?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qemu-option: Remove qemu_opts_create_nofail
[qemu.git] / exec.c
blob7e49e8e555e5e63ec4db6d36738e7c422b7128d7
1 /*
2 * Virtual page mapping
3 *
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
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18 */
19 #include "config.h"
20 #ifdef _WIN32
21 #include <windows.h>
22 #else
23 #include <sys/types.h>
24 #include <sys/mman.h>
25 #endif
26
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "tcg.h"
30 #include "hw/hw.h"
31 #include "hw/qdev.h"
32 #include "qemu/osdep.h"
33 #include "sysemu/kvm.h"
34 #include "sysemu/sysemu.h"
35 #include "hw/xen/xen.h"
36 #include "qemu/timer.h"
37 #include "qemu/config-file.h"
38 #include "exec/memory.h"
39 #include "sysemu/dma.h"
40 #include "exec/address-spaces.h"
41 #if defined(CONFIG_USER_ONLY)
42 #include <qemu.h>
43 #else /* !CONFIG_USER_ONLY */
44 #include "sysemu/xen-mapcache.h"
45 #include "trace.h"
46 #endif
47 #include "exec/cpu-all.h"
48
49 #include "exec/cputlb.h"
50 #include "translate-all.h"
51
52 #include "exec/memory-internal.h"
53 #include "qemu/cache-utils.h"
54
55 #include "qemu/range.h"
56
57 //#define DEBUG_SUBPAGE
58
59 #if !defined(CONFIG_USER_ONLY)
60 static int in_migration;
61
62 RAMList ram_list = { .blocks = QTAILQ_HEAD_INITIALIZER(ram_list.blocks) };
63
64 static MemoryRegion *system_memory;
65 static MemoryRegion *system_io;
66
67 AddressSpace address_space_io;
68 AddressSpace address_space_memory;
69
70 MemoryRegion io_mem_rom, io_mem_notdirty;
71 static MemoryRegion io_mem_unassigned;
72
73 #endif
74
75 struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus);
76 /* current CPU in the current thread. It is only valid inside
77 cpu_exec() */
78 DEFINE_TLS(CPUState *, current_cpu);
79 /* 0 = Do not count executed instructions.
80 1 = Precise instruction counting.
81 2 = Adaptive rate instruction counting. */
82 int use_icount;
83
84 #if !defined(CONFIG_USER_ONLY)
85
86 typedef struct PhysPageEntry PhysPageEntry;
87
88 struct PhysPageEntry {
89 /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */
90 uint32_t skip : 6;
91 /* index into phys_sections (!skip) or phys_map_nodes (skip) */
92 uint32_t ptr : 26;
93 };
94
95 #define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6)
96
97 /* Size of the L2 (and L3, etc) page tables. */
98 #define ADDR_SPACE_BITS 64
99
100 #define P_L2_BITS 9
101 #define P_L2_SIZE (1 << P_L2_BITS)
102
103 #define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1)
104
105 typedef PhysPageEntry Node[P_L2_SIZE];
106
107 typedef struct PhysPageMap {
108 unsigned sections_nb;
109 unsigned sections_nb_alloc;
110 unsigned nodes_nb;
111 unsigned nodes_nb_alloc;
112 Node *nodes;
113 MemoryRegionSection *sections;
114 } PhysPageMap;
115
116 struct AddressSpaceDispatch {
117 /* This is a multi-level map on the physical address space.
118 * The bottom level has pointers to MemoryRegionSections.
119 */
120 PhysPageEntry phys_map;
121 PhysPageMap map;
122 AddressSpace *as;
123 };
124
125 #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
126 typedef struct subpage_t {
127 MemoryRegion iomem;
128 AddressSpace *as;
129 hwaddr base;
130 uint16_t sub_section[TARGET_PAGE_SIZE];
131 } subpage_t;
132
133 #define PHYS_SECTION_UNASSIGNED 0
134 #define PHYS_SECTION_NOTDIRTY 1
135 #define PHYS_SECTION_ROM 2
136 #define PHYS_SECTION_WATCH 3
137
138 static void io_mem_init(void);
139 static void memory_map_init(void);
140
141 static MemoryRegion io_mem_watch;
142 #endif
143
144 #if !defined(CONFIG_USER_ONLY)
145
146 static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes)
147 {
148 if (map->nodes_nb + nodes > map->nodes_nb_alloc) {
149 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc * 2, 16);
150 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, map->nodes_nb + nodes);
151 map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc);
152 }
153 }
154
155 static uint32_t phys_map_node_alloc(PhysPageMap *map)
156 {
157 unsigned i;
158 uint32_t ret;
159
160 ret = map->nodes_nb++;
161 assert(ret != PHYS_MAP_NODE_NIL);
162 assert(ret != map->nodes_nb_alloc);
163 for (i = 0; i < P_L2_SIZE; ++i) {
164 map->nodes[ret][i].skip = 1;
165 map->nodes[ret][i].ptr = PHYS_MAP_NODE_NIL;
166 }
167 return ret;
168 }
169
170 static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp,
171 hwaddr *index, hwaddr *nb, uint16_t leaf,
172 int level)
173 {
174 PhysPageEntry *p;
175 int i;
176 hwaddr step = (hwaddr)1 << (level * P_L2_BITS);
177
178 if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) {
179 lp->ptr = phys_map_node_alloc(map);
180 p = map->nodes[lp->ptr];
181 if (level == 0) {
182 for (i = 0; i < P_L2_SIZE; i++) {
183 p[i].skip = 0;
184 p[i].ptr = PHYS_SECTION_UNASSIGNED;
185 }
186 }
187 } else {
188 p = map->nodes[lp->ptr];
189 }
190 lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)];
191
192 while (*nb && lp < &p[P_L2_SIZE]) {
193 if ((*index & (step - 1)) == 0 && *nb >= step) {
194 lp->skip = 0;
195 lp->ptr = leaf;
196 *index += step;
197 *nb -= step;
198 } else {
199 phys_page_set_level(map, lp, index, nb, leaf, level - 1);
200 }
201 ++lp;
202 }
203 }
204
205 static void phys_page_set(AddressSpaceDispatch *d,
206 hwaddr index, hwaddr nb,
207 uint16_t leaf)
208 {
209 /* Wildly overreserve - it doesn't matter much. */
210 phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS);
211
212 phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1);
213 }
214
215 /* Compact a non leaf page entry. Simply detect that the entry has a single child,
216 * and update our entry so we can skip it and go directly to the destination.
217 */
218 static void phys_page_compact(PhysPageEntry *lp, Node *nodes, unsigned long *compacted)
219 {
220 unsigned valid_ptr = P_L2_SIZE;
221 int valid = 0;
222 PhysPageEntry *p;
223 int i;
224
225 if (lp->ptr == PHYS_MAP_NODE_NIL) {
226 return;
227 }
228
229 p = nodes[lp->ptr];
230 for (i = 0; i < P_L2_SIZE; i++) {
231 if (p[i].ptr == PHYS_MAP_NODE_NIL) {
232 continue;
233 }
234
235 valid_ptr = i;
236 valid++;
237 if (p[i].skip) {
238 phys_page_compact(&p[i], nodes, compacted);
239 }
240 }
241
242 /* We can only compress if there's only one child. */
243 if (valid != 1) {
244 return;
245 }
246
247 assert(valid_ptr < P_L2_SIZE);
248
249 /* Don't compress if it won't fit in the # of bits we have. */
250 if (lp->skip + p[valid_ptr].skip >= (1 << 3)) {
251 return;
252 }
253
254 lp->ptr = p[valid_ptr].ptr;
255 if (!p[valid_ptr].skip) {
256 /* If our only child is a leaf, make this a leaf. */
257 /* By design, we should have made this node a leaf to begin with so we
258 * should never reach here.
259 * But since it's so simple to handle this, let's do it just in case we
260 * change this rule.
261 */
262 lp->skip = 0;
263 } else {
264 lp->skip += p[valid_ptr].skip;
265 }
266 }
267
268 static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb)
269 {
270 DECLARE_BITMAP(compacted, nodes_nb);
271
272 if (d->phys_map.skip) {
273 phys_page_compact(&d->phys_map, d->map.nodes, compacted);
274 }
275 }
276
277 static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr addr,
278 Node *nodes, MemoryRegionSection *sections)
279 {
280 PhysPageEntry *p;
281 hwaddr index = addr >> TARGET_PAGE_BITS;
282 int i;
283
284 for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) {
285 if (lp.ptr == PHYS_MAP_NODE_NIL) {
286 return &sections[PHYS_SECTION_UNASSIGNED];
287 }
288 p = nodes[lp.ptr];
289 lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)];
290 }
291
292 if (sections[lp.ptr].size.hi ||
293 range_covers_byte(sections[lp.ptr].offset_within_address_space,
294 sections[lp.ptr].size.lo, addr)) {
295 return &sections[lp.ptr];
296 } else {
297 return &sections[PHYS_SECTION_UNASSIGNED];
298 }
299 }
300
301 bool memory_region_is_unassigned(MemoryRegion *mr)
302 {
303 return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device
304 && mr != &io_mem_watch;
305 }
306
307 static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
308 hwaddr addr,
309 bool resolve_subpage)
310 {
311 MemoryRegionSection *section;
312 subpage_t *subpage;
313
314 section = phys_page_find(d->phys_map, addr, d->map.nodes, d->map.sections);
315 if (resolve_subpage && section->mr->subpage) {
316 subpage = container_of(section->mr, subpage_t, iomem);
317 section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]];
318 }
319 return section;
320 }
321
322 static MemoryRegionSection *
323 address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
324 hwaddr *plen, bool resolve_subpage)
325 {
326 MemoryRegionSection *section;
327 Int128 diff;
328
329 section = address_space_lookup_region(d, addr, resolve_subpage);
330 /* Compute offset within MemoryRegionSection */
331 addr -= section->offset_within_address_space;
332
333 /* Compute offset within MemoryRegion */
334 *xlat = addr + section->offset_within_region;
335
336 diff = int128_sub(section->mr->size, int128_make64(addr));
337 *plen = int128_get64(int128_min(diff, int128_make64(*plen)));
338 return section;
339 }
340
341 MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
342 hwaddr *xlat, hwaddr *plen,
343 bool is_write)
344 {
345 IOMMUTLBEntry iotlb;
346 MemoryRegionSection *section;
347 MemoryRegion *mr;
348 hwaddr len = *plen;
349
350 for (;;) {
351 section = address_space_translate_internal(as->dispatch, addr, &addr, plen, true);
352 mr = section->mr;
353
354 if (!mr->iommu_ops) {
355 break;
356 }
357
358 iotlb = mr->iommu_ops->translate(mr, addr);
359 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
360 | (addr & iotlb.addr_mask));
361 len = MIN(len, (addr | iotlb.addr_mask) - addr + 1);
362 if (!(iotlb.perm & (1 << is_write))) {
363 mr = &io_mem_unassigned;
364 break;
365 }
366
367 as = iotlb.target_as;
368 }
369
370 *plen = len;
371 *xlat = addr;
372 return mr;
373 }
374
375 MemoryRegionSection *
376 address_space_translate_for_iotlb(AddressSpace *as, hwaddr addr, hwaddr *xlat,
377 hwaddr *plen)
378 {
379 MemoryRegionSection *section;
380 section = address_space_translate_internal(as->dispatch, addr, xlat, plen, false);
381
382 assert(!section->mr->iommu_ops);
383 return section;
384 }
385 #endif
386
387 void cpu_exec_init_all(void)
388 {
389 #if !defined(CONFIG_USER_ONLY)
390 qemu_mutex_init(&ram_list.mutex);
391 memory_map_init();
392 io_mem_init();
393 #endif
394 }
395
396 #if !defined(CONFIG_USER_ONLY)
397
398 static int cpu_common_post_load(void *opaque, int version_id)
399 {
400 CPUState *cpu = opaque;
401
402 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
403 version_id is increased. */
404 cpu->interrupt_request &= ~0x01;
405 tlb_flush(cpu->env_ptr, 1);
406
407 return 0;
408 }
409
410 const VMStateDescription vmstate_cpu_common = {
411 .name = "cpu_common",
412 .version_id = 1,
413 .minimum_version_id = 1,
414 .minimum_version_id_old = 1,
415 .post_load = cpu_common_post_load,
416 .fields = (VMStateField []) {
417 VMSTATE_UINT32(halted, CPUState),
418 VMSTATE_UINT32(interrupt_request, CPUState),
419 VMSTATE_END_OF_LIST()
420 }
421 };
422
423 #endif
424
425 CPUState *qemu_get_cpu(int index)
426 {
427 CPUState *cpu;
428
429 CPU_FOREACH(cpu) {
430 if (cpu->cpu_index == index) {
431 return cpu;
432 }
433 }
434
435 return NULL;
436 }
437
438 void cpu_exec_init(CPUArchState *env)
439 {
440 CPUState *cpu = ENV_GET_CPU(env);
441 CPUClass *cc = CPU_GET_CLASS(cpu);
442 CPUState *some_cpu;
443 int cpu_index;
444
445 #if defined(CONFIG_USER_ONLY)
446 cpu_list_lock();
447 #endif
448 cpu_index = 0;
449 CPU_FOREACH(some_cpu) {
450 cpu_index++;
451 }
452 cpu->cpu_index = cpu_index;
453 cpu->numa_node = 0;
454 QTAILQ_INIT(&env->breakpoints);
455 QTAILQ_INIT(&env->watchpoints);
456 #ifndef CONFIG_USER_ONLY
457 cpu->thread_id = qemu_get_thread_id();
458 #endif
459 QTAILQ_INSERT_TAIL(&cpus, cpu, node);
460 #if defined(CONFIG_USER_ONLY)
461 cpu_list_unlock();
462 #endif
463 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
464 vmstate_register(NULL, cpu_index, &vmstate_cpu_common, cpu);
465 }
466 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
467 register_savevm(NULL, "cpu", cpu_index, CPU_SAVE_VERSION,
468 cpu_save, cpu_load, env);
469 assert(cc->vmsd == NULL);
470 assert(qdev_get_vmsd(DEVICE(cpu)) == NULL);
471 #endif
472 if (cc->vmsd != NULL) {
473 vmstate_register(NULL, cpu_index, cc->vmsd, cpu);
474 }
475 }
476
477 #if defined(TARGET_HAS_ICE)
478 #if defined(CONFIG_USER_ONLY)
479 static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
480 {
481 tb_invalidate_phys_page_range(pc, pc + 1, 0);
482 }
483 #else
484 static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
485 {
486 hwaddr phys = cpu_get_phys_page_debug(cpu, pc);
487 if (phys != -1) {
488 tb_invalidate_phys_addr(phys | (pc & ~TARGET_PAGE_MASK));
489 }
490 }
491 #endif
492 #endif /* TARGET_HAS_ICE */
493
494 #if defined(CONFIG_USER_ONLY)
495 void cpu_watchpoint_remove_all(CPUArchState *env, int mask)
496
497 {
498 }
499
500 int cpu_watchpoint_insert(CPUArchState *env, target_ulong addr, target_ulong len,
501 int flags, CPUWatchpoint **watchpoint)
502 {
503 return -ENOSYS;
504 }
505 #else
506 /* Add a watchpoint. */
507 int cpu_watchpoint_insert(CPUArchState *env, target_ulong addr, target_ulong len,
508 int flags, CPUWatchpoint **watchpoint)
509 {
510 target_ulong len_mask = ~(len - 1);
511 CPUWatchpoint *wp;
512
513 /* sanity checks: allow power-of-2 lengths, deny unaligned watchpoints */
514 if ((len & (len - 1)) || (addr & ~len_mask) ||
515 len == 0 || len > TARGET_PAGE_SIZE) {
516 fprintf(stderr, "qemu: tried to set invalid watchpoint at "
517 TARGET_FMT_lx ", len=" TARGET_FMT_lu "\n", addr, len);
518 return -EINVAL;
519 }
520 wp = g_malloc(sizeof(*wp));
521
522 wp->vaddr = addr;
523 wp->len_mask = len_mask;
524 wp->flags = flags;
525
526 /* keep all GDB-injected watchpoints in front */
527 if (flags & BP_GDB)
528 QTAILQ_INSERT_HEAD(&env->watchpoints, wp, entry);
529 else
530 QTAILQ_INSERT_TAIL(&env->watchpoints, wp, entry);
531
532 tlb_flush_page(env, addr);
533
534 if (watchpoint)
535 *watchpoint = wp;
536 return 0;
537 }
538
539 /* Remove a specific watchpoint. */
540 int cpu_watchpoint_remove(CPUArchState *env, target_ulong addr, target_ulong len,
541 int flags)
542 {
543 target_ulong len_mask = ~(len - 1);
544 CPUWatchpoint *wp;
545
546 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
547 if (addr == wp->vaddr && len_mask == wp->len_mask
548 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
549 cpu_watchpoint_remove_by_ref(env, wp);
550 return 0;
551 }
552 }
553 return -ENOENT;
554 }
555
556 /* Remove a specific watchpoint by reference. */
557 void cpu_watchpoint_remove_by_ref(CPUArchState *env, CPUWatchpoint *watchpoint)
558 {
559 QTAILQ_REMOVE(&env->watchpoints, watchpoint, entry);
560
561 tlb_flush_page(env, watchpoint->vaddr);
562
563 g_free(watchpoint);
564 }
565
566 /* Remove all matching watchpoints. */
567 void cpu_watchpoint_remove_all(CPUArchState *env, int mask)
568 {
569 CPUWatchpoint *wp, *next;
570
571 QTAILQ_FOREACH_SAFE(wp, &env->watchpoints, entry, next) {
572 if (wp->flags & mask)
573 cpu_watchpoint_remove_by_ref(env, wp);
574 }
575 }
576 #endif
577
578 /* Add a breakpoint. */
579 int cpu_breakpoint_insert(CPUArchState *env, target_ulong pc, int flags,
580 CPUBreakpoint **breakpoint)
581 {
582 #if defined(TARGET_HAS_ICE)
583 CPUBreakpoint *bp;
584
585 bp = g_malloc(sizeof(*bp));
586
587 bp->pc = pc;
588 bp->flags = flags;
589
590 /* keep all GDB-injected breakpoints in front */
591 if (flags & BP_GDB) {
592 QTAILQ_INSERT_HEAD(&env->breakpoints, bp, entry);
593 } else {
594 QTAILQ_INSERT_TAIL(&env->breakpoints, bp, entry);
595 }
596
597 breakpoint_invalidate(ENV_GET_CPU(env), pc);
598
599 if (breakpoint) {
600 *breakpoint = bp;
601 }
602 return 0;
603 #else
604 return -ENOSYS;
605 #endif
606 }
607
608 /* Remove a specific breakpoint. */
609 int cpu_breakpoint_remove(CPUArchState *env, target_ulong pc, int flags)
610 {
611 #if defined(TARGET_HAS_ICE)
612 CPUBreakpoint *bp;
613
614 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
615 if (bp->pc == pc && bp->flags == flags) {
616 cpu_breakpoint_remove_by_ref(env, bp);
617 return 0;
618 }
619 }
620 return -ENOENT;
621 #else
622 return -ENOSYS;
623 #endif
624 }
625
626 /* Remove a specific breakpoint by reference. */
627 void cpu_breakpoint_remove_by_ref(CPUArchState *env, CPUBreakpoint *breakpoint)
628 {
629 #if defined(TARGET_HAS_ICE)
630 QTAILQ_REMOVE(&env->breakpoints, breakpoint, entry);
631
632 breakpoint_invalidate(ENV_GET_CPU(env), breakpoint->pc);
633
634 g_free(breakpoint);
635 #endif
636 }
637
638 /* Remove all matching breakpoints. */
639 void cpu_breakpoint_remove_all(CPUArchState *env, int mask)
640 {
641 #if defined(TARGET_HAS_ICE)
642 CPUBreakpoint *bp, *next;
643
644 QTAILQ_FOREACH_SAFE(bp, &env->breakpoints, entry, next) {
645 if (bp->flags & mask)
646 cpu_breakpoint_remove_by_ref(env, bp);
647 }
648 #endif
649 }
650
651 /* enable or disable single step mode. EXCP_DEBUG is returned by the
652 CPU loop after each instruction */
653 void cpu_single_step(CPUState *cpu, int enabled)
654 {
655 #if defined(TARGET_HAS_ICE)
656 if (cpu->singlestep_enabled != enabled) {
657 cpu->singlestep_enabled = enabled;
658 if (kvm_enabled()) {
659 kvm_update_guest_debug(cpu, 0);
660 } else {
661 /* must flush all the translated code to avoid inconsistencies */
662 /* XXX: only flush what is necessary */
663 CPUArchState *env = cpu->env_ptr;
664 tb_flush(env);
665 }
666 }
667 #endif
668 }
669
670 void cpu_abort(CPUArchState *env, const char *fmt, ...)
671 {
672 CPUState *cpu = ENV_GET_CPU(env);
673 va_list ap;
674 va_list ap2;
675
676 va_start(ap, fmt);
677 va_copy(ap2, ap);
678 fprintf(stderr, "qemu: fatal: ");
679 vfprintf(stderr, fmt, ap);
680 fprintf(stderr, "\n");
681 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP);
682 if (qemu_log_enabled()) {
683 qemu_log("qemu: fatal: ");
684 qemu_log_vprintf(fmt, ap2);
685 qemu_log("\n");
686 log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP);
687 qemu_log_flush();
688 qemu_log_close();
689 }
690 va_end(ap2);
691 va_end(ap);
692 #if defined(CONFIG_USER_ONLY)
693 {
694 struct sigaction act;
695 sigfillset(&act.sa_mask);
696 act.sa_handler = SIG_DFL;
697 sigaction(SIGABRT, &act, NULL);
698 }
699 #endif
700 abort();
701 }
702
703 #if !defined(CONFIG_USER_ONLY)
704 static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
705 {
706 RAMBlock *block;
707
708 /* The list is protected by the iothread lock here. */
709 block = ram_list.mru_block;
710 if (block && addr - block->offset < block->length) {
711 goto found;
712 }
713 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
714 if (addr - block->offset < block->length) {
715 goto found;
716 }
717 }
718
719 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
720 abort();
721
722 found:
723 ram_list.mru_block = block;
724 return block;
725 }
726
727 static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t end,
728 uintptr_t length)
729 {
730 RAMBlock *block;
731 ram_addr_t start1;
732
733 block = qemu_get_ram_block(start);
734 assert(block == qemu_get_ram_block(end - 1));
735 start1 = (uintptr_t)block->host + (start - block->offset);
736 cpu_tlb_reset_dirty_all(start1, length);
737 }
738
739 /* Note: start and end must be within the same ram block. */
740 void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t end,
741 int dirty_flags)
742 {
743 uintptr_t length;
744
745 start &= TARGET_PAGE_MASK;
746 end = TARGET_PAGE_ALIGN(end);
747
748 length = end - start;
749 if (length == 0)
750 return;
751 cpu_physical_memory_mask_dirty_range(start, length, dirty_flags);
752
753 if (tcg_enabled()) {
754 tlb_reset_dirty_range_all(start, end, length);
755 }
756 }
757
758 static int cpu_physical_memory_set_dirty_tracking(int enable)
759 {
760 int ret = 0;
761 in_migration = enable;
762 return ret;
763 }
764
765 hwaddr memory_region_section_get_iotlb(CPUArchState *env,
766 MemoryRegionSection *section,
767 target_ulong vaddr,
768 hwaddr paddr, hwaddr xlat,
769 int prot,
770 target_ulong *address)
771 {
772 hwaddr iotlb;
773 CPUWatchpoint *wp;
774
775 if (memory_region_is_ram(section->mr)) {
776 /* Normal RAM. */
777 iotlb = (memory_region_get_ram_addr(section->mr) & TARGET_PAGE_MASK)
778 + xlat;
779 if (!section->readonly) {
780 iotlb |= PHYS_SECTION_NOTDIRTY;
781 } else {
782 iotlb |= PHYS_SECTION_ROM;
783 }
784 } else {
785 iotlb = section - address_space_memory.dispatch->map.sections;
786 iotlb += xlat;
787 }
788
789 /* Make accesses to pages with watchpoints go via the
790 watchpoint trap routines. */
791 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
792 if (vaddr == (wp->vaddr & TARGET_PAGE_MASK)) {
793 /* Avoid trapping reads of pages with a write breakpoint. */
794 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
795 iotlb = PHYS_SECTION_WATCH + paddr;
796 *address |= TLB_MMIO;
797 break;
798 }
799 }
800 }
801
802 return iotlb;
803 }
804 #endif /* defined(CONFIG_USER_ONLY) */
805
806 #if !defined(CONFIG_USER_ONLY)
807
808 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
809 uint16_t section);
810 static subpage_t *subpage_init(AddressSpace *as, hwaddr base);
811
812 static void *(*phys_mem_alloc)(size_t size) = qemu_anon_ram_alloc;
813
814 /*
815 * Set a custom physical guest memory alloator.
816 * Accelerators with unusual needs may need this. Hopefully, we can
817 * get rid of it eventually.
818 */
819 void phys_mem_set_alloc(void *(*alloc)(size_t))
820 {
821 phys_mem_alloc = alloc;
822 }
823
824 static uint16_t phys_section_add(PhysPageMap *map,
825 MemoryRegionSection *section)
826 {
827 /* The physical section number is ORed with a page-aligned
828 * pointer to produce the iotlb entries. Thus it should
829 * never overflow into the page-aligned value.
830 */
831 assert(map->sections_nb < TARGET_PAGE_SIZE);
832
833 if (map->sections_nb == map->sections_nb_alloc) {
834 map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16);
835 map->sections = g_renew(MemoryRegionSection, map->sections,
836 map->sections_nb_alloc);
837 }
838 map->sections[map->sections_nb] = *section;
839 memory_region_ref(section->mr);
840 return map->sections_nb++;
841 }
842
843 static void phys_section_destroy(MemoryRegion *mr)
844 {
845 memory_region_unref(mr);
846
847 if (mr->subpage) {
848 subpage_t *subpage = container_of(mr, subpage_t, iomem);
849 memory_region_destroy(&subpage->iomem);
850 g_free(subpage);
851 }
852 }
853
854 static void phys_sections_free(PhysPageMap *map)
855 {
856 while (map->sections_nb > 0) {
857 MemoryRegionSection *section = &map->sections[--map->sections_nb];
858 phys_section_destroy(section->mr);
859 }
860 g_free(map->sections);
861 g_free(map->nodes);
862 }
863
864 static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section)
865 {
866 subpage_t *subpage;
867 hwaddr base = section->offset_within_address_space
868 & TARGET_PAGE_MASK;
869 MemoryRegionSection *existing = phys_page_find(d->phys_map, base,
870 d->map.nodes, d->map.sections);
871 MemoryRegionSection subsection = {
872 .offset_within_address_space = base,
873 .size = int128_make64(TARGET_PAGE_SIZE),
874 };
875 hwaddr start, end;
876
877 assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
878
879 if (!(existing->mr->subpage)) {
880 subpage = subpage_init(d->as, base);
881 subsection.mr = &subpage->iomem;
882 phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
883 phys_section_add(&d->map, &subsection));
884 } else {
885 subpage = container_of(existing->mr, subpage_t, iomem);
886 }
887 start = section->offset_within_address_space & ~TARGET_PAGE_MASK;
888 end = start + int128_get64(section->size) - 1;
889 subpage_register(subpage, start, end,
890 phys_section_add(&d->map, section));
891 }
892
893
894 static void register_multipage(AddressSpaceDispatch *d,
895 MemoryRegionSection *section)
896 {
897 hwaddr start_addr = section->offset_within_address_space;
898 uint16_t section_index = phys_section_add(&d->map, section);
899 uint64_t num_pages = int128_get64(int128_rshift(section->size,
900 TARGET_PAGE_BITS));
901
902 assert(num_pages);
903 phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
904 }
905
906 static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
907 {
908 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
909 AddressSpaceDispatch *d = as->next_dispatch;
910 MemoryRegionSection now = *section, remain = *section;
911 Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
912
913 if (now.offset_within_address_space & ~TARGET_PAGE_MASK) {
914 uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space)
915 - now.offset_within_address_space;
916
917 now.size = int128_min(int128_make64(left), now.size);
918 register_subpage(d, &now);
919 } else {
920 now.size = int128_zero();
921 }
922 while (int128_ne(remain.size, now.size)) {
923 remain.size = int128_sub(remain.size, now.size);
924 remain.offset_within_address_space += int128_get64(now.size);
925 remain.offset_within_region += int128_get64(now.size);
926 now = remain;
927 if (int128_lt(remain.size, page_size)) {
928 register_subpage(d, &now);
929 } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) {
930 now.size = page_size;
931 register_subpage(d, &now);
932 } else {
933 now.size = int128_and(now.size, int128_neg(page_size));
934 register_multipage(d, &now);
935 }
936 }
937 }
938
939 void qemu_flush_coalesced_mmio_buffer(void)
940 {
941 if (kvm_enabled())
942 kvm_flush_coalesced_mmio_buffer();
943 }
944
945 void qemu_mutex_lock_ramlist(void)
946 {
947 qemu_mutex_lock(&ram_list.mutex);
948 }
949
950 void qemu_mutex_unlock_ramlist(void)
951 {
952 qemu_mutex_unlock(&ram_list.mutex);
953 }
954
955 #ifdef __linux__
956
957 #include <sys/vfs.h>
958
959 #define HUGETLBFS_MAGIC 0x958458f6
960
961 static long gethugepagesize(const char *path)
962 {
963 struct statfs fs;
964 int ret;
965
966 do {
967 ret = statfs(path, &fs);
968 } while (ret != 0 && errno == EINTR);
969
970 if (ret != 0) {
971 perror(path);
972 return 0;
973 }
974
975 if (fs.f_type != HUGETLBFS_MAGIC)
976 fprintf(stderr, "Warning: path not on HugeTLBFS: %s\n", path);
977
978 return fs.f_bsize;
979 }
980
981 static sigjmp_buf sigjump;
982
983 static void sigbus_handler(int signal)
984 {
985 siglongjmp(sigjump, 1);
986 }
987
988 static void *file_ram_alloc(RAMBlock *block,
989 ram_addr_t memory,
990 const char *path)
991 {
992 char *filename;
993 char *sanitized_name;
994 char *c;
995 void *area;
996 int fd;
997 unsigned long hpagesize;
998
999 hpagesize = gethugepagesize(path);
1000 if (!hpagesize) {
1001 return NULL;
1002 }
1003
1004 if (memory < hpagesize) {
1005 return NULL;
1006 }
1007
1008 if (kvm_enabled() && !kvm_has_sync_mmu()) {
1009 fprintf(stderr, "host lacks kvm mmu notifiers, -mem-path unsupported\n");
1010 return NULL;
1011 }
1012
1013 /* Make name safe to use with mkstemp by replacing '/' with '_'. */
1014 sanitized_name = g_strdup(block->mr->name);
1015 for (c = sanitized_name; *c != '\0'; c++) {
1016 if (*c == '/')
1017 *c = '_';
1018 }
1019
1020 filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path,
1021 sanitized_name);
1022 g_free(sanitized_name);
1023
1024 fd = mkstemp(filename);
1025 if (fd < 0) {
1026 perror("unable to create backing store for hugepages");
1027 g_free(filename);
1028 return NULL;
1029 }
1030 unlink(filename);
1031 g_free(filename);
1032
1033 memory = (memory+hpagesize-1) & ~(hpagesize-1);
1034
1035 /*
1036 * ftruncate is not supported by hugetlbfs in older
1037 * hosts, so don't bother bailing out on errors.
1038 * If anything goes wrong with it under other filesystems,
1039 * mmap will fail.
1040 */
1041 if (ftruncate(fd, memory))
1042 perror("ftruncate");
1043
1044 area = mmap(0, memory, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1045 if (area == MAP_FAILED) {
1046 perror("file_ram_alloc: can't mmap RAM pages");
1047 close(fd);
1048 return (NULL);
1049 }
1050
1051 if (mem_prealloc) {
1052 int ret, i;
1053 struct sigaction act, oldact;
1054 sigset_t set, oldset;
1055
1056 memset(&act, 0, sizeof(act));
1057 act.sa_handler = &sigbus_handler;
1058 act.sa_flags = 0;
1059
1060 ret = sigaction(SIGBUS, &act, &oldact);
1061 if (ret) {
1062 perror("file_ram_alloc: failed to install signal handler");
1063 exit(1);
1064 }
1065
1066 /* unblock SIGBUS */
1067 sigemptyset(&set);
1068 sigaddset(&set, SIGBUS);
1069 pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
1070
1071 if (sigsetjmp(sigjump, 1)) {
1072 fprintf(stderr, "file_ram_alloc: failed to preallocate pages\n");
1073 exit(1);
1074 }
1075
1076 /* MAP_POPULATE silently ignores failures */
1077 for (i = 0; i < (memory/hpagesize)-1; i++) {
1078 memset(area + (hpagesize*i), 0, 1);
1079 }
1080
1081 ret = sigaction(SIGBUS, &oldact, NULL);
1082 if (ret) {
1083 perror("file_ram_alloc: failed to reinstall signal handler");
1084 exit(1);
1085 }
1086
1087 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
1088 }
1089
1090 block->fd = fd;
1091 return area;
1092 }
1093 #else
1094 static void *file_ram_alloc(RAMBlock *block,
1095 ram_addr_t memory,
1096 const char *path)
1097 {
1098 fprintf(stderr, "-mem-path not supported on this host\n");
1099 exit(1);
1100 }
1101 #endif
1102
1103 static ram_addr_t find_ram_offset(ram_addr_t size)
1104 {
1105 RAMBlock *block, *next_block;
1106 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
1107
1108 assert(size != 0); /* it would hand out same offset multiple times */
1109
1110 if (QTAILQ_EMPTY(&ram_list.blocks))
1111 return 0;
1112
1113 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1114 ram_addr_t end, next = RAM_ADDR_MAX;
1115
1116 end = block->offset + block->length;
1117
1118 QTAILQ_FOREACH(next_block, &ram_list.blocks, next) {
1119 if (next_block->offset >= end) {
1120 next = MIN(next, next_block->offset);
1121 }
1122 }
1123 if (next - end >= size && next - end < mingap) {
1124 offset = end;
1125 mingap = next - end;
1126 }
1127 }
1128
1129 if (offset == RAM_ADDR_MAX) {
1130 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
1131 (uint64_t)size);
1132 abort();
1133 }
1134
1135 return offset;
1136 }
1137
1138 ram_addr_t last_ram_offset(void)
1139 {
1140 RAMBlock *block;
1141 ram_addr_t last = 0;
1142
1143 QTAILQ_FOREACH(block, &ram_list.blocks, next)
1144 last = MAX(last, block->offset + block->length);
1145
1146 return last;
1147 }
1148
1149 static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
1150 {
1151 int ret;
1152
1153 /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */
1154 if (!qemu_opt_get_bool(qemu_get_machine_opts(),
1155 "dump-guest-core", true)) {
1156 ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP);
1157 if (ret) {
1158 perror("qemu_madvise");
1159 fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, "
1160 "but dump_guest_core=off specified\n");
1161 }
1162 }
1163 }
1164
1165 void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
1166 {
1167 RAMBlock *new_block, *block;
1168
1169 new_block = NULL;
1170 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1171 if (block->offset == addr) {
1172 new_block = block;
1173 break;
1174 }
1175 }
1176 assert(new_block);
1177 assert(!new_block->idstr[0]);
1178
1179 if (dev) {
1180 char *id = qdev_get_dev_path(dev);
1181 if (id) {
1182 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
1183 g_free(id);
1184 }
1185 }
1186 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
1187
1188 /* This assumes the iothread lock is taken here too. */
1189 qemu_mutex_lock_ramlist();
1190 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1191 if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
1192 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
1193 new_block->idstr);
1194 abort();
1195 }
1196 }
1197 qemu_mutex_unlock_ramlist();
1198 }
1199
1200 static int memory_try_enable_merging(void *addr, size_t len)
1201 {
1202 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "mem-merge", true)) {
1203 /* disabled by the user */
1204 return 0;
1205 }
1206
1207 return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE);
1208 }
1209
1210 ram_addr_t qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
1211 MemoryRegion *mr)
1212 {
1213 RAMBlock *block, *new_block;
1214
1215 size = TARGET_PAGE_ALIGN(size);
1216 new_block = g_malloc0(sizeof(*new_block));
1217 new_block->fd = -1;
1218
1219 /* This assumes the iothread lock is taken here too. */
1220 qemu_mutex_lock_ramlist();
1221 new_block->mr = mr;
1222 new_block->offset = find_ram_offset(size);
1223 if (host) {
1224 new_block->host = host;
1225 new_block->flags |= RAM_PREALLOC_MASK;
1226 } else if (xen_enabled()) {
1227 if (mem_path) {
1228 fprintf(stderr, "-mem-path not supported with Xen\n");
1229 exit(1);
1230 }
1231 xen_ram_alloc(new_block->offset, size, mr);
1232 } else {
1233 if (mem_path) {
1234 if (phys_mem_alloc != qemu_anon_ram_alloc) {
1235 /*
1236 * file_ram_alloc() needs to allocate just like
1237 * phys_mem_alloc, but we haven't bothered to provide
1238 * a hook there.
1239 */
1240 fprintf(stderr,
1241 "-mem-path not supported with this accelerator\n");
1242 exit(1);
1243 }
1244 new_block->host = file_ram_alloc(new_block, size, mem_path);
1245 }
1246 if (!new_block->host) {
1247 new_block->host = phys_mem_alloc(size);
1248 if (!new_block->host) {
1249 fprintf(stderr, "Cannot set up guest memory '%s': %s\n",
1250 new_block->mr->name, strerror(errno));
1251 exit(1);
1252 }
1253 memory_try_enable_merging(new_block->host, size);
1254 }
1255 }
1256 new_block->length = size;
1257
1258 /* Keep the list sorted from biggest to smallest block. */
1259 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1260 if (block->length < new_block->length) {
1261 break;
1262 }
1263 }
1264 if (block) {
1265 QTAILQ_INSERT_BEFORE(block, new_block, next);
1266 } else {
1267 QTAILQ_INSERT_TAIL(&ram_list.blocks, new_block, next);
1268 }
1269 ram_list.mru_block = NULL;
1270
1271 ram_list.version++;
1272 qemu_mutex_unlock_ramlist();
1273
1274 ram_list.phys_dirty = g_realloc(ram_list.phys_dirty,
1275 last_ram_offset() >> TARGET_PAGE_BITS);
1276 memset(ram_list.phys_dirty + (new_block->offset >> TARGET_PAGE_BITS),
1277 0, size >> TARGET_PAGE_BITS);
1278 cpu_physical_memory_set_dirty_range(new_block->offset, size, 0xff);
1279
1280 qemu_ram_setup_dump(new_block->host, size);
1281 qemu_madvise(new_block->host, size, QEMU_MADV_HUGEPAGE);
1282 qemu_madvise(new_block->host, size, QEMU_MADV_DONTFORK);
1283
1284 if (kvm_enabled())
1285 kvm_setup_guest_memory(new_block->host, size);
1286
1287 return new_block->offset;
1288 }
1289
1290 ram_addr_t qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr)
1291 {
1292 return qemu_ram_alloc_from_ptr(size, NULL, mr);
1293 }
1294
1295 void qemu_ram_free_from_ptr(ram_addr_t addr)
1296 {
1297 RAMBlock *block;
1298
1299 /* This assumes the iothread lock is taken here too. */
1300 qemu_mutex_lock_ramlist();
1301 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1302 if (addr == block->offset) {
1303 QTAILQ_REMOVE(&ram_list.blocks, block, next);
1304 ram_list.mru_block = NULL;
1305 ram_list.version++;
1306 g_free(block);
1307 break;
1308 }
1309 }
1310 qemu_mutex_unlock_ramlist();
1311 }
1312
1313 void qemu_ram_free(ram_addr_t addr)
1314 {
1315 RAMBlock *block;
1316
1317 /* This assumes the iothread lock is taken here too. */
1318 qemu_mutex_lock_ramlist();
1319 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1320 if (addr == block->offset) {
1321 QTAILQ_REMOVE(&ram_list.blocks, block, next);
1322 ram_list.mru_block = NULL;
1323 ram_list.version++;
1324 if (block->flags & RAM_PREALLOC_MASK) {
1325 ;
1326 } else if (xen_enabled()) {
1327 xen_invalidate_map_cache_entry(block->host);
1328 #ifndef _WIN32
1329 } else if (block->fd >= 0) {
1330 munmap(block->host, block->length);
1331 close(block->fd);
1332 #endif
1333 } else {
1334 qemu_anon_ram_free(block->host, block->length);
1335 }
1336 g_free(block);
1337 break;
1338 }
1339 }
1340 qemu_mutex_unlock_ramlist();
1341
1342 }
1343
1344 #ifndef _WIN32
1345 void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
1346 {
1347 RAMBlock *block;
1348 ram_addr_t offset;
1349 int flags;
1350 void *area, *vaddr;
1351
1352 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1353 offset = addr - block->offset;
1354 if (offset < block->length) {
1355 vaddr = block->host + offset;
1356 if (block->flags & RAM_PREALLOC_MASK) {
1357 ;
1358 } else if (xen_enabled()) {
1359 abort();
1360 } else {
1361 flags = MAP_FIXED;
1362 munmap(vaddr, length);
1363 if (block->fd >= 0) {
1364 #ifdef MAP_POPULATE
1365 flags |= mem_prealloc ? MAP_POPULATE | MAP_SHARED :
1366 MAP_PRIVATE;
1367 #else
1368 flags |= MAP_PRIVATE;
1369 #endif
1370 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1371 flags, block->fd, offset);
1372 } else {
1373 /*
1374 * Remap needs to match alloc. Accelerators that
1375 * set phys_mem_alloc never remap. If they did,
1376 * we'd need a remap hook here.
1377 */
1378 assert(phys_mem_alloc == qemu_anon_ram_alloc);
1379
1380 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
1381 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1382 flags, -1, 0);
1383 }
1384 if (area != vaddr) {
1385 fprintf(stderr, "Could not remap addr: "
1386 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
1387 length, addr);
1388 exit(1);
1389 }
1390 memory_try_enable_merging(vaddr, length);
1391 qemu_ram_setup_dump(vaddr, length);
1392 }
1393 return;
1394 }
1395 }
1396 }
1397 #endif /* !_WIN32 */
1398
1399 /* Return a host pointer to ram allocated with qemu_ram_alloc.
1400 With the exception of the softmmu code in this file, this should
1401 only be used for local memory (e.g. video ram) that the device owns,
1402 and knows it isn't going to access beyond the end of the block.
1403
1404 It should not be used for general purpose DMA.
1405 Use cpu_physical_memory_map/cpu_physical_memory_rw instead.
1406 */
1407 void *qemu_get_ram_ptr(ram_addr_t addr)
1408 {
1409 RAMBlock *block = qemu_get_ram_block(addr);
1410
1411 if (xen_enabled()) {
1412 /* We need to check if the requested address is in the RAM
1413 * because we don't want to map the entire memory in QEMU.
1414 * In that case just map until the end of the page.
1415 */
1416 if (block->offset == 0) {
1417 return xen_map_cache(addr, 0, 0);
1418 } else if (block->host == NULL) {
1419 block->host =
1420 xen_map_cache(block->offset, block->length, 1);
1421 }
1422 }
1423 return block->host + (addr - block->offset);
1424 }
1425
1426 /* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr
1427 * but takes a size argument */
1428 static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
1429 {
1430 if (*size == 0) {
1431 return NULL;
1432 }
1433 if (xen_enabled()) {
1434 return xen_map_cache(addr, *size, 1);
1435 } else {
1436 RAMBlock *block;
1437
1438 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1439 if (addr - block->offset < block->length) {
1440 if (addr - block->offset + *size > block->length)
1441 *size = block->length - addr + block->offset;
1442 return block->host + (addr - block->offset);
1443 }
1444 }
1445
1446 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1447 abort();
1448 }
1449 }
1450
1451 /* Some of the softmmu routines need to translate from a host pointer
1452 (typically a TLB entry) back to a ram offset. */
1453 MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
1454 {
1455 RAMBlock *block;
1456 uint8_t *host = ptr;
1457
1458 if (xen_enabled()) {
1459 *ram_addr = xen_ram_addr_from_mapcache(ptr);
1460 return qemu_get_ram_block(*ram_addr)->mr;
1461 }
1462
1463 block = ram_list.mru_block;
1464 if (block && block->host && host - block->host < block->length) {
1465 goto found;
1466 }
1467
1468 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1469 /* This case append when the block is not mapped. */
1470 if (block->host == NULL) {
1471 continue;
1472 }
1473 if (host - block->host < block->length) {
1474 goto found;
1475 }
1476 }
1477
1478 return NULL;
1479
1480 found:
1481 *ram_addr = block->offset + (host - block->host);
1482 return block->mr;
1483 }
1484
1485 static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
1486 uint64_t val, unsigned size)
1487 {
1488 int dirty_flags;
1489 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
1490 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
1491 tb_invalidate_phys_page_fast(ram_addr, size);
1492 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
1493 }
1494 switch (size) {
1495 case 1:
1496 stb_p(qemu_get_ram_ptr(ram_addr), val);
1497 break;
1498 case 2:
1499 stw_p(qemu_get_ram_ptr(ram_addr), val);
1500 break;
1501 case 4:
1502 stl_p(qemu_get_ram_ptr(ram_addr), val);
1503 break;
1504 default:
1505 abort();
1506 }
1507 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
1508 cpu_physical_memory_set_dirty_flags(ram_addr, dirty_flags);
1509 /* we remove the notdirty callback only if the code has been
1510 flushed */
1511 if (dirty_flags == 0xff) {
1512 CPUArchState *env = current_cpu->env_ptr;
1513 tlb_set_dirty(env, env->mem_io_vaddr);
1514 }
1515 }
1516
1517 static bool notdirty_mem_accepts(void *opaque, hwaddr addr,
1518 unsigned size, bool is_write)
1519 {
1520 return is_write;
1521 }
1522
1523 static const MemoryRegionOps notdirty_mem_ops = {
1524 .write = notdirty_mem_write,
1525 .valid.accepts = notdirty_mem_accepts,
1526 .endianness = DEVICE_NATIVE_ENDIAN,
1527 };
1528
1529 /* Generate a debug exception if a watchpoint has been hit. */
1530 static void check_watchpoint(int offset, int len_mask, int flags)
1531 {
1532 CPUArchState *env = current_cpu->env_ptr;
1533 target_ulong pc, cs_base;
1534 target_ulong vaddr;
1535 CPUWatchpoint *wp;
1536 int cpu_flags;
1537
1538 if (env->watchpoint_hit) {
1539 /* We re-entered the check after replacing the TB. Now raise
1540 * the debug interrupt so that is will trigger after the
1541 * current instruction. */
1542 cpu_interrupt(ENV_GET_CPU(env), CPU_INTERRUPT_DEBUG);
1543 return;
1544 }
1545 vaddr = (env->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
1546 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1547 if ((vaddr == (wp->vaddr & len_mask) ||
1548 (vaddr & wp->len_mask) == wp->vaddr) && (wp->flags & flags)) {
1549 wp->flags |= BP_WATCHPOINT_HIT;
1550 if (!env->watchpoint_hit) {
1551 env->watchpoint_hit = wp;
1552 tb_check_watchpoint(env);
1553 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
1554 env->exception_index = EXCP_DEBUG;
1555 cpu_loop_exit(env);
1556 } else {
1557 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
1558 tb_gen_code(env, pc, cs_base, cpu_flags, 1);
1559 cpu_resume_from_signal(env, NULL);
1560 }
1561 }
1562 } else {
1563 wp->flags &= ~BP_WATCHPOINT_HIT;
1564 }
1565 }
1566 }
1567
1568 /* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
1569 so these check for a hit then pass through to the normal out-of-line
1570 phys routines. */
1571 static uint64_t watch_mem_read(void *opaque, hwaddr addr,
1572 unsigned size)
1573 {
1574 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~(size - 1), BP_MEM_READ);
1575 switch (size) {
1576 case 1: return ldub_phys(addr);
1577 case 2: return lduw_phys(addr);
1578 case 4: return ldl_phys(addr);
1579 default: abort();
1580 }
1581 }
1582
1583 static void watch_mem_write(void *opaque, hwaddr addr,
1584 uint64_t val, unsigned size)
1585 {
1586 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~(size - 1), BP_MEM_WRITE);
1587 switch (size) {
1588 case 1:
1589 stb_phys(addr, val);
1590 break;
1591 case 2:
1592 stw_phys(addr, val);
1593 break;
1594 case 4:
1595 stl_phys(addr, val);
1596 break;
1597 default: abort();
1598 }
1599 }
1600
1601 static const MemoryRegionOps watch_mem_ops = {
1602 .read = watch_mem_read,
1603 .write = watch_mem_write,
1604 .endianness = DEVICE_NATIVE_ENDIAN,
1605 };
1606
1607 static uint64_t subpage_read(void *opaque, hwaddr addr,
1608 unsigned len)
1609 {
1610 subpage_t *subpage = opaque;
1611 uint8_t buf[4];
1612
1613 #if defined(DEBUG_SUBPAGE)
1614 printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__,
1615 subpage, len, addr);
1616 #endif
1617 address_space_read(subpage->as, addr + subpage->base, buf, len);
1618 switch (len) {
1619 case 1:
1620 return ldub_p(buf);
1621 case 2:
1622 return lduw_p(buf);
1623 case 4:
1624 return ldl_p(buf);
1625 default:
1626 abort();
1627 }
1628 }
1629
1630 static void subpage_write(void *opaque, hwaddr addr,
1631 uint64_t value, unsigned len)
1632 {
1633 subpage_t *subpage = opaque;
1634 uint8_t buf[4];
1635
1636 #if defined(DEBUG_SUBPAGE)
1637 printf("%s: subpage %p len %u addr " TARGET_FMT_plx
1638 " value %"PRIx64"\n",
1639 __func__, subpage, len, addr, value);
1640 #endif
1641 switch (len) {
1642 case 1:
1643 stb_p(buf, value);
1644 break;
1645 case 2:
1646 stw_p(buf, value);
1647 break;
1648 case 4:
1649 stl_p(buf, value);
1650 break;
1651 default:
1652 abort();
1653 }
1654 address_space_write(subpage->as, addr + subpage->base, buf, len);
1655 }
1656
1657 static bool subpage_accepts(void *opaque, hwaddr addr,
1658 unsigned len, bool is_write)
1659 {
1660 subpage_t *subpage = opaque;
1661 #if defined(DEBUG_SUBPAGE)
1662 printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n",
1663 __func__, subpage, is_write ? 'w' : 'r', len, addr);
1664 #endif
1665
1666 return address_space_access_valid(subpage->as, addr + subpage->base,
1667 len, is_write);
1668 }
1669
1670 static const MemoryRegionOps subpage_ops = {
1671 .read = subpage_read,
1672 .write = subpage_write,
1673 .valid.accepts = subpage_accepts,
1674 .endianness = DEVICE_NATIVE_ENDIAN,
1675 };
1676
1677 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
1678 uint16_t section)
1679 {
1680 int idx, eidx;
1681
1682 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
1683 return -1;
1684 idx = SUBPAGE_IDX(start);
1685 eidx = SUBPAGE_IDX(end);
1686 #if defined(DEBUG_SUBPAGE)
1687 printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n",
1688 __func__, mmio, start, end, idx, eidx, section);
1689 #endif
1690 for (; idx <= eidx; idx++) {
1691 mmio->sub_section[idx] = section;
1692 }
1693
1694 return 0;
1695 }
1696
1697 static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
1698 {
1699 subpage_t *mmio;
1700
1701 mmio = g_malloc0(sizeof(subpage_t));
1702
1703 mmio->as = as;
1704 mmio->base = base;
1705 memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
1706 "subpage", TARGET_PAGE_SIZE);
1707 mmio->iomem.subpage = true;
1708 #if defined(DEBUG_SUBPAGE)
1709 printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__,
1710 mmio, base, TARGET_PAGE_SIZE);
1711 #endif
1712 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED);
1713
1714 return mmio;
1715 }
1716
1717 static uint16_t dummy_section(PhysPageMap *map, MemoryRegion *mr)
1718 {
1719 MemoryRegionSection section = {
1720 .mr = mr,
1721 .offset_within_address_space = 0,
1722 .offset_within_region = 0,
1723 .size = int128_2_64(),
1724 };
1725
1726 return phys_section_add(map, &section);
1727 }
1728
1729 MemoryRegion *iotlb_to_region(hwaddr index)
1730 {
1731 return address_space_memory.dispatch->map.sections[
1732 index & ~TARGET_PAGE_MASK].mr;
1733 }
1734
1735 static void io_mem_init(void)
1736 {
1737 memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, "rom", UINT64_MAX);
1738 memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
1739 "unassigned", UINT64_MAX);
1740 memory_region_init_io(&io_mem_notdirty, NULL, &notdirty_mem_ops, NULL,
1741 "notdirty", UINT64_MAX);
1742 memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL,
1743 "watch", UINT64_MAX);
1744 }
1745
1746 static void mem_begin(MemoryListener *listener)
1747 {
1748 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
1749 AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1);
1750 uint16_t n;
1751
1752 n = dummy_section(&d->map, &io_mem_unassigned);
1753 assert(n == PHYS_SECTION_UNASSIGNED);
1754 n = dummy_section(&d->map, &io_mem_notdirty);
1755 assert(n == PHYS_SECTION_NOTDIRTY);
1756 n = dummy_section(&d->map, &io_mem_rom);
1757 assert(n == PHYS_SECTION_ROM);
1758 n = dummy_section(&d->map, &io_mem_watch);
1759 assert(n == PHYS_SECTION_WATCH);
1760
1761 d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
1762 d->as = as;
1763 as->next_dispatch = d;
1764 }
1765
1766 static void mem_commit(MemoryListener *listener)
1767 {
1768 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
1769 AddressSpaceDispatch *cur = as->dispatch;
1770 AddressSpaceDispatch *next = as->next_dispatch;
1771
1772 phys_page_compact_all(next, next->map.nodes_nb);
1773
1774 as->dispatch = next;
1775
1776 if (cur) {
1777 phys_sections_free(&cur->map);
1778 g_free(cur);
1779 }
1780 }
1781
1782 static void tcg_commit(MemoryListener *listener)
1783 {
1784 CPUState *cpu;
1785
1786 /* since each CPU stores ram addresses in its TLB cache, we must
1787 reset the modified entries */
1788 /* XXX: slow ! */
1789 CPU_FOREACH(cpu) {
1790 CPUArchState *env = cpu->env_ptr;
1791
1792 tlb_flush(env, 1);
1793 }
1794 }
1795
1796 static void core_log_global_start(MemoryListener *listener)
1797 {
1798 cpu_physical_memory_set_dirty_tracking(1);
1799 }
1800
1801 static void core_log_global_stop(MemoryListener *listener)
1802 {
1803 cpu_physical_memory_set_dirty_tracking(0);
1804 }
1805
1806 static MemoryListener core_memory_listener = {
1807 .log_global_start = core_log_global_start,
1808 .log_global_stop = core_log_global_stop,
1809 .priority = 1,
1810 };
1811
1812 static MemoryListener tcg_memory_listener = {
1813 .commit = tcg_commit,
1814 };
1815
1816 void address_space_init_dispatch(AddressSpace *as)
1817 {
1818 as->dispatch = NULL;
1819 as->dispatch_listener = (MemoryListener) {
1820 .begin = mem_begin,
1821 .commit = mem_commit,
1822 .region_add = mem_add,
1823 .region_nop = mem_add,
1824 .priority = 0,
1825 };
1826 memory_listener_register(&as->dispatch_listener, as);
1827 }
1828
1829 void address_space_destroy_dispatch(AddressSpace *as)
1830 {
1831 AddressSpaceDispatch *d = as->dispatch;
1832
1833 memory_listener_unregister(&as->dispatch_listener);
1834 g_free(d);
1835 as->dispatch = NULL;
1836 }
1837
1838 static void memory_map_init(void)
1839 {
1840 system_memory = g_malloc(sizeof(*system_memory));
1841
1842 memory_region_init(system_memory, NULL, "system", UINT64_MAX);
1843 address_space_init(&address_space_memory, system_memory, "memory");
1844
1845 system_io = g_malloc(sizeof(*system_io));
1846 memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io",
1847 65536);
1848 address_space_init(&address_space_io, system_io, "I/O");
1849
1850 memory_listener_register(&core_memory_listener, &address_space_memory);
1851 if (tcg_enabled()) {
1852 memory_listener_register(&tcg_memory_listener, &address_space_memory);
1853 }
1854 }
1855
1856 MemoryRegion *get_system_memory(void)
1857 {
1858 return system_memory;
1859 }
1860
1861 MemoryRegion *get_system_io(void)
1862 {
1863 return system_io;
1864 }
1865
1866 #endif /* !defined(CONFIG_USER_ONLY) */
1867
1868 /* physical memory access (slow version, mainly for debug) */
1869 #if defined(CONFIG_USER_ONLY)
1870 int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
1871 uint8_t *buf, int len, int is_write)
1872 {
1873 int l, flags;
1874 target_ulong page;
1875 void * p;
1876
1877 while (len > 0) {
1878 page = addr & TARGET_PAGE_MASK;
1879 l = (page + TARGET_PAGE_SIZE) - addr;
1880 if (l > len)
1881 l = len;
1882 flags = page_get_flags(page);
1883 if (!(flags & PAGE_VALID))
1884 return -1;
1885 if (is_write) {
1886 if (!(flags & PAGE_WRITE))
1887 return -1;
1888 /* XXX: this code should not depend on lock_user */
1889 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
1890 return -1;
1891 memcpy(p, buf, l);
1892 unlock_user(p, addr, l);
1893 } else {
1894 if (!(flags & PAGE_READ))
1895 return -1;
1896 /* XXX: this code should not depend on lock_user */
1897 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
1898 return -1;
1899 memcpy(buf, p, l);
1900 unlock_user(p, addr, 0);
1901 }
1902 len -= l;
1903 buf += l;
1904 addr += l;
1905 }
1906 return 0;
1907 }
1908
1909 #else
1910
1911 static void invalidate_and_set_dirty(hwaddr addr,
1912 hwaddr length)
1913 {
1914 if (!cpu_physical_memory_is_dirty(addr)) {
1915 /* invalidate code */
1916 tb_invalidate_phys_page_range(addr, addr + length, 0);
1917 /* set dirty bit */
1918 cpu_physical_memory_set_dirty_flags(addr, (0xff & ~CODE_DIRTY_FLAG));
1919 }
1920 xen_modified_memory(addr, length);
1921 }
1922
1923 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
1924 {
1925 if (memory_region_is_ram(mr)) {
1926 return !(is_write && mr->readonly);
1927 }
1928 if (memory_region_is_romd(mr)) {
1929 return !is_write;
1930 }
1931
1932 return false;
1933 }
1934
1935 static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr)
1936 {
1937 unsigned access_size_max = mr->ops->valid.max_access_size;
1938
1939 /* Regions are assumed to support 1-4 byte accesses unless
1940 otherwise specified. */
1941 if (access_size_max == 0) {
1942 access_size_max = 4;
1943 }
1944
1945 /* Bound the maximum access by the alignment of the address. */
1946 if (!mr->ops->impl.unaligned) {
1947 unsigned align_size_max = addr & -addr;
1948 if (align_size_max != 0 && align_size_max < access_size_max) {
1949 access_size_max = align_size_max;
1950 }
1951 }
1952
1953 /* Don't attempt accesses larger than the maximum. */
1954 if (l > access_size_max) {
1955 l = access_size_max;
1956 }
1957 if (l & (l - 1)) {
1958 l = 1 << (qemu_fls(l) - 1);
1959 }
1960
1961 return l;
1962 }
1963
1964 bool address_space_rw(AddressSpace *as, hwaddr addr, uint8_t *buf,
1965 int len, bool is_write)
1966 {
1967 hwaddr l;
1968 uint8_t *ptr;
1969 uint64_t val;
1970 hwaddr addr1;
1971 MemoryRegion *mr;
1972 bool error = false;
1973
1974 while (len > 0) {
1975 l = len;
1976 mr = address_space_translate(as, addr, &addr1, &l, is_write);
1977
1978 if (is_write) {
1979 if (!memory_access_is_direct(mr, is_write)) {
1980 l = memory_access_size(mr, l, addr1);
1981 /* XXX: could force current_cpu to NULL to avoid
1982 potential bugs */
1983 switch (l) {
1984 case 8:
1985 /* 64 bit write access */
1986 val = ldq_p(buf);
1987 error |= io_mem_write(mr, addr1, val, 8);
1988 break;
1989 case 4:
1990 /* 32 bit write access */
1991 val = ldl_p(buf);
1992 error |= io_mem_write(mr, addr1, val, 4);
1993 break;
1994 case 2:
1995 /* 16 bit write access */
1996 val = lduw_p(buf);
1997 error |= io_mem_write(mr, addr1, val, 2);
1998 break;
1999 case 1:
2000 /* 8 bit write access */
2001 val = ldub_p(buf);
2002 error |= io_mem_write(mr, addr1, val, 1);
2003 break;
2004 default:
2005 abort();
2006 }
2007 } else {
2008 addr1 += memory_region_get_ram_addr(mr);
2009 /* RAM case */
2010 ptr = qemu_get_ram_ptr(addr1);
2011 memcpy(ptr, buf, l);
2012 invalidate_and_set_dirty(addr1, l);
2013 }
2014 } else {
2015 if (!memory_access_is_direct(mr, is_write)) {
2016 /* I/O case */
2017 l = memory_access_size(mr, l, addr1);
2018 switch (l) {
2019 case 8:
2020 /* 64 bit read access */
2021 error |= io_mem_read(mr, addr1, &val, 8);
2022 stq_p(buf, val);
2023 break;
2024 case 4:
2025 /* 32 bit read access */
2026 error |= io_mem_read(mr, addr1, &val, 4);
2027 stl_p(buf, val);
2028 break;
2029 case 2:
2030 /* 16 bit read access */
2031 error |= io_mem_read(mr, addr1, &val, 2);
2032 stw_p(buf, val);
2033 break;
2034 case 1:
2035 /* 8 bit read access */
2036 error |= io_mem_read(mr, addr1, &val, 1);
2037 stb_p(buf, val);
2038 break;
2039 default:
2040 abort();
2041 }
2042 } else {
2043 /* RAM case */
2044 ptr = qemu_get_ram_ptr(mr->ram_addr + addr1);
2045 memcpy(buf, ptr, l);
2046 }
2047 }
2048 len -= l;
2049 buf += l;
2050 addr += l;
2051 }
2052
2053 return error;
2054 }
2055
2056 bool address_space_write(AddressSpace *as, hwaddr addr,
2057 const uint8_t *buf, int len)
2058 {
2059 return address_space_rw(as, addr, (uint8_t *)buf, len, true);
2060 }
2061
2062 bool address_space_read(AddressSpace *as, hwaddr addr, uint8_t *buf, int len)
2063 {
2064 return address_space_rw(as, addr, buf, len, false);
2065 }
2066
2067
2068 void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
2069 int len, int is_write)
2070 {
2071 address_space_rw(&address_space_memory, addr, buf, len, is_write);
2072 }
2073
2074 enum write_rom_type {
2075 WRITE_DATA,
2076 FLUSH_CACHE,
2077 };
2078
2079 static inline void cpu_physical_memory_write_rom_internal(
2080 hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type)
2081 {
2082 hwaddr l;
2083 uint8_t *ptr;
2084 hwaddr addr1;
2085 MemoryRegion *mr;
2086
2087 while (len > 0) {
2088 l = len;
2089 mr = address_space_translate(&address_space_memory,
2090 addr, &addr1, &l, true);
2091
2092 if (!(memory_region_is_ram(mr) ||
2093 memory_region_is_romd(mr))) {
2094 /* do nothing */
2095 } else {
2096 addr1 += memory_region_get_ram_addr(mr);
2097 /* ROM/RAM case */
2098 ptr = qemu_get_ram_ptr(addr1);
2099 switch (type) {
2100 case WRITE_DATA:
2101 memcpy(ptr, buf, l);
2102 invalidate_and_set_dirty(addr1, l);
2103 break;
2104 case FLUSH_CACHE:
2105 flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l);
2106 break;
2107 }
2108 }
2109 len -= l;
2110 buf += l;
2111 addr += l;
2112 }
2113 }
2114
2115 /* used for ROM loading : can write in RAM and ROM */
2116 void cpu_physical_memory_write_rom(hwaddr addr,
2117 const uint8_t *buf, int len)
2118 {
2119 cpu_physical_memory_write_rom_internal(addr, buf, len, WRITE_DATA);
2120 }
2121
2122 void cpu_flush_icache_range(hwaddr start, int len)
2123 {
2124 /*
2125 * This function should do the same thing as an icache flush that was
2126 * triggered from within the guest. For TCG we are always cache coherent,
2127 * so there is no need to flush anything. For KVM / Xen we need to flush
2128 * the host's instruction cache at least.
2129 */
2130 if (tcg_enabled()) {
2131 return;
2132 }
2133
2134 cpu_physical_memory_write_rom_internal(start, NULL, len, FLUSH_CACHE);
2135 }
2136
2137 typedef struct {
2138 MemoryRegion *mr;
2139 void *buffer;
2140 hwaddr addr;
2141 hwaddr len;
2142 } BounceBuffer;
2143
2144 static BounceBuffer bounce;
2145
2146 typedef struct MapClient {
2147 void *opaque;
2148 void (*callback)(void *opaque);
2149 QLIST_ENTRY(MapClient) link;
2150 } MapClient;
2151
2152 static QLIST_HEAD(map_client_list, MapClient) map_client_list
2153 = QLIST_HEAD_INITIALIZER(map_client_list);
2154
2155 void *cpu_register_map_client(void *opaque, void (*callback)(void *opaque))
2156 {
2157 MapClient *client = g_malloc(sizeof(*client));
2158
2159 client->opaque = opaque;
2160 client->callback = callback;
2161 QLIST_INSERT_HEAD(&map_client_list, client, link);
2162 return client;
2163 }
2164
2165 static void cpu_unregister_map_client(void *_client)
2166 {
2167 MapClient *client = (MapClient *)_client;
2168
2169 QLIST_REMOVE(client, link);
2170 g_free(client);
2171 }
2172
2173 static void cpu_notify_map_clients(void)
2174 {
2175 MapClient *client;
2176
2177 while (!QLIST_EMPTY(&map_client_list)) {
2178 client = QLIST_FIRST(&map_client_list);
2179 client->callback(client->opaque);
2180 cpu_unregister_map_client(client);
2181 }
2182 }
2183
2184 bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write)
2185 {
2186 MemoryRegion *mr;
2187 hwaddr l, xlat;
2188
2189 while (len > 0) {
2190 l = len;
2191 mr = address_space_translate(as, addr, &xlat, &l, is_write);
2192 if (!memory_access_is_direct(mr, is_write)) {
2193 l = memory_access_size(mr, l, addr);
2194 if (!memory_region_access_valid(mr, xlat, l, is_write)) {
2195 return false;
2196 }
2197 }
2198
2199 len -= l;
2200 addr += l;
2201 }
2202 return true;
2203 }
2204
2205 /* Map a physical memory region into a host virtual address.
2206 * May map a subset of the requested range, given by and returned in *plen.
2207 * May return NULL if resources needed to perform the mapping are exhausted.
2208 * Use only for reads OR writes - not for read-modify-write operations.
2209 * Use cpu_register_map_client() to know when retrying the map operation is
2210 * likely to succeed.
2211 */
2212 void *address_space_map(AddressSpace *as,
2213 hwaddr addr,
2214 hwaddr *plen,
2215 bool is_write)
2216 {
2217 hwaddr len = *plen;
2218 hwaddr done = 0;
2219 hwaddr l, xlat, base;
2220 MemoryRegion *mr, *this_mr;
2221 ram_addr_t raddr;
2222
2223 if (len == 0) {
2224 return NULL;
2225 }
2226
2227 l = len;
2228 mr = address_space_translate(as, addr, &xlat, &l, is_write);
2229 if (!memory_access_is_direct(mr, is_write)) {
2230 if (bounce.buffer) {
2231 return NULL;
2232 }
2233 /* Avoid unbounded allocations */
2234 l = MIN(l, TARGET_PAGE_SIZE);
2235 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l);
2236 bounce.addr = addr;
2237 bounce.len = l;
2238
2239 memory_region_ref(mr);
2240 bounce.mr = mr;
2241 if (!is_write) {
2242 address_space_read(as, addr, bounce.buffer, l);
2243 }
2244
2245 *plen = l;
2246 return bounce.buffer;
2247 }
2248
2249 base = xlat;
2250 raddr = memory_region_get_ram_addr(mr);
2251
2252 for (;;) {
2253 len -= l;
2254 addr += l;
2255 done += l;
2256 if (len == 0) {
2257 break;
2258 }
2259
2260 l = len;
2261 this_mr = address_space_translate(as, addr, &xlat, &l, is_write);
2262 if (this_mr != mr || xlat != base + done) {
2263 break;
2264 }
2265 }
2266
2267 memory_region_ref(mr);
2268 *plen = done;
2269 return qemu_ram_ptr_length(raddr + base, plen);
2270 }
2271
2272 /* Unmaps a memory region previously mapped by address_space_map().
2273 * Will also mark the memory as dirty if is_write == 1. access_len gives
2274 * the amount of memory that was actually read or written by the caller.
2275 */
2276 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2277 int is_write, hwaddr access_len)
2278 {
2279 if (buffer != bounce.buffer) {
2280 MemoryRegion *mr;
2281 ram_addr_t addr1;
2282
2283 mr = qemu_ram_addr_from_host(buffer, &addr1);
2284 assert(mr != NULL);
2285 if (is_write) {
2286 while (access_len) {
2287 unsigned l;
2288 l = TARGET_PAGE_SIZE;
2289 if (l > access_len)
2290 l = access_len;
2291 invalidate_and_set_dirty(addr1, l);
2292 addr1 += l;
2293 access_len -= l;
2294 }
2295 }
2296 if (xen_enabled()) {
2297 xen_invalidate_map_cache_entry(buffer);
2298 }
2299 memory_region_unref(mr);
2300 return;
2301 }
2302 if (is_write) {
2303 address_space_write(as, bounce.addr, bounce.buffer, access_len);
2304 }
2305 qemu_vfree(bounce.buffer);
2306 bounce.buffer = NULL;
2307 memory_region_unref(bounce.mr);
2308 cpu_notify_map_clients();
2309 }
2310
2311 void *cpu_physical_memory_map(hwaddr addr,
2312 hwaddr *plen,
2313 int is_write)
2314 {
2315 return address_space_map(&address_space_memory, addr, plen, is_write);
2316 }
2317
2318 void cpu_physical_memory_unmap(void *buffer, hwaddr len,
2319 int is_write, hwaddr access_len)
2320 {
2321 return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len);
2322 }
2323
2324 /* warning: addr must be aligned */
2325 static inline uint32_t ldl_phys_internal(hwaddr addr,
2326 enum device_endian endian)
2327 {
2328 uint8_t *ptr;
2329 uint64_t val;
2330 MemoryRegion *mr;
2331 hwaddr l = 4;
2332 hwaddr addr1;
2333
2334 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2335 false);
2336 if (l < 4 || !memory_access_is_direct(mr, false)) {
2337 /* I/O case */
2338 io_mem_read(mr, addr1, &val, 4);
2339 #if defined(TARGET_WORDS_BIGENDIAN)
2340 if (endian == DEVICE_LITTLE_ENDIAN) {
2341 val = bswap32(val);
2342 }
2343 #else
2344 if (endian == DEVICE_BIG_ENDIAN) {
2345 val = bswap32(val);
2346 }
2347 #endif
2348 } else {
2349 /* RAM case */
2350 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
2351 & TARGET_PAGE_MASK)
2352 + addr1);
2353 switch (endian) {
2354 case DEVICE_LITTLE_ENDIAN:
2355 val = ldl_le_p(ptr);
2356 break;
2357 case DEVICE_BIG_ENDIAN:
2358 val = ldl_be_p(ptr);
2359 break;
2360 default:
2361 val = ldl_p(ptr);
2362 break;
2363 }
2364 }
2365 return val;
2366 }
2367
2368 uint32_t ldl_phys(hwaddr addr)
2369 {
2370 return ldl_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
2371 }
2372
2373 uint32_t ldl_le_phys(hwaddr addr)
2374 {
2375 return ldl_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
2376 }
2377
2378 uint32_t ldl_be_phys(hwaddr addr)
2379 {
2380 return ldl_phys_internal(addr, DEVICE_BIG_ENDIAN);
2381 }
2382
2383 /* warning: addr must be aligned */
2384 static inline uint64_t ldq_phys_internal(hwaddr addr,
2385 enum device_endian endian)
2386 {
2387 uint8_t *ptr;
2388 uint64_t val;
2389 MemoryRegion *mr;
2390 hwaddr l = 8;
2391 hwaddr addr1;
2392
2393 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2394 false);
2395 if (l < 8 || !memory_access_is_direct(mr, false)) {
2396 /* I/O case */
2397 io_mem_read(mr, addr1, &val, 8);
2398 #if defined(TARGET_WORDS_BIGENDIAN)
2399 if (endian == DEVICE_LITTLE_ENDIAN) {
2400 val = bswap64(val);
2401 }
2402 #else
2403 if (endian == DEVICE_BIG_ENDIAN) {
2404 val = bswap64(val);
2405 }
2406 #endif
2407 } else {
2408 /* RAM case */
2409 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
2410 & TARGET_PAGE_MASK)
2411 + addr1);
2412 switch (endian) {
2413 case DEVICE_LITTLE_ENDIAN:
2414 val = ldq_le_p(ptr);
2415 break;
2416 case DEVICE_BIG_ENDIAN:
2417 val = ldq_be_p(ptr);
2418 break;
2419 default:
2420 val = ldq_p(ptr);
2421 break;
2422 }
2423 }
2424 return val;
2425 }
2426
2427 uint64_t ldq_phys(hwaddr addr)
2428 {
2429 return ldq_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
2430 }
2431
2432 uint64_t ldq_le_phys(hwaddr addr)
2433 {
2434 return ldq_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
2435 }
2436
2437 uint64_t ldq_be_phys(hwaddr addr)
2438 {
2439 return ldq_phys_internal(addr, DEVICE_BIG_ENDIAN);
2440 }
2441
2442 /* XXX: optimize */
2443 uint32_t ldub_phys(hwaddr addr)
2444 {
2445 uint8_t val;
2446 cpu_physical_memory_read(addr, &val, 1);
2447 return val;
2448 }
2449
2450 /* warning: addr must be aligned */
2451 static inline uint32_t lduw_phys_internal(hwaddr addr,
2452 enum device_endian endian)
2453 {
2454 uint8_t *ptr;
2455 uint64_t val;
2456 MemoryRegion *mr;
2457 hwaddr l = 2;
2458 hwaddr addr1;
2459
2460 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2461 false);
2462 if (l < 2 || !memory_access_is_direct(mr, false)) {
2463 /* I/O case */
2464 io_mem_read(mr, addr1, &val, 2);
2465 #if defined(TARGET_WORDS_BIGENDIAN)
2466 if (endian == DEVICE_LITTLE_ENDIAN) {
2467 val = bswap16(val);
2468 }
2469 #else
2470 if (endian == DEVICE_BIG_ENDIAN) {
2471 val = bswap16(val);
2472 }
2473 #endif
2474 } else {
2475 /* RAM case */
2476 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
2477 & TARGET_PAGE_MASK)
2478 + addr1);
2479 switch (endian) {
2480 case DEVICE_LITTLE_ENDIAN:
2481 val = lduw_le_p(ptr);
2482 break;
2483 case DEVICE_BIG_ENDIAN:
2484 val = lduw_be_p(ptr);
2485 break;
2486 default:
2487 val = lduw_p(ptr);
2488 break;
2489 }
2490 }
2491 return val;
2492 }
2493
2494 uint32_t lduw_phys(hwaddr addr)
2495 {
2496 return lduw_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
2497 }
2498
2499 uint32_t lduw_le_phys(hwaddr addr)
2500 {
2501 return lduw_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
2502 }
2503
2504 uint32_t lduw_be_phys(hwaddr addr)
2505 {
2506 return lduw_phys_internal(addr, DEVICE_BIG_ENDIAN);
2507 }
2508
2509 /* warning: addr must be aligned. The ram page is not masked as dirty
2510 and the code inside is not invalidated. It is useful if the dirty
2511 bits are used to track modified PTEs */
2512 void stl_phys_notdirty(hwaddr addr, uint32_t val)
2513 {
2514 uint8_t *ptr;
2515 MemoryRegion *mr;
2516 hwaddr l = 4;
2517 hwaddr addr1;
2518
2519 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2520 true);
2521 if (l < 4 || !memory_access_is_direct(mr, true)) {
2522 io_mem_write(mr, addr1, val, 4);
2523 } else {
2524 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
2525 ptr = qemu_get_ram_ptr(addr1);
2526 stl_p(ptr, val);
2527
2528 if (unlikely(in_migration)) {
2529 if (!cpu_physical_memory_is_dirty(addr1)) {
2530 /* invalidate code */
2531 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
2532 /* set dirty bit */
2533 cpu_physical_memory_set_dirty_flags(
2534 addr1, (0xff & ~CODE_DIRTY_FLAG));
2535 }
2536 }
2537 }
2538 }
2539
2540 /* warning: addr must be aligned */
2541 static inline void stl_phys_internal(hwaddr addr, uint32_t val,
2542 enum device_endian endian)
2543 {
2544 uint8_t *ptr;
2545 MemoryRegion *mr;
2546 hwaddr l = 4;
2547 hwaddr addr1;
2548
2549 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2550 true);
2551 if (l < 4 || !memory_access_is_direct(mr, true)) {
2552 #if defined(TARGET_WORDS_BIGENDIAN)
2553 if (endian == DEVICE_LITTLE_ENDIAN) {
2554 val = bswap32(val);
2555 }
2556 #else
2557 if (endian == DEVICE_BIG_ENDIAN) {
2558 val = bswap32(val);
2559 }
2560 #endif
2561 io_mem_write(mr, addr1, val, 4);
2562 } else {
2563 /* RAM case */
2564 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
2565 ptr = qemu_get_ram_ptr(addr1);
2566 switch (endian) {
2567 case DEVICE_LITTLE_ENDIAN:
2568 stl_le_p(ptr, val);
2569 break;
2570 case DEVICE_BIG_ENDIAN:
2571 stl_be_p(ptr, val);
2572 break;
2573 default:
2574 stl_p(ptr, val);
2575 break;
2576 }
2577 invalidate_and_set_dirty(addr1, 4);
2578 }
2579 }
2580
2581 void stl_phys(hwaddr addr, uint32_t val)
2582 {
2583 stl_phys_internal(addr, val, DEVICE_NATIVE_ENDIAN);
2584 }
2585
2586 void stl_le_phys(hwaddr addr, uint32_t val)
2587 {
2588 stl_phys_internal(addr, val, DEVICE_LITTLE_ENDIAN);
2589 }
2590
2591 void stl_be_phys(hwaddr addr, uint32_t val)
2592 {
2593 stl_phys_internal(addr, val, DEVICE_BIG_ENDIAN);
2594 }
2595
2596 /* XXX: optimize */
2597 void stb_phys(hwaddr addr, uint32_t val)
2598 {
2599 uint8_t v = val;
2600 cpu_physical_memory_write(addr, &v, 1);
2601 }
2602
2603 /* warning: addr must be aligned */
2604 static inline void stw_phys_internal(hwaddr addr, uint32_t val,
2605 enum device_endian endian)
2606 {
2607 uint8_t *ptr;
2608 MemoryRegion *mr;
2609 hwaddr l = 2;
2610 hwaddr addr1;
2611
2612 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2613 true);
2614 if (l < 2 || !memory_access_is_direct(mr, true)) {
2615 #if defined(TARGET_WORDS_BIGENDIAN)
2616 if (endian == DEVICE_LITTLE_ENDIAN) {
2617 val = bswap16(val);
2618 }
2619 #else
2620 if (endian == DEVICE_BIG_ENDIAN) {
2621 val = bswap16(val);
2622 }
2623 #endif
2624 io_mem_write(mr, addr1, val, 2);
2625 } else {
2626 /* RAM case */
2627 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
2628 ptr = qemu_get_ram_ptr(addr1);
2629 switch (endian) {
2630 case DEVICE_LITTLE_ENDIAN:
2631 stw_le_p(ptr, val);
2632 break;
2633 case DEVICE_BIG_ENDIAN:
2634 stw_be_p(ptr, val);
2635 break;
2636 default:
2637 stw_p(ptr, val);
2638 break;
2639 }
2640 invalidate_and_set_dirty(addr1, 2);
2641 }
2642 }
2643
2644 void stw_phys(hwaddr addr, uint32_t val)
2645 {
2646 stw_phys_internal(addr, val, DEVICE_NATIVE_ENDIAN);
2647 }
2648
2649 void stw_le_phys(hwaddr addr, uint32_t val)
2650 {
2651 stw_phys_internal(addr, val, DEVICE_LITTLE_ENDIAN);
2652 }
2653
2654 void stw_be_phys(hwaddr addr, uint32_t val)
2655 {
2656 stw_phys_internal(addr, val, DEVICE_BIG_ENDIAN);
2657 }
2658
2659 /* XXX: optimize */
2660 void stq_phys(hwaddr addr, uint64_t val)
2661 {
2662 val = tswap64(val);
2663 cpu_physical_memory_write(addr, &val, 8);
2664 }
2665
2666 void stq_le_phys(hwaddr addr, uint64_t val)
2667 {
2668 val = cpu_to_le64(val);
2669 cpu_physical_memory_write(addr, &val, 8);
2670 }
2671
2672 void stq_be_phys(hwaddr addr, uint64_t val)
2673 {
2674 val = cpu_to_be64(val);
2675 cpu_physical_memory_write(addr, &val, 8);
2676 }
2677
2678 /* virtual memory access for debug (includes writing to ROM) */
2679 int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
2680 uint8_t *buf, int len, int is_write)
2681 {
2682 int l;
2683 hwaddr phys_addr;
2684 target_ulong page;
2685
2686 while (len > 0) {
2687 page = addr & TARGET_PAGE_MASK;
2688 phys_addr = cpu_get_phys_page_debug(cpu, page);
2689 /* if no physical page mapped, return an error */
2690 if (phys_addr == -1)
2691 return -1;
2692 l = (page + TARGET_PAGE_SIZE) - addr;
2693 if (l > len)
2694 l = len;
2695 phys_addr += (addr & ~TARGET_PAGE_MASK);
2696 if (is_write)
2697 cpu_physical_memory_write_rom(phys_addr, buf, l);
2698 else
2699 cpu_physical_memory_rw(phys_addr, buf, l, is_write);
2700 len -= l;
2701 buf += l;
2702 addr += l;
2703 }
2704 return 0;
2705 }
2706 #endif
2707
2708 #if !defined(CONFIG_USER_ONLY)
2709
2710 /*
2711 * A helper function for the _utterly broken_ virtio device model to find out if
2712 * it's running on a big endian machine. Don't do this at home kids!
2713 */
2714 bool virtio_is_big_endian(void);
2715 bool virtio_is_big_endian(void)
2716 {
2717 #if defined(TARGET_WORDS_BIGENDIAN)
2718 return true;
2719 #else
2720 return false;
2721 #endif
2722 }
2723
2724 #endif
2725
2726 #ifndef CONFIG_USER_ONLY
2727 bool cpu_physical_memory_is_io(hwaddr phys_addr)
2728 {
2729 MemoryRegion*mr;
2730 hwaddr l = 1;
2731
2732 mr = address_space_translate(&address_space_memory,
2733 phys_addr, &phys_addr, &l, false);
2734
2735 return !(memory_region_is_ram(mr) ||
2736 memory_region_is_romd(mr));
2737 }
2738
2739 void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
2740 {
2741 RAMBlock *block;
2742
2743 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
2744 func(block->host, block->offset, block->length, opaque);
2745 }
2746 }
2747 #endif
QEmu