryRegion *mr,
962 DeviceState *dev,
963 const char *name,
964 uint64_t size,
965 void *ptr)
966 {
967 memory_region_init(mr, name, size);
968 mr->terminates = true;
969 mr->destructor = memory_region_destructor_ram_from_ptr;
970 mr->ram_addr = qemu_ram_alloc_from_ptr(dev, name, size, ptr);
971 mr->backend_registered = true;
972 }
973
974 void memory_region_init_alias(MemoryRegion *mr,
975 const char *name,
976 MemoryRegion *orig,
977 target_phys_addr_t offset,
978 uint64_t size)
979 {
980 memory_region_init(mr, name, size);
981 mr->alias = orig;
982 mr->alias_offset = offset;
983 }
984
985 void memory_region_init_rom_device(MemoryRegion *mr,
986 const MemoryRegionOps *ops,
987 void *opaque,
988 DeviceState *dev,
989 const char *name,
990 uint64_t size)
991 {
992 memory_region_init(mr, name, size);
993 mr->ops = ops;
994 mr->opaque = opaque;
995 mr->terminates = true;
996 mr->destructor = memory_region_destructor_rom_device;
997 mr->ram_addr = qemu_ram_alloc(dev, name, size);
998 mr->ram_addr |= cpu_register_io_memory(memory_region_read_thunk,
999 memory_region_write_thunk,
1000 mr,
1001 mr->ops->endianness);
1002 mr->ram_addr |= IO_MEM_ROMD;
1003 mr->backend_registered = true;
1004 }
1005
1006 void memory_region_destroy(MemoryRegion *mr)
1007 {
1008 assert(QTAILQ_EMPTY(&mr->subregions));
1009 mr->destructor(mr);
1010 memory_region_clear_coalescing(mr);
1011 g_free((char *)mr->name);
1012 g_free(mr->ioeventfds);
1013 }
1014
1015 uint64_t memory_region_size(MemoryRegion *mr)
1016 {
1017 return mr->size;
1018 }
1019
1020 void memory_region_set_offset(MemoryRegion *mr, target_phys_addr_t offset)
1021 {
1022 mr->offset = offset;
1023 }
1024
1025 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1026 {
1027 uint8_t mask = 1 << client;
1028
1029 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1030 memory_region_update_topology();
1031 }
1032
1033 bool memory_region_get_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1034 unsigned client)
1035 {
1036 assert(mr->terminates);
1037 return cpu_physical_memory_get_dirty(mr->ram_addr + addr, 1 << client);
1038 }
1039
1040 void memory_region_set_dirty(MemoryRegion *mr, target_phys_addr_t addr)
1041 {
1042 assert(mr->terminates);
1043 return cpu_physical_memory_set_dirty(mr->ram_addr + addr);
1044 }
1045
1046 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1047 {
1048 FlatRange *fr;
1049
1050 FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1051 if (fr->mr == mr) {
1052 cpu_physical_sync_dirty_bitmap(fr->addr.start,
1053 fr->addr.start + fr->addr.size);
1054 }
1055 }
1056 }
1057
1058 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1059 {
1060 if (mr->readonly != readonly) {
1061 mr->readonly = readonly;
1062 memory_region_update_topology();
1063 }
1064 }
1065
1066 void memory_region_rom_device_set_readable(MemoryRegion *mr, bool readable)
1067 {
1068 if (mr->readable != readable) {
1069 mr->readable = readable;
1070 memory_region_update_topology();
1071 }
1072 }
1073
1074 void memory_region_reset_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1075 target_phys_addr_t size, unsigned client)
1076 {
1077 assert(mr->terminates);
1078 cpu_physical_memory_reset_dirty(mr->ram_addr + addr,
1079 mr->ram_addr + addr + size,
1080 1 << client);
1081 }
1082
1083 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1084 {
1085 if (mr->alias) {
1086 return memory_region_get_ram_ptr(mr->alias) + mr->alias_offset;
1087 }
1088
1089 assert(mr->terminates);
1090
1091 return qemu_get_ram_ptr(mr->ram_addr & TARGET_PAGE_MASK);
1092 }
1093
1094 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1095 {
1096 FlatRange *fr;
1097 CoalescedMemoryRange *cmr;
1098 AddrRange tmp;
1099
1100 FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1101 if (fr->mr == mr) {
1102 qemu_unregister_coalesced_mmio(fr->addr.start, fr->addr.size);
1103 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1104 tmp = addrrange_shift(cmr->addr,
1105 fr->addr.start - fr->offset_in_region);
1106 if (!addrrange_intersects(tmp, fr->addr)) {
1107 continue;
1108 }
1109 tmp = addrrange_intersection(tmp, fr->addr);
1110 qemu_register_coalesced_mmio(tmp.start, tmp.size);
1111 }
1112 }
1113 }
1114 }
1115
1116 void memory_region_set_coalescing(MemoryRegion *mr)
1117 {
1118 memory_region_clear_coalescing(mr);
1119 memory_region_add_coalescing(mr, 0, mr->size);
1120 }
1121
1122 void memory_region_add_coalescing(MemoryRegion *mr,
1123 target_phys_addr_t offset,
1124 uint64_t size)
1125 {
1126 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1127
1128 cmr->addr = addrrange_make(offset, size);
1129 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1130 memory_region_update_coalesced_range(mr);
1131 }
1132
1133 void memory_region_clear_coalescing(MemoryRegion *mr)
1134 {
1135 CoalescedMemoryRange *cmr;
1136
1137 while (!QTAILQ_EMPTY(&mr->coalesced)) {
1138 cmr = QTAILQ_FIRST(&mr->coalesced);
1139 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1140 g_free(cmr);
1141 }
1142 memory_region_update_coalesced_range(mr);
1143 }
1144
1145 void memory_region_add_eventfd(MemoryRegion *mr,
1146 target_phys_addr_t addr,
1147 unsigned size,
1148 bool match_data,
1149 uint64_t data,
1150 int fd)
1151 {
1152 MemoryRegionIoeventfd mrfd = {
1153 .addr.start = addr,
1154 .addr.size = size,
1155 .match_data = match_data,
1156 .data = data,
1157 .fd = fd,
1158 };
1159 unsigned i;
1160
1161 for (i = 0; i < mr->ioeventfd_nb; ++i) {
1162 if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1163 break;
1164 }
1165 }
1166 ++mr->ioeventfd_nb;
1167 mr->ioeventfds = g_realloc(mr->ioeventfds,
1168 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1169 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1170 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1171 mr->ioeventfds[i] = mrfd;
1172 memory_region_update_topology();
1173 }
1174
1175 void memory_region_del_eventfd(MemoryRegion *mr,
1176 target_phys_addr_t addr,
1177 unsigned size,
1178 bool match_data,
1179 uint64_t data,
1180 int fd)
1181 {
1182 MemoryRegionIoeventfd mrfd = {
1183 .addr.start = addr,
1184 .addr.size = size,
1185 .match_data = match_data,
1186 .data = data,
1187 .fd = fd,
1188 };
1189 unsigned i;
1190
1191 for (i = 0; i < mr->ioeventfd_nb; ++i) {
1192 if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1193 break;
1194 }
1195 }
1196 assert(i != mr->ioeventfd_nb);
1197 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1198 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1199 --mr->ioeventfd_nb;
1200 mr->ioeventfds = g_realloc(mr->ioeventfds,
1201 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1202 memory_region_update_topology();
1203 }
1204
1205 static void memory_region_add_subregion_common(MemoryRegion *mr,
1206 target_phys_addr_t offset,
1207 MemoryRegion *subregion)
1208 {
1209 MemoryRegion *other;
1210
1211 assert(!subregion->parent);
1212 subregion->parent = mr;
1213 subregion->addr = offset;
1214 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1215 if (subregion->may_overlap || other->may_overlap) {
1216 continue;
1217 }
1218 if (offset >= other->addr + other->size
1219 || offset + subregion->size <= other->addr) {
1220 continue;
1221 }
1222 #if 0
1223 printf("warning: subregion collision %llx/%llx (%s) "
1224 "vs %llx/%llx (%s)\n",
1225 (unsigned long long)offset,
1226 (unsigned long long)subregion->size,
1227 subregion->name,
1228 (unsigned long long)other->addr,
1229 (unsigned long long)other->size,
1230 other->name);
1231 #endif
1232 }
1233 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1234 if (subregion->priority >= other->priority) {
1235 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1236 goto done;
1237 }
1238 }
1239 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1240 done:
1241 memory_region_update_topology();
1242 }
1243
1244
1245 void memory_region_add_subregion(MemoryRegion *mr,
1246 target_phys_addr_t offset,
1247 MemoryRegion *subregion)
1248 {
1249 subregion->may_overlap = false;
1250 subregion->priority = 0;
1251 memory_region_add_subregion_common(mr, offset, subregion);
1252 }
1253
1254 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1255 target_phys_addr_t offset,
1256 MemoryRegion *subregion,
1257 unsigned priority)
1258 {
1259 subregion->may_overlap = true;
1260 subregion->priority = priority;
1261 memory_region_add_subregion_common(mr, offset, subregion);
1262 }
1263
1264 void memory_region_del_subregion(MemoryRegion *mr,
1265 MemoryRegion *subregion)
1266 {
1267 assert(subregion->parent == mr);
1268 subregion->parent = NULL;
1269 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1270 memory_region_update_topology();
1271 }
1272
1273 void set_system_memory_map(MemoryRegion *mr)
1274 {
1275 address_space_memory.root = mr;
1276 memory_region_update_topology();
1277 }
1278
1279 void set_system_io_map(MemoryRegion *mr)
1280 {
1281 address_space_io.root = mr;
1282 memory_region_update_topology();
1283 }
1284
1285 typedef struct MemoryRegionList MemoryRegionList;
1286
1287 struct MemoryRegionList {
1288 const MemoryRegion *mr;
1289 bool printed;
1290 QTAILQ_ENTRY(MemoryRegionList) queue;
1291 };
1292
1293 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
1294
1295 static void mtree_print_mr(fprintf_function mon_printf, void *f,
1296 const MemoryRegion *mr, unsigned int level,
1297 target_phys_addr_t base,
1298 MemoryRegionListHead *alias_print_queue)
1299 {
1300 MemoryRegionList *new_ml, *ml, *next_ml;
1301 MemoryRegionListHead submr_print_queue;
1302 const MemoryRegion *submr;
1303 unsigned int i;
1304
1305 if (!mr) {
1306 return;
1307 }
1308
1309 for (i = 0; i < level; i++) {
1310 mon_printf(f, " ");
1311 }
1312
1313 if (mr->alias) {
1314 MemoryRegionList *ml;
1315 bool found = false;
1316
1317 /* check if the alias is already in the queue */
1318 QTAILQ_FOREACH(ml, alias_print_queue, queue) {
1319 if (ml->mr == mr->alias && !ml->printed) {
1320 found = true;
1321 }
1322 }
1323
1324 if (!found) {
1325 ml = g_new(MemoryRegionList, 1);
1326 ml->mr = mr->alias;
1327 ml->printed = false;
1328 QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
1329 }
1330 mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d): alias %s @%s "
1331 TARGET_FMT_plx "-" TARGET_FMT_plx "\n",
1332 base + mr->addr,
1333 base + mr->addr + (target_phys_addr_t)mr->size - 1,
1334 mr->priority,
1335 mr->name,
1336 mr->alias->name,
1337 mr->alias_offset,
1338 mr->alias_offset + (target_phys_addr_t)mr->size - 1);
1339 } else {
1340 mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d): %s\n",
1341 base + mr->addr,
1342 base + mr->addr + (target_phys_addr_t)mr->size - 1,
1343 mr->priority,
1344 mr->name);
1345 }
1346
1347 QTAILQ_INIT(&submr_print_queue);
1348
1349 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
1350 new_ml = g_new(MemoryRegionList, 1);
1351 new_ml->mr = submr;
1352 QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
1353 if (new_ml->mr->addr < ml->mr->addr ||
1354 (new_ml->mr->addr == ml->mr->addr &&
1355 new_ml->mr->priority > ml->mr->priority)) {
1356 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
1357 new_ml = NULL;
1358 break;
1359 }
1360 }
1361 if (new_ml) {
1362 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
1363 }
1364 }
1365
1366 QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
1367 mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
1368 alias_print_queue);
1369 }
1370
1371 QTAILQ_FOREACH_SAFE(next_ml, &submr_print_queue, queue, ml) {
1372 g_free(ml);
1373 }
1374 }
1375
1376 void mtree_info(fprintf_function mon_printf, void *f)
1377 {
1378 MemoryRegionListHead ml_head;
1379 MemoryRegionList *ml, *ml2;
1380
1381 QTAILQ_INIT(&ml_head);
1382
1383 mon_printf(f, "memory\n");
1384 mtree_print_mr(mon_printf, f, address_space_memory.root, 0, 0, &ml_head);
1385
1386 /* print aliased regions */
1387 QTAILQ_FOREACH(ml, &ml_head, queue) {
1388 if (!ml->printed) {
1389 mon_printf(f, "%s\n", ml->mr->name);
1390 mtree_print_mr(mon_printf, f, ml->mr, 0, 0, &ml_head);
1391 }
1392 }
1393
1394 QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
1395 g_free(ml2);
1396 }
1397
1398 if (address_space_io.root &&
1399 !QTAILQ_EMPTY(&address_space_io.root->subregions)) {
1400 QTAILQ_INIT(&ml_head);
1401 mon_printf(f, "I/O\n");
1402 mtree_print_mr(mon_printf, f, address_space_io.root, 0, 0, &ml_head);
1403 }
1404 }
QEmu