?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
block: Introduce bdrv_co_flush_to_os
[qemu.git] / memory.c
blobc0c1d12d79a44bef142c5eb45973ee0da987c765
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 Int128 start;
32 Int128 size;
33 };
34
35 static AddrRange addrrange_make(Int128 start, Int128 size)
36 {
37 return (AddrRange) { start, size };
38 }
39
40 static bool addrrange_equal(AddrRange r1, AddrRange r2)
41 {
42 return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
43 }
44
45 static Int128 addrrange_end(AddrRange r)
46 {
47 return int128_add(r.start, r.size);
48 }
49
50 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
51 {
52 int128_addto(&range.start, delta);
53 return range;
54 }
55
56 static bool addrrange_contains(AddrRange range, Int128 addr)
57 {
58 return int128_ge(addr, range.start)
59 && int128_lt(addr, addrrange_end(range));
60 }
61
62 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
63 {
64 return addrrange_contains(r1, r2.start)
65 || addrrange_contains(r2, r1.start);
66 }
67
68 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
69 {
70 Int128 start = int128_max(r1.start, r2.start);
71 Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
72 return addrrange_make(start, int128_sub(end, start));
73 }
74
75 struct CoalescedMemoryRange {
76 AddrRange addr;
77 QTAILQ_ENTRY(CoalescedMemoryRange) link;
78 };
79
80 struct MemoryRegionIoeventfd {
81 AddrRange addr;
82 bool match_data;
83 uint64_t data;
84 int fd;
85 };
86
87 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
88 MemoryRegionIoeventfd b)
89 {
90 if (int128_lt(a.addr.start, b.addr.start)) {
91 return true;
92 } else if (int128_gt(a.addr.start, b.addr.start)) {
93 return false;
94 } else if (int128_lt(a.addr.size, b.addr.size)) {
95 return true;
96 } else if (int128_gt(a.addr.size, b.addr.size)) {
97 return false;
98 } else if (a.match_data < b.match_data) {
99 return true;
100 } else if (a.match_data > b.match_data) {
101 return false;
102 } else if (a.match_data) {
103 if (a.data < b.data) {
104 return true;
105 } else if (a.data > b.data) {
106 return false;
107 }
108 }
109 if (a.fd < b.fd) {
110 return true;
111 } else if (a.fd > b.fd) {
112 return false;
113 }
114 return false;
115 }
116
117 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
118 MemoryRegionIoeventfd b)
119 {
120 return !memory_region_ioeventfd_before(a, b)
121 && !memory_region_ioeventfd_before(b, a);
122 }
123
124 typedef struct FlatRange FlatRange;
125 typedef struct FlatView FlatView;
126
127 /* Range of memory in the global map. Addresses are absolute. */
128 struct FlatRange {
129 MemoryRegion *mr;
130 target_phys_addr_t offset_in_region;
131 AddrRange addr;
132 uint8_t dirty_log_mask;
133 bool readable;
134 bool readonly;
135 };
136
137 /* Flattened global view of current active memory hierarchy. Kept in sorted
138 * order.
139 */
140 struct FlatView {
141 FlatRange *ranges;
142 unsigned nr;
143 unsigned nr_allocated;
144 };
145
146 typedef struct AddressSpace AddressSpace;
147 typedef struct AddressSpaceOps AddressSpaceOps;
148
149 /* A system address space - I/O, memory, etc. */
150 struct AddressSpace {
151 const AddressSpaceOps *ops;
152 MemoryRegion *root;
153 FlatView current_map;
154 int ioeventfd_nb;
155 MemoryRegionIoeventfd *ioeventfds;
156 };
157
158 struct AddressSpaceOps {
159 void (*range_add)(AddressSpace *as, FlatRange *fr);
160 void (*range_del)(AddressSpace *as, FlatRange *fr);
161 void (*log_start)(AddressSpace *as, FlatRange *fr);
162 void (*log_stop)(AddressSpace *as, FlatRange *fr);
163 void (*ioeventfd_add)(AddressSpace *as, MemoryRegionIoeventfd *fd);
164 void (*ioeventfd_del)(AddressSpace *as, MemoryRegionIoeventfd *fd);
165 };
166
167 #define FOR_EACH_FLAT_RANGE(var, view) \
168 for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
169
170 static bool flatrange_equal(FlatRange *a, FlatRange *b)
171 {
172 return a->mr == b->mr
173 && addrrange_equal(a->addr, b->addr)
174 && a->offset_in_region == b->offset_in_region
175 && a->readable == b->readable
176 && a->readonly == b->readonly;
177 }
178
179 static void flatview_init(FlatView *view)
180 {
181 view->ranges = NULL;
182 view->nr = 0;
183 view->nr_allocated = 0;
184 }
185
186 /* Insert a range into a given position. Caller is responsible for maintaining
187 * sorting order.
188 */
189 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
190 {
191 if (view->nr == view->nr_allocated) {
192 view->nr_allocated = MAX(2 * view->nr, 10);
193 view->ranges = g_realloc(view->ranges,
194 view->nr_allocated * sizeof(*view->ranges));
195 }
196 memmove(view->ranges + pos + 1, view->ranges + pos,
197 (view->nr - pos) * sizeof(FlatRange));
198 view->ranges[pos] = *range;
199 ++view->nr;
200 }
201
202 static void flatview_destroy(FlatView *view)
203 {
204 g_free(view->ranges);
205 }
206
207 static bool can_merge(FlatRange *r1, FlatRange *r2)
208 {
209 return int128_eq(addrrange_end(r1->addr), r2->addr.start)
210 && r1->mr == r2->mr
211 && int128_eq(int128_add(int128_make64(r1->offset_in_region),
212 r1->addr.size),
213 int128_make64(r2->offset_in_region))
214 && r1->dirty_log_mask == r2->dirty_log_mask
215 && r1->readable == r2->readable
216 && r1->readonly == r2->readonly;
217 }
218
219 /* Attempt to simplify a view by merging ajacent ranges */
220 static void flatview_simplify(FlatView *view)
221 {
222 unsigned i, j;
223
224 i = 0;
225 while (i < view->nr) {
226 j = i + 1;
227 while (j < view->nr
228 && can_merge(&view->ranges[j-1], &view->ranges[j])) {
229 int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
230 ++j;
231 }
232 ++i;
233 memmove(&view->ranges[i], &view->ranges[j],
234 (view->nr - j) * sizeof(view->ranges[j]));
235 view->nr -= j - i;
236 }
237 }
238
239 static void memory_region_read_accessor(void *opaque,
240 target_phys_addr_t addr,
241 uint64_t *value,
242 unsigned size,
243 unsigned shift,
244 uint64_t mask)
245 {
246 MemoryRegion *mr = opaque;
247 uint64_t tmp;
248
249 tmp = mr->ops->read(mr->opaque, addr, size);
250 *value |= (tmp & mask) << shift;
251 }
252
253 static void memory_region_write_accessor(void *opaque,
254 target_phys_addr_t addr,
255 uint64_t *value,
256 unsigned size,
257 unsigned shift,
258 uint64_t mask)
259 {
260 MemoryRegion *mr = opaque;
261 uint64_t tmp;
262
263 tmp = (*value >> shift) & mask;
264 mr->ops->write(mr->opaque, addr, tmp, size);
265 }
266
267 static void access_with_adjusted_size(target_phys_addr_t addr,
268 uint64_t *value,
269 unsigned size,
270 unsigned access_size_min,
271 unsigned access_size_max,
272 void (*access)(void *opaque,
273 target_phys_addr_t addr,
274 uint64_t *value,
275 unsigned size,
276 unsigned shift,
277 uint64_t mask),
278 void *opaque)
279 {
280 uint64_t access_mask;
281 unsigned access_size;
282 unsigned i;
283
284 if (!access_size_min) {
285 access_size_min = 1;
286 }
287 if (!access_size_max) {
288 access_size_max = 4;
289 }
290 access_size = MAX(MIN(size, access_size_max), access_size_min);
291 access_mask = -1ULL >> (64 - access_size * 8);
292 for (i = 0; i < size; i += access_size) {
293 /* FIXME: big-endian support */
294 access(opaque, addr + i, value, access_size, i * 8, access_mask);
295 }
296 }
297
298 static void memory_region_prepare_ram_addr(MemoryRegion *mr);
299
300 static void as_memory_range_add(AddressSpace *as, FlatRange *fr)
301 {
302 ram_addr_t phys_offset, region_offset;
303
304 memory_region_prepare_ram_addr(fr->mr);
305
306 phys_offset = fr->mr->ram_addr;
307 region_offset = fr->offset_in_region;
308 /* cpu_register_physical_memory_log() wants region_offset for
309 * mmio, but prefers offseting phys_offset for RAM. Humour it.
310 */
311 if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
312 phys_offset += region_offset;
313 region_offset = 0;
314 }
315
316 if (!fr->readable) {
317 phys_offset &= ~TARGET_PAGE_MASK & ~IO_MEM_ROMD;
318 }
319
320 if (fr->readonly) {
321 phys_offset |= IO_MEM_ROM;
322 }
323
324 cpu_register_physical_memory_log(int128_get64(fr->addr.start),
325 int128_get64(fr->addr.size),
326 phys_offset,
327 region_offset,
328 fr->dirty_log_mask);
329 }
330
331 static void as_memory_range_del(AddressSpace *as, FlatRange *fr)
332 {
333 if (fr->dirty_log_mask) {
334 Int128 end = addrrange_end(fr->addr);
335 cpu_physical_sync_dirty_bitmap(int128_get64(fr->addr.start),
336 int128_get64(end));
337 }
338 cpu_register_physical_memory(int128_get64(fr->addr.start),
339 int128_get64(fr->addr.size),
340 IO_MEM_UNASSIGNED);
341 }
342
343 static void as_memory_log_start(AddressSpace *as, FlatRange *fr)
344 {
345 cpu_physical_log_start(int128_get64(fr->addr.start),
346 int128_get64(fr->addr.size));
347 }
348
349 static void as_memory_log_stop(AddressSpace *as, FlatRange *fr)
350 {
351 cpu_physical_log_stop(int128_get64(fr->addr.start),
352 int128_get64(fr->addr.size));
353 }
354
355 static void as_memory_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
356 {
357 int r;
358
359 assert(fd->match_data && int128_get64(fd->addr.size) == 4);
360
361 r = kvm_set_ioeventfd_mmio_long(fd->fd, int128_get64(fd->addr.start),
362 fd->data, true);
363 if (r < 0) {
364 abort();
365 }
366 }
367
368 static void as_memory_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
369 {
370 int r;
371
372 r = kvm_set_ioeventfd_mmio_long(fd->fd, int128_get64(fd->addr.start),
373 fd->data, false);
374 if (r < 0) {
375 abort();
376 }
377 }
378
379 static const AddressSpaceOps address_space_ops_memory = {
380 .range_add = as_memory_range_add,
381 .range_del = as_memory_range_del,
382 .log_start = as_memory_log_start,
383 .log_stop = as_memory_log_stop,
384 .ioeventfd_add = as_memory_ioeventfd_add,
385 .ioeventfd_del = as_memory_ioeventfd_del,
386 };
387
388 static AddressSpace address_space_memory = {
389 .ops = &address_space_ops_memory,
390 };
391
392 static const MemoryRegionPortio *find_portio(MemoryRegion *mr, uint64_t offset,
393 unsigned width, bool write)
394 {
395 const MemoryRegionPortio *mrp;
396
397 for (mrp = mr->ops->old_portio; mrp->size; ++mrp) {
398 if (offset >= mrp->offset && offset < mrp->offset + mrp->len
399 && width == mrp->size
400 && (write ? (bool)mrp->write : (bool)mrp->read)) {
401 return mrp;
402 }
403 }
404 return NULL;
405 }
406
407 static void memory_region_iorange_read(IORange *iorange,
408 uint64_t offset,
409 unsigned width,
410 uint64_t *data)
411 {
412 MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
413
414 if (mr->ops->old_portio) {
415 const MemoryRegionPortio *mrp = find_portio(mr, offset, width, false);
416
417 *data = ((uint64_t)1 << (width * 8)) - 1;
418 if (mrp) {
419 *data = mrp->read(mr->opaque, offset + mr->offset);
420 } else if (width == 2) {
421 mrp = find_portio(mr, offset, 1, false);
422 assert(mrp);
423 *data = mrp->read(mr->opaque, offset + mr->offset) |
424 (mrp->read(mr->opaque, offset + mr->offset + 1) << 8);
425 }
426 return;
427 }
428 *data = 0;
429 access_with_adjusted_size(offset + mr->offset, data, width,
430 mr->ops->impl.min_access_size,
431 mr->ops->impl.max_access_size,
432 memory_region_read_accessor, mr);
433 }
434
435 static void memory_region_iorange_write(IORange *iorange,
436 uint64_t offset,
437 unsigned width,
438 uint64_t data)
439 {
440 MemoryRegion *mr = container_of(iorange, MemoryRegion, iorange);
441
442 if (mr->ops->old_portio) {
443 const MemoryRegionPortio *mrp = find_portio(mr, offset, width, true);
444
445 if (mrp) {
446 mrp->write(mr->opaque, offset + mr->offset, data);
447 } else if (width == 2) {
448 mrp = find_portio(mr, offset, 1, false);
449 assert(mrp);
450 mrp->write(mr->opaque, offset + mr->offset, data & 0xff);
451 mrp->write(mr->opaque, offset + mr->offset + 1, data >> 8);
452 }
453 return;
454 }
455 access_with_adjusted_size(offset + mr->offset, &data, width,
456 mr->ops->impl.min_access_size,
457 mr->ops->impl.max_access_size,
458 memory_region_write_accessor, mr);
459 }
460
461 static const IORangeOps memory_region_iorange_ops = {
462 .read = memory_region_iorange_read,
463 .write = memory_region_iorange_write,
464 };
465
466 static void as_io_range_add(AddressSpace *as, FlatRange *fr)
467 {
468 iorange_init(&fr->mr->iorange, &memory_region_iorange_ops,
469 int128_get64(fr->addr.start), int128_get64(fr->addr.size));
470 ioport_register(&fr->mr->iorange);
471 }
472
473 static void as_io_range_del(AddressSpace *as, FlatRange *fr)
474 {
475 isa_unassign_ioport(int128_get64(fr->addr.start),
476 int128_get64(fr->addr.size));
477 }
478
479 static void as_io_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
480 {
481 int r;
482
483 assert(fd->match_data && int128_get64(fd->addr.size) == 2);
484
485 r = kvm_set_ioeventfd_pio_word(fd->fd, int128_get64(fd->addr.start),
486 fd->data, true);
487 if (r < 0) {
488 abort();
489 }
490 }
491
492 static void as_io_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
493 {
494 int r;
495
496 r = kvm_set_ioeventfd_pio_word(fd->fd, int128_get64(fd->addr.start),
497 fd->data, false);
498 if (r < 0) {
499 abort();
500 }
501 }
502
503 static const AddressSpaceOps address_space_ops_io = {
504 .range_add = as_io_range_add,
505 .range_del = as_io_range_del,
506 .ioeventfd_add = as_io_ioeventfd_add,
507 .ioeventfd_del = as_io_ioeventfd_del,
508 };
509
510 static AddressSpace address_space_io = {
511 .ops = &address_space_ops_io,
512 };
513
514 /* Render a memory region into the global view. Ranges in @view obscure
515 * ranges in @mr.
516 */
517 static void render_memory_region(FlatView *view,
518 MemoryRegion *mr,
519 Int128 base,
520 AddrRange clip,
521 bool readonly)
522 {
523 MemoryRegion *subregion;
524 unsigned i;
525 target_phys_addr_t offset_in_region;
526 Int128 remain;
527 Int128 now;
528 FlatRange fr;
529 AddrRange tmp;
530
531 int128_addto(&base, int128_make64(mr->addr));
532 readonly |= mr->readonly;
533
534 tmp = addrrange_make(base, mr->size);
535
536 if (!addrrange_intersects(tmp, clip)) {
537 return;
538 }
539
540 clip = addrrange_intersection(tmp, clip);
541
542 if (mr->alias) {
543 int128_subfrom(&base, int128_make64(mr->alias->addr));
544 int128_subfrom(&base, int128_make64(mr->alias_offset));
545 render_memory_region(view, mr->alias, base, clip, readonly);
546 return;
547 }
548
549 /* Render subregions in priority order. */
550 QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
551 render_memory_region(view, subregion, base, clip, readonly);
552 }
553
554 if (!mr->terminates) {
555 return;
556 }
557
558 offset_in_region = int128_get64(int128_sub(clip.start, base));
559 base = clip.start;
560 remain = clip.size;
561
562 /* Render the region itself into any gaps left by the current view. */
563 for (i = 0; i < view->nr && int128_nz(remain); ++i) {
564 if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
565 continue;
566 }
567 if (int128_lt(base, view->ranges[i].addr.start)) {
568 now = int128_min(remain,
569 int128_sub(view->ranges[i].addr.start, base));
570 fr.mr = mr;
571 fr.offset_in_region = offset_in_region;
572 fr.addr = addrrange_make(base, now);
573 fr.dirty_log_mask = mr->dirty_log_mask;
574 fr.readable = mr->readable;
575 fr.readonly = readonly;
576 flatview_insert(view, i, &fr);
577 ++i;
578 int128_addto(&base, now);
579 offset_in_region += int128_get64(now);
580 int128_subfrom(&remain, now);
581 }
582 if (int128_eq(base, view->ranges[i].addr.start)) {
583 now = int128_min(remain, view->ranges[i].addr.size);
584 int128_addto(&base, now);
585 offset_in_region += int128_get64(now);
586 int128_subfrom(&remain, now);
587 }
588 }
589 if (int128_nz(remain)) {
590 fr.mr = mr;
591 fr.offset_in_region = offset_in_region;
592 fr.addr = addrrange_make(base, remain);
593 fr.dirty_log_mask = mr->dirty_log_mask;
594 fr.readable = mr->readable;
595 fr.readonly = readonly;
596 flatview_insert(view, i, &fr);
597 }
598 }
599
600 /* Render a memory topology into a list of disjoint absolute ranges. */
601 static FlatView generate_memory_topology(MemoryRegion *mr)
602 {
603 FlatView view;
604
605 flatview_init(&view);
606
607 render_memory_region(&view, mr, int128_zero(),
608 addrrange_make(int128_zero(), int128_2_64()), false);
609 flatview_simplify(&view);
610
611 return view;
612 }
613
614 static void address_space_add_del_ioeventfds(AddressSpace *as,
615 MemoryRegionIoeventfd *fds_new,
616 unsigned fds_new_nb,
617 MemoryRegionIoeventfd *fds_old,
618 unsigned fds_old_nb)
619 {
620 unsigned iold, inew;
621
622 /* Generate a symmetric difference of the old and new fd sets, adding
623 * and deleting as necessary.
624 */
625
626 iold = inew = 0;
627 while (iold < fds_old_nb || inew < fds_new_nb) {
628 if (iold < fds_old_nb
629 && (inew == fds_new_nb
630 || memory_region_ioeventfd_before(fds_old[iold],
631 fds_new[inew]))) {
632 as->ops->ioeventfd_del(as, &fds_old[iold]);
633 ++iold;
634 } else if (inew < fds_new_nb
635 && (iold == fds_old_nb
636 || memory_region_ioeventfd_before(fds_new[inew],
637 fds_old[iold]))) {
638 as->ops->ioeventfd_add(as, &fds_new[inew]);
639 ++inew;
640 } else {
641 ++iold;
642 ++inew;
643 }
644 }
645 }
646
647 static void address_space_update_ioeventfds(AddressSpace *as)
648 {
649 FlatRange *fr;
650 unsigned ioeventfd_nb = 0;
651 MemoryRegionIoeventfd *ioeventfds = NULL;
652 AddrRange tmp;
653 unsigned i;
654
655 FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
656 for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
657 tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
658 int128_sub(fr->addr.start,
659 int128_make64(fr->offset_in_region)));
660 if (addrrange_intersects(fr->addr, tmp)) {
661 ++ioeventfd_nb;
662 ioeventfds = g_realloc(ioeventfds,
663 ioeventfd_nb * sizeof(*ioeventfds));
664 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
665 ioeventfds[ioeventfd_nb-1].addr = tmp;
666 }
667 }
668 }
669
670 address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
671 as->ioeventfds, as->ioeventfd_nb);
672
673 g_free(as->ioeventfds);
674 as->ioeventfds = ioeventfds;
675 as->ioeventfd_nb = ioeventfd_nb;
676 }
677
678 static void address_space_update_topology_pass(AddressSpace *as,
679 FlatView old_view,
680 FlatView new_view,
681 bool adding)
682 {
683 unsigned iold, inew;
684 FlatRange *frold, *frnew;
685
686 /* Generate a symmetric difference of the old and new memory maps.
687 * Kill ranges in the old map, and instantiate ranges in the new map.
688 */
689 iold = inew = 0;
690 while (iold < old_view.nr || inew < new_view.nr) {
691 if (iold < old_view.nr) {
692 frold = &old_view.ranges[iold];
693 } else {
694 frold = NULL;
695 }
696 if (inew < new_view.nr) {
697 frnew = &new_view.ranges[inew];
698 } else {
699 frnew = NULL;
700 }
701
702 if (frold
703 && (!frnew
704 || int128_lt(frold->addr.start, frnew->addr.start)
705 || (int128_eq(frold->addr.start, frnew->addr.start)
706 && !flatrange_equal(frold, frnew)))) {
707 /* In old, but (not in new, or in new but attributes changed). */
708
709 if (!adding) {
710 as->ops->range_del(as, frold);
711 }
712
713 ++iold;
714 } else if (frold && frnew && flatrange_equal(frold, frnew)) {
715 /* In both (logging may have changed) */
716
717 if (adding) {
718 if (frold->dirty_log_mask && !frnew->dirty_log_mask) {
719 as->ops->log_stop(as, frnew);
720 } else if (frnew->dirty_log_mask && !frold->dirty_log_mask) {
721 as->ops->log_start(as, frnew);
722 }
723 }
724
725 ++iold;
726 ++inew;
727 } else {
728 /* In new */
729
730 if (adding) {
731 as->ops->range_add(as, frnew);
732 }
733
734 ++inew;
735 }
736 }
737 }
738
739
740 static void address_space_update_topology(AddressSpace *as)
741 {
742 FlatView old_view = as->current_map;
743 FlatView new_view = generate_memory_topology(as->root);
744
745 address_space_update_topology_pass(as, old_view, new_view, false);
746 address_space_update_topology_pass(as, old_view, new_view, true);
747
748 as->current_map = new_view;
749 flatview_destroy(&old_view);
750 address_space_update_ioeventfds(as);
751 }
752
753 static void memory_region_update_topology(void)
754 {
755 if (memory_region_transaction_depth) {
756 return;
757 }
758
759 if (address_space_memory.root) {
760 address_space_update_topology(&address_space_memory);
761 }
762 if (address_space_io.root) {
763 address_space_update_topology(&address_space_io);
764 }
765 }
766
767 void memory_region_transaction_begin(void)
768 {
769 ++memory_region_transaction_depth;
770 }
771
772 void memory_region_transaction_commit(void)
773 {
774 assert(memory_region_transaction_depth);
775 --memory_region_transaction_depth;
776 memory_region_update_topology();
777 }
778
779 static void memory_region_destructor_none(MemoryRegion *mr)
780 {
781 }
782
783 static void memory_region_destructor_ram(MemoryRegion *mr)
784 {
785 qemu_ram_free(mr->ram_addr);
786 }
787
788 static void memory_region_destructor_ram_from_ptr(MemoryRegion *mr)
789 {
790 qemu_ram_free_from_ptr(mr->ram_addr);
791 }
792
793 static void memory_region_destructor_iomem(MemoryRegion *mr)
794 {
795 cpu_unregister_io_memory(mr->ram_addr);
796 }
797
798 static void memory_region_destructor_rom_device(MemoryRegion *mr)
799 {
800 qemu_ram_free(mr->ram_addr & TARGET_PAGE_MASK);
801 cpu_unregister_io_memory(mr->ram_addr & ~(TARGET_PAGE_MASK | IO_MEM_ROMD));
802 }
803
804 void memory_region_init(MemoryRegion *mr,
805 const char *name,
806 uint64_t size)
807 {
808 mr->ops = NULL;
809 mr->parent = NULL;
810 mr->size = int128_make64(size);
811 if (size == UINT64_MAX) {
812 mr->size = int128_2_64();
813 }
814 mr->addr = 0;
815 mr->offset = 0;
816 mr->terminates = false;
817 mr->readable = true;
818 mr->readonly = false;
819 mr->destructor = memory_region_destructor_none;
820 mr->priority = 0;
821 mr->may_overlap = false;
822 mr->alias = NULL;
823 QTAILQ_INIT(&mr->subregions);
824 memset(&mr->subregions_link, 0, sizeof mr->subregions_link);
825 QTAILQ_INIT(&mr->coalesced);
826 mr->name = g_strdup(name);
827 mr->dirty_log_mask = 0;
828 mr->ioeventfd_nb = 0;
829 mr->ioeventfds = NULL;
830 }
831
832 static bool memory_region_access_valid(MemoryRegion *mr,
833 target_phys_addr_t addr,
834 unsigned size)
835 {
836 if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
837 return false;
838 }
839
840 /* Treat zero as compatibility all valid */
841 if (!mr->ops->valid.max_access_size) {
842 return true;
843 }
844
845 if (size > mr->ops->valid.max_access_size
846 || size < mr->ops->valid.min_access_size) {
847 return false;
848 }
849 return true;
850 }
851
852 static uint32_t memory_region_read_thunk_n(void *_mr,
853 target_phys_addr_t addr,
854 unsigned size)
855 {
856 MemoryRegion *mr = _mr;
857 uint64_t data = 0;
858
859 if (!memory_region_access_valid(mr, addr, size)) {
860 return -1U; /* FIXME: better signalling */
861 }
862
863 if (!mr->ops->read) {
864 return mr->ops->old_mmio.read[bitops_ffsl(size)](mr->opaque, addr);
865 }
866
867 /* FIXME: support unaligned access */
868 access_with_adjusted_size(addr + mr->offset, &data, size,
869 mr->ops->impl.min_access_size,
870 mr->ops->impl.max_access_size,
871 memory_region_read_accessor, mr);
872
873 return data;
874 }
875
876 static void memory_region_write_thunk_n(void *_mr,
877 target_phys_addr_t addr,
878 unsigned size,
879 uint64_t data)
880 {
881 MemoryRegion *mr = _mr;
882
883 if (!memory_region_access_valid(mr, addr, size)) {
884 return; /* FIXME: better signalling */
885 }
886
887 if (!mr->ops->write) {
888 mr->ops->old_mmio.write[bitops_ffsl(size)](mr->opaque, addr, data);
889 return;
890 }
891
892 /* FIXME: support unaligned access */
893 access_with_adjusted_size(addr + mr->offset, &data, size,
894 mr->ops->impl.min_access_size,
895 mr->ops->impl.max_access_size,
896 memory_region_write_accessor, mr);
897 }
898
899 static uint32_t memory_region_read_thunk_b(void *mr, target_phys_addr_t addr)
900 {
901 return memory_region_read_thunk_n(mr, addr, 1);
902 }
903
904 static uint32_t memory_region_read_thunk_w(void *mr, target_phys_addr_t addr)
905 {
906 return memory_region_read_thunk_n(mr, addr, 2);
907 }
908
909 static uint32_t memory_region_read_thunk_l(void *mr, target_phys_addr_t addr)
910 {
911 return memory_region_read_thunk_n(mr, addr, 4);
912 }
913
914 static void memory_region_write_thunk_b(void *mr, target_phys_addr_t addr,
915 uint32_t data)
916 {
917 memory_region_write_thunk_n(mr, addr, 1, data);
918 }
919
920 static void memory_region_write_thunk_w(void *mr, target_phys_addr_t addr,
921 uint32_t data)
922 {
923 memory_region_write_thunk_n(mr, addr, 2, data);
924 }
925
926 static void memory_region_write_thunk_l(void *mr, target_phys_addr_t addr,
927 uint32_t data)
928 {
929 memory_region_write_thunk_n(mr, addr, 4, data);
930 }
931
932 static CPUReadMemoryFunc * const memory_region_read_thunk[] = {
933 memory_region_read_thunk_b,
934 memory_region_read_thunk_w,
935 memory_region_read_thunk_l,
936 };
937
938 static CPUWriteMemoryFunc * const memory_region_write_thunk[] = {
939 memory_region_write_thunk_b,
940 memory_region_write_thunk_w,
941 memory_region_write_thunk_l,
942 };
943
944 static void memory_region_prepare_ram_addr(MemoryRegion *mr)
945 {
946 if (mr->backend_registered) {
947 return;
948 }
949
950 mr->destructor = memory_region_destructor_iomem;
951 mr->ram_addr = cpu_register_io_memory(memory_region_read_thunk,
952 memory_region_write_thunk,
953 mr,
954 mr->ops->endianness);
955 mr->backend_registered = true;
956 }
957
958 void memory_region_init_io(MemoryRegion *mr,
959 const MemoryRegionOps *ops,
960 void *opaque,
961 const char *name,
962 uint64_t size)
963 {
964 memory_region_init(mr, name, size);
965 mr->ops = ops;
966 mr->opaque = opaque;
967 mr->terminates = true;
968 mr->backend_registered = false;
969 }
970
971 void memory_region_init_ram(MemoryRegion *mr,
972 DeviceState *dev,
973 const char *name,
974 uint64_t size)
975 {
976 memory_region_init(mr, name, size);
977 mr->terminates = true;
978 mr->destructor = memory_region_destructor_ram;
979 mr->ram_addr = qemu_ram_alloc(dev, name, size);
980 mr->backend_registered = true;
981 }
982
983 void memory_region_init_ram_ptr(MemoryRegion *mr,
984 DeviceState *dev,
985 const char *name,
986 uint64_t size,
987 void *ptr)
988 {
989 memory_region_init(mr, name, size);
990 mr->terminates = true;
991 mr->destructor = memory_region_destructor_ram_from_ptr;
992 mr->ram_addr = qemu_ram_alloc_from_ptr(dev, name, size, ptr);
993 mr->backend_registered = true;
994 }
995
996 void memory_region_init_alias(MemoryRegion *mr,
997 const char *name,
998 MemoryRegion *orig,
999 target_phys_addr_t offset,
1000 uint64_t size)
1001 {
1002 memory_region_init(mr, name, size);
1003 mr->alias = orig;
1004 mr->alias_offset = offset;
1005 }
1006
1007 void memory_region_init_rom_device(MemoryRegion *mr,
1008 const MemoryRegionOps *ops,
1009 void *opaque,
1010 DeviceState *dev,
1011 const char *name,
1012 uint64_t size)
1013 {
1014 memory_region_init(mr, name, size);
1015 mr->ops = ops;
1016 mr->opaque = opaque;
1017 mr->terminates = true;
1018 mr->destructor = memory_region_destructor_rom_device;
1019 mr->ram_addr = qemu_ram_alloc(dev, name, size);
1020 mr->ram_addr |= cpu_register_io_memory(memory_region_read_thunk,
1021 memory_region_write_thunk,
1022 mr,
1023 mr->ops->endianness);
1024 mr->ram_addr |= IO_MEM_ROMD;
1025 mr->backend_registered = true;
1026 }
1027
1028 void memory_region_destroy(MemoryRegion *mr)
1029 {
1030 assert(QTAILQ_EMPTY(&mr->subregions));
1031 mr->destructor(mr);
1032 memory_region_clear_coalescing(mr);
1033 g_free((char *)mr->name);
1034 g_free(mr->ioeventfds);
1035 }
1036
1037 uint64_t memory_region_size(MemoryRegion *mr)
1038 {
1039 if (int128_eq(mr->size, int128_2_64())) {
1040 return UINT64_MAX;
1041 }
1042 return int128_get64(mr->size);
1043 }
1044
1045 void memory_region_set_offset(MemoryRegion *mr, target_phys_addr_t offset)
1046 {
1047 mr->offset = offset;
1048 }
1049
1050 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1051 {
1052 uint8_t mask = 1 << client;
1053
1054 mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1055 memory_region_update_topology();
1056 }
1057
1058 bool memory_region_get_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1059 unsigned client)
1060 {
1061 assert(mr->terminates);
1062 return cpu_physical_memory_get_dirty(mr->ram_addr + addr, 1 << client);
1063 }
1064
1065 void memory_region_set_dirty(MemoryRegion *mr, target_phys_addr_t addr)
1066 {
1067 assert(mr->terminates);
1068 return cpu_physical_memory_set_dirty(mr->ram_addr + addr);
1069 }
1070
1071 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1072 {
1073 FlatRange *fr;
1074
1075 FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1076 if (fr->mr == mr) {
1077 cpu_physical_sync_dirty_bitmap(int128_get64(fr->addr.start),
1078 int128_get64(addrrange_end(fr->addr)));
1079 }
1080 }
1081 }
1082
1083 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1084 {
1085 if (mr->readonly != readonly) {
1086 mr->readonly = readonly;
1087 memory_region_update_topology();
1088 }
1089 }
1090
1091 void memory_region_rom_device_set_readable(MemoryRegion *mr, bool readable)
1092 {
1093 if (mr->readable != readable) {
1094 mr->readable = readable;
1095 memory_region_update_topology();
1096 }
1097 }
1098
1099 void memory_region_reset_dirty(MemoryRegion *mr, target_phys_addr_t addr,
1100 target_phys_addr_t size, unsigned client)
1101 {
1102 assert(mr->terminates);
1103 cpu_physical_memory_reset_dirty(mr->ram_addr + addr,
1104 mr->ram_addr + addr + size,
1105 1 << client);
1106 }
1107
1108 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1109 {
1110 if (mr->alias) {
1111 return memory_region_get_ram_ptr(mr->alias) + mr->alias_offset;
1112 }
1113
1114 assert(mr->terminates);
1115
1116 return qemu_get_ram_ptr(mr->ram_addr & TARGET_PAGE_MASK);
1117 }
1118
1119 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1120 {
1121 FlatRange *fr;
1122 CoalescedMemoryRange *cmr;
1123 AddrRange tmp;
1124
1125 FOR_EACH_FLAT_RANGE(fr, &address_space_memory.current_map) {
1126 if (fr->mr == mr) {
1127 qemu_unregister_coalesced_mmio(int128_get64(fr->addr.start),
1128 int128_get64(fr->addr.size));
1129 QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1130 tmp = addrrange_shift(cmr->addr,
1131 int128_sub(fr->addr.start,
1132 int128_make64(fr->offset_in_region)));
1133 if (!addrrange_intersects(tmp, fr->addr)) {
1134 continue;
1135 }
1136 tmp = addrrange_intersection(tmp, fr->addr);
1137 qemu_register_coalesced_mmio(int128_get64(tmp.start),
1138 int128_get64(tmp.size));
1139 }
1140 }
1141 }
1142 }
1143
1144 void memory_region_set_coalescing(MemoryRegion *mr)
1145 {
1146 memory_region_clear_coalescing(mr);
1147 memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1148 }
1149
1150 void memory_region_add_coalescing(MemoryRegion *mr,
1151 target_phys_addr_t offset,
1152 uint64_t size)
1153 {
1154 CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1155
1156 cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1157 QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1158 memory_region_update_coalesced_range(mr);
1159 }
1160
1161 void memory_region_clear_coalescing(MemoryRegion *mr)
1162 {
1163 CoalescedMemoryRange *cmr;
1164
1165 while (!QTAILQ_EMPTY(&mr->coalesced)) {
1166 cmr = QTAILQ_FIRST(&mr->coalesced);
1167 QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1168 g_free(cmr);
1169 }
1170 memory_region_update_coalesced_range(mr);
1171 }
1172
1173 void memory_region_add_eventfd(MemoryRegion *mr,
1174 target_phys_addr_t addr,
1175 unsigned size,
1176 bool match_data,
1177 uint64_t data,
1178 int fd)
1179 {
1180 MemoryRegionIoeventfd mrfd = {
1181 .addr.start = int128_make64(addr),
1182 .addr.size = int128_make64(size),
1183 .match_data = match_data,
1184 .data = data,
1185 .fd = fd,
1186 };
1187 unsigned i;
1188
1189 for (i = 0; i < mr->ioeventfd_nb; ++i) {
1190 if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1191 break;
1192 }
1193 }
1194 ++mr->ioeventfd_nb;
1195 mr->ioeventfds = g_realloc(mr->ioeventfds,
1196 sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1197 memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1198 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1199 mr->ioeventfds[i] = mrfd;
1200 memory_region_update_topology();
1201 }
1202
1203 void memory_region_del_eventfd(MemoryRegion *mr,
1204 target_phys_addr_t addr,
1205 unsigned size,
1206 bool match_data,
1207 uint64_t data,
1208 int fd)
1209 {
1210 MemoryRegionIoeventfd mrfd = {
1211 .addr.start = int128_make64(addr),
1212 .addr.size = int128_make64(size),
1213 .match_data = match_data,
1214 .data = data,
1215 .fd = fd,
1216 };
1217 unsigned i;
1218
1219 for (i = 0; i < mr->ioeventfd_nb; ++i) {
1220 if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1221 break;
1222 }
1223 }
1224 assert(i != mr->ioeventfd_nb);
1225 memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1226 sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1227 --mr->ioeventfd_nb;
1228 mr->ioeventfds = g_realloc(mr->ioeventfds,
1229 sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1230 memory_region_update_topology();
1231 }
1232
1233 static void memory_region_add_subregion_common(MemoryRegion *mr,
1234 target_phys_addr_t offset,
1235 MemoryRegion *subregion)
1236 {
1237 MemoryRegion *other;
1238
1239 assert(!subregion->parent);
1240 subregion->parent = mr;
1241 subregion->addr = offset;
1242 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1243 if (subregion->may_overlap || other->may_overlap) {
1244 continue;
1245 }
1246 if (int128_gt(int128_make64(offset),
1247 int128_add(int128_make64(other->addr), other->size))
1248 || int128_le(int128_add(int128_make64(offset), subregion->size),
1249 int128_make64(other->addr))) {
1250 continue;
1251 }
1252 #if 0
1253 printf("warning: subregion collision %llx/%llx (%s) "
1254 "vs %llx/%llx (%s)\n",
1255 (unsigned long long)offset,
1256 (unsigned long long)int128_get64(subregion->size),
1257 subregion->name,
1258 (unsigned long long)other->addr,
1259 (unsigned long long)int128_get64(other->size),
1260 other->name);
1261 #endif
1262 }
1263 QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1264 if (subregion->priority >= other->priority) {
1265 QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1266 goto done;
1267 }
1268 }
1269 QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1270 done:
1271 memory_region_update_topology();
1272 }
1273
1274
1275 void memory_region_add_subregion(MemoryRegion *mr,
1276 target_phys_addr_t offset,
1277 MemoryRegion *subregion)
1278 {
1279 subregion->may_overlap = false;
1280 subregion->priority = 0;
1281 memory_region_add_subregion_common(mr, offset, subregion);
1282 }
1283
1284 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1285 target_phys_addr_t offset,
1286 MemoryRegion *subregion,
1287 unsigned priority)
1288 {
1289 subregion->may_overlap = true;
1290 subregion->priority = priority;
1291 memory_region_add_subregion_common(mr, offset, subregion);
1292 }
1293
1294 void memory_region_del_subregion(MemoryRegion *mr,
1295 MemoryRegion *subregion)
1296 {
1297 assert(subregion->parent == mr);
1298 subregion->parent = NULL;
1299 QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1300 memory_region_update_topology();
1301 }
1302
1303 void set_system_memory_map(MemoryRegion *mr)
1304 {
1305 address_space_memory.root = mr;
1306 memory_region_update_topology();
1307 }
1308
1309 void set_system_io_map(MemoryRegion *mr)
1310 {
1311 address_space_io.root = mr;
1312 memory_region_update_topology();
1313 }
1314
1315 typedef struct MemoryRegionList MemoryRegionList;
1316
1317 struct MemoryRegionList {
1318 const MemoryRegion *mr;
1319 bool printed;
1320 QTAILQ_ENTRY(MemoryRegionList) queue;
1321 };
1322
1323 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
1324
1325 static void mtree_print_mr(fprintf_function mon_printf, void *f,
1326 const MemoryRegion *mr, unsigned int level,
1327 target_phys_addr_t base,
1328 MemoryRegionListHead *alias_print_queue)
1329 {
1330 MemoryRegionList *new_ml, *ml, *next_ml;
1331 MemoryRegionListHead submr_print_queue;
1332 const MemoryRegion *submr;
1333 unsigned int i;
1334
1335 if (!mr) {
1336 return;
1337 }
1338
1339 for (i = 0; i < level; i++) {
1340 mon_printf(f, " ");
1341 }
1342
1343 if (mr->alias) {
1344 MemoryRegionList *ml;
1345 bool found = false;
1346
1347 /* check if the alias is already in the queue */
1348 QTAILQ_FOREACH(ml, alias_print_queue, queue) {
1349 if (ml->mr == mr->alias && !ml->printed) {
1350 found = true;
1351 }
1352 }
1353
1354 if (!found) {
1355 ml = g_new(MemoryRegionList, 1);
1356 ml->mr = mr->alias;
1357 ml->printed = false;
1358 QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
1359 }
1360 mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d): alias %s @%s "
1361 TARGET_FMT_plx "-" TARGET_FMT_plx "\n",
1362 base + mr->addr,
1363 base + mr->addr
1364 + (target_phys_addr_t)int128_get64(mr->size) - 1,
1365 mr->priority,
1366 mr->name,
1367 mr->alias->name,
1368 mr->alias_offset,
1369 mr->alias_offset
1370 + (target_phys_addr_t)int128_get64(mr->size) - 1);
1371 } else {
1372 mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d): %s\n",
1373 base + mr->addr,
1374 base + mr->addr
1375 + (target_phys_addr_t)int128_get64(mr->size) - 1,
1376 mr->priority,
1377 mr->name);
1378 }
1379
1380 QTAILQ_INIT(&submr_print_queue);
1381
1382 QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
1383 new_ml = g_new(MemoryRegionList, 1);
1384 new_ml->mr = submr;
1385 QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
1386 if (new_ml->mr->addr < ml->mr->addr ||
1387 (new_ml->mr->addr == ml->mr->addr &&
1388 new_ml->mr->priority > ml->mr->priority)) {
1389 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
1390 new_ml = NULL;
1391 break;
1392 }
1393 }
1394 if (new_ml) {
1395 QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
1396 }
1397 }
1398
1399 QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
1400 mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
1401 alias_print_queue);
1402 }
1403
1404 QTAILQ_FOREACH_SAFE(next_ml, &submr_print_queue, queue, ml) {
1405 g_free(ml);
1406 }
1407 }
1408
1409 void mtree_info(fprintf_function mon_printf, void *f)
1410 {
1411 MemoryRegionListHead ml_head;
1412 MemoryRegionList *ml, *ml2;
1413
1414 QTAILQ_INIT(&ml_head);
1415
1416 mon_printf(f, "memory\n");
1417 mtree_print_mr(mon_printf, f, address_space_memory.root, 0, 0, &ml_head);
1418
1419 /* print aliased regions */
1420 QTAILQ_FOREACH(ml, &ml_head, queue) {
1421 if (!ml->printed) {
1422 mon_printf(f, "%s\n", ml->mr->name);
1423 mtree_print_mr(mon_printf, f, ml->mr, 0, 0, &ml_head);
1424 }
1425 }
1426
1427 QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
1428 g_free(ml2);
1429 }
1430
1431 if (address_space_io.root &&
1432 !QTAILQ_EMPTY(&address_space_io.root->subregions)) {
1433 QTAILQ_INIT(&ml_head);
1434 mon_printf(f, "I/O\n");
1435 mtree_print_mr(mon_printf, f, address_space_io.root, 0, 0, &ml_head);
1436 }
1437 }
QEmu