]> git.proxmox.com Git - mirror_qemu.git/blame - exec.c
exec.c: Add new address_space_ld*/st* functions
[mirror_qemu.git] / exec.c
CommitLineData
54936004 1/*
5b6dd868 2 * Virtual page mapping
5fafdf24 3 *
54936004
FB
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
8167ee88 17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
54936004 18 */
67b915a5 19#include "config.h"
777872e5 20#ifndef _WIN32
a98d49b1 21#include <sys/types.h>
d5a8f07c
FB
22#include <sys/mman.h>
23#endif
54936004 24
055403b2 25#include "qemu-common.h"
6180a181 26#include "cpu.h"
b67d9a52 27#include "tcg.h"
b3c7724c 28#include "hw/hw.h"
4485bd26 29#if !defined(CONFIG_USER_ONLY)
47c8ca53 30#include "hw/boards.h"
4485bd26 31#endif
cc9e98cb 32#include "hw/qdev.h"
1de7afc9 33#include "qemu/osdep.h"
9c17d615 34#include "sysemu/kvm.h"
2ff3de68 35#include "sysemu/sysemu.h"
0d09e41a 36#include "hw/xen/xen.h"
1de7afc9
PB
37#include "qemu/timer.h"
38#include "qemu/config-file.h"
75a34036 39#include "qemu/error-report.h"
022c62cb 40#include "exec/memory.h"
9c17d615 41#include "sysemu/dma.h"
022c62cb 42#include "exec/address-spaces.h"
53a5960a
PB
43#if defined(CONFIG_USER_ONLY)
44#include <qemu.h>
432d268c 45#else /* !CONFIG_USER_ONLY */
9c17d615 46#include "sysemu/xen-mapcache.h"
6506e4f9 47#include "trace.h"
53a5960a 48#endif
0d6d3c87 49#include "exec/cpu-all.h"
0dc3f44a 50#include "qemu/rcu_queue.h"
022c62cb 51#include "exec/cputlb.h"
5b6dd868 52#include "translate-all.h"
0cac1b66 53
022c62cb 54#include "exec/memory-internal.h"
220c3ebd 55#include "exec/ram_addr.h"
67d95c15 56
b35ba30f
MT
57#include "qemu/range.h"
58
db7b5426 59//#define DEBUG_SUBPAGE
1196be37 60
e2eef170 61#if !defined(CONFIG_USER_ONLY)
981fdf23 62static bool in_migration;
94a6b54f 63
0dc3f44a
MD
64/* ram_list is read under rcu_read_lock()/rcu_read_unlock(). Writes
65 * are protected by the ramlist lock.
66 */
0d53d9fe 67RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) };
62152b8a
AK
68
69static MemoryRegion *system_memory;
309cb471 70static MemoryRegion *system_io;
62152b8a 71
f6790af6
AK
72AddressSpace address_space_io;
73AddressSpace address_space_memory;
2673a5da 74
0844e007 75MemoryRegion io_mem_rom, io_mem_notdirty;
acc9d80b 76static MemoryRegion io_mem_unassigned;
0e0df1e2 77
7bd4f430
PB
78/* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
79#define RAM_PREALLOC (1 << 0)
80
dbcb8981
PB
81/* RAM is mmap-ed with MAP_SHARED */
82#define RAM_SHARED (1 << 1)
83
62be4e3a
MT
84/* Only a portion of RAM (used_length) is actually used, and migrated.
85 * This used_length size can change across reboots.
86 */
87#define RAM_RESIZEABLE (1 << 2)
88
e2eef170 89#endif
9fa3e853 90
bdc44640 91struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus);
6a00d601
FB
92/* current CPU in the current thread. It is only valid inside
93 cpu_exec() */
4917cf44 94DEFINE_TLS(CPUState *, current_cpu);
2e70f6ef 95/* 0 = Do not count executed instructions.
bf20dc07 96 1 = Precise instruction counting.
2e70f6ef 97 2 = Adaptive rate instruction counting. */
5708fc66 98int use_icount;
6a00d601 99
e2eef170 100#if !defined(CONFIG_USER_ONLY)
4346ae3e 101
1db8abb1
PB
102typedef struct PhysPageEntry PhysPageEntry;
103
104struct PhysPageEntry {
9736e55b 105 /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */
8b795765 106 uint32_t skip : 6;
9736e55b 107 /* index into phys_sections (!skip) or phys_map_nodes (skip) */
8b795765 108 uint32_t ptr : 26;
1db8abb1
PB
109};
110
8b795765
MT
111#define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6)
112
03f49957 113/* Size of the L2 (and L3, etc) page tables. */
57271d63 114#define ADDR_SPACE_BITS 64
03f49957 115
026736ce 116#define P_L2_BITS 9
03f49957
PB
117#define P_L2_SIZE (1 << P_L2_BITS)
118
119#define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1)
120
121typedef PhysPageEntry Node[P_L2_SIZE];
0475d94f 122
53cb28cb 123typedef struct PhysPageMap {
79e2b9ae
PB
124 struct rcu_head rcu;
125
53cb28cb
MA
126 unsigned sections_nb;
127 unsigned sections_nb_alloc;
128 unsigned nodes_nb;
129 unsigned nodes_nb_alloc;
130 Node *nodes;
131 MemoryRegionSection *sections;
132} PhysPageMap;
133
1db8abb1 134struct AddressSpaceDispatch {
79e2b9ae
PB
135 struct rcu_head rcu;
136
1db8abb1
PB
137 /* This is a multi-level map on the physical address space.
138 * The bottom level has pointers to MemoryRegionSections.
139 */
140 PhysPageEntry phys_map;
53cb28cb 141 PhysPageMap map;
acc9d80b 142 AddressSpace *as;
1db8abb1
PB
143};
144
90260c6c
JK
145#define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
146typedef struct subpage_t {
147 MemoryRegion iomem;
acc9d80b 148 AddressSpace *as;
90260c6c
JK
149 hwaddr base;
150 uint16_t sub_section[TARGET_PAGE_SIZE];
151} subpage_t;
152
b41aac4f
LPF
153#define PHYS_SECTION_UNASSIGNED 0
154#define PHYS_SECTION_NOTDIRTY 1
155#define PHYS_SECTION_ROM 2
156#define PHYS_SECTION_WATCH 3
5312bd8b 157
e2eef170 158static void io_mem_init(void);
62152b8a 159static void memory_map_init(void);
09daed84 160static void tcg_commit(MemoryListener *listener);
e2eef170 161
1ec9b909 162static MemoryRegion io_mem_watch;
6658ffb8 163#endif
fd6ce8f6 164
6d9a1304 165#if !defined(CONFIG_USER_ONLY)
d6f2ea22 166
53cb28cb 167static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes)
d6f2ea22 168{
53cb28cb
MA
169 if (map->nodes_nb + nodes > map->nodes_nb_alloc) {
170 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc * 2, 16);
171 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, map->nodes_nb + nodes);
172 map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc);
d6f2ea22 173 }
f7bf5461
AK
174}
175
53cb28cb 176static uint32_t phys_map_node_alloc(PhysPageMap *map)
f7bf5461
AK
177{
178 unsigned i;
8b795765 179 uint32_t ret;
f7bf5461 180
53cb28cb 181 ret = map->nodes_nb++;
f7bf5461 182 assert(ret != PHYS_MAP_NODE_NIL);
53cb28cb 183 assert(ret != map->nodes_nb_alloc);
03f49957 184 for (i = 0; i < P_L2_SIZE; ++i) {
53cb28cb
MA
185 map->nodes[ret][i].skip = 1;
186 map->nodes[ret][i].ptr = PHYS_MAP_NODE_NIL;
d6f2ea22 187 }
f7bf5461 188 return ret;
d6f2ea22
AK
189}
190
53cb28cb
MA
191static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp,
192 hwaddr *index, hwaddr *nb, uint16_t leaf,
2999097b 193 int level)
f7bf5461
AK
194{
195 PhysPageEntry *p;
196 int i;
03f49957 197 hwaddr step = (hwaddr)1 << (level * P_L2_BITS);
108c49b8 198
9736e55b 199 if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) {
53cb28cb
MA
200 lp->ptr = phys_map_node_alloc(map);
201 p = map->nodes[lp->ptr];
f7bf5461 202 if (level == 0) {
03f49957 203 for (i = 0; i < P_L2_SIZE; i++) {
9736e55b 204 p[i].skip = 0;
b41aac4f 205 p[i].ptr = PHYS_SECTION_UNASSIGNED;
4346ae3e 206 }
67c4d23c 207 }
f7bf5461 208 } else {
53cb28cb 209 p = map->nodes[lp->ptr];
92e873b9 210 }
03f49957 211 lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)];
f7bf5461 212
03f49957 213 while (*nb && lp < &p[P_L2_SIZE]) {
07f07b31 214 if ((*index & (step - 1)) == 0 && *nb >= step) {
9736e55b 215 lp->skip = 0;
c19e8800 216 lp->ptr = leaf;
07f07b31
AK
217 *index += step;
218 *nb -= step;
2999097b 219 } else {
53cb28cb 220 phys_page_set_level(map, lp, index, nb, leaf, level - 1);
2999097b
AK
221 }
222 ++lp;
f7bf5461
AK
223 }
224}
225
ac1970fb 226static void phys_page_set(AddressSpaceDispatch *d,
a8170e5e 227 hwaddr index, hwaddr nb,
2999097b 228 uint16_t leaf)
f7bf5461 229{
2999097b 230 /* Wildly overreserve - it doesn't matter much. */
53cb28cb 231 phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS);
5cd2c5b6 232
53cb28cb 233 phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1);
92e873b9
FB
234}
235
b35ba30f
MT
236/* Compact a non leaf page entry. Simply detect that the entry has a single child,
237 * and update our entry so we can skip it and go directly to the destination.
238 */
239static void phys_page_compact(PhysPageEntry *lp, Node *nodes, unsigned long *compacted)
240{
241 unsigned valid_ptr = P_L2_SIZE;
242 int valid = 0;
243 PhysPageEntry *p;
244 int i;
245
246 if (lp->ptr == PHYS_MAP_NODE_NIL) {
247 return;
248 }
249
250 p = nodes[lp->ptr];
251 for (i = 0; i < P_L2_SIZE; i++) {
252 if (p[i].ptr == PHYS_MAP_NODE_NIL) {
253 continue;
254 }
255
256 valid_ptr = i;
257 valid++;
258 if (p[i].skip) {
259 phys_page_compact(&p[i], nodes, compacted);
260 }
261 }
262
263 /* We can only compress if there's only one child. */
264 if (valid != 1) {
265 return;
266 }
267
268 assert(valid_ptr < P_L2_SIZE);
269
270 /* Don't compress if it won't fit in the # of bits we have. */
271 if (lp->skip + p[valid_ptr].skip >= (1 << 3)) {
272 return;
273 }
274
275 lp->ptr = p[valid_ptr].ptr;
276 if (!p[valid_ptr].skip) {
277 /* If our only child is a leaf, make this a leaf. */
278 /* By design, we should have made this node a leaf to begin with so we
279 * should never reach here.
280 * But since it's so simple to handle this, let's do it just in case we
281 * change this rule.
282 */
283 lp->skip = 0;
284 } else {
285 lp->skip += p[valid_ptr].skip;
286 }
287}
288
289static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb)
290{
291 DECLARE_BITMAP(compacted, nodes_nb);
292
293 if (d->phys_map.skip) {
53cb28cb 294 phys_page_compact(&d->phys_map, d->map.nodes, compacted);
b35ba30f
MT
295 }
296}
297
97115a8d 298static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr addr,
9affd6fc 299 Node *nodes, MemoryRegionSection *sections)
92e873b9 300{
31ab2b4a 301 PhysPageEntry *p;
97115a8d 302 hwaddr index = addr >> TARGET_PAGE_BITS;
31ab2b4a 303 int i;
f1f6e3b8 304
9736e55b 305 for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) {
c19e8800 306 if (lp.ptr == PHYS_MAP_NODE_NIL) {
9affd6fc 307 return &sections[PHYS_SECTION_UNASSIGNED];
31ab2b4a 308 }
9affd6fc 309 p = nodes[lp.ptr];
03f49957 310 lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)];
5312bd8b 311 }
b35ba30f
MT
312
313 if (sections[lp.ptr].size.hi ||
314 range_covers_byte(sections[lp.ptr].offset_within_address_space,
315 sections[lp.ptr].size.lo, addr)) {
316 return &sections[lp.ptr];
317 } else {
318 return &sections[PHYS_SECTION_UNASSIGNED];
319 }
f3705d53
AK
320}
321
e5548617
BS
322bool memory_region_is_unassigned(MemoryRegion *mr)
323{
2a8e7499 324 return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device
5b6dd868 325 && mr != &io_mem_watch;
fd6ce8f6 326}
149f54b5 327
79e2b9ae 328/* Called from RCU critical section */
c7086b4a 329static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
90260c6c
JK
330 hwaddr addr,
331 bool resolve_subpage)
9f029603 332{
90260c6c
JK
333 MemoryRegionSection *section;
334 subpage_t *subpage;
335
53cb28cb 336 section = phys_page_find(d->phys_map, addr, d->map.nodes, d->map.sections);
90260c6c
JK
337 if (resolve_subpage && section->mr->subpage) {
338 subpage = container_of(section->mr, subpage_t, iomem);
53cb28cb 339 section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]];
90260c6c
JK
340 }
341 return section;
9f029603
JK
342}
343
79e2b9ae 344/* Called from RCU critical section */
90260c6c 345static MemoryRegionSection *
c7086b4a 346address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
90260c6c 347 hwaddr *plen, bool resolve_subpage)
149f54b5
PB
348{
349 MemoryRegionSection *section;
a87f3954 350 Int128 diff;
149f54b5 351
c7086b4a 352 section = address_space_lookup_region(d, addr, resolve_subpage);
149f54b5
PB
353 /* Compute offset within MemoryRegionSection */
354 addr -= section->offset_within_address_space;
355
356 /* Compute offset within MemoryRegion */
357 *xlat = addr + section->offset_within_region;
358
359 diff = int128_sub(section->mr->size, int128_make64(addr));
3752a036 360 *plen = int128_get64(int128_min(diff, int128_make64(*plen)));
149f54b5
PB
361 return section;
362}
90260c6c 363
a87f3954
PB
364static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
365{
366 if (memory_region_is_ram(mr)) {
367 return !(is_write && mr->readonly);
368 }
369 if (memory_region_is_romd(mr)) {
370 return !is_write;
371 }
372
373 return false;
374}
375
5c8a00ce
PB
376MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
377 hwaddr *xlat, hwaddr *plen,
378 bool is_write)
90260c6c 379{
30951157
AK
380 IOMMUTLBEntry iotlb;
381 MemoryRegionSection *section;
382 MemoryRegion *mr;
4025446f 383 hwaddr len = *plen;
30951157 384
79e2b9ae 385 rcu_read_lock();
30951157 386 for (;;) {
79e2b9ae
PB
387 AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
388 section = address_space_translate_internal(d, addr, &addr, plen, true);
30951157
AK
389 mr = section->mr;
390
391 if (!mr->iommu_ops) {
392 break;
393 }
394
8d7b8cb9 395 iotlb = mr->iommu_ops->translate(mr, addr, is_write);
30951157
AK
396 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
397 | (addr & iotlb.addr_mask));
4025446f 398 len = MIN(len, (addr | iotlb.addr_mask) - addr + 1);
30951157
AK
399 if (!(iotlb.perm & (1 << is_write))) {
400 mr = &io_mem_unassigned;
401 break;
402 }
403
404 as = iotlb.target_as;
405 }
406
fe680d0d 407 if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
a87f3954 408 hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
4025446f 409 len = MIN(page, len);
a87f3954
PB
410 }
411
4025446f 412 *plen = len;
30951157 413 *xlat = addr;
79e2b9ae 414 rcu_read_unlock();
30951157 415 return mr;
90260c6c
JK
416}
417
79e2b9ae 418/* Called from RCU critical section */
90260c6c 419MemoryRegionSection *
9d82b5a7
PB
420address_space_translate_for_iotlb(CPUState *cpu, hwaddr addr,
421 hwaddr *xlat, hwaddr *plen)
90260c6c 422{
30951157 423 MemoryRegionSection *section;
9d82b5a7
PB
424 section = address_space_translate_internal(cpu->memory_dispatch,
425 addr, xlat, plen, false);
30951157
AK
426
427 assert(!section->mr->iommu_ops);
428 return section;
90260c6c 429}
5b6dd868 430#endif
fd6ce8f6 431
5b6dd868 432void cpu_exec_init_all(void)
fdbb84d1 433{
5b6dd868 434#if !defined(CONFIG_USER_ONLY)
b2a8658e 435 qemu_mutex_init(&ram_list.mutex);
5b6dd868
BS
436 memory_map_init();
437 io_mem_init();
fdbb84d1 438#endif
5b6dd868 439}
fdbb84d1 440
b170fce3 441#if !defined(CONFIG_USER_ONLY)
5b6dd868
BS
442
443static int cpu_common_post_load(void *opaque, int version_id)
fd6ce8f6 444{
259186a7 445 CPUState *cpu = opaque;
a513fe19 446
5b6dd868
BS
447 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
448 version_id is increased. */
259186a7 449 cpu->interrupt_request &= ~0x01;
c01a71c1 450 tlb_flush(cpu, 1);
5b6dd868
BS
451
452 return 0;
a513fe19 453}
7501267e 454
6c3bff0e
PD
455static int cpu_common_pre_load(void *opaque)
456{
457 CPUState *cpu = opaque;
458
adee6424 459 cpu->exception_index = -1;
6c3bff0e
PD
460
461 return 0;
462}
463
464static bool cpu_common_exception_index_needed(void *opaque)
465{
466 CPUState *cpu = opaque;
467
adee6424 468 return tcg_enabled() && cpu->exception_index != -1;
6c3bff0e
PD
469}
470
471static const VMStateDescription vmstate_cpu_common_exception_index = {
472 .name = "cpu_common/exception_index",
473 .version_id = 1,
474 .minimum_version_id = 1,
475 .fields = (VMStateField[]) {
476 VMSTATE_INT32(exception_index, CPUState),
477 VMSTATE_END_OF_LIST()
478 }
479};
480
1a1562f5 481const VMStateDescription vmstate_cpu_common = {
5b6dd868
BS
482 .name = "cpu_common",
483 .version_id = 1,
484 .minimum_version_id = 1,
6c3bff0e 485 .pre_load = cpu_common_pre_load,
5b6dd868 486 .post_load = cpu_common_post_load,
35d08458 487 .fields = (VMStateField[]) {
259186a7
AF
488 VMSTATE_UINT32(halted, CPUState),
489 VMSTATE_UINT32(interrupt_request, CPUState),
5b6dd868 490 VMSTATE_END_OF_LIST()
6c3bff0e
PD
491 },
492 .subsections = (VMStateSubsection[]) {
493 {
494 .vmsd = &vmstate_cpu_common_exception_index,
495 .needed = cpu_common_exception_index_needed,
496 } , {
497 /* empty */
498 }
5b6dd868
BS
499 }
500};
1a1562f5 501
5b6dd868 502#endif
ea041c0e 503
38d8f5c8 504CPUState *qemu_get_cpu(int index)
ea041c0e 505{
bdc44640 506 CPUState *cpu;
ea041c0e 507
bdc44640 508 CPU_FOREACH(cpu) {
55e5c285 509 if (cpu->cpu_index == index) {
bdc44640 510 return cpu;
55e5c285 511 }
ea041c0e 512 }
5b6dd868 513
bdc44640 514 return NULL;
ea041c0e
FB
515}
516
09daed84
EI
517#if !defined(CONFIG_USER_ONLY)
518void tcg_cpu_address_space_init(CPUState *cpu, AddressSpace *as)
519{
520 /* We only support one address space per cpu at the moment. */
521 assert(cpu->as == as);
522
523 if (cpu->tcg_as_listener) {
524 memory_listener_unregister(cpu->tcg_as_listener);
525 } else {
526 cpu->tcg_as_listener = g_new0(MemoryListener, 1);
527 }
528 cpu->tcg_as_listener->commit = tcg_commit;
529 memory_listener_register(cpu->tcg_as_listener, as);
530}
531#endif
532
5b6dd868 533void cpu_exec_init(CPUArchState *env)
ea041c0e 534{
5b6dd868 535 CPUState *cpu = ENV_GET_CPU(env);
b170fce3 536 CPUClass *cc = CPU_GET_CLASS(cpu);
bdc44640 537 CPUState *some_cpu;
5b6dd868
BS
538 int cpu_index;
539
540#if defined(CONFIG_USER_ONLY)
541 cpu_list_lock();
542#endif
5b6dd868 543 cpu_index = 0;
bdc44640 544 CPU_FOREACH(some_cpu) {
5b6dd868
BS
545 cpu_index++;
546 }
55e5c285 547 cpu->cpu_index = cpu_index;
1b1ed8dc 548 cpu->numa_node = 0;
f0c3c505 549 QTAILQ_INIT(&cpu->breakpoints);
ff4700b0 550 QTAILQ_INIT(&cpu->watchpoints);
5b6dd868 551#ifndef CONFIG_USER_ONLY
09daed84 552 cpu->as = &address_space_memory;
5b6dd868 553 cpu->thread_id = qemu_get_thread_id();
cba70549 554 cpu_reload_memory_map(cpu);
5b6dd868 555#endif
bdc44640 556 QTAILQ_INSERT_TAIL(&cpus, cpu, node);
5b6dd868
BS
557#if defined(CONFIG_USER_ONLY)
558 cpu_list_unlock();
559#endif
e0d47944
AF
560 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
561 vmstate_register(NULL, cpu_index, &vmstate_cpu_common, cpu);
562 }
5b6dd868 563#if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
5b6dd868
BS
564 register_savevm(NULL, "cpu", cpu_index, CPU_SAVE_VERSION,
565 cpu_save, cpu_load, env);
b170fce3 566 assert(cc->vmsd == NULL);
e0d47944 567 assert(qdev_get_vmsd(DEVICE(cpu)) == NULL);
5b6dd868 568#endif
b170fce3
AF
569 if (cc->vmsd != NULL) {
570 vmstate_register(NULL, cpu_index, cc->vmsd, cpu);
571 }
ea041c0e
FB
572}
573
94df27fd 574#if defined(CONFIG_USER_ONLY)
00b941e5 575static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
94df27fd
PB
576{
577 tb_invalidate_phys_page_range(pc, pc + 1, 0);
578}
579#else
00b941e5 580static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
1e7855a5 581{
e8262a1b
MF
582 hwaddr phys = cpu_get_phys_page_debug(cpu, pc);
583 if (phys != -1) {
09daed84 584 tb_invalidate_phys_addr(cpu->as,
29d8ec7b 585 phys | (pc & ~TARGET_PAGE_MASK));
e8262a1b 586 }
1e7855a5 587}
c27004ec 588#endif
d720b93d 589
c527ee8f 590#if defined(CONFIG_USER_ONLY)
75a34036 591void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
c527ee8f
PB
592
593{
594}
595
3ee887e8
PM
596int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
597 int flags)
598{
599 return -ENOSYS;
600}
601
602void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
603{
604}
605
75a34036 606int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
c527ee8f
PB
607 int flags, CPUWatchpoint **watchpoint)
608{
609 return -ENOSYS;
610}
611#else
6658ffb8 612/* Add a watchpoint. */
75a34036 613int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
a1d1bb31 614 int flags, CPUWatchpoint **watchpoint)
6658ffb8 615{
c0ce998e 616 CPUWatchpoint *wp;
6658ffb8 617
05068c0d 618 /* forbid ranges which are empty or run off the end of the address space */
07e2863d 619 if (len == 0 || (addr + len - 1) < addr) {
75a34036
AF
620 error_report("tried to set invalid watchpoint at %"
621 VADDR_PRIx ", len=%" VADDR_PRIu, addr, len);
b4051334
AL
622 return -EINVAL;
623 }
7267c094 624 wp = g_malloc(sizeof(*wp));
a1d1bb31
AL
625
626 wp->vaddr = addr;
05068c0d 627 wp->len = len;
a1d1bb31
AL
628 wp->flags = flags;
629
2dc9f411 630 /* keep all GDB-injected watchpoints in front */
ff4700b0
AF
631 if (flags & BP_GDB) {
632 QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry);
633 } else {
634 QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry);
635 }
6658ffb8 636
31b030d4 637 tlb_flush_page(cpu, addr);
a1d1bb31
AL
638
639 if (watchpoint)
640 *watchpoint = wp;
641 return 0;
6658ffb8
PB
642}
643
a1d1bb31 644/* Remove a specific watchpoint. */
75a34036 645int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
a1d1bb31 646 int flags)
6658ffb8 647{
a1d1bb31 648 CPUWatchpoint *wp;
6658ffb8 649
ff4700b0 650 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
05068c0d 651 if (addr == wp->vaddr && len == wp->len
6e140f28 652 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
75a34036 653 cpu_watchpoint_remove_by_ref(cpu, wp);
6658ffb8
PB
654 return 0;
655 }
656 }
a1d1bb31 657 return -ENOENT;
6658ffb8
PB
658}
659
a1d1bb31 660/* Remove a specific watchpoint by reference. */
75a34036 661void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
a1d1bb31 662{
ff4700b0 663 QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry);
7d03f82f 664
31b030d4 665 tlb_flush_page(cpu, watchpoint->vaddr);
a1d1bb31 666
7267c094 667 g_free(watchpoint);
a1d1bb31
AL
668}
669
670/* Remove all matching watchpoints. */
75a34036 671void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
a1d1bb31 672{
c0ce998e 673 CPUWatchpoint *wp, *next;
a1d1bb31 674
ff4700b0 675 QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) {
75a34036
AF
676 if (wp->flags & mask) {
677 cpu_watchpoint_remove_by_ref(cpu, wp);
678 }
c0ce998e 679 }
7d03f82f 680}
05068c0d
PM
681
682/* Return true if this watchpoint address matches the specified
683 * access (ie the address range covered by the watchpoint overlaps
684 * partially or completely with the address range covered by the
685 * access).
686 */
687static inline bool cpu_watchpoint_address_matches(CPUWatchpoint *wp,
688 vaddr addr,
689 vaddr len)
690{
691 /* We know the lengths are non-zero, but a little caution is
692 * required to avoid errors in the case where the range ends
693 * exactly at the top of the address space and so addr + len
694 * wraps round to zero.
695 */
696 vaddr wpend = wp->vaddr + wp->len - 1;
697 vaddr addrend = addr + len - 1;
698
699 return !(addr > wpend || wp->vaddr > addrend);
700}
701
c527ee8f 702#endif
7d03f82f 703
a1d1bb31 704/* Add a breakpoint. */
b3310ab3 705int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags,
a1d1bb31 706 CPUBreakpoint **breakpoint)
4c3a88a2 707{
c0ce998e 708 CPUBreakpoint *bp;
3b46e624 709
7267c094 710 bp = g_malloc(sizeof(*bp));
4c3a88a2 711
a1d1bb31
AL
712 bp->pc = pc;
713 bp->flags = flags;
714
2dc9f411 715 /* keep all GDB-injected breakpoints in front */
00b941e5 716 if (flags & BP_GDB) {
f0c3c505 717 QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry);
00b941e5 718 } else {
f0c3c505 719 QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry);
00b941e5 720 }
3b46e624 721
f0c3c505 722 breakpoint_invalidate(cpu, pc);
a1d1bb31 723
00b941e5 724 if (breakpoint) {
a1d1bb31 725 *breakpoint = bp;
00b941e5 726 }
4c3a88a2 727 return 0;
4c3a88a2
FB
728}
729
a1d1bb31 730/* Remove a specific breakpoint. */
b3310ab3 731int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags)
a1d1bb31 732{
a1d1bb31
AL
733 CPUBreakpoint *bp;
734
f0c3c505 735 QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) {
a1d1bb31 736 if (bp->pc == pc && bp->flags == flags) {
b3310ab3 737 cpu_breakpoint_remove_by_ref(cpu, bp);
a1d1bb31
AL
738 return 0;
739 }
7d03f82f 740 }
a1d1bb31 741 return -ENOENT;
7d03f82f
EI
742}
743
a1d1bb31 744/* Remove a specific breakpoint by reference. */
b3310ab3 745void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint)
4c3a88a2 746{
f0c3c505
AF
747 QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry);
748
749 breakpoint_invalidate(cpu, breakpoint->pc);
a1d1bb31 750
7267c094 751 g_free(breakpoint);
a1d1bb31
AL
752}
753
754/* Remove all matching breakpoints. */
b3310ab3 755void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
a1d1bb31 756{
c0ce998e 757 CPUBreakpoint *bp, *next;
a1d1bb31 758
f0c3c505 759 QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) {
b3310ab3
AF
760 if (bp->flags & mask) {
761 cpu_breakpoint_remove_by_ref(cpu, bp);
762 }
c0ce998e 763 }
4c3a88a2
FB
764}
765
c33a346e
FB
766/* enable or disable single step mode. EXCP_DEBUG is returned by the
767 CPU loop after each instruction */
3825b28f 768void cpu_single_step(CPUState *cpu, int enabled)
c33a346e 769{
ed2803da
AF
770 if (cpu->singlestep_enabled != enabled) {
771 cpu->singlestep_enabled = enabled;
772 if (kvm_enabled()) {
38e478ec 773 kvm_update_guest_debug(cpu, 0);
ed2803da 774 } else {
ccbb4d44 775 /* must flush all the translated code to avoid inconsistencies */
e22a25c9 776 /* XXX: only flush what is necessary */
38e478ec 777 CPUArchState *env = cpu->env_ptr;
e22a25c9
AL
778 tb_flush(env);
779 }
c33a346e 780 }
c33a346e
FB
781}
782
a47dddd7 783void cpu_abort(CPUState *cpu, const char *fmt, ...)
7501267e
FB
784{
785 va_list ap;
493ae1f0 786 va_list ap2;
7501267e
FB
787
788 va_start(ap, fmt);
493ae1f0 789 va_copy(ap2, ap);
7501267e
FB
790 fprintf(stderr, "qemu: fatal: ");
791 vfprintf(stderr, fmt, ap);
792 fprintf(stderr, "\n");
878096ee 793 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP);
93fcfe39
AL
794 if (qemu_log_enabled()) {
795 qemu_log("qemu: fatal: ");
796 qemu_log_vprintf(fmt, ap2);
797 qemu_log("\n");
a0762859 798 log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP);
31b1a7b4 799 qemu_log_flush();
93fcfe39 800 qemu_log_close();
924edcae 801 }
493ae1f0 802 va_end(ap2);
f9373291 803 va_end(ap);
fd052bf6
RV
804#if defined(CONFIG_USER_ONLY)
805 {
806 struct sigaction act;
807 sigfillset(&act.sa_mask);
808 act.sa_handler = SIG_DFL;
809 sigaction(SIGABRT, &act, NULL);
810 }
811#endif
7501267e
FB
812 abort();
813}
814
0124311e 815#if !defined(CONFIG_USER_ONLY)
0dc3f44a 816/* Called from RCU critical section */
041603fe
PB
817static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
818{
819 RAMBlock *block;
820
43771539 821 block = atomic_rcu_read(&ram_list.mru_block);
9b8424d5 822 if (block && addr - block->offset < block->max_length) {
041603fe
PB
823 goto found;
824 }
0dc3f44a 825 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
9b8424d5 826 if (addr - block->offset < block->max_length) {
041603fe
PB
827 goto found;
828 }
829 }
830
831 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
832 abort();
833
834found:
43771539
PB
835 /* It is safe to write mru_block outside the iothread lock. This
836 * is what happens:
837 *
838 * mru_block = xxx
839 * rcu_read_unlock()
840 * xxx removed from list
841 * rcu_read_lock()
842 * read mru_block
843 * mru_block = NULL;
844 * call_rcu(reclaim_ramblock, xxx);
845 * rcu_read_unlock()
846 *
847 * atomic_rcu_set is not needed here. The block was already published
848 * when it was placed into the list. Here we're just making an extra
849 * copy of the pointer.
850 */
041603fe
PB
851 ram_list.mru_block = block;
852 return block;
853}
854
a2f4d5be 855static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length)
d24981d3 856{
041603fe 857 ram_addr_t start1;
a2f4d5be
JQ
858 RAMBlock *block;
859 ram_addr_t end;
860
861 end = TARGET_PAGE_ALIGN(start + length);
862 start &= TARGET_PAGE_MASK;
d24981d3 863
0dc3f44a 864 rcu_read_lock();
041603fe
PB
865 block = qemu_get_ram_block(start);
866 assert(block == qemu_get_ram_block(end - 1));
1240be24 867 start1 = (uintptr_t)ramblock_ptr(block, start - block->offset);
041603fe 868 cpu_tlb_reset_dirty_all(start1, length);
0dc3f44a 869 rcu_read_unlock();
d24981d3
JQ
870}
871
5579c7f3 872/* Note: start and end must be within the same ram block. */
a2f4d5be 873void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t length,
52159192 874 unsigned client)
1ccde1cb 875{
1ccde1cb
FB
876 if (length == 0)
877 return;
c8d6f66a 878 cpu_physical_memory_clear_dirty_range_type(start, length, client);
f23db169 879
d24981d3 880 if (tcg_enabled()) {
a2f4d5be 881 tlb_reset_dirty_range_all(start, length);
5579c7f3 882 }
1ccde1cb
FB
883}
884
981fdf23 885static void cpu_physical_memory_set_dirty_tracking(bool enable)
74576198
AL
886{
887 in_migration = enable;
74576198
AL
888}
889
79e2b9ae 890/* Called from RCU critical section */
bb0e627a 891hwaddr memory_region_section_get_iotlb(CPUState *cpu,
149f54b5
PB
892 MemoryRegionSection *section,
893 target_ulong vaddr,
894 hwaddr paddr, hwaddr xlat,
895 int prot,
896 target_ulong *address)
e5548617 897{
a8170e5e 898 hwaddr iotlb;
e5548617
BS
899 CPUWatchpoint *wp;
900
cc5bea60 901 if (memory_region_is_ram(section->mr)) {
e5548617
BS
902 /* Normal RAM. */
903 iotlb = (memory_region_get_ram_addr(section->mr) & TARGET_PAGE_MASK)
149f54b5 904 + xlat;
e5548617 905 if (!section->readonly) {
b41aac4f 906 iotlb |= PHYS_SECTION_NOTDIRTY;
e5548617 907 } else {
b41aac4f 908 iotlb |= PHYS_SECTION_ROM;
e5548617
BS
909 }
910 } else {
1b3fb98f 911 iotlb = section - section->address_space->dispatch->map.sections;
149f54b5 912 iotlb += xlat;
e5548617
BS
913 }
914
915 /* Make accesses to pages with watchpoints go via the
916 watchpoint trap routines. */
ff4700b0 917 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
05068c0d 918 if (cpu_watchpoint_address_matches(wp, vaddr, TARGET_PAGE_SIZE)) {
e5548617
BS
919 /* Avoid trapping reads of pages with a write breakpoint. */
920 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
b41aac4f 921 iotlb = PHYS_SECTION_WATCH + paddr;
e5548617
BS
922 *address |= TLB_MMIO;
923 break;
924 }
925 }
926 }
927
928 return iotlb;
929}
9fa3e853
FB
930#endif /* defined(CONFIG_USER_ONLY) */
931
e2eef170 932#if !defined(CONFIG_USER_ONLY)
8da3ff18 933
c227f099 934static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
5312bd8b 935 uint16_t section);
acc9d80b 936static subpage_t *subpage_init(AddressSpace *as, hwaddr base);
54688b1e 937
a2b257d6
IM
938static void *(*phys_mem_alloc)(size_t size, uint64_t *align) =
939 qemu_anon_ram_alloc;
91138037
MA
940
941/*
942 * Set a custom physical guest memory alloator.
943 * Accelerators with unusual needs may need this. Hopefully, we can
944 * get rid of it eventually.
945 */
a2b257d6 946void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align))
91138037
MA
947{
948 phys_mem_alloc = alloc;
949}
950
53cb28cb
MA
951static uint16_t phys_section_add(PhysPageMap *map,
952 MemoryRegionSection *section)
5312bd8b 953{
68f3f65b
PB
954 /* The physical section number is ORed with a page-aligned
955 * pointer to produce the iotlb entries. Thus it should
956 * never overflow into the page-aligned value.
957 */
53cb28cb 958 assert(map->sections_nb < TARGET_PAGE_SIZE);
68f3f65b 959
53cb28cb
MA
960 if (map->sections_nb == map->sections_nb_alloc) {
961 map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16);
962 map->sections = g_renew(MemoryRegionSection, map->sections,
963 map->sections_nb_alloc);
5312bd8b 964 }
53cb28cb 965 map->sections[map->sections_nb] = *section;
dfde4e6e 966 memory_region_ref(section->mr);
53cb28cb 967 return map->sections_nb++;
5312bd8b
AK
968}
969
058bc4b5
PB
970static void phys_section_destroy(MemoryRegion *mr)
971{
dfde4e6e
PB
972 memory_region_unref(mr);
973
058bc4b5
PB
974 if (mr->subpage) {
975 subpage_t *subpage = container_of(mr, subpage_t, iomem);
b4fefef9 976 object_unref(OBJECT(&subpage->iomem));
058bc4b5
PB
977 g_free(subpage);
978 }
979}
980
6092666e 981static void phys_sections_free(PhysPageMap *map)
5312bd8b 982{
9affd6fc
PB
983 while (map->sections_nb > 0) {
984 MemoryRegionSection *section = &map->sections[--map->sections_nb];
058bc4b5
PB
985 phys_section_destroy(section->mr);
986 }
9affd6fc
PB
987 g_free(map->sections);
988 g_free(map->nodes);
5312bd8b
AK
989}
990
ac1970fb 991static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section)
0f0cb164
AK
992{
993 subpage_t *subpage;
a8170e5e 994 hwaddr base = section->offset_within_address_space
0f0cb164 995 & TARGET_PAGE_MASK;
97115a8d 996 MemoryRegionSection *existing = phys_page_find(d->phys_map, base,
53cb28cb 997 d->map.nodes, d->map.sections);
0f0cb164
AK
998 MemoryRegionSection subsection = {
999 .offset_within_address_space = base,
052e87b0 1000 .size = int128_make64(TARGET_PAGE_SIZE),
0f0cb164 1001 };
a8170e5e 1002 hwaddr start, end;
0f0cb164 1003
f3705d53 1004 assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
0f0cb164 1005
f3705d53 1006 if (!(existing->mr->subpage)) {
acc9d80b 1007 subpage = subpage_init(d->as, base);
3be91e86 1008 subsection.address_space = d->as;
0f0cb164 1009 subsection.mr = &subpage->iomem;
ac1970fb 1010 phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
53cb28cb 1011 phys_section_add(&d->map, &subsection));
0f0cb164 1012 } else {
f3705d53 1013 subpage = container_of(existing->mr, subpage_t, iomem);
0f0cb164
AK
1014 }
1015 start = section->offset_within_address_space & ~TARGET_PAGE_MASK;
052e87b0 1016 end = start + int128_get64(section->size) - 1;
53cb28cb
MA
1017 subpage_register(subpage, start, end,
1018 phys_section_add(&d->map, section));
0f0cb164
AK
1019}
1020
1021
052e87b0
PB
1022static void register_multipage(AddressSpaceDispatch *d,
1023 MemoryRegionSection *section)
33417e70 1024{
a8170e5e 1025 hwaddr start_addr = section->offset_within_address_space;
53cb28cb 1026 uint16_t section_index = phys_section_add(&d->map, section);
052e87b0
PB
1027 uint64_t num_pages = int128_get64(int128_rshift(section->size,
1028 TARGET_PAGE_BITS));
dd81124b 1029
733d5ef5
PB
1030 assert(num_pages);
1031 phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
33417e70
FB
1032}
1033
ac1970fb 1034static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
0f0cb164 1035{
89ae337a 1036 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
00752703 1037 AddressSpaceDispatch *d = as->next_dispatch;
99b9cc06 1038 MemoryRegionSection now = *section, remain = *section;
052e87b0 1039 Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
0f0cb164 1040
733d5ef5
PB
1041 if (now.offset_within_address_space & ~TARGET_PAGE_MASK) {
1042 uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space)
1043 - now.offset_within_address_space;
1044
052e87b0 1045 now.size = int128_min(int128_make64(left), now.size);
ac1970fb 1046 register_subpage(d, &now);
733d5ef5 1047 } else {
052e87b0 1048 now.size = int128_zero();
733d5ef5 1049 }
052e87b0
PB
1050 while (int128_ne(remain.size, now.size)) {
1051 remain.size = int128_sub(remain.size, now.size);
1052 remain.offset_within_address_space += int128_get64(now.size);
1053 remain.offset_within_region += int128_get64(now.size);
69b67646 1054 now = remain;
052e87b0 1055 if (int128_lt(remain.size, page_size)) {
733d5ef5 1056 register_subpage(d, &now);
88266249 1057 } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) {
052e87b0 1058 now.size = page_size;
ac1970fb 1059 register_subpage(d, &now);
69b67646 1060 } else {
052e87b0 1061 now.size = int128_and(now.size, int128_neg(page_size));
ac1970fb 1062 register_multipage(d, &now);
69b67646 1063 }
0f0cb164
AK
1064 }
1065}
1066
62a2744c
SY
1067void qemu_flush_coalesced_mmio_buffer(void)
1068{
1069 if (kvm_enabled())
1070 kvm_flush_coalesced_mmio_buffer();
1071}
1072
b2a8658e
UD
1073void qemu_mutex_lock_ramlist(void)
1074{
1075 qemu_mutex_lock(&ram_list.mutex);
1076}
1077
1078void qemu_mutex_unlock_ramlist(void)
1079{
1080 qemu_mutex_unlock(&ram_list.mutex);
1081}
1082
e1e84ba0 1083#ifdef __linux__
c902760f
MT
1084
1085#include <sys/vfs.h>
1086
1087#define HUGETLBFS_MAGIC 0x958458f6
1088
fc7a5800 1089static long gethugepagesize(const char *path, Error **errp)
c902760f
MT
1090{
1091 struct statfs fs;
1092 int ret;
1093
1094 do {
9742bf26 1095 ret = statfs(path, &fs);
c902760f
MT
1096 } while (ret != 0 && errno == EINTR);
1097
1098 if (ret != 0) {
fc7a5800
HT
1099 error_setg_errno(errp, errno, "failed to get page size of file %s",
1100 path);
9742bf26 1101 return 0;
c902760f
MT
1102 }
1103
1104 if (fs.f_type != HUGETLBFS_MAGIC)
9742bf26 1105 fprintf(stderr, "Warning: path not on HugeTLBFS: %s\n", path);
c902760f
MT
1106
1107 return fs.f_bsize;
1108}
1109
04b16653
AW
1110static void *file_ram_alloc(RAMBlock *block,
1111 ram_addr_t memory,
7f56e740
PB
1112 const char *path,
1113 Error **errp)
c902760f
MT
1114{
1115 char *filename;
8ca761f6
PF
1116 char *sanitized_name;
1117 char *c;
557529dd 1118 void *area = NULL;
c902760f 1119 int fd;
557529dd 1120 uint64_t hpagesize;
fc7a5800 1121 Error *local_err = NULL;
c902760f 1122
fc7a5800
HT
1123 hpagesize = gethugepagesize(path, &local_err);
1124 if (local_err) {
1125 error_propagate(errp, local_err);
f9a49dfa 1126 goto error;
c902760f 1127 }
a2b257d6 1128 block->mr->align = hpagesize;
c902760f
MT
1129
1130 if (memory < hpagesize) {
557529dd
HT
1131 error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "
1132 "or larger than huge page size 0x%" PRIx64,
1133 memory, hpagesize);
1134 goto error;
c902760f
MT
1135 }
1136
1137 if (kvm_enabled() && !kvm_has_sync_mmu()) {
7f56e740
PB
1138 error_setg(errp,
1139 "host lacks kvm mmu notifiers, -mem-path unsupported");
f9a49dfa 1140 goto error;
c902760f
MT
1141 }
1142
8ca761f6 1143 /* Make name safe to use with mkstemp by replacing '/' with '_'. */
83234bf2 1144 sanitized_name = g_strdup(memory_region_name(block->mr));
8ca761f6
PF
1145 for (c = sanitized_name; *c != '\0'; c++) {
1146 if (*c == '/')
1147 *c = '_';
1148 }
1149
1150 filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path,
1151 sanitized_name);
1152 g_free(sanitized_name);
c902760f
MT
1153
1154 fd = mkstemp(filename);
1155 if (fd < 0) {
7f56e740
PB
1156 error_setg_errno(errp, errno,
1157 "unable to create backing store for hugepages");
e4ada482 1158 g_free(filename);
f9a49dfa 1159 goto error;
c902760f
MT
1160 }
1161 unlink(filename);
e4ada482 1162 g_free(filename);
c902760f
MT
1163
1164 memory = (memory+hpagesize-1) & ~(hpagesize-1);
1165
1166 /*
1167 * ftruncate is not supported by hugetlbfs in older
1168 * hosts, so don't bother bailing out on errors.
1169 * If anything goes wrong with it under other filesystems,
1170 * mmap will fail.
1171 */
7f56e740 1172 if (ftruncate(fd, memory)) {
9742bf26 1173 perror("ftruncate");
7f56e740 1174 }
c902760f 1175
dbcb8981
PB
1176 area = mmap(0, memory, PROT_READ | PROT_WRITE,
1177 (block->flags & RAM_SHARED ? MAP_SHARED : MAP_PRIVATE),
1178 fd, 0);
c902760f 1179 if (area == MAP_FAILED) {
7f56e740
PB
1180 error_setg_errno(errp, errno,
1181 "unable to map backing store for hugepages");
9742bf26 1182 close(fd);
f9a49dfa 1183 goto error;
c902760f 1184 }
ef36fa14
MT
1185
1186 if (mem_prealloc) {
38183310 1187 os_mem_prealloc(fd, area, memory);
ef36fa14
MT
1188 }
1189
04b16653 1190 block->fd = fd;
c902760f 1191 return area;
f9a49dfa
MT
1192
1193error:
1194 if (mem_prealloc) {
81b07353 1195 error_report("%s", error_get_pretty(*errp));
f9a49dfa
MT
1196 exit(1);
1197 }
1198 return NULL;
c902760f
MT
1199}
1200#endif
1201
0dc3f44a 1202/* Called with the ramlist lock held. */
d17b5288 1203static ram_addr_t find_ram_offset(ram_addr_t size)
04b16653
AW
1204{
1205 RAMBlock *block, *next_block;
3e837b2c 1206 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
04b16653 1207
49cd9ac6
SH
1208 assert(size != 0); /* it would hand out same offset multiple times */
1209
0dc3f44a 1210 if (QLIST_EMPTY_RCU(&ram_list.blocks)) {
04b16653 1211 return 0;
0d53d9fe 1212 }
04b16653 1213
0dc3f44a 1214 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
f15fbc4b 1215 ram_addr_t end, next = RAM_ADDR_MAX;
04b16653 1216
62be4e3a 1217 end = block->offset + block->max_length;
04b16653 1218
0dc3f44a 1219 QLIST_FOREACH_RCU(next_block, &ram_list.blocks, next) {
04b16653
AW
1220 if (next_block->offset >= end) {
1221 next = MIN(next, next_block->offset);
1222 }
1223 }
1224 if (next - end >= size && next - end < mingap) {
3e837b2c 1225 offset = end;
04b16653
AW
1226 mingap = next - end;
1227 }
1228 }
3e837b2c
AW
1229
1230 if (offset == RAM_ADDR_MAX) {
1231 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
1232 (uint64_t)size);
1233 abort();
1234 }
1235
04b16653
AW
1236 return offset;
1237}
1238
652d7ec2 1239ram_addr_t last_ram_offset(void)
d17b5288
AW
1240{
1241 RAMBlock *block;
1242 ram_addr_t last = 0;
1243
0dc3f44a
MD
1244 rcu_read_lock();
1245 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
62be4e3a 1246 last = MAX(last, block->offset + block->max_length);
0d53d9fe 1247 }
0dc3f44a 1248 rcu_read_unlock();
d17b5288
AW
1249 return last;
1250}
1251
ddb97f1d
JB
1252static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
1253{
1254 int ret;
ddb97f1d
JB
1255
1256 /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */
47c8ca53 1257 if (!machine_dump_guest_core(current_machine)) {
ddb97f1d
JB
1258 ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP);
1259 if (ret) {
1260 perror("qemu_madvise");
1261 fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, "
1262 "but dump_guest_core=off specified\n");
1263 }
1264 }
1265}
1266
0dc3f44a
MD
1267/* Called within an RCU critical section, or while the ramlist lock
1268 * is held.
1269 */
20cfe881 1270static RAMBlock *find_ram_block(ram_addr_t addr)
84b89d78 1271{
20cfe881 1272 RAMBlock *block;
84b89d78 1273
0dc3f44a 1274 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
c5705a77 1275 if (block->offset == addr) {
20cfe881 1276 return block;
c5705a77
AK
1277 }
1278 }
20cfe881
HT
1279
1280 return NULL;
1281}
1282
ae3a7047 1283/* Called with iothread lock held. */
20cfe881
HT
1284void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
1285{
ae3a7047 1286 RAMBlock *new_block, *block;
20cfe881 1287
0dc3f44a 1288 rcu_read_lock();
ae3a7047 1289 new_block = find_ram_block(addr);
c5705a77
AK
1290 assert(new_block);
1291 assert(!new_block->idstr[0]);
84b89d78 1292
09e5ab63
AL
1293 if (dev) {
1294 char *id = qdev_get_dev_path(dev);
84b89d78
CM
1295 if (id) {
1296 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
7267c094 1297 g_free(id);
84b89d78
CM
1298 }
1299 }
1300 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
1301
0dc3f44a 1302 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
c5705a77 1303 if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
84b89d78
CM
1304 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
1305 new_block->idstr);
1306 abort();
1307 }
1308 }
0dc3f44a 1309 rcu_read_unlock();
c5705a77
AK
1310}
1311
ae3a7047 1312/* Called with iothread lock held. */
20cfe881
HT
1313void qemu_ram_unset_idstr(ram_addr_t addr)
1314{
ae3a7047 1315 RAMBlock *block;
20cfe881 1316
ae3a7047
MD
1317 /* FIXME: arch_init.c assumes that this is not called throughout
1318 * migration. Ignore the problem since hot-unplug during migration
1319 * does not work anyway.
1320 */
1321
0dc3f44a 1322 rcu_read_lock();
ae3a7047 1323 block = find_ram_block(addr);
20cfe881
HT
1324 if (block) {
1325 memset(block->idstr, 0, sizeof(block->idstr));
1326 }
0dc3f44a 1327 rcu_read_unlock();
20cfe881
HT
1328}
1329
8490fc78
LC
1330static int memory_try_enable_merging(void *addr, size_t len)
1331{
75cc7f01 1332 if (!machine_mem_merge(current_machine)) {
8490fc78
LC
1333 /* disabled by the user */
1334 return 0;
1335 }
1336
1337 return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE);
1338}
1339
62be4e3a
MT
1340/* Only legal before guest might have detected the memory size: e.g. on
1341 * incoming migration, or right after reset.
1342 *
1343 * As memory core doesn't know how is memory accessed, it is up to
1344 * resize callback to update device state and/or add assertions to detect
1345 * misuse, if necessary.
1346 */
1347int qemu_ram_resize(ram_addr_t base, ram_addr_t newsize, Error **errp)
1348{
1349 RAMBlock *block = find_ram_block(base);
1350
1351 assert(block);
1352
129ddaf3
MT
1353 newsize = TARGET_PAGE_ALIGN(newsize);
1354
62be4e3a
MT
1355 if (block->used_length == newsize) {
1356 return 0;
1357 }
1358
1359 if (!(block->flags & RAM_RESIZEABLE)) {
1360 error_setg_errno(errp, EINVAL,
1361 "Length mismatch: %s: 0x" RAM_ADDR_FMT
1362 " in != 0x" RAM_ADDR_FMT, block->idstr,
1363 newsize, block->used_length);
1364 return -EINVAL;
1365 }
1366
1367 if (block->max_length < newsize) {
1368 error_setg_errno(errp, EINVAL,
1369 "Length too large: %s: 0x" RAM_ADDR_FMT
1370 " > 0x" RAM_ADDR_FMT, block->idstr,
1371 newsize, block->max_length);
1372 return -EINVAL;
1373 }
1374
1375 cpu_physical_memory_clear_dirty_range(block->offset, block->used_length);
1376 block->used_length = newsize;
1377 cpu_physical_memory_set_dirty_range(block->offset, block->used_length);
1378 memory_region_set_size(block->mr, newsize);
1379 if (block->resized) {
1380 block->resized(block->idstr, newsize, block->host);
1381 }
1382 return 0;
1383}
1384
ef701d7b 1385static ram_addr_t ram_block_add(RAMBlock *new_block, Error **errp)
c5705a77 1386{
e1c57ab8 1387 RAMBlock *block;
0d53d9fe 1388 RAMBlock *last_block = NULL;
2152f5ca
JQ
1389 ram_addr_t old_ram_size, new_ram_size;
1390
1391 old_ram_size = last_ram_offset() >> TARGET_PAGE_BITS;
c5705a77 1392
b2a8658e 1393 qemu_mutex_lock_ramlist();
9b8424d5 1394 new_block->offset = find_ram_offset(new_block->max_length);
e1c57ab8
PB
1395
1396 if (!new_block->host) {
1397 if (xen_enabled()) {
9b8424d5
MT
1398 xen_ram_alloc(new_block->offset, new_block->max_length,
1399 new_block->mr);
e1c57ab8 1400 } else {
9b8424d5 1401 new_block->host = phys_mem_alloc(new_block->max_length,
a2b257d6 1402 &new_block->mr->align);
39228250 1403 if (!new_block->host) {
ef701d7b
HT
1404 error_setg_errno(errp, errno,
1405 "cannot set up guest memory '%s'",
1406 memory_region_name(new_block->mr));
1407 qemu_mutex_unlock_ramlist();
1408 return -1;
39228250 1409 }
9b8424d5 1410 memory_try_enable_merging(new_block->host, new_block->max_length);
6977dfe6 1411 }
c902760f 1412 }
94a6b54f 1413
0d53d9fe
MD
1414 /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ,
1415 * QLIST (which has an RCU-friendly variant) does not have insertion at
1416 * tail, so save the last element in last_block.
1417 */
0dc3f44a 1418 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
0d53d9fe 1419 last_block = block;
9b8424d5 1420 if (block->max_length < new_block->max_length) {
abb26d63
PB
1421 break;
1422 }
1423 }
1424 if (block) {
0dc3f44a 1425 QLIST_INSERT_BEFORE_RCU(block, new_block, next);
0d53d9fe 1426 } else if (last_block) {
0dc3f44a 1427 QLIST_INSERT_AFTER_RCU(last_block, new_block, next);
0d53d9fe 1428 } else { /* list is empty */
0dc3f44a 1429 QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next);
abb26d63 1430 }
0d6d3c87 1431 ram_list.mru_block = NULL;
94a6b54f 1432
0dc3f44a
MD
1433 /* Write list before version */
1434 smp_wmb();
f798b07f 1435 ram_list.version++;
b2a8658e 1436 qemu_mutex_unlock_ramlist();
f798b07f 1437
2152f5ca
JQ
1438 new_ram_size = last_ram_offset() >> TARGET_PAGE_BITS;
1439
1440 if (new_ram_size > old_ram_size) {
1ab4c8ce 1441 int i;
ae3a7047
MD
1442
1443 /* ram_list.dirty_memory[] is protected by the iothread lock. */
1ab4c8ce
JQ
1444 for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
1445 ram_list.dirty_memory[i] =
1446 bitmap_zero_extend(ram_list.dirty_memory[i],
1447 old_ram_size, new_ram_size);
1448 }
2152f5ca 1449 }
9b8424d5
MT
1450 cpu_physical_memory_set_dirty_range(new_block->offset,
1451 new_block->used_length);
94a6b54f 1452
a904c911
PB
1453 if (new_block->host) {
1454 qemu_ram_setup_dump(new_block->host, new_block->max_length);
1455 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE);
1456 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK);
1457 if (kvm_enabled()) {
1458 kvm_setup_guest_memory(new_block->host, new_block->max_length);
1459 }
e1c57ab8 1460 }
6f0437e8 1461
94a6b54f
PB
1462 return new_block->offset;
1463}
e9a1ab19 1464
0b183fc8 1465#ifdef __linux__
e1c57ab8 1466ram_addr_t qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
dbcb8981 1467 bool share, const char *mem_path,
7f56e740 1468 Error **errp)
e1c57ab8
PB
1469{
1470 RAMBlock *new_block;
ef701d7b
HT
1471 ram_addr_t addr;
1472 Error *local_err = NULL;
e1c57ab8
PB
1473
1474 if (xen_enabled()) {
7f56e740
PB
1475 error_setg(errp, "-mem-path not supported with Xen");
1476 return -1;
e1c57ab8
PB
1477 }
1478
1479 if (phys_mem_alloc != qemu_anon_ram_alloc) {
1480 /*
1481 * file_ram_alloc() needs to allocate just like
1482 * phys_mem_alloc, but we haven't bothered to provide
1483 * a hook there.
1484 */
7f56e740
PB
1485 error_setg(errp,
1486 "-mem-path not supported with this accelerator");
1487 return -1;
e1c57ab8
PB
1488 }
1489
1490 size = TARGET_PAGE_ALIGN(size);
1491 new_block = g_malloc0(sizeof(*new_block));
1492 new_block->mr = mr;
9b8424d5
MT
1493 new_block->used_length = size;
1494 new_block->max_length = size;
dbcb8981 1495 new_block->flags = share ? RAM_SHARED : 0;
7f56e740
PB
1496 new_block->host = file_ram_alloc(new_block, size,
1497 mem_path, errp);
1498 if (!new_block->host) {
1499 g_free(new_block);
1500 return -1;
1501 }
1502
ef701d7b
HT
1503 addr = ram_block_add(new_block, &local_err);
1504 if (local_err) {
1505 g_free(new_block);
1506 error_propagate(errp, local_err);
1507 return -1;
1508 }
1509 return addr;
e1c57ab8 1510}
0b183fc8 1511#endif
e1c57ab8 1512
62be4e3a
MT
1513static
1514ram_addr_t qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size,
1515 void (*resized)(const char*,
1516 uint64_t length,
1517 void *host),
1518 void *host, bool resizeable,
ef701d7b 1519 MemoryRegion *mr, Error **errp)
e1c57ab8
PB
1520{
1521 RAMBlock *new_block;
ef701d7b
HT
1522 ram_addr_t addr;
1523 Error *local_err = NULL;
e1c57ab8
PB
1524
1525 size = TARGET_PAGE_ALIGN(size);
62be4e3a 1526 max_size = TARGET_PAGE_ALIGN(max_size);
e1c57ab8
PB
1527 new_block = g_malloc0(sizeof(*new_block));
1528 new_block->mr = mr;
62be4e3a 1529 new_block->resized = resized;
9b8424d5
MT
1530 new_block->used_length = size;
1531 new_block->max_length = max_size;
62be4e3a 1532 assert(max_size >= size);
e1c57ab8
PB
1533 new_block->fd = -1;
1534 new_block->host = host;
1535 if (host) {
7bd4f430 1536 new_block->flags |= RAM_PREALLOC;
e1c57ab8 1537 }
62be4e3a
MT
1538 if (resizeable) {
1539 new_block->flags |= RAM_RESIZEABLE;
1540 }
ef701d7b
HT
1541 addr = ram_block_add(new_block, &local_err);
1542 if (local_err) {
1543 g_free(new_block);
1544 error_propagate(errp, local_err);
1545 return -1;
1546 }
1547 return addr;
e1c57ab8
PB
1548}
1549
62be4e3a
MT
1550ram_addr_t qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
1551 MemoryRegion *mr, Error **errp)
1552{
1553 return qemu_ram_alloc_internal(size, size, NULL, host, false, mr, errp);
1554}
1555
ef701d7b 1556ram_addr_t qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp)
6977dfe6 1557{
62be4e3a
MT
1558 return qemu_ram_alloc_internal(size, size, NULL, NULL, false, mr, errp);
1559}
1560
1561ram_addr_t qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz,
1562 void (*resized)(const char*,
1563 uint64_t length,
1564 void *host),
1565 MemoryRegion *mr, Error **errp)
1566{
1567 return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, mr, errp);
6977dfe6
YT
1568}
1569
1f2e98b6
AW
1570void qemu_ram_free_from_ptr(ram_addr_t addr)
1571{
1572 RAMBlock *block;
1573
b2a8658e 1574 qemu_mutex_lock_ramlist();
0dc3f44a 1575 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1f2e98b6 1576 if (addr == block->offset) {
0dc3f44a 1577 QLIST_REMOVE_RCU(block, next);
0d6d3c87 1578 ram_list.mru_block = NULL;
0dc3f44a
MD
1579 /* Write list before version */
1580 smp_wmb();
f798b07f 1581 ram_list.version++;
43771539 1582 g_free_rcu(block, rcu);
b2a8658e 1583 break;
1f2e98b6
AW
1584 }
1585 }
b2a8658e 1586 qemu_mutex_unlock_ramlist();
1f2e98b6
AW
1587}
1588
43771539
PB
1589static void reclaim_ramblock(RAMBlock *block)
1590{
1591 if (block->flags & RAM_PREALLOC) {
1592 ;
1593 } else if (xen_enabled()) {
1594 xen_invalidate_map_cache_entry(block->host);
1595#ifndef _WIN32
1596 } else if (block->fd >= 0) {
1597 munmap(block->host, block->max_length);
1598 close(block->fd);
1599#endif
1600 } else {
1601 qemu_anon_ram_free(block->host, block->max_length);
1602 }
1603 g_free(block);
1604}
1605
c227f099 1606void qemu_ram_free(ram_addr_t addr)
e9a1ab19 1607{
04b16653
AW
1608 RAMBlock *block;
1609
b2a8658e 1610 qemu_mutex_lock_ramlist();
0dc3f44a 1611 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
04b16653 1612 if (addr == block->offset) {
0dc3f44a 1613 QLIST_REMOVE_RCU(block, next);
0d6d3c87 1614 ram_list.mru_block = NULL;
0dc3f44a
MD
1615 /* Write list before version */
1616 smp_wmb();
f798b07f 1617 ram_list.version++;
43771539 1618 call_rcu(block, reclaim_ramblock, rcu);
b2a8658e 1619 break;
04b16653
AW
1620 }
1621 }
b2a8658e 1622 qemu_mutex_unlock_ramlist();
e9a1ab19
FB
1623}
1624
cd19cfa2
HY
1625#ifndef _WIN32
1626void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
1627{
1628 RAMBlock *block;
1629 ram_addr_t offset;
1630 int flags;
1631 void *area, *vaddr;
1632
0dc3f44a 1633 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
cd19cfa2 1634 offset = addr - block->offset;
9b8424d5 1635 if (offset < block->max_length) {
1240be24 1636 vaddr = ramblock_ptr(block, offset);
7bd4f430 1637 if (block->flags & RAM_PREALLOC) {
cd19cfa2 1638 ;
dfeaf2ab
MA
1639 } else if (xen_enabled()) {
1640 abort();
cd19cfa2
HY
1641 } else {
1642 flags = MAP_FIXED;
3435f395 1643 if (block->fd >= 0) {
dbcb8981
PB
1644 flags |= (block->flags & RAM_SHARED ?
1645 MAP_SHARED : MAP_PRIVATE);
3435f395
MA
1646 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1647 flags, block->fd, offset);
cd19cfa2 1648 } else {
2eb9fbaa
MA
1649 /*
1650 * Remap needs to match alloc. Accelerators that
1651 * set phys_mem_alloc never remap. If they did,
1652 * we'd need a remap hook here.
1653 */
1654 assert(phys_mem_alloc == qemu_anon_ram_alloc);
1655
cd19cfa2
HY
1656 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
1657 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1658 flags, -1, 0);
cd19cfa2
HY
1659 }
1660 if (area != vaddr) {
f15fbc4b
AP
1661 fprintf(stderr, "Could not remap addr: "
1662 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
cd19cfa2
HY
1663 length, addr);
1664 exit(1);
1665 }
8490fc78 1666 memory_try_enable_merging(vaddr, length);
ddb97f1d 1667 qemu_ram_setup_dump(vaddr, length);
cd19cfa2 1668 }
cd19cfa2
HY
1669 }
1670 }
1671}
1672#endif /* !_WIN32 */
1673
a35ba7be
PB
1674int qemu_get_ram_fd(ram_addr_t addr)
1675{
ae3a7047
MD
1676 RAMBlock *block;
1677 int fd;
a35ba7be 1678
0dc3f44a 1679 rcu_read_lock();
ae3a7047
MD
1680 block = qemu_get_ram_block(addr);
1681 fd = block->fd;
0dc3f44a 1682 rcu_read_unlock();
ae3a7047 1683 return fd;
a35ba7be
PB
1684}
1685
3fd74b84
DM
1686void *qemu_get_ram_block_host_ptr(ram_addr_t addr)
1687{
ae3a7047
MD
1688 RAMBlock *block;
1689 void *ptr;
3fd74b84 1690
0dc3f44a 1691 rcu_read_lock();
ae3a7047
MD
1692 block = qemu_get_ram_block(addr);
1693 ptr = ramblock_ptr(block, 0);
0dc3f44a 1694 rcu_read_unlock();
ae3a7047 1695 return ptr;
3fd74b84
DM
1696}
1697
1b5ec234 1698/* Return a host pointer to ram allocated with qemu_ram_alloc.
ae3a7047
MD
1699 * This should not be used for general purpose DMA. Use address_space_map
1700 * or address_space_rw instead. For local memory (e.g. video ram) that the
1701 * device owns, use memory_region_get_ram_ptr.
0dc3f44a
MD
1702 *
1703 * By the time this function returns, the returned pointer is not protected
1704 * by RCU anymore. If the caller is not within an RCU critical section and
1705 * does not hold the iothread lock, it must have other means of protecting the
1706 * pointer, such as a reference to the region that includes the incoming
1707 * ram_addr_t.
1b5ec234
PB
1708 */
1709void *qemu_get_ram_ptr(ram_addr_t addr)
1710{
ae3a7047
MD
1711 RAMBlock *block;
1712 void *ptr;
1b5ec234 1713
0dc3f44a 1714 rcu_read_lock();
ae3a7047
MD
1715 block = qemu_get_ram_block(addr);
1716
1717 if (xen_enabled() && block->host == NULL) {
0d6d3c87
PB
1718 /* We need to check if the requested address is in the RAM
1719 * because we don't want to map the entire memory in QEMU.
1720 * In that case just map until the end of the page.
1721 */
1722 if (block->offset == 0) {
ae3a7047 1723 ptr = xen_map_cache(addr, 0, 0);
0dc3f44a 1724 goto unlock;
0d6d3c87 1725 }
ae3a7047
MD
1726
1727 block->host = xen_map_cache(block->offset, block->max_length, 1);
0d6d3c87 1728 }
ae3a7047
MD
1729 ptr = ramblock_ptr(block, addr - block->offset);
1730
0dc3f44a
MD
1731unlock:
1732 rcu_read_unlock();
ae3a7047 1733 return ptr;
dc828ca1
PB
1734}
1735
38bee5dc 1736/* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr
ae3a7047 1737 * but takes a size argument.
0dc3f44a
MD
1738 *
1739 * By the time this function returns, the returned pointer is not protected
1740 * by RCU anymore. If the caller is not within an RCU critical section and
1741 * does not hold the iothread lock, it must have other means of protecting the
1742 * pointer, such as a reference to the region that includes the incoming
1743 * ram_addr_t.
ae3a7047 1744 */
cb85f7ab 1745static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
38bee5dc 1746{
ae3a7047 1747 void *ptr;
8ab934f9
SS
1748 if (*size == 0) {
1749 return NULL;
1750 }
868bb33f 1751 if (xen_enabled()) {
e41d7c69 1752 return xen_map_cache(addr, *size, 1);
868bb33f 1753 } else {
38bee5dc 1754 RAMBlock *block;
0dc3f44a
MD
1755 rcu_read_lock();
1756 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
9b8424d5
MT
1757 if (addr - block->offset < block->max_length) {
1758 if (addr - block->offset + *size > block->max_length)
1759 *size = block->max_length - addr + block->offset;
ae3a7047 1760 ptr = ramblock_ptr(block, addr - block->offset);
0dc3f44a 1761 rcu_read_unlock();
ae3a7047 1762 return ptr;
38bee5dc
SS
1763 }
1764 }
1765
1766 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1767 abort();
38bee5dc
SS
1768 }
1769}
1770
7443b437 1771/* Some of the softmmu routines need to translate from a host pointer
ae3a7047
MD
1772 * (typically a TLB entry) back to a ram offset.
1773 *
1774 * By the time this function returns, the returned pointer is not protected
1775 * by RCU anymore. If the caller is not within an RCU critical section and
1776 * does not hold the iothread lock, it must have other means of protecting the
1777 * pointer, such as a reference to the region that includes the incoming
1778 * ram_addr_t.
1779 */
1b5ec234 1780MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
5579c7f3 1781{
94a6b54f
PB
1782 RAMBlock *block;
1783 uint8_t *host = ptr;
ae3a7047 1784 MemoryRegion *mr;
94a6b54f 1785
868bb33f 1786 if (xen_enabled()) {
0dc3f44a 1787 rcu_read_lock();
e41d7c69 1788 *ram_addr = xen_ram_addr_from_mapcache(ptr);
ae3a7047 1789 mr = qemu_get_ram_block(*ram_addr)->mr;
0dc3f44a 1790 rcu_read_unlock();
ae3a7047 1791 return mr;
712c2b41
SS
1792 }
1793
0dc3f44a
MD
1794 rcu_read_lock();
1795 block = atomic_rcu_read(&ram_list.mru_block);
9b8424d5 1796 if (block && block->host && host - block->host < block->max_length) {
23887b79
PB
1797 goto found;
1798 }
1799
0dc3f44a 1800 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
432d268c
JN
1801 /* This case append when the block is not mapped. */
1802 if (block->host == NULL) {
1803 continue;
1804 }
9b8424d5 1805 if (host - block->host < block->max_length) {
23887b79 1806 goto found;
f471a17e 1807 }
94a6b54f 1808 }
432d268c 1809
0dc3f44a 1810 rcu_read_unlock();
1b5ec234 1811 return NULL;
23887b79
PB
1812
1813found:
1814 *ram_addr = block->offset + (host - block->host);
ae3a7047 1815 mr = block->mr;
0dc3f44a 1816 rcu_read_unlock();
ae3a7047 1817 return mr;
e890261f 1818}
f471a17e 1819
a8170e5e 1820static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
0e0df1e2 1821 uint64_t val, unsigned size)
9fa3e853 1822{
52159192 1823 if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) {
0e0df1e2 1824 tb_invalidate_phys_page_fast(ram_addr, size);
3a7d929e 1825 }
0e0df1e2
AK
1826 switch (size) {
1827 case 1:
1828 stb_p(qemu_get_ram_ptr(ram_addr), val);
1829 break;
1830 case 2:
1831 stw_p(qemu_get_ram_ptr(ram_addr), val);
1832 break;
1833 case 4:
1834 stl_p(qemu_get_ram_ptr(ram_addr), val);
1835 break;
1836 default:
1837 abort();
3a7d929e 1838 }
6886867e 1839 cpu_physical_memory_set_dirty_range_nocode(ram_addr, size);
f23db169
FB
1840 /* we remove the notdirty callback only if the code has been
1841 flushed */
a2cd8c85 1842 if (!cpu_physical_memory_is_clean(ram_addr)) {
4917cf44 1843 CPUArchState *env = current_cpu->env_ptr;
93afeade 1844 tlb_set_dirty(env, current_cpu->mem_io_vaddr);
4917cf44 1845 }
9fa3e853
FB
1846}
1847
b018ddf6
PB
1848static bool notdirty_mem_accepts(void *opaque, hwaddr addr,
1849 unsigned size, bool is_write)
1850{
1851 return is_write;
1852}
1853
0e0df1e2 1854static const MemoryRegionOps notdirty_mem_ops = {
0e0df1e2 1855 .write = notdirty_mem_write,
b018ddf6 1856 .valid.accepts = notdirty_mem_accepts,
0e0df1e2 1857 .endianness = DEVICE_NATIVE_ENDIAN,
1ccde1cb
FB
1858};
1859
0f459d16 1860/* Generate a debug exception if a watchpoint has been hit. */
05068c0d 1861static void check_watchpoint(int offset, int len, int flags)
0f459d16 1862{
93afeade
AF
1863 CPUState *cpu = current_cpu;
1864 CPUArchState *env = cpu->env_ptr;
06d55cc1 1865 target_ulong pc, cs_base;
0f459d16 1866 target_ulong vaddr;
a1d1bb31 1867 CPUWatchpoint *wp;
06d55cc1 1868 int cpu_flags;
0f459d16 1869
ff4700b0 1870 if (cpu->watchpoint_hit) {
06d55cc1
AL
1871 /* We re-entered the check after replacing the TB. Now raise
1872 * the debug interrupt so that is will trigger after the
1873 * current instruction. */
93afeade 1874 cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG);
06d55cc1
AL
1875 return;
1876 }
93afeade 1877 vaddr = (cpu->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
ff4700b0 1878 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
05068c0d
PM
1879 if (cpu_watchpoint_address_matches(wp, vaddr, len)
1880 && (wp->flags & flags)) {
08225676
PM
1881 if (flags == BP_MEM_READ) {
1882 wp->flags |= BP_WATCHPOINT_HIT_READ;
1883 } else {
1884 wp->flags |= BP_WATCHPOINT_HIT_WRITE;
1885 }
1886 wp->hitaddr = vaddr;
ff4700b0
AF
1887 if (!cpu->watchpoint_hit) {
1888 cpu->watchpoint_hit = wp;
239c51a5 1889 tb_check_watchpoint(cpu);
6e140f28 1890 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
27103424 1891 cpu->exception_index = EXCP_DEBUG;
5638d180 1892 cpu_loop_exit(cpu);
6e140f28
AL
1893 } else {
1894 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
648f034c 1895 tb_gen_code(cpu, pc, cs_base, cpu_flags, 1);
0ea8cb88 1896 cpu_resume_from_signal(cpu, NULL);
6e140f28 1897 }
06d55cc1 1898 }
6e140f28
AL
1899 } else {
1900 wp->flags &= ~BP_WATCHPOINT_HIT;
0f459d16
PB
1901 }
1902 }
1903}
1904
6658ffb8
PB
1905/* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
1906 so these check for a hit then pass through to the normal out-of-line
1907 phys routines. */
a8170e5e 1908static uint64_t watch_mem_read(void *opaque, hwaddr addr,
1ec9b909 1909 unsigned size)
6658ffb8 1910{
05068c0d 1911 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, BP_MEM_READ);
1ec9b909 1912 switch (size) {
2c17449b 1913 case 1: return ldub_phys(&address_space_memory, addr);
41701aa4 1914 case 2: return lduw_phys(&address_space_memory, addr);
fdfba1a2 1915 case 4: return ldl_phys(&address_space_memory, addr);
1ec9b909
AK
1916 default: abort();
1917 }
6658ffb8
PB
1918}
1919
a8170e5e 1920static void watch_mem_write(void *opaque, hwaddr addr,
1ec9b909 1921 uint64_t val, unsigned size)
6658ffb8 1922{
05068c0d 1923 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, BP_MEM_WRITE);
1ec9b909 1924 switch (size) {
67364150 1925 case 1:
db3be60d 1926 stb_phys(&address_space_memory, addr, val);
67364150
MF
1927 break;
1928 case 2:
5ce5944d 1929 stw_phys(&address_space_memory, addr, val);
67364150
MF
1930 break;
1931 case 4:
ab1da857 1932 stl_phys(&address_space_memory, addr, val);
67364150 1933 break;
1ec9b909
AK
1934 default: abort();
1935 }
6658ffb8
PB
1936}
1937
1ec9b909
AK
1938static const MemoryRegionOps watch_mem_ops = {
1939 .read = watch_mem_read,
1940 .write = watch_mem_write,
1941 .endianness = DEVICE_NATIVE_ENDIAN,
6658ffb8 1942};
6658ffb8 1943
f25a49e0
PM
1944static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data,
1945 unsigned len, MemTxAttrs attrs)
db7b5426 1946{
acc9d80b 1947 subpage_t *subpage = opaque;
ff6cff75 1948 uint8_t buf[8];
5c9eb028 1949 MemTxResult res;
791af8c8 1950
db7b5426 1951#if defined(DEBUG_SUBPAGE)
016e9d62 1952 printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__,
acc9d80b 1953 subpage, len, addr);
db7b5426 1954#endif
5c9eb028
PM
1955 res = address_space_read(subpage->as, addr + subpage->base,
1956 attrs, buf, len);
1957 if (res) {
1958 return res;
f25a49e0 1959 }
acc9d80b
JK
1960 switch (len) {
1961 case 1:
f25a49e0
PM
1962 *data = ldub_p(buf);
1963 return MEMTX_OK;
acc9d80b 1964 case 2:
f25a49e0
PM
1965 *data = lduw_p(buf);
1966 return MEMTX_OK;
acc9d80b 1967 case 4:
f25a49e0
PM
1968 *data = ldl_p(buf);
1969 return MEMTX_OK;
ff6cff75 1970 case 8:
f25a49e0
PM
1971 *data = ldq_p(buf);
1972 return MEMTX_OK;
acc9d80b
JK
1973 default:
1974 abort();
1975 }
db7b5426
BS
1976}
1977
f25a49e0
PM
1978static MemTxResult subpage_write(void *opaque, hwaddr addr,
1979 uint64_t value, unsigned len, MemTxAttrs attrs)
db7b5426 1980{
acc9d80b 1981 subpage_t *subpage = opaque;
ff6cff75 1982 uint8_t buf[8];
acc9d80b 1983
db7b5426 1984#if defined(DEBUG_SUBPAGE)
016e9d62 1985 printf("%s: subpage %p len %u addr " TARGET_FMT_plx
acc9d80b
JK
1986 " value %"PRIx64"\n",
1987 __func__, subpage, len, addr, value);
db7b5426 1988#endif
acc9d80b
JK
1989 switch (len) {
1990 case 1:
1991 stb_p(buf, value);
1992 break;
1993 case 2:
1994 stw_p(buf, value);
1995 break;
1996 case 4:
1997 stl_p(buf, value);
1998 break;
ff6cff75
PB
1999 case 8:
2000 stq_p(buf, value);
2001 break;
acc9d80b
JK
2002 default:
2003 abort();
2004 }
5c9eb028
PM
2005 return address_space_write(subpage->as, addr + subpage->base,
2006 attrs, buf, len);
db7b5426
BS
2007}
2008
c353e4cc 2009static bool subpage_accepts(void *opaque, hwaddr addr,
016e9d62 2010 unsigned len, bool is_write)
c353e4cc 2011{
acc9d80b 2012 subpage_t *subpage = opaque;
c353e4cc 2013#if defined(DEBUG_SUBPAGE)
016e9d62 2014 printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n",
acc9d80b 2015 __func__, subpage, is_write ? 'w' : 'r', len, addr);
c353e4cc
PB
2016#endif
2017
acc9d80b 2018 return address_space_access_valid(subpage->as, addr + subpage->base,
016e9d62 2019 len, is_write);
c353e4cc
PB
2020}
2021
70c68e44 2022static const MemoryRegionOps subpage_ops = {
f25a49e0
PM
2023 .read_with_attrs = subpage_read,
2024 .write_with_attrs = subpage_write,
ff6cff75
PB
2025 .impl.min_access_size = 1,
2026 .impl.max_access_size = 8,
2027 .valid.min_access_size = 1,
2028 .valid.max_access_size = 8,
c353e4cc 2029 .valid.accepts = subpage_accepts,
70c68e44 2030 .endianness = DEVICE_NATIVE_ENDIAN,
db7b5426
BS
2031};
2032
c227f099 2033static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
5312bd8b 2034 uint16_t section)
db7b5426
BS
2035{
2036 int idx, eidx;
2037
2038 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
2039 return -1;
2040 idx = SUBPAGE_IDX(start);
2041 eidx = SUBPAGE_IDX(end);
2042#if defined(DEBUG_SUBPAGE)
016e9d62
AK
2043 printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n",
2044 __func__, mmio, start, end, idx, eidx, section);
db7b5426 2045#endif
db7b5426 2046 for (; idx <= eidx; idx++) {
5312bd8b 2047 mmio->sub_section[idx] = section;
db7b5426
BS
2048 }
2049
2050 return 0;
2051}
2052
acc9d80b 2053static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
db7b5426 2054{
c227f099 2055 subpage_t *mmio;
db7b5426 2056
7267c094 2057 mmio = g_malloc0(sizeof(subpage_t));
1eec614b 2058
acc9d80b 2059 mmio->as = as;
1eec614b 2060 mmio->base = base;
2c9b15ca 2061 memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
b4fefef9 2062 NULL, TARGET_PAGE_SIZE);
b3b00c78 2063 mmio->iomem.subpage = true;
db7b5426 2064#if defined(DEBUG_SUBPAGE)
016e9d62
AK
2065 printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__,
2066 mmio, base, TARGET_PAGE_SIZE);
db7b5426 2067#endif
b41aac4f 2068 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED);
db7b5426
BS
2069
2070 return mmio;
2071}
2072
a656e22f
PC
2073static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
2074 MemoryRegion *mr)
5312bd8b 2075{
a656e22f 2076 assert(as);
5312bd8b 2077 MemoryRegionSection section = {
a656e22f 2078 .address_space = as,
5312bd8b
AK
2079 .mr = mr,
2080 .offset_within_address_space = 0,
2081 .offset_within_region = 0,
052e87b0 2082 .size = int128_2_64(),
5312bd8b
AK
2083 };
2084
53cb28cb 2085 return phys_section_add(map, &section);
5312bd8b
AK
2086}
2087
9d82b5a7 2088MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index)
aa102231 2089{
79e2b9ae
PB
2090 AddressSpaceDispatch *d = atomic_rcu_read(&cpu->memory_dispatch);
2091 MemoryRegionSection *sections = d->map.sections;
9d82b5a7
PB
2092
2093 return sections[index & ~TARGET_PAGE_MASK].mr;
aa102231
AK
2094}
2095
e9179ce1
AK
2096static void io_mem_init(void)
2097{
1f6245e5 2098 memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, NULL, UINT64_MAX);
2c9b15ca 2099 memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
1f6245e5 2100 NULL, UINT64_MAX);
2c9b15ca 2101 memory_region_init_io(&io_mem_notdirty, NULL, &notdirty_mem_ops, NULL,
1f6245e5 2102 NULL, UINT64_MAX);
2c9b15ca 2103 memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL,
1f6245e5 2104 NULL, UINT64_MAX);
e9179ce1
AK
2105}
2106
ac1970fb 2107static void mem_begin(MemoryListener *listener)
00752703
PB
2108{
2109 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
53cb28cb
MA
2110 AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1);
2111 uint16_t n;
2112
a656e22f 2113 n = dummy_section(&d->map, as, &io_mem_unassigned);
53cb28cb 2114 assert(n == PHYS_SECTION_UNASSIGNED);
a656e22f 2115 n = dummy_section(&d->map, as, &io_mem_notdirty);
53cb28cb 2116 assert(n == PHYS_SECTION_NOTDIRTY);
a656e22f 2117 n = dummy_section(&d->map, as, &io_mem_rom);
53cb28cb 2118 assert(n == PHYS_SECTION_ROM);
a656e22f 2119 n = dummy_section(&d->map, as, &io_mem_watch);
53cb28cb 2120 assert(n == PHYS_SECTION_WATCH);
00752703 2121
9736e55b 2122 d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
00752703
PB
2123 d->as = as;
2124 as->next_dispatch = d;
2125}
2126
79e2b9ae
PB
2127static void address_space_dispatch_free(AddressSpaceDispatch *d)
2128{
2129 phys_sections_free(&d->map);
2130 g_free(d);
2131}
2132
00752703 2133static void mem_commit(MemoryListener *listener)
ac1970fb 2134{
89ae337a 2135 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
0475d94f
PB
2136 AddressSpaceDispatch *cur = as->dispatch;
2137 AddressSpaceDispatch *next = as->next_dispatch;
2138
53cb28cb 2139 phys_page_compact_all(next, next->map.nodes_nb);
b35ba30f 2140
79e2b9ae 2141 atomic_rcu_set(&as->dispatch, next);
53cb28cb 2142 if (cur) {
79e2b9ae 2143 call_rcu(cur, address_space_dispatch_free, rcu);
53cb28cb 2144 }
9affd6fc
PB
2145}
2146
1d71148e 2147static void tcg_commit(MemoryListener *listener)
50c1e149 2148{
182735ef 2149 CPUState *cpu;
117712c3
AK
2150
2151 /* since each CPU stores ram addresses in its TLB cache, we must
2152 reset the modified entries */
2153 /* XXX: slow ! */
bdc44640 2154 CPU_FOREACH(cpu) {
33bde2e1
EI
2155 /* FIXME: Disentangle the cpu.h circular files deps so we can
2156 directly get the right CPU from listener. */
2157 if (cpu->tcg_as_listener != listener) {
2158 continue;
2159 }
76e5c76f 2160 cpu_reload_memory_map(cpu);
117712c3 2161 }
50c1e149
AK
2162}
2163
93632747
AK
2164static void core_log_global_start(MemoryListener *listener)
2165{
981fdf23 2166 cpu_physical_memory_set_dirty_tracking(true);
93632747
AK
2167}
2168
2169static void core_log_global_stop(MemoryListener *listener)
2170{
981fdf23 2171 cpu_physical_memory_set_dirty_tracking(false);
93632747
AK
2172}
2173
93632747 2174static MemoryListener core_memory_listener = {
93632747
AK
2175 .log_global_start = core_log_global_start,
2176 .log_global_stop = core_log_global_stop,
ac1970fb 2177 .priority = 1,
93632747
AK
2178};
2179
ac1970fb
AK
2180void address_space_init_dispatch(AddressSpace *as)
2181{
00752703 2182 as->dispatch = NULL;
89ae337a 2183 as->dispatch_listener = (MemoryListener) {
ac1970fb 2184 .begin = mem_begin,
00752703 2185 .commit = mem_commit,
ac1970fb
AK
2186 .region_add = mem_add,
2187 .region_nop = mem_add,
2188 .priority = 0,
2189 };
89ae337a 2190 memory_listener_register(&as->dispatch_listener, as);
ac1970fb
AK
2191}
2192
6e48e8f9
PB
2193void address_space_unregister(AddressSpace *as)
2194{
2195 memory_listener_unregister(&as->dispatch_listener);
2196}
2197
83f3c251
AK
2198void address_space_destroy_dispatch(AddressSpace *as)
2199{
2200 AddressSpaceDispatch *d = as->dispatch;
2201
79e2b9ae
PB
2202 atomic_rcu_set(&as->dispatch, NULL);
2203 if (d) {
2204 call_rcu(d, address_space_dispatch_free, rcu);
2205 }
83f3c251
AK
2206}
2207
62152b8a
AK
2208static void memory_map_init(void)
2209{
7267c094 2210 system_memory = g_malloc(sizeof(*system_memory));
03f49957 2211
57271d63 2212 memory_region_init(system_memory, NULL, "system", UINT64_MAX);
7dca8043 2213 address_space_init(&address_space_memory, system_memory, "memory");
309cb471 2214
7267c094 2215 system_io = g_malloc(sizeof(*system_io));
3bb28b72
JK
2216 memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io",
2217 65536);
7dca8043 2218 address_space_init(&address_space_io, system_io, "I/O");
93632747 2219
f6790af6 2220 memory_listener_register(&core_memory_listener, &address_space_memory);
62152b8a
AK
2221}
2222
2223MemoryRegion *get_system_memory(void)
2224{
2225 return system_memory;
2226}
2227
309cb471
AK
2228MemoryRegion *get_system_io(void)
2229{
2230 return system_io;
2231}
2232
e2eef170
PB
2233#endif /* !defined(CONFIG_USER_ONLY) */
2234
13eb76e0
FB
2235/* physical memory access (slow version, mainly for debug) */
2236#if defined(CONFIG_USER_ONLY)
f17ec444 2237int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
a68fe89c 2238 uint8_t *buf, int len, int is_write)
13eb76e0
FB
2239{
2240 int l, flags;
2241 target_ulong page;
53a5960a 2242 void * p;
13eb76e0
FB
2243
2244 while (len > 0) {
2245 page = addr & TARGET_PAGE_MASK;
2246 l = (page + TARGET_PAGE_SIZE) - addr;
2247 if (l > len)
2248 l = len;
2249 flags = page_get_flags(page);
2250 if (!(flags & PAGE_VALID))
a68fe89c 2251 return -1;
13eb76e0
FB
2252 if (is_write) {
2253 if (!(flags & PAGE_WRITE))
a68fe89c 2254 return -1;
579a97f7 2255 /* XXX: this code should not depend on lock_user */
72fb7daa 2256 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
a68fe89c 2257 return -1;
72fb7daa
AJ
2258 memcpy(p, buf, l);
2259 unlock_user(p, addr, l);
13eb76e0
FB
2260 } else {
2261 if (!(flags & PAGE_READ))
a68fe89c 2262 return -1;
579a97f7 2263 /* XXX: this code should not depend on lock_user */
72fb7daa 2264 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
a68fe89c 2265 return -1;
72fb7daa 2266 memcpy(buf, p, l);
5b257578 2267 unlock_user(p, addr, 0);
13eb76e0
FB
2268 }
2269 len -= l;
2270 buf += l;
2271 addr += l;
2272 }
a68fe89c 2273 return 0;
13eb76e0 2274}
8df1cd07 2275
13eb76e0 2276#else
51d7a9eb 2277
a8170e5e
AK
2278static void invalidate_and_set_dirty(hwaddr addr,
2279 hwaddr length)
51d7a9eb 2280{
f874bf90
PM
2281 if (cpu_physical_memory_range_includes_clean(addr, length)) {
2282 tb_invalidate_phys_range(addr, addr + length, 0);
6886867e 2283 cpu_physical_memory_set_dirty_range_nocode(addr, length);
51d7a9eb 2284 }
e226939d 2285 xen_modified_memory(addr, length);
51d7a9eb
AP
2286}
2287
23326164 2288static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr)
82f2563f 2289{
e1622f4b 2290 unsigned access_size_max = mr->ops->valid.max_access_size;
23326164
RH
2291
2292 /* Regions are assumed to support 1-4 byte accesses unless
2293 otherwise specified. */
23326164
RH
2294 if (access_size_max == 0) {
2295 access_size_max = 4;
2296 }
2297
2298 /* Bound the maximum access by the alignment of the address. */
2299 if (!mr->ops->impl.unaligned) {
2300 unsigned align_size_max = addr & -addr;
2301 if (align_size_max != 0 && align_size_max < access_size_max) {
2302 access_size_max = align_size_max;
2303 }
82f2563f 2304 }
23326164
RH
2305
2306 /* Don't attempt accesses larger than the maximum. */
2307 if (l > access_size_max) {
2308 l = access_size_max;
82f2563f 2309 }
098178f2
PB
2310 if (l & (l - 1)) {
2311 l = 1 << (qemu_fls(l) - 1);
2312 }
23326164
RH
2313
2314 return l;
82f2563f
PB
2315}
2316
5c9eb028
PM
2317MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
2318 uint8_t *buf, int len, bool is_write)
13eb76e0 2319{
149f54b5 2320 hwaddr l;
13eb76e0 2321 uint8_t *ptr;
791af8c8 2322 uint64_t val;
149f54b5 2323 hwaddr addr1;
5c8a00ce 2324 MemoryRegion *mr;
3b643495 2325 MemTxResult result = MEMTX_OK;
3b46e624 2326
13eb76e0 2327 while (len > 0) {
149f54b5 2328 l = len;
5c8a00ce 2329 mr = address_space_translate(as, addr, &addr1, &l, is_write);
3b46e624 2330
13eb76e0 2331 if (is_write) {
5c8a00ce
PB
2332 if (!memory_access_is_direct(mr, is_write)) {
2333 l = memory_access_size(mr, l, addr1);
4917cf44 2334 /* XXX: could force current_cpu to NULL to avoid
6a00d601 2335 potential bugs */
23326164
RH
2336 switch (l) {
2337 case 8:
2338 /* 64 bit write access */
2339 val = ldq_p(buf);
3b643495
PM
2340 result |= memory_region_dispatch_write(mr, addr1, val, 8,
2341 attrs);
23326164
RH
2342 break;
2343 case 4:
1c213d19 2344 /* 32 bit write access */
c27004ec 2345 val = ldl_p(buf);
3b643495
PM
2346 result |= memory_region_dispatch_write(mr, addr1, val, 4,
2347 attrs);
23326164
RH
2348 break;
2349 case 2:
1c213d19 2350 /* 16 bit write access */
c27004ec 2351 val = lduw_p(buf);
3b643495
PM
2352 result |= memory_region_dispatch_write(mr, addr1, val, 2,
2353 attrs);
23326164
RH
2354 break;
2355 case 1:
1c213d19 2356 /* 8 bit write access */
c27004ec 2357 val = ldub_p(buf);
3b643495
PM
2358 result |= memory_region_dispatch_write(mr, addr1, val, 1,
2359 attrs);
23326164
RH
2360 break;
2361 default:
2362 abort();
13eb76e0 2363 }
2bbfa05d 2364 } else {
5c8a00ce 2365 addr1 += memory_region_get_ram_addr(mr);
13eb76e0 2366 /* RAM case */
5579c7f3 2367 ptr = qemu_get_ram_ptr(addr1);
13eb76e0 2368 memcpy(ptr, buf, l);
51d7a9eb 2369 invalidate_and_set_dirty(addr1, l);
13eb76e0
FB
2370 }
2371 } else {
5c8a00ce 2372 if (!memory_access_is_direct(mr, is_write)) {
13eb76e0 2373 /* I/O case */
5c8a00ce 2374 l = memory_access_size(mr, l, addr1);
23326164
RH
2375 switch (l) {
2376 case 8:
2377 /* 64 bit read access */
3b643495
PM
2378 result |= memory_region_dispatch_read(mr, addr1, &val, 8,
2379 attrs);
23326164
RH
2380 stq_p(buf, val);
2381 break;
2382 case 4:
13eb76e0 2383 /* 32 bit read access */
3b643495
PM
2384 result |= memory_region_dispatch_read(mr, addr1, &val, 4,
2385 attrs);
c27004ec 2386 stl_p(buf, val);
23326164
RH
2387 break;
2388 case 2:
13eb76e0 2389 /* 16 bit read access */
3b643495
PM
2390 result |= memory_region_dispatch_read(mr, addr1, &val, 2,
2391 attrs);
c27004ec 2392 stw_p(buf, val);
23326164
RH
2393 break;
2394 case 1:
1c213d19 2395 /* 8 bit read access */
3b643495
PM
2396 result |= memory_region_dispatch_read(mr, addr1, &val, 1,
2397 attrs);
c27004ec 2398 stb_p(buf, val);
23326164
RH
2399 break;
2400 default:
2401 abort();
13eb76e0
FB
2402 }
2403 } else {
2404 /* RAM case */
5c8a00ce 2405 ptr = qemu_get_ram_ptr(mr->ram_addr + addr1);
f3705d53 2406 memcpy(buf, ptr, l);
13eb76e0
FB
2407 }
2408 }
2409 len -= l;
2410 buf += l;
2411 addr += l;
2412 }
fd8aaa76 2413
3b643495 2414 return result;
13eb76e0 2415}
8df1cd07 2416
5c9eb028
PM
2417MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
2418 const uint8_t *buf, int len)
ac1970fb 2419{
5c9eb028 2420 return address_space_rw(as, addr, attrs, (uint8_t *)buf, len, true);
ac1970fb
AK
2421}
2422
5c9eb028
PM
2423MemTxResult address_space_read(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
2424 uint8_t *buf, int len)
ac1970fb 2425{
5c9eb028 2426 return address_space_rw(as, addr, attrs, buf, len, false);
ac1970fb
AK
2427}
2428
2429
a8170e5e 2430void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
ac1970fb
AK
2431 int len, int is_write)
2432{
5c9eb028
PM
2433 address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED,
2434 buf, len, is_write);
ac1970fb
AK
2435}
2436
582b55a9
AG
2437enum write_rom_type {
2438 WRITE_DATA,
2439 FLUSH_CACHE,
2440};
2441
2a221651 2442static inline void cpu_physical_memory_write_rom_internal(AddressSpace *as,
582b55a9 2443 hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type)
d0ecd2aa 2444{
149f54b5 2445 hwaddr l;
d0ecd2aa 2446 uint8_t *ptr;
149f54b5 2447 hwaddr addr1;
5c8a00ce 2448 MemoryRegion *mr;
3b46e624 2449
d0ecd2aa 2450 while (len > 0) {
149f54b5 2451 l = len;
2a221651 2452 mr = address_space_translate(as, addr, &addr1, &l, true);
3b46e624 2453
5c8a00ce
PB
2454 if (!(memory_region_is_ram(mr) ||
2455 memory_region_is_romd(mr))) {
d0ecd2aa
FB
2456 /* do nothing */
2457 } else {
5c8a00ce 2458 addr1 += memory_region_get_ram_addr(mr);
d0ecd2aa 2459 /* ROM/RAM case */
5579c7f3 2460 ptr = qemu_get_ram_ptr(addr1);
582b55a9
AG
2461 switch (type) {
2462 case WRITE_DATA:
2463 memcpy(ptr, buf, l);
2464 invalidate_and_set_dirty(addr1, l);
2465 break;
2466 case FLUSH_CACHE:
2467 flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l);
2468 break;
2469 }
d0ecd2aa
FB
2470 }
2471 len -= l;
2472 buf += l;
2473 addr += l;
2474 }
2475}
2476
582b55a9 2477/* used for ROM loading : can write in RAM and ROM */
2a221651 2478void cpu_physical_memory_write_rom(AddressSpace *as, hwaddr addr,
582b55a9
AG
2479 const uint8_t *buf, int len)
2480{
2a221651 2481 cpu_physical_memory_write_rom_internal(as, addr, buf, len, WRITE_DATA);
582b55a9
AG
2482}
2483
2484void cpu_flush_icache_range(hwaddr start, int len)
2485{
2486 /*
2487 * This function should do the same thing as an icache flush that was
2488 * triggered from within the guest. For TCG we are always cache coherent,
2489 * so there is no need to flush anything. For KVM / Xen we need to flush
2490 * the host's instruction cache at least.
2491 */
2492 if (tcg_enabled()) {
2493 return;
2494 }
2495
2a221651
EI
2496 cpu_physical_memory_write_rom_internal(&address_space_memory,
2497 start, NULL, len, FLUSH_CACHE);
582b55a9
AG
2498}
2499
6d16c2f8 2500typedef struct {
d3e71559 2501 MemoryRegion *mr;
6d16c2f8 2502 void *buffer;
a8170e5e
AK
2503 hwaddr addr;
2504 hwaddr len;
6d16c2f8
AL
2505} BounceBuffer;
2506
2507static BounceBuffer bounce;
2508
ba223c29
AL
2509typedef struct MapClient {
2510 void *opaque;
2511 void (*callback)(void *opaque);
72cf2d4f 2512 QLIST_ENTRY(MapClient) link;
ba223c29
AL
2513} MapClient;
2514
72cf2d4f
BS
2515static QLIST_HEAD(map_client_list, MapClient) map_client_list
2516 = QLIST_HEAD_INITIALIZER(map_client_list);
ba223c29
AL
2517
2518void *cpu_register_map_client(void *opaque, void (*callback)(void *opaque))
2519{
7267c094 2520 MapClient *client = g_malloc(sizeof(*client));
ba223c29
AL
2521
2522 client->opaque = opaque;
2523 client->callback = callback;
72cf2d4f 2524 QLIST_INSERT_HEAD(&map_client_list, client, link);
ba223c29
AL
2525 return client;
2526}
2527
8b9c99d9 2528static void cpu_unregister_map_client(void *_client)
ba223c29
AL
2529{
2530 MapClient *client = (MapClient *)_client;
2531
72cf2d4f 2532 QLIST_REMOVE(client, link);
7267c094 2533 g_free(client);
ba223c29
AL
2534}
2535
2536static void cpu_notify_map_clients(void)
2537{
2538 MapClient *client;
2539
72cf2d4f
BS
2540 while (!QLIST_EMPTY(&map_client_list)) {
2541 client = QLIST_FIRST(&map_client_list);
ba223c29 2542 client->callback(client->opaque);
34d5e948 2543 cpu_unregister_map_client(client);
ba223c29
AL
2544 }
2545}
2546
51644ab7
PB
2547bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write)
2548{
5c8a00ce 2549 MemoryRegion *mr;
51644ab7
PB
2550 hwaddr l, xlat;
2551
2552 while (len > 0) {
2553 l = len;
5c8a00ce
PB
2554 mr = address_space_translate(as, addr, &xlat, &l, is_write);
2555 if (!memory_access_is_direct(mr, is_write)) {
2556 l = memory_access_size(mr, l, addr);
2557 if (!memory_region_access_valid(mr, xlat, l, is_write)) {
51644ab7
PB
2558 return false;
2559 }
2560 }
2561
2562 len -= l;
2563 addr += l;
2564 }
2565 return true;
2566}
2567
6d16c2f8
AL
2568/* Map a physical memory region into a host virtual address.
2569 * May map a subset of the requested range, given by and returned in *plen.
2570 * May return NULL if resources needed to perform the mapping are exhausted.
2571 * Use only for reads OR writes - not for read-modify-write operations.
ba223c29
AL
2572 * Use cpu_register_map_client() to know when retrying the map operation is
2573 * likely to succeed.
6d16c2f8 2574 */
ac1970fb 2575void *address_space_map(AddressSpace *as,
a8170e5e
AK
2576 hwaddr addr,
2577 hwaddr *plen,
ac1970fb 2578 bool is_write)
6d16c2f8 2579{
a8170e5e 2580 hwaddr len = *plen;
e3127ae0
PB
2581 hwaddr done = 0;
2582 hwaddr l, xlat, base;
2583 MemoryRegion *mr, *this_mr;
2584 ram_addr_t raddr;
6d16c2f8 2585
e3127ae0
PB
2586 if (len == 0) {
2587 return NULL;
2588 }
38bee5dc 2589
e3127ae0
PB
2590 l = len;
2591 mr = address_space_translate(as, addr, &xlat, &l, is_write);
2592 if (!memory_access_is_direct(mr, is_write)) {
2593 if (bounce.buffer) {
2594 return NULL;
6d16c2f8 2595 }
e85d9db5
KW
2596 /* Avoid unbounded allocations */
2597 l = MIN(l, TARGET_PAGE_SIZE);
2598 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l);
e3127ae0
PB
2599 bounce.addr = addr;
2600 bounce.len = l;
d3e71559
PB
2601
2602 memory_region_ref(mr);
2603 bounce.mr = mr;
e3127ae0 2604 if (!is_write) {
5c9eb028
PM
2605 address_space_read(as, addr, MEMTXATTRS_UNSPECIFIED,
2606 bounce.buffer, l);
8ab934f9 2607 }
6d16c2f8 2608
e3127ae0
PB
2609 *plen = l;
2610 return bounce.buffer;
2611 }
2612
2613 base = xlat;
2614 raddr = memory_region_get_ram_addr(mr);
2615
2616 for (;;) {
6d16c2f8
AL
2617 len -= l;
2618 addr += l;
e3127ae0
PB
2619 done += l;
2620 if (len == 0) {
2621 break;
2622 }
2623
2624 l = len;
2625 this_mr = address_space_translate(as, addr, &xlat, &l, is_write);
2626 if (this_mr != mr || xlat != base + done) {
2627 break;
2628 }
6d16c2f8 2629 }
e3127ae0 2630
d3e71559 2631 memory_region_ref(mr);
e3127ae0
PB
2632 *plen = done;
2633 return qemu_ram_ptr_length(raddr + base, plen);
6d16c2f8
AL
2634}
2635
ac1970fb 2636/* Unmaps a memory region previously mapped by address_space_map().
6d16c2f8
AL
2637 * Will also mark the memory as dirty if is_write == 1. access_len gives
2638 * the amount of memory that was actually read or written by the caller.
2639 */
a8170e5e
AK
2640void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2641 int is_write, hwaddr access_len)
6d16c2f8
AL
2642{
2643 if (buffer != bounce.buffer) {
d3e71559
PB
2644 MemoryRegion *mr;
2645 ram_addr_t addr1;
2646
2647 mr = qemu_ram_addr_from_host(buffer, &addr1);
2648 assert(mr != NULL);
6d16c2f8 2649 if (is_write) {
6886867e 2650 invalidate_and_set_dirty(addr1, access_len);
6d16c2f8 2651 }
868bb33f 2652 if (xen_enabled()) {
e41d7c69 2653 xen_invalidate_map_cache_entry(buffer);
050a0ddf 2654 }
d3e71559 2655 memory_region_unref(mr);
6d16c2f8
AL
2656 return;
2657 }
2658 if (is_write) {
5c9eb028
PM
2659 address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED,
2660 bounce.buffer, access_len);
6d16c2f8 2661 }
f8a83245 2662 qemu_vfree(bounce.buffer);
6d16c2f8 2663 bounce.buffer = NULL;
d3e71559 2664 memory_region_unref(bounce.mr);
ba223c29 2665 cpu_notify_map_clients();
6d16c2f8 2666}
d0ecd2aa 2667
a8170e5e
AK
2668void *cpu_physical_memory_map(hwaddr addr,
2669 hwaddr *plen,
ac1970fb
AK
2670 int is_write)
2671{
2672 return address_space_map(&address_space_memory, addr, plen, is_write);
2673}
2674
a8170e5e
AK
2675void cpu_physical_memory_unmap(void *buffer, hwaddr len,
2676 int is_write, hwaddr access_len)
ac1970fb
AK
2677{
2678 return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len);
2679}
2680
8df1cd07 2681/* warning: addr must be aligned */
50013115
PM
2682static inline uint32_t address_space_ldl_internal(AddressSpace *as, hwaddr addr,
2683 MemTxAttrs attrs,
2684 MemTxResult *result,
2685 enum device_endian endian)
8df1cd07 2686{
8df1cd07 2687 uint8_t *ptr;
791af8c8 2688 uint64_t val;
5c8a00ce 2689 MemoryRegion *mr;
149f54b5
PB
2690 hwaddr l = 4;
2691 hwaddr addr1;
50013115 2692 MemTxResult r;
8df1cd07 2693
fdfba1a2 2694 mr = address_space_translate(as, addr, &addr1, &l, false);
5c8a00ce 2695 if (l < 4 || !memory_access_is_direct(mr, false)) {
8df1cd07 2696 /* I/O case */
50013115 2697 r = memory_region_dispatch_read(mr, addr1, &val, 4, attrs);
1e78bcc1
AG
2698#if defined(TARGET_WORDS_BIGENDIAN)
2699 if (endian == DEVICE_LITTLE_ENDIAN) {
2700 val = bswap32(val);
2701 }
2702#else
2703 if (endian == DEVICE_BIG_ENDIAN) {
2704 val = bswap32(val);
2705 }
2706#endif
8df1cd07
FB
2707 } else {
2708 /* RAM case */
5c8a00ce 2709 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
06ef3525 2710 & TARGET_PAGE_MASK)
149f54b5 2711 + addr1);
1e78bcc1
AG
2712 switch (endian) {
2713 case DEVICE_LITTLE_ENDIAN:
2714 val = ldl_le_p(ptr);
2715 break;
2716 case DEVICE_BIG_ENDIAN:
2717 val = ldl_be_p(ptr);
2718 break;
2719 default:
2720 val = ldl_p(ptr);
2721 break;
2722 }
50013115
PM
2723 r = MEMTX_OK;
2724 }
2725 if (result) {
2726 *result = r;
8df1cd07
FB
2727 }
2728 return val;
2729}
2730
50013115
PM
2731uint32_t address_space_ldl(AddressSpace *as, hwaddr addr,
2732 MemTxAttrs attrs, MemTxResult *result)
2733{
2734 return address_space_ldl_internal(as, addr, attrs, result,
2735 DEVICE_NATIVE_ENDIAN);
2736}
2737
2738uint32_t address_space_ldl_le(AddressSpace *as, hwaddr addr,
2739 MemTxAttrs attrs, MemTxResult *result)
2740{
2741 return address_space_ldl_internal(as, addr, attrs, result,
2742 DEVICE_LITTLE_ENDIAN);
2743}
2744
2745uint32_t address_space_ldl_be(AddressSpace *as, hwaddr addr,
2746 MemTxAttrs attrs, MemTxResult *result)
2747{
2748 return address_space_ldl_internal(as, addr, attrs, result,
2749 DEVICE_BIG_ENDIAN);
2750}
2751
fdfba1a2 2752uint32_t ldl_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2753{
50013115 2754 return address_space_ldl(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2755}
2756
fdfba1a2 2757uint32_t ldl_le_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2758{
50013115 2759 return address_space_ldl_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2760}
2761
fdfba1a2 2762uint32_t ldl_be_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2763{
50013115 2764 return address_space_ldl_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2765}
2766
84b7b8e7 2767/* warning: addr must be aligned */
50013115
PM
2768static inline uint64_t address_space_ldq_internal(AddressSpace *as, hwaddr addr,
2769 MemTxAttrs attrs,
2770 MemTxResult *result,
2771 enum device_endian endian)
84b7b8e7 2772{
84b7b8e7
FB
2773 uint8_t *ptr;
2774 uint64_t val;
5c8a00ce 2775 MemoryRegion *mr;
149f54b5
PB
2776 hwaddr l = 8;
2777 hwaddr addr1;
50013115 2778 MemTxResult r;
84b7b8e7 2779
2c17449b 2780 mr = address_space_translate(as, addr, &addr1, &l,
5c8a00ce
PB
2781 false);
2782 if (l < 8 || !memory_access_is_direct(mr, false)) {
84b7b8e7 2783 /* I/O case */
50013115 2784 r = memory_region_dispatch_read(mr, addr1, &val, 8, attrs);
968a5627
PB
2785#if defined(TARGET_WORDS_BIGENDIAN)
2786 if (endian == DEVICE_LITTLE_ENDIAN) {
2787 val = bswap64(val);
2788 }
2789#else
2790 if (endian == DEVICE_BIG_ENDIAN) {
2791 val = bswap64(val);
2792 }
84b7b8e7
FB
2793#endif
2794 } else {
2795 /* RAM case */
5c8a00ce 2796 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
06ef3525 2797 & TARGET_PAGE_MASK)
149f54b5 2798 + addr1);
1e78bcc1
AG
2799 switch (endian) {
2800 case DEVICE_LITTLE_ENDIAN:
2801 val = ldq_le_p(ptr);
2802 break;
2803 case DEVICE_BIG_ENDIAN:
2804 val = ldq_be_p(ptr);
2805 break;
2806 default:
2807 val = ldq_p(ptr);
2808 break;
2809 }
50013115
PM
2810 r = MEMTX_OK;
2811 }
2812 if (result) {
2813 *result = r;
84b7b8e7
FB
2814 }
2815 return val;
2816}
2817
50013115
PM
2818uint64_t address_space_ldq(AddressSpace *as, hwaddr addr,
2819 MemTxAttrs attrs, MemTxResult *result)
2820{
2821 return address_space_ldq_internal(as, addr, attrs, result,
2822 DEVICE_NATIVE_ENDIAN);
2823}
2824
2825uint64_t address_space_ldq_le(AddressSpace *as, hwaddr addr,
2826 MemTxAttrs attrs, MemTxResult *result)
2827{
2828 return address_space_ldq_internal(as, addr, attrs, result,
2829 DEVICE_LITTLE_ENDIAN);
2830}
2831
2832uint64_t address_space_ldq_be(AddressSpace *as, hwaddr addr,
2833 MemTxAttrs attrs, MemTxResult *result)
2834{
2835 return address_space_ldq_internal(as, addr, attrs, result,
2836 DEVICE_BIG_ENDIAN);
2837}
2838
2c17449b 2839uint64_t ldq_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2840{
50013115 2841 return address_space_ldq(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2842}
2843
2c17449b 2844uint64_t ldq_le_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2845{
50013115 2846 return address_space_ldq_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2847}
2848
2c17449b 2849uint64_t ldq_be_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2850{
50013115 2851 return address_space_ldq_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2852}
2853
aab33094 2854/* XXX: optimize */
50013115
PM
2855uint32_t address_space_ldub(AddressSpace *as, hwaddr addr,
2856 MemTxAttrs attrs, MemTxResult *result)
aab33094
FB
2857{
2858 uint8_t val;
50013115
PM
2859 MemTxResult r;
2860
2861 r = address_space_rw(as, addr, attrs, &val, 1, 0);
2862 if (result) {
2863 *result = r;
2864 }
aab33094
FB
2865 return val;
2866}
2867
50013115
PM
2868uint32_t ldub_phys(AddressSpace *as, hwaddr addr)
2869{
2870 return address_space_ldub(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
2871}
2872
733f0b02 2873/* warning: addr must be aligned */
50013115
PM
2874static inline uint32_t address_space_lduw_internal(AddressSpace *as,
2875 hwaddr addr,
2876 MemTxAttrs attrs,
2877 MemTxResult *result,
2878 enum device_endian endian)
aab33094 2879{
733f0b02
MT
2880 uint8_t *ptr;
2881 uint64_t val;
5c8a00ce 2882 MemoryRegion *mr;
149f54b5
PB
2883 hwaddr l = 2;
2884 hwaddr addr1;
50013115 2885 MemTxResult r;
733f0b02 2886
41701aa4 2887 mr = address_space_translate(as, addr, &addr1, &l,
5c8a00ce
PB
2888 false);
2889 if (l < 2 || !memory_access_is_direct(mr, false)) {
733f0b02 2890 /* I/O case */
50013115 2891 r = memory_region_dispatch_read(mr, addr1, &val, 2, attrs);
1e78bcc1
AG
2892#if defined(TARGET_WORDS_BIGENDIAN)
2893 if (endian == DEVICE_LITTLE_ENDIAN) {
2894 val = bswap16(val);
2895 }
2896#else
2897 if (endian == DEVICE_BIG_ENDIAN) {
2898 val = bswap16(val);
2899 }
2900#endif
733f0b02
MT
2901 } else {
2902 /* RAM case */
5c8a00ce 2903 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
06ef3525 2904 & TARGET_PAGE_MASK)
149f54b5 2905 + addr1);
1e78bcc1
AG
2906 switch (endian) {
2907 case DEVICE_LITTLE_ENDIAN:
2908 val = lduw_le_p(ptr);
2909 break;
2910 case DEVICE_BIG_ENDIAN:
2911 val = lduw_be_p(ptr);
2912 break;
2913 default:
2914 val = lduw_p(ptr);
2915 break;
2916 }
50013115
PM
2917 r = MEMTX_OK;
2918 }
2919 if (result) {
2920 *result = r;
733f0b02
MT
2921 }
2922 return val;
aab33094
FB
2923}
2924
50013115
PM
2925uint32_t address_space_lduw(AddressSpace *as, hwaddr addr,
2926 MemTxAttrs attrs, MemTxResult *result)
2927{
2928 return address_space_lduw_internal(as, addr, attrs, result,
2929 DEVICE_NATIVE_ENDIAN);
2930}
2931
2932uint32_t address_space_lduw_le(AddressSpace *as, hwaddr addr,
2933 MemTxAttrs attrs, MemTxResult *result)
2934{
2935 return address_space_lduw_internal(as, addr, attrs, result,
2936 DEVICE_LITTLE_ENDIAN);
2937}
2938
2939uint32_t address_space_lduw_be(AddressSpace *as, hwaddr addr,
2940 MemTxAttrs attrs, MemTxResult *result)
2941{
2942 return address_space_lduw_internal(as, addr, attrs, result,
2943 DEVICE_BIG_ENDIAN);
2944}
2945
41701aa4 2946uint32_t lduw_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2947{
50013115 2948 return address_space_lduw(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2949}
2950
41701aa4 2951uint32_t lduw_le_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2952{
50013115 2953 return address_space_lduw_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2954}
2955
41701aa4 2956uint32_t lduw_be_phys(AddressSpace *as, hwaddr addr)
1e78bcc1 2957{
50013115 2958 return address_space_lduw_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
2959}
2960
8df1cd07
FB
2961/* warning: addr must be aligned. The ram page is not masked as dirty
2962 and the code inside is not invalidated. It is useful if the dirty
2963 bits are used to track modified PTEs */
50013115
PM
2964void address_space_stl_notdirty(AddressSpace *as, hwaddr addr, uint32_t val,
2965 MemTxAttrs attrs, MemTxResult *result)
8df1cd07 2966{
8df1cd07 2967 uint8_t *ptr;
5c8a00ce 2968 MemoryRegion *mr;
149f54b5
PB
2969 hwaddr l = 4;
2970 hwaddr addr1;
50013115 2971 MemTxResult r;
8df1cd07 2972
2198a121 2973 mr = address_space_translate(as, addr, &addr1, &l,
5c8a00ce
PB
2974 true);
2975 if (l < 4 || !memory_access_is_direct(mr, true)) {
50013115 2976 r = memory_region_dispatch_write(mr, addr1, val, 4, attrs);
8df1cd07 2977 } else {
5c8a00ce 2978 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
5579c7f3 2979 ptr = qemu_get_ram_ptr(addr1);
8df1cd07 2980 stl_p(ptr, val);
74576198
AL
2981
2982 if (unlikely(in_migration)) {
a2cd8c85 2983 if (cpu_physical_memory_is_clean(addr1)) {
74576198
AL
2984 /* invalidate code */
2985 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
2986 /* set dirty bit */
6886867e 2987 cpu_physical_memory_set_dirty_range_nocode(addr1, 4);
74576198
AL
2988 }
2989 }
50013115
PM
2990 r = MEMTX_OK;
2991 }
2992 if (result) {
2993 *result = r;
8df1cd07
FB
2994 }
2995}
2996
50013115
PM
2997void stl_phys_notdirty(AddressSpace *as, hwaddr addr, uint32_t val)
2998{
2999 address_space_stl_notdirty(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
3000}
3001
8df1cd07 3002/* warning: addr must be aligned */
50013115
PM
3003static inline void address_space_stl_internal(AddressSpace *as,
3004 hwaddr addr, uint32_t val,
3005 MemTxAttrs attrs,
3006 MemTxResult *result,
3007 enum device_endian endian)
8df1cd07 3008{
8df1cd07 3009 uint8_t *ptr;
5c8a00ce 3010 MemoryRegion *mr;
149f54b5
PB
3011 hwaddr l = 4;
3012 hwaddr addr1;
50013115 3013 MemTxResult r;
8df1cd07 3014
ab1da857 3015 mr = address_space_translate(as, addr, &addr1, &l,
5c8a00ce
PB
3016 true);
3017 if (l < 4 || !memory_access_is_direct(mr, true)) {
1e78bcc1
AG
3018#if defined(TARGET_WORDS_BIGENDIAN)
3019 if (endian == DEVICE_LITTLE_ENDIAN) {
3020 val = bswap32(val);
3021 }
3022#else
3023 if (endian == DEVICE_BIG_ENDIAN) {
3024 val = bswap32(val);
3025 }
3026#endif
50013115 3027 r = memory_region_dispatch_write(mr, addr1, val, 4, attrs);
8df1cd07 3028 } else {
8df1cd07 3029 /* RAM case */
5c8a00ce 3030 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
5579c7f3 3031 ptr = qemu_get_ram_ptr(addr1);
1e78bcc1
AG
3032 switch (endian) {
3033 case DEVICE_LITTLE_ENDIAN:
3034 stl_le_p(ptr, val);
3035 break;
3036 case DEVICE_BIG_ENDIAN:
3037 stl_be_p(ptr, val);
3038 break;
3039 default:
3040 stl_p(ptr, val);
3041 break;
3042 }
51d7a9eb 3043 invalidate_and_set_dirty(addr1, 4);
50013115
PM
3044 r = MEMTX_OK;
3045 }
3046 if (result) {
3047 *result = r;
8df1cd07
FB
3048 }
3049}
3050
50013115
PM
3051void address_space_stl(AddressSpace *as, hwaddr addr, uint32_t val,
3052 MemTxAttrs attrs, MemTxResult *result)
3053{
3054 address_space_stl_internal(as, addr, val, attrs, result,
3055 DEVICE_NATIVE_ENDIAN);
3056}
3057
3058void address_space_stl_le(AddressSpace *as, hwaddr addr, uint32_t val,
3059 MemTxAttrs attrs, MemTxResult *result)
3060{
3061 address_space_stl_internal(as, addr, val, attrs, result,
3062 DEVICE_LITTLE_ENDIAN);
3063}
3064
3065void address_space_stl_be(AddressSpace *as, hwaddr addr, uint32_t val,
3066 MemTxAttrs attrs, MemTxResult *result)
3067{
3068 address_space_stl_internal(as, addr, val, attrs, result,
3069 DEVICE_BIG_ENDIAN);
3070}
3071
ab1da857 3072void stl_phys(AddressSpace *as, hwaddr addr, uint32_t val)
1e78bcc1 3073{
50013115 3074 address_space_stl(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3075}
3076
ab1da857 3077void stl_le_phys(AddressSpace *as, hwaddr addr, uint32_t val)
1e78bcc1 3078{
50013115 3079 address_space_stl_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3080}
3081
ab1da857 3082void stl_be_phys(AddressSpace *as, hwaddr addr, uint32_t val)
1e78bcc1 3083{
50013115 3084 address_space_stl_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3085}
3086
aab33094 3087/* XXX: optimize */
50013115
PM
3088void address_space_stb(AddressSpace *as, hwaddr addr, uint32_t val,
3089 MemTxAttrs attrs, MemTxResult *result)
aab33094
FB
3090{
3091 uint8_t v = val;
50013115
PM
3092 MemTxResult r;
3093
3094 r = address_space_rw(as, addr, attrs, &v, 1, 1);
3095 if (result) {
3096 *result = r;
3097 }
3098}
3099
3100void stb_phys(AddressSpace *as, hwaddr addr, uint32_t val)
3101{
3102 address_space_stb(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
aab33094
FB
3103}
3104
733f0b02 3105/* warning: addr must be aligned */
50013115
PM
3106static inline void address_space_stw_internal(AddressSpace *as,
3107 hwaddr addr, uint32_t val,
3108 MemTxAttrs attrs,
3109 MemTxResult *result,
3110 enum device_endian endian)
aab33094 3111{
733f0b02 3112 uint8_t *ptr;
5c8a00ce 3113 MemoryRegion *mr;
149f54b5
PB
3114 hwaddr l = 2;
3115 hwaddr addr1;
50013115 3116 MemTxResult r;
733f0b02 3117
5ce5944d 3118 mr = address_space_translate(as, addr, &addr1, &l, true);
5c8a00ce 3119 if (l < 2 || !memory_access_is_direct(mr, true)) {
1e78bcc1
AG
3120#if defined(TARGET_WORDS_BIGENDIAN)
3121 if (endian == DEVICE_LITTLE_ENDIAN) {
3122 val = bswap16(val);
3123 }
3124#else
3125 if (endian == DEVICE_BIG_ENDIAN) {
3126 val = bswap16(val);
3127 }
3128#endif
50013115 3129 r = memory_region_dispatch_write(mr, addr1, val, 2, attrs);
733f0b02 3130 } else {
733f0b02 3131 /* RAM case */
5c8a00ce 3132 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
733f0b02 3133 ptr = qemu_get_ram_ptr(addr1);
1e78bcc1
AG
3134 switch (endian) {
3135 case DEVICE_LITTLE_ENDIAN:
3136 stw_le_p(ptr, val);
3137 break;
3138 case DEVICE_BIG_ENDIAN:
3139 stw_be_p(ptr, val);
3140 break;
3141 default:
3142 stw_p(ptr, val);
3143 break;
3144 }
51d7a9eb 3145 invalidate_and_set_dirty(addr1, 2);
50013115
PM
3146 r = MEMTX_OK;
3147 }
3148 if (result) {
3149 *result = r;
733f0b02 3150 }
aab33094
FB
3151}
3152
50013115
PM
3153void address_space_stw(AddressSpace *as, hwaddr addr, uint32_t val,
3154 MemTxAttrs attrs, MemTxResult *result)
3155{
3156 address_space_stw_internal(as, addr, val, attrs, result,
3157 DEVICE_NATIVE_ENDIAN);
3158}
3159
3160void address_space_stw_le(AddressSpace *as, hwaddr addr, uint32_t val,
3161 MemTxAttrs attrs, MemTxResult *result)
3162{
3163 address_space_stw_internal(as, addr, val, attrs, result,
3164 DEVICE_LITTLE_ENDIAN);
3165}
3166
3167void address_space_stw_be(AddressSpace *as, hwaddr addr, uint32_t val,
3168 MemTxAttrs attrs, MemTxResult *result)
3169{
3170 address_space_stw_internal(as, addr, val, attrs, result,
3171 DEVICE_BIG_ENDIAN);
3172}
3173
5ce5944d 3174void stw_phys(AddressSpace *as, hwaddr addr, uint32_t val)
1e78bcc1 3175{
50013115 3176 address_space_stw(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3177}
3178
5ce5944d 3179void stw_le_phys(AddressSpace *as, hwaddr addr, uint32_t val)
1e78bcc1 3180{
50013115 3181 address_space_stw_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3182}
3183
5ce5944d 3184void stw_be_phys(AddressSpace *as, hwaddr addr, uint32_t val)
1e78bcc1 3185{
50013115 3186 address_space_stw_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3187}
3188
aab33094 3189/* XXX: optimize */
50013115
PM
3190void address_space_stq(AddressSpace *as, hwaddr addr, uint64_t val,
3191 MemTxAttrs attrs, MemTxResult *result)
aab33094 3192{
50013115 3193 MemTxResult r;
aab33094 3194 val = tswap64(val);
50013115
PM
3195 r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1);
3196 if (result) {
3197 *result = r;
3198 }
aab33094
FB
3199}
3200
50013115
PM
3201void address_space_stq_le(AddressSpace *as, hwaddr addr, uint64_t val,
3202 MemTxAttrs attrs, MemTxResult *result)
1e78bcc1 3203{
50013115 3204 MemTxResult r;
1e78bcc1 3205 val = cpu_to_le64(val);
50013115
PM
3206 r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1);
3207 if (result) {
3208 *result = r;
3209 }
3210}
3211void address_space_stq_be(AddressSpace *as, hwaddr addr, uint64_t val,
3212 MemTxAttrs attrs, MemTxResult *result)
3213{
3214 MemTxResult r;
3215 val = cpu_to_be64(val);
3216 r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1);
3217 if (result) {
3218 *result = r;
3219 }
3220}
3221
3222void stq_phys(AddressSpace *as, hwaddr addr, uint64_t val)
3223{
3224 address_space_stq(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
3225}
3226
3227void stq_le_phys(AddressSpace *as, hwaddr addr, uint64_t val)
3228{
3229 address_space_stq_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3230}
3231
f606604f 3232void stq_be_phys(AddressSpace *as, hwaddr addr, uint64_t val)
1e78bcc1 3233{
50013115 3234 address_space_stq_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL);
1e78bcc1
AG
3235}
3236
5e2972fd 3237/* virtual memory access for debug (includes writing to ROM) */
f17ec444 3238int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
b448f2f3 3239 uint8_t *buf, int len, int is_write)
13eb76e0
FB
3240{
3241 int l;
a8170e5e 3242 hwaddr phys_addr;
9b3c35e0 3243 target_ulong page;
13eb76e0
FB
3244
3245 while (len > 0) {
3246 page = addr & TARGET_PAGE_MASK;
f17ec444 3247 phys_addr = cpu_get_phys_page_debug(cpu, page);
13eb76e0
FB
3248 /* if no physical page mapped, return an error */
3249 if (phys_addr == -1)
3250 return -1;
3251 l = (page + TARGET_PAGE_SIZE) - addr;
3252 if (l > len)
3253 l = len;
5e2972fd 3254 phys_addr += (addr & ~TARGET_PAGE_MASK);
2e38847b
EI
3255 if (is_write) {
3256 cpu_physical_memory_write_rom(cpu->as, phys_addr, buf, l);
3257 } else {
5c9eb028
PM
3258 address_space_rw(cpu->as, phys_addr, MEMTXATTRS_UNSPECIFIED,
3259 buf, l, 0);
2e38847b 3260 }
13eb76e0
FB
3261 len -= l;
3262 buf += l;
3263 addr += l;
3264 }
3265 return 0;
3266}
a68fe89c 3267#endif
13eb76e0 3268
8e4a424b
BS
3269/*
3270 * A helper function for the _utterly broken_ virtio device model to find out if
3271 * it's running on a big endian machine. Don't do this at home kids!
3272 */
98ed8ecf
GK
3273bool target_words_bigendian(void);
3274bool target_words_bigendian(void)
8e4a424b
BS
3275{
3276#if defined(TARGET_WORDS_BIGENDIAN)
3277 return true;
3278#else
3279 return false;
3280#endif
3281}
3282
76f35538 3283#ifndef CONFIG_USER_ONLY
a8170e5e 3284bool cpu_physical_memory_is_io(hwaddr phys_addr)
76f35538 3285{
5c8a00ce 3286 MemoryRegion*mr;
149f54b5 3287 hwaddr l = 1;
76f35538 3288
5c8a00ce
PB
3289 mr = address_space_translate(&address_space_memory,
3290 phys_addr, &phys_addr, &l, false);
76f35538 3291
5c8a00ce
PB
3292 return !(memory_region_is_ram(mr) ||
3293 memory_region_is_romd(mr));
76f35538 3294}
bd2fa51f
MH
3295
3296void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
3297{
3298 RAMBlock *block;
3299
0dc3f44a
MD
3300 rcu_read_lock();
3301 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
9b8424d5 3302 func(block->host, block->offset, block->used_length, opaque);
bd2fa51f 3303 }
0dc3f44a 3304 rcu_read_unlock();
bd2fa51f 3305}
ec3f8c99 3306#endif