4 * Copyright (c) 2003 Fabrice Bellard
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.
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.
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/>.
23 #include <sys/types.h>
27 #include "qemu-common.h"
32 #include "qemu/osdep.h"
33 #include "sysemu/kvm.h"
34 #include "hw/xen/xen.h"
35 #include "qemu/timer.h"
36 #include "qemu/config-file.h"
37 #include "exec/memory.h"
38 #include "sysemu/dma.h"
39 #include "exec/address-spaces.h"
40 #if defined(CONFIG_USER_ONLY)
42 #else /* !CONFIG_USER_ONLY */
43 #include "sysemu/xen-mapcache.h"
46 #include "exec/cpu-all.h"
48 #include "exec/cputlb.h"
49 #include "translate-all.h"
51 #include "exec/memory-internal.h"
53 //#define DEBUG_SUBPAGE
55 #if !defined(CONFIG_USER_ONLY)
57 static int in_migration;
59 RAMList ram_list = { .blocks = QTAILQ_HEAD_INITIALIZER(ram_list.blocks) };
61 static MemoryRegion *system_memory;
62 static MemoryRegion *system_io;
64 AddressSpace address_space_io;
65 AddressSpace address_space_memory;
67 MemoryRegion io_mem_rom, io_mem_notdirty;
68 static MemoryRegion io_mem_unassigned;
73 /* current CPU in the current thread. It is only valid inside
75 DEFINE_TLS(CPUState *, current_cpu);
76 /* 0 = Do not count executed instructions.
77 1 = Precise instruction counting.
78 2 = Adaptive rate instruction counting. */
81 #if !defined(CONFIG_USER_ONLY)
83 typedef struct PhysPageEntry PhysPageEntry;
85 struct PhysPageEntry {
87 /* index into phys_sections (is_leaf) or phys_map_nodes (!is_leaf) */
91 typedef PhysPageEntry Node[L2_SIZE];
93 struct AddressSpaceDispatch {
94 /* This is a multi-level map on the physical address space.
95 * The bottom level has pointers to MemoryRegionSections.
97 PhysPageEntry phys_map;
99 MemoryRegionSection *sections;
103 #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
104 typedef struct subpage_t {
108 uint16_t sub_section[TARGET_PAGE_SIZE];
111 #define PHYS_SECTION_UNASSIGNED 0
112 #define PHYS_SECTION_NOTDIRTY 1
113 #define PHYS_SECTION_ROM 2
114 #define PHYS_SECTION_WATCH 3
116 typedef struct PhysPageMap {
117 unsigned sections_nb;
118 unsigned sections_nb_alloc;
120 unsigned nodes_nb_alloc;
122 MemoryRegionSection *sections;
125 static PhysPageMap *prev_map;
126 static PhysPageMap next_map;
128 #define PHYS_MAP_NODE_NIL (((uint16_t)~0) >> 1)
130 static void io_mem_init(void);
131 static void memory_map_init(void);
132 static void *qemu_safe_ram_ptr(ram_addr_t addr);
134 static MemoryRegion io_mem_watch;
137 #if !defined(CONFIG_USER_ONLY)
139 static void phys_map_node_reserve(unsigned nodes)
141 if (next_map.nodes_nb + nodes > next_map.nodes_nb_alloc) {
142 next_map.nodes_nb_alloc = MAX(next_map.nodes_nb_alloc * 2,
144 next_map.nodes_nb_alloc = MAX(next_map.nodes_nb_alloc,
145 next_map.nodes_nb + nodes);
146 next_map.nodes = g_renew(Node, next_map.nodes,
147 next_map.nodes_nb_alloc);
151 static uint16_t phys_map_node_alloc(void)
156 ret = next_map.nodes_nb++;
157 assert(ret != PHYS_MAP_NODE_NIL);
158 assert(ret != next_map.nodes_nb_alloc);
159 for (i = 0; i < L2_SIZE; ++i) {
160 next_map.nodes[ret][i].is_leaf = 0;
161 next_map.nodes[ret][i].ptr = PHYS_MAP_NODE_NIL;
166 static void phys_page_set_level(PhysPageEntry *lp, hwaddr *index,
167 hwaddr *nb, uint16_t leaf,
172 hwaddr step = (hwaddr)1 << (level * L2_BITS);
174 if (!lp->is_leaf && lp->ptr == PHYS_MAP_NODE_NIL) {
175 lp->ptr = phys_map_node_alloc();
176 p = next_map.nodes[lp->ptr];
178 for (i = 0; i < L2_SIZE; i++) {
180 p[i].ptr = PHYS_SECTION_UNASSIGNED;
184 p = next_map.nodes[lp->ptr];
186 lp = &p[(*index >> (level * L2_BITS)) & (L2_SIZE - 1)];
188 while (*nb && lp < &p[L2_SIZE]) {
189 if ((*index & (step - 1)) == 0 && *nb >= step) {
195 phys_page_set_level(lp, index, nb, leaf, level - 1);
201 static void phys_page_set(AddressSpaceDispatch *d,
202 hwaddr index, hwaddr nb,
205 /* Wildly overreserve - it doesn't matter much. */
206 phys_map_node_reserve(3 * P_L2_LEVELS);
208 phys_page_set_level(&d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1);
211 static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr index,
212 Node *nodes, MemoryRegionSection *sections)
217 for (i = P_L2_LEVELS - 1; i >= 0 && !lp.is_leaf; i--) {
218 if (lp.ptr == PHYS_MAP_NODE_NIL) {
219 return §ions[PHYS_SECTION_UNASSIGNED];
222 lp = p[(index >> (i * L2_BITS)) & (L2_SIZE - 1)];
224 return §ions[lp.ptr];
227 bool memory_region_is_unassigned(MemoryRegion *mr)
229 return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device
230 && mr != &io_mem_watch;
233 static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
235 bool resolve_subpage)
237 MemoryRegionSection *section;
240 section = phys_page_find(d->phys_map, addr >> TARGET_PAGE_BITS,
241 d->nodes, d->sections);
242 if (resolve_subpage && section->mr->subpage) {
243 subpage = container_of(section->mr, subpage_t, iomem);
244 section = &d->sections[subpage->sub_section[SUBPAGE_IDX(addr)]];
249 static MemoryRegionSection *
250 address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
251 hwaddr *plen, bool resolve_subpage)
253 MemoryRegionSection *section;
256 section = address_space_lookup_region(d, addr, resolve_subpage);
257 /* Compute offset within MemoryRegionSection */
258 addr -= section->offset_within_address_space;
260 /* Compute offset within MemoryRegion */
261 *xlat = addr + section->offset_within_region;
263 diff = int128_sub(section->mr->size, int128_make64(addr));
264 *plen = int128_get64(int128_min(diff, int128_make64(*plen)));
268 MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
269 hwaddr *xlat, hwaddr *plen,
273 MemoryRegionSection *section;
278 section = address_space_translate_internal(as->dispatch, addr, &addr, plen, true);
281 if (!mr->iommu_ops) {
285 iotlb = mr->iommu_ops->translate(mr, addr);
286 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
287 | (addr & iotlb.addr_mask));
288 len = MIN(len, (addr | iotlb.addr_mask) - addr + 1);
289 if (!(iotlb.perm & (1 << is_write))) {
290 mr = &io_mem_unassigned;
294 as = iotlb.target_as;
302 MemoryRegionSection *
303 address_space_translate_for_iotlb(AddressSpace *as, hwaddr addr, hwaddr *xlat,
306 MemoryRegionSection *section;
307 section = address_space_translate_internal(as->dispatch, addr, xlat, plen, false);
309 assert(!section->mr->iommu_ops);
314 void cpu_exec_init_all(void)
316 #if !defined(CONFIG_USER_ONLY)
317 qemu_mutex_init(&ram_list.mutex);
323 #if !defined(CONFIG_USER_ONLY)
325 static int cpu_common_post_load(void *opaque, int version_id)
327 CPUState *cpu = opaque;
329 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
330 version_id is increased. */
331 cpu->interrupt_request &= ~0x01;
332 tlb_flush(cpu->env_ptr, 1);
337 const VMStateDescription vmstate_cpu_common = {
338 .name = "cpu_common",
340 .minimum_version_id = 1,
341 .minimum_version_id_old = 1,
342 .post_load = cpu_common_post_load,
343 .fields = (VMStateField []) {
344 VMSTATE_UINT32(halted, CPUState),
345 VMSTATE_UINT32(interrupt_request, CPUState),
346 VMSTATE_END_OF_LIST()
352 CPUState *qemu_get_cpu(int index)
354 CPUState *cpu = first_cpu;
357 if (cpu->cpu_index == index) {
366 void qemu_for_each_cpu(void (*func)(CPUState *cpu, void *data), void *data)
377 void cpu_exec_init(CPUArchState *env)
379 CPUState *cpu = ENV_GET_CPU(env);
380 CPUClass *cc = CPU_GET_CLASS(cpu);
384 #if defined(CONFIG_USER_ONLY)
387 cpu->next_cpu = NULL;
390 while (*pcpu != NULL) {
391 pcpu = &(*pcpu)->next_cpu;
394 cpu->cpu_index = cpu_index;
396 QTAILQ_INIT(&env->breakpoints);
397 QTAILQ_INIT(&env->watchpoints);
398 #ifndef CONFIG_USER_ONLY
399 cpu->thread_id = qemu_get_thread_id();
402 #if defined(CONFIG_USER_ONLY)
405 vmstate_register(NULL, cpu_index, &vmstate_cpu_common, cpu);
406 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
407 register_savevm(NULL, "cpu", cpu_index, CPU_SAVE_VERSION,
408 cpu_save, cpu_load, env);
409 assert(cc->vmsd == NULL);
411 if (cc->vmsd != NULL) {
412 vmstate_register(NULL, cpu_index, cc->vmsd, cpu);
416 #if defined(TARGET_HAS_ICE)
417 #if defined(CONFIG_USER_ONLY)
418 static void breakpoint_invalidate(CPUArchState *env, target_ulong pc)
420 tb_invalidate_phys_page_range(pc, pc + 1, 0);
423 static void breakpoint_invalidate(CPUArchState *env, target_ulong pc)
425 tb_invalidate_phys_addr(cpu_get_phys_page_debug(env, pc) |
426 (pc & ~TARGET_PAGE_MASK));
429 #endif /* TARGET_HAS_ICE */
431 #if defined(CONFIG_USER_ONLY)
432 void cpu_watchpoint_remove_all(CPUArchState *env, int mask)
437 int cpu_watchpoint_insert(CPUArchState *env, target_ulong addr, target_ulong len,
438 int flags, CPUWatchpoint **watchpoint)
443 /* Add a watchpoint. */
444 int cpu_watchpoint_insert(CPUArchState *env, target_ulong addr, target_ulong len,
445 int flags, CPUWatchpoint **watchpoint)
447 target_ulong len_mask = ~(len - 1);
450 /* sanity checks: allow power-of-2 lengths, deny unaligned watchpoints */
451 if ((len & (len - 1)) || (addr & ~len_mask) ||
452 len == 0 || len > TARGET_PAGE_SIZE) {
453 fprintf(stderr, "qemu: tried to set invalid watchpoint at "
454 TARGET_FMT_lx ", len=" TARGET_FMT_lu "\n", addr, len);
457 wp = g_malloc(sizeof(*wp));
460 wp->len_mask = len_mask;
463 /* keep all GDB-injected watchpoints in front */
465 QTAILQ_INSERT_HEAD(&env->watchpoints, wp, entry);
467 QTAILQ_INSERT_TAIL(&env->watchpoints, wp, entry);
469 tlb_flush_page(env, addr);
476 /* Remove a specific watchpoint. */
477 int cpu_watchpoint_remove(CPUArchState *env, target_ulong addr, target_ulong len,
480 target_ulong len_mask = ~(len - 1);
483 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
484 if (addr == wp->vaddr && len_mask == wp->len_mask
485 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
486 cpu_watchpoint_remove_by_ref(env, wp);
493 /* Remove a specific watchpoint by reference. */
494 void cpu_watchpoint_remove_by_ref(CPUArchState *env, CPUWatchpoint *watchpoint)
496 QTAILQ_REMOVE(&env->watchpoints, watchpoint, entry);
498 tlb_flush_page(env, watchpoint->vaddr);
503 /* Remove all matching watchpoints. */
504 void cpu_watchpoint_remove_all(CPUArchState *env, int mask)
506 CPUWatchpoint *wp, *next;
508 QTAILQ_FOREACH_SAFE(wp, &env->watchpoints, entry, next) {
509 if (wp->flags & mask)
510 cpu_watchpoint_remove_by_ref(env, wp);
515 /* Add a breakpoint. */
516 int cpu_breakpoint_insert(CPUArchState *env, target_ulong pc, int flags,
517 CPUBreakpoint **breakpoint)
519 #if defined(TARGET_HAS_ICE)
522 bp = g_malloc(sizeof(*bp));
527 /* keep all GDB-injected breakpoints in front */
529 QTAILQ_INSERT_HEAD(&env->breakpoints, bp, entry);
531 QTAILQ_INSERT_TAIL(&env->breakpoints, bp, entry);
533 breakpoint_invalidate(env, pc);
543 /* Remove a specific breakpoint. */
544 int cpu_breakpoint_remove(CPUArchState *env, target_ulong pc, int flags)
546 #if defined(TARGET_HAS_ICE)
549 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
550 if (bp->pc == pc && bp->flags == flags) {
551 cpu_breakpoint_remove_by_ref(env, bp);
561 /* Remove a specific breakpoint by reference. */
562 void cpu_breakpoint_remove_by_ref(CPUArchState *env, CPUBreakpoint *breakpoint)
564 #if defined(TARGET_HAS_ICE)
565 QTAILQ_REMOVE(&env->breakpoints, breakpoint, entry);
567 breakpoint_invalidate(env, breakpoint->pc);
573 /* Remove all matching breakpoints. */
574 void cpu_breakpoint_remove_all(CPUArchState *env, int mask)
576 #if defined(TARGET_HAS_ICE)
577 CPUBreakpoint *bp, *next;
579 QTAILQ_FOREACH_SAFE(bp, &env->breakpoints, entry, next) {
580 if (bp->flags & mask)
581 cpu_breakpoint_remove_by_ref(env, bp);
586 /* enable or disable single step mode. EXCP_DEBUG is returned by the
587 CPU loop after each instruction */
588 void cpu_single_step(CPUArchState *env, int enabled)
590 #if defined(TARGET_HAS_ICE)
591 if (env->singlestep_enabled != enabled) {
592 env->singlestep_enabled = enabled;
594 kvm_update_guest_debug(env, 0);
596 /* must flush all the translated code to avoid inconsistencies */
597 /* XXX: only flush what is necessary */
604 void cpu_abort(CPUArchState *env, const char *fmt, ...)
606 CPUState *cpu = ENV_GET_CPU(env);
612 fprintf(stderr, "qemu: fatal: ");
613 vfprintf(stderr, fmt, ap);
614 fprintf(stderr, "\n");
615 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP);
616 if (qemu_log_enabled()) {
617 qemu_log("qemu: fatal: ");
618 qemu_log_vprintf(fmt, ap2);
620 log_cpu_state(env, CPU_DUMP_FPU | CPU_DUMP_CCOP);
626 #if defined(CONFIG_USER_ONLY)
628 struct sigaction act;
629 sigfillset(&act.sa_mask);
630 act.sa_handler = SIG_DFL;
631 sigaction(SIGABRT, &act, NULL);
637 CPUArchState *cpu_copy(CPUArchState *env)
639 CPUArchState *new_env = cpu_init(env->cpu_model_str);
640 #if defined(TARGET_HAS_ICE)
645 memcpy(new_env, env, sizeof(CPUArchState));
647 /* Clone all break/watchpoints.
648 Note: Once we support ptrace with hw-debug register access, make sure
649 BP_CPU break/watchpoints are handled correctly on clone. */
650 QTAILQ_INIT(&env->breakpoints);
651 QTAILQ_INIT(&env->watchpoints);
652 #if defined(TARGET_HAS_ICE)
653 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
654 cpu_breakpoint_insert(new_env, bp->pc, bp->flags, NULL);
656 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
657 cpu_watchpoint_insert(new_env, wp->vaddr, (~wp->len_mask) + 1,
665 #if !defined(CONFIG_USER_ONLY)
666 static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t end,
671 /* we modify the TLB cache so that the dirty bit will be set again
672 when accessing the range */
673 start1 = (uintptr_t)qemu_safe_ram_ptr(start);
674 /* Check that we don't span multiple blocks - this breaks the
675 address comparisons below. */
676 if ((uintptr_t)qemu_safe_ram_ptr(end - 1) - start1
677 != (end - 1) - start) {
680 cpu_tlb_reset_dirty_all(start1, length);
684 /* Note: start and end must be within the same ram block. */
685 void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t end,
690 start &= TARGET_PAGE_MASK;
691 end = TARGET_PAGE_ALIGN(end);
693 length = end - start;
696 cpu_physical_memory_mask_dirty_range(start, length, dirty_flags);
699 tlb_reset_dirty_range_all(start, end, length);
703 static int cpu_physical_memory_set_dirty_tracking(int enable)
706 in_migration = enable;
710 hwaddr memory_region_section_get_iotlb(CPUArchState *env,
711 MemoryRegionSection *section,
713 hwaddr paddr, hwaddr xlat,
715 target_ulong *address)
720 if (memory_region_is_ram(section->mr)) {
722 iotlb = (memory_region_get_ram_addr(section->mr) & TARGET_PAGE_MASK)
724 if (!section->readonly) {
725 iotlb |= PHYS_SECTION_NOTDIRTY;
727 iotlb |= PHYS_SECTION_ROM;
730 iotlb = section - address_space_memory.dispatch->sections;
734 /* Make accesses to pages with watchpoints go via the
735 watchpoint trap routines. */
736 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
737 if (vaddr == (wp->vaddr & TARGET_PAGE_MASK)) {
738 /* Avoid trapping reads of pages with a write breakpoint. */
739 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
740 iotlb = PHYS_SECTION_WATCH + paddr;
741 *address |= TLB_MMIO;
749 #endif /* defined(CONFIG_USER_ONLY) */
751 #if !defined(CONFIG_USER_ONLY)
753 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
755 static subpage_t *subpage_init(AddressSpace *as, hwaddr base);
757 static uint16_t phys_section_add(MemoryRegionSection *section)
759 /* The physical section number is ORed with a page-aligned
760 * pointer to produce the iotlb entries. Thus it should
761 * never overflow into the page-aligned value.
763 assert(next_map.sections_nb < TARGET_PAGE_SIZE);
765 if (next_map.sections_nb == next_map.sections_nb_alloc) {
766 next_map.sections_nb_alloc = MAX(next_map.sections_nb_alloc * 2,
768 next_map.sections = g_renew(MemoryRegionSection, next_map.sections,
769 next_map.sections_nb_alloc);
771 next_map.sections[next_map.sections_nb] = *section;
772 memory_region_ref(section->mr);
773 return next_map.sections_nb++;
776 static void phys_section_destroy(MemoryRegion *mr)
778 memory_region_unref(mr);
781 subpage_t *subpage = container_of(mr, subpage_t, iomem);
782 memory_region_destroy(&subpage->iomem);
787 static void phys_sections_free(PhysPageMap *map)
789 while (map->sections_nb > 0) {
790 MemoryRegionSection *section = &map->sections[--map->sections_nb];
791 phys_section_destroy(section->mr);
793 g_free(map->sections);
798 static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section)
801 hwaddr base = section->offset_within_address_space
803 MemoryRegionSection *existing = phys_page_find(d->phys_map, base >> TARGET_PAGE_BITS,
804 next_map.nodes, next_map.sections);
805 MemoryRegionSection subsection = {
806 .offset_within_address_space = base,
807 .size = int128_make64(TARGET_PAGE_SIZE),
811 assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
813 if (!(existing->mr->subpage)) {
814 subpage = subpage_init(d->as, base);
815 subsection.mr = &subpage->iomem;
816 phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
817 phys_section_add(&subsection));
819 subpage = container_of(existing->mr, subpage_t, iomem);
821 start = section->offset_within_address_space & ~TARGET_PAGE_MASK;
822 end = start + int128_get64(section->size) - 1;
823 subpage_register(subpage, start, end, phys_section_add(section));
827 static void register_multipage(AddressSpaceDispatch *d,
828 MemoryRegionSection *section)
830 hwaddr start_addr = section->offset_within_address_space;
831 uint16_t section_index = phys_section_add(section);
832 uint64_t num_pages = int128_get64(int128_rshift(section->size,
836 phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
839 static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
841 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
842 AddressSpaceDispatch *d = as->next_dispatch;
843 MemoryRegionSection now = *section, remain = *section;
844 Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
846 if (now.offset_within_address_space & ~TARGET_PAGE_MASK) {
847 uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space)
848 - now.offset_within_address_space;
850 now.size = int128_min(int128_make64(left), now.size);
851 register_subpage(d, &now);
853 now.size = int128_zero();
855 while (int128_ne(remain.size, now.size)) {
856 remain.size = int128_sub(remain.size, now.size);
857 remain.offset_within_address_space += int128_get64(now.size);
858 remain.offset_within_region += int128_get64(now.size);
860 if (int128_lt(remain.size, page_size)) {
861 register_subpage(d, &now);
862 } else if (remain.offset_within_region & ~TARGET_PAGE_MASK) {
863 now.size = page_size;
864 register_subpage(d, &now);
866 now.size = int128_and(now.size, int128_neg(page_size));
867 register_multipage(d, &now);
872 void qemu_flush_coalesced_mmio_buffer(void)
875 kvm_flush_coalesced_mmio_buffer();
878 void qemu_mutex_lock_ramlist(void)
880 qemu_mutex_lock(&ram_list.mutex);
883 void qemu_mutex_unlock_ramlist(void)
885 qemu_mutex_unlock(&ram_list.mutex);
888 #if defined(__linux__) && !defined(TARGET_S390X)
892 #define HUGETLBFS_MAGIC 0x958458f6
894 static long gethugepagesize(const char *path)
900 ret = statfs(path, &fs);
901 } while (ret != 0 && errno == EINTR);
908 if (fs.f_type != HUGETLBFS_MAGIC)
909 fprintf(stderr, "Warning: path not on HugeTLBFS: %s\n", path);
914 static void *file_ram_alloc(RAMBlock *block,
919 char *sanitized_name;
926 unsigned long hpagesize;
928 hpagesize = gethugepagesize(path);
933 if (memory < hpagesize) {
937 if (kvm_enabled() && !kvm_has_sync_mmu()) {
938 fprintf(stderr, "host lacks kvm mmu notifiers, -mem-path unsupported\n");
942 /* Make name safe to use with mkstemp by replacing '/' with '_'. */
943 sanitized_name = g_strdup(block->mr->name);
944 for (c = sanitized_name; *c != '\0'; c++) {
949 filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path,
951 g_free(sanitized_name);
953 fd = mkstemp(filename);
955 perror("unable to create backing store for hugepages");
962 memory = (memory+hpagesize-1) & ~(hpagesize-1);
965 * ftruncate is not supported by hugetlbfs in older
966 * hosts, so don't bother bailing out on errors.
967 * If anything goes wrong with it under other filesystems,
970 if (ftruncate(fd, memory))
974 /* NB: MAP_POPULATE won't exhaustively alloc all phys pages in the case
975 * MAP_PRIVATE is requested. For mem_prealloc we mmap as MAP_SHARED
976 * to sidestep this quirk.
978 flags = mem_prealloc ? MAP_POPULATE | MAP_SHARED : MAP_PRIVATE;
979 area = mmap(0, memory, PROT_READ | PROT_WRITE, flags, fd, 0);
981 area = mmap(0, memory, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
983 if (area == MAP_FAILED) {
984 perror("file_ram_alloc: can't mmap RAM pages");
993 static ram_addr_t find_ram_offset(ram_addr_t size)
995 RAMBlock *block, *next_block;
996 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
998 assert(size != 0); /* it would hand out same offset multiple times */
1000 if (QTAILQ_EMPTY(&ram_list.blocks))
1003 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1004 ram_addr_t end, next = RAM_ADDR_MAX;
1006 end = block->offset + block->length;
1008 QTAILQ_FOREACH(next_block, &ram_list.blocks, next) {
1009 if (next_block->offset >= end) {
1010 next = MIN(next, next_block->offset);
1013 if (next - end >= size && next - end < mingap) {
1015 mingap = next - end;
1019 if (offset == RAM_ADDR_MAX) {
1020 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
1028 ram_addr_t last_ram_offset(void)
1031 ram_addr_t last = 0;
1033 QTAILQ_FOREACH(block, &ram_list.blocks, next)
1034 last = MAX(last, block->offset + block->length);
1039 static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
1042 QemuOpts *machine_opts;
1044 /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */
1045 machine_opts = qemu_opts_find(qemu_find_opts("machine"), 0);
1047 !qemu_opt_get_bool(machine_opts, "dump-guest-core", true)) {
1048 ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP);
1050 perror("qemu_madvise");
1051 fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, "
1052 "but dump_guest_core=off specified\n");
1057 void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
1059 RAMBlock *new_block, *block;
1062 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1063 if (block->offset == addr) {
1069 assert(!new_block->idstr[0]);
1072 char *id = qdev_get_dev_path(dev);
1074 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
1078 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
1080 /* This assumes the iothread lock is taken here too. */
1081 qemu_mutex_lock_ramlist();
1082 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1083 if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
1084 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
1089 qemu_mutex_unlock_ramlist();
1092 static int memory_try_enable_merging(void *addr, size_t len)
1096 opts = qemu_opts_find(qemu_find_opts("machine"), 0);
1097 if (opts && !qemu_opt_get_bool(opts, "mem-merge", true)) {
1098 /* disabled by the user */
1102 return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE);
1105 ram_addr_t qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
1108 RAMBlock *block, *new_block;
1110 size = TARGET_PAGE_ALIGN(size);
1111 new_block = g_malloc0(sizeof(*new_block));
1113 /* This assumes the iothread lock is taken here too. */
1114 qemu_mutex_lock_ramlist();
1116 new_block->offset = find_ram_offset(size);
1118 new_block->host = host;
1119 new_block->flags |= RAM_PREALLOC_MASK;
1122 #if defined (__linux__) && !defined(TARGET_S390X)
1123 new_block->host = file_ram_alloc(new_block, size, mem_path);
1124 if (!new_block->host) {
1125 new_block->host = qemu_anon_ram_alloc(size);
1126 memory_try_enable_merging(new_block->host, size);
1129 fprintf(stderr, "-mem-path option unsupported\n");
1133 if (xen_enabled()) {
1134 xen_ram_alloc(new_block->offset, size, mr);
1135 } else if (kvm_enabled()) {
1136 /* some s390/kvm configurations have special constraints */
1137 new_block->host = kvm_ram_alloc(size);
1139 new_block->host = qemu_anon_ram_alloc(size);
1141 memory_try_enable_merging(new_block->host, size);
1144 new_block->length = size;
1146 /* Keep the list sorted from biggest to smallest block. */
1147 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1148 if (block->length < new_block->length) {
1153 QTAILQ_INSERT_BEFORE(block, new_block, next);
1155 QTAILQ_INSERT_TAIL(&ram_list.blocks, new_block, next);
1157 ram_list.mru_block = NULL;
1160 qemu_mutex_unlock_ramlist();
1162 ram_list.phys_dirty = g_realloc(ram_list.phys_dirty,
1163 last_ram_offset() >> TARGET_PAGE_BITS);
1164 memset(ram_list.phys_dirty + (new_block->offset >> TARGET_PAGE_BITS),
1165 0, size >> TARGET_PAGE_BITS);
1166 cpu_physical_memory_set_dirty_range(new_block->offset, size, 0xff);
1168 qemu_ram_setup_dump(new_block->host, size);
1169 qemu_madvise(new_block->host, size, QEMU_MADV_HUGEPAGE);
1172 kvm_setup_guest_memory(new_block->host, size);
1174 return new_block->offset;
1177 ram_addr_t qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr)
1179 return qemu_ram_alloc_from_ptr(size, NULL, mr);
1182 void qemu_ram_free_from_ptr(ram_addr_t addr)
1186 /* This assumes the iothread lock is taken here too. */
1187 qemu_mutex_lock_ramlist();
1188 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1189 if (addr == block->offset) {
1190 QTAILQ_REMOVE(&ram_list.blocks, block, next);
1191 ram_list.mru_block = NULL;
1197 qemu_mutex_unlock_ramlist();
1200 void qemu_ram_free(ram_addr_t addr)
1204 /* This assumes the iothread lock is taken here too. */
1205 qemu_mutex_lock_ramlist();
1206 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1207 if (addr == block->offset) {
1208 QTAILQ_REMOVE(&ram_list.blocks, block, next);
1209 ram_list.mru_block = NULL;
1211 if (block->flags & RAM_PREALLOC_MASK) {
1213 } else if (mem_path) {
1214 #if defined (__linux__) && !defined(TARGET_S390X)
1216 munmap(block->host, block->length);
1219 qemu_anon_ram_free(block->host, block->length);
1225 if (xen_enabled()) {
1226 xen_invalidate_map_cache_entry(block->host);
1228 qemu_anon_ram_free(block->host, block->length);
1235 qemu_mutex_unlock_ramlist();
1240 void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
1247 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1248 offset = addr - block->offset;
1249 if (offset < block->length) {
1250 vaddr = block->host + offset;
1251 if (block->flags & RAM_PREALLOC_MASK) {
1255 munmap(vaddr, length);
1257 #if defined(__linux__) && !defined(TARGET_S390X)
1260 flags |= mem_prealloc ? MAP_POPULATE | MAP_SHARED :
1263 flags |= MAP_PRIVATE;
1265 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1266 flags, block->fd, offset);
1268 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
1269 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1276 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
1277 flags |= MAP_SHARED | MAP_ANONYMOUS;
1278 area = mmap(vaddr, length, PROT_EXEC|PROT_READ|PROT_WRITE,
1281 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
1282 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
1286 if (area != vaddr) {
1287 fprintf(stderr, "Could not remap addr: "
1288 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
1292 memory_try_enable_merging(vaddr, length);
1293 qemu_ram_setup_dump(vaddr, length);
1299 #endif /* !_WIN32 */
1301 static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
1305 /* The list is protected by the iothread lock here. */
1306 block = ram_list.mru_block;
1307 if (block && addr - block->offset < block->length) {
1310 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1311 if (addr - block->offset < block->length) {
1316 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1320 ram_list.mru_block = block;
1324 /* Return a host pointer to ram allocated with qemu_ram_alloc.
1325 With the exception of the softmmu code in this file, this should
1326 only be used for local memory (e.g. video ram) that the device owns,
1327 and knows it isn't going to access beyond the end of the block.
1329 It should not be used for general purpose DMA.
1330 Use cpu_physical_memory_map/cpu_physical_memory_rw instead.
1332 void *qemu_get_ram_ptr(ram_addr_t addr)
1334 RAMBlock *block = qemu_get_ram_block(addr);
1336 if (xen_enabled()) {
1337 /* We need to check if the requested address is in the RAM
1338 * because we don't want to map the entire memory in QEMU.
1339 * In that case just map until the end of the page.
1341 if (block->offset == 0) {
1342 return xen_map_cache(addr, 0, 0);
1343 } else if (block->host == NULL) {
1345 xen_map_cache(block->offset, block->length, 1);
1348 return block->host + (addr - block->offset);
1351 /* Return a host pointer to ram allocated with qemu_ram_alloc. Same as
1352 * qemu_get_ram_ptr but do not touch ram_list.mru_block.
1354 * ??? Is this still necessary?
1356 static void *qemu_safe_ram_ptr(ram_addr_t addr)
1360 /* The list is protected by the iothread lock here. */
1361 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1362 if (addr - block->offset < block->length) {
1363 if (xen_enabled()) {
1364 /* We need to check if the requested address is in the RAM
1365 * because we don't want to map the entire memory in QEMU.
1366 * In that case just map until the end of the page.
1368 if (block->offset == 0) {
1369 return xen_map_cache(addr, 0, 0);
1370 } else if (block->host == NULL) {
1372 xen_map_cache(block->offset, block->length, 1);
1375 return block->host + (addr - block->offset);
1379 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1385 /* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr
1386 * but takes a size argument */
1387 static void *qemu_ram_ptr_length(ram_addr_t addr, ram_addr_t *size)
1392 if (xen_enabled()) {
1393 return xen_map_cache(addr, *size, 1);
1397 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1398 if (addr - block->offset < block->length) {
1399 if (addr - block->offset + *size > block->length)
1400 *size = block->length - addr + block->offset;
1401 return block->host + (addr - block->offset);
1405 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1410 /* Some of the softmmu routines need to translate from a host pointer
1411 (typically a TLB entry) back to a ram offset. */
1412 MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
1415 uint8_t *host = ptr;
1417 if (xen_enabled()) {
1418 *ram_addr = xen_ram_addr_from_mapcache(ptr);
1419 return qemu_get_ram_block(*ram_addr)->mr;
1422 block = ram_list.mru_block;
1423 if (block && block->host && host - block->host < block->length) {
1427 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
1428 /* This case append when the block is not mapped. */
1429 if (block->host == NULL) {
1432 if (host - block->host < block->length) {
1440 *ram_addr = block->offset + (host - block->host);
1444 static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
1445 uint64_t val, unsigned size)
1448 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
1449 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
1450 tb_invalidate_phys_page_fast(ram_addr, size);
1451 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
1455 stb_p(qemu_get_ram_ptr(ram_addr), val);
1458 stw_p(qemu_get_ram_ptr(ram_addr), val);
1461 stl_p(qemu_get_ram_ptr(ram_addr), val);
1466 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
1467 cpu_physical_memory_set_dirty_flags(ram_addr, dirty_flags);
1468 /* we remove the notdirty callback only if the code has been
1470 if (dirty_flags == 0xff) {
1471 CPUArchState *env = current_cpu->env_ptr;
1472 tlb_set_dirty(env, env->mem_io_vaddr);
1476 static bool notdirty_mem_accepts(void *opaque, hwaddr addr,
1477 unsigned size, bool is_write)
1482 static const MemoryRegionOps notdirty_mem_ops = {
1483 .write = notdirty_mem_write,
1484 .valid.accepts = notdirty_mem_accepts,
1485 .endianness = DEVICE_NATIVE_ENDIAN,
1488 /* Generate a debug exception if a watchpoint has been hit. */
1489 static void check_watchpoint(int offset, int len_mask, int flags)
1491 CPUArchState *env = current_cpu->env_ptr;
1492 target_ulong pc, cs_base;
1497 if (env->watchpoint_hit) {
1498 /* We re-entered the check after replacing the TB. Now raise
1499 * the debug interrupt so that is will trigger after the
1500 * current instruction. */
1501 cpu_interrupt(ENV_GET_CPU(env), CPU_INTERRUPT_DEBUG);
1504 vaddr = (env->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
1505 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1506 if ((vaddr == (wp->vaddr & len_mask) ||
1507 (vaddr & wp->len_mask) == wp->vaddr) && (wp->flags & flags)) {
1508 wp->flags |= BP_WATCHPOINT_HIT;
1509 if (!env->watchpoint_hit) {
1510 env->watchpoint_hit = wp;
1511 tb_check_watchpoint(env);
1512 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
1513 env->exception_index = EXCP_DEBUG;
1516 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
1517 tb_gen_code(env, pc, cs_base, cpu_flags, 1);
1518 cpu_resume_from_signal(env, NULL);
1522 wp->flags &= ~BP_WATCHPOINT_HIT;
1527 /* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
1528 so these check for a hit then pass through to the normal out-of-line
1530 static uint64_t watch_mem_read(void *opaque, hwaddr addr,
1533 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~(size - 1), BP_MEM_READ);
1535 case 1: return ldub_phys(addr);
1536 case 2: return lduw_phys(addr);
1537 case 4: return ldl_phys(addr);
1542 static void watch_mem_write(void *opaque, hwaddr addr,
1543 uint64_t val, unsigned size)
1545 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~(size - 1), BP_MEM_WRITE);
1548 stb_phys(addr, val);
1551 stw_phys(addr, val);
1554 stl_phys(addr, val);
1560 static const MemoryRegionOps watch_mem_ops = {
1561 .read = watch_mem_read,
1562 .write = watch_mem_write,
1563 .endianness = DEVICE_NATIVE_ENDIAN,
1566 static uint64_t subpage_read(void *opaque, hwaddr addr,
1569 subpage_t *subpage = opaque;
1572 #if defined(DEBUG_SUBPAGE)
1573 printf("%s: subpage %p len %d addr " TARGET_FMT_plx "\n", __func__,
1574 subpage, len, addr);
1576 address_space_read(subpage->as, addr + subpage->base, buf, len);
1589 static void subpage_write(void *opaque, hwaddr addr,
1590 uint64_t value, unsigned len)
1592 subpage_t *subpage = opaque;
1595 #if defined(DEBUG_SUBPAGE)
1596 printf("%s: subpage %p len %d addr " TARGET_FMT_plx
1597 " value %"PRIx64"\n",
1598 __func__, subpage, len, addr, value);
1613 address_space_write(subpage->as, addr + subpage->base, buf, len);
1616 static bool subpage_accepts(void *opaque, hwaddr addr,
1617 unsigned size, bool is_write)
1619 subpage_t *subpage = opaque;
1620 #if defined(DEBUG_SUBPAGE)
1621 printf("%s: subpage %p %c len %d addr " TARGET_FMT_plx "\n",
1622 __func__, subpage, is_write ? 'w' : 'r', len, addr);
1625 return address_space_access_valid(subpage->as, addr + subpage->base,
1629 static const MemoryRegionOps subpage_ops = {
1630 .read = subpage_read,
1631 .write = subpage_write,
1632 .valid.accepts = subpage_accepts,
1633 .endianness = DEVICE_NATIVE_ENDIAN,
1636 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
1641 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
1643 idx = SUBPAGE_IDX(start);
1644 eidx = SUBPAGE_IDX(end);
1645 #if defined(DEBUG_SUBPAGE)
1646 printf("%s: %p start %08x end %08x idx %08x eidx %08x mem %ld\n", __func__,
1647 mmio, start, end, idx, eidx, memory);
1649 for (; idx <= eidx; idx++) {
1650 mmio->sub_section[idx] = section;
1656 static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
1660 mmio = g_malloc0(sizeof(subpage_t));
1664 memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
1665 "subpage", TARGET_PAGE_SIZE);
1666 mmio->iomem.subpage = true;
1667 #if defined(DEBUG_SUBPAGE)
1668 printf("%s: %p base " TARGET_FMT_plx " len %08x %d\n", __func__,
1669 mmio, base, TARGET_PAGE_SIZE, subpage_memory);
1671 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED);
1676 static uint16_t dummy_section(MemoryRegion *mr)
1678 MemoryRegionSection section = {
1680 .offset_within_address_space = 0,
1681 .offset_within_region = 0,
1682 .size = int128_2_64(),
1685 return phys_section_add(§ion);
1688 MemoryRegion *iotlb_to_region(hwaddr index)
1690 return address_space_memory.dispatch->sections[index & ~TARGET_PAGE_MASK].mr;
1693 static void io_mem_init(void)
1695 memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, "rom", UINT64_MAX);
1696 memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
1697 "unassigned", UINT64_MAX);
1698 memory_region_init_io(&io_mem_notdirty, NULL, ¬dirty_mem_ops, NULL,
1699 "notdirty", UINT64_MAX);
1700 memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL,
1701 "watch", UINT64_MAX);
1704 static void mem_begin(MemoryListener *listener)
1706 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
1707 AddressSpaceDispatch *d = g_new(AddressSpaceDispatch, 1);
1709 d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .is_leaf = 0 };
1711 as->next_dispatch = d;
1714 static void mem_commit(MemoryListener *listener)
1716 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
1717 AddressSpaceDispatch *cur = as->dispatch;
1718 AddressSpaceDispatch *next = as->next_dispatch;
1720 next->nodes = next_map.nodes;
1721 next->sections = next_map.sections;
1723 as->dispatch = next;
1727 static void core_begin(MemoryListener *listener)
1731 prev_map = g_new(PhysPageMap, 1);
1732 *prev_map = next_map;
1734 memset(&next_map, 0, sizeof(next_map));
1735 n = dummy_section(&io_mem_unassigned);
1736 assert(n == PHYS_SECTION_UNASSIGNED);
1737 n = dummy_section(&io_mem_notdirty);
1738 assert(n == PHYS_SECTION_NOTDIRTY);
1739 n = dummy_section(&io_mem_rom);
1740 assert(n == PHYS_SECTION_ROM);
1741 n = dummy_section(&io_mem_watch);
1742 assert(n == PHYS_SECTION_WATCH);
1745 /* This listener's commit run after the other AddressSpaceDispatch listeners'.
1746 * All AddressSpaceDispatch instances have switched to the next map.
1748 static void core_commit(MemoryListener *listener)
1750 phys_sections_free(prev_map);
1753 static void tcg_commit(MemoryListener *listener)
1757 /* since each CPU stores ram addresses in its TLB cache, we must
1758 reset the modified entries */
1760 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
1761 CPUArchState *env = cpu->env_ptr;
1767 static void core_log_global_start(MemoryListener *listener)
1769 cpu_physical_memory_set_dirty_tracking(1);
1772 static void core_log_global_stop(MemoryListener *listener)
1774 cpu_physical_memory_set_dirty_tracking(0);
1777 static MemoryListener core_memory_listener = {
1778 .begin = core_begin,
1779 .commit = core_commit,
1780 .log_global_start = core_log_global_start,
1781 .log_global_stop = core_log_global_stop,
1785 static MemoryListener tcg_memory_listener = {
1786 .commit = tcg_commit,
1789 void address_space_init_dispatch(AddressSpace *as)
1791 as->dispatch = NULL;
1792 as->dispatch_listener = (MemoryListener) {
1794 .commit = mem_commit,
1795 .region_add = mem_add,
1796 .region_nop = mem_add,
1799 memory_listener_register(&as->dispatch_listener, as);
1802 void address_space_destroy_dispatch(AddressSpace *as)
1804 AddressSpaceDispatch *d = as->dispatch;
1806 memory_listener_unregister(&as->dispatch_listener);
1808 as->dispatch = NULL;
1811 static void memory_map_init(void)
1813 system_memory = g_malloc(sizeof(*system_memory));
1814 memory_region_init(system_memory, NULL, "system", INT64_MAX);
1815 address_space_init(&address_space_memory, system_memory, "memory");
1817 system_io = g_malloc(sizeof(*system_io));
1818 memory_region_init(system_io, NULL, "io", 65536);
1819 address_space_init(&address_space_io, system_io, "I/O");
1821 memory_listener_register(&core_memory_listener, &address_space_memory);
1822 memory_listener_register(&tcg_memory_listener, &address_space_memory);
1825 MemoryRegion *get_system_memory(void)
1827 return system_memory;
1830 MemoryRegion *get_system_io(void)
1835 #endif /* !defined(CONFIG_USER_ONLY) */
1837 /* physical memory access (slow version, mainly for debug) */
1838 #if defined(CONFIG_USER_ONLY)
1839 int cpu_memory_rw_debug(CPUArchState *env, target_ulong addr,
1840 uint8_t *buf, int len, int is_write)
1847 page = addr & TARGET_PAGE_MASK;
1848 l = (page + TARGET_PAGE_SIZE) - addr;
1851 flags = page_get_flags(page);
1852 if (!(flags & PAGE_VALID))
1855 if (!(flags & PAGE_WRITE))
1857 /* XXX: this code should not depend on lock_user */
1858 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
1861 unlock_user(p, addr, l);
1863 if (!(flags & PAGE_READ))
1865 /* XXX: this code should not depend on lock_user */
1866 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
1869 unlock_user(p, addr, 0);
1880 static void invalidate_and_set_dirty(hwaddr addr,
1883 if (!cpu_physical_memory_is_dirty(addr)) {
1884 /* invalidate code */
1885 tb_invalidate_phys_page_range(addr, addr + length, 0);
1887 cpu_physical_memory_set_dirty_flags(addr, (0xff & ~CODE_DIRTY_FLAG));
1889 xen_modified_memory(addr, length);
1892 static inline bool memory_access_is_direct(MemoryRegion *mr, bool is_write)
1894 if (memory_region_is_ram(mr)) {
1895 return !(is_write && mr->readonly);
1897 if (memory_region_is_romd(mr)) {
1904 static inline int memory_access_size(MemoryRegion *mr, int l, hwaddr addr)
1906 if (l >= 4 && (((addr & 3) == 0 || mr->ops->impl.unaligned))) {
1909 if (l >= 2 && (((addr & 1) == 0) || mr->ops->impl.unaligned)) {
1915 bool address_space_rw(AddressSpace *as, hwaddr addr, uint8_t *buf,
1916 int len, bool is_write)
1927 mr = address_space_translate(as, addr, &addr1, &l, is_write);
1930 if (!memory_access_is_direct(mr, is_write)) {
1931 l = memory_access_size(mr, l, addr1);
1932 /* XXX: could force current_cpu to NULL to avoid
1935 /* 32 bit write access */
1937 error |= io_mem_write(mr, addr1, val, 4);
1938 } else if (l == 2) {
1939 /* 16 bit write access */
1941 error |= io_mem_write(mr, addr1, val, 2);
1943 /* 8 bit write access */
1945 error |= io_mem_write(mr, addr1, val, 1);
1948 addr1 += memory_region_get_ram_addr(mr);
1950 ptr = qemu_get_ram_ptr(addr1);
1951 memcpy(ptr, buf, l);
1952 invalidate_and_set_dirty(addr1, l);
1955 if (!memory_access_is_direct(mr, is_write)) {
1957 l = memory_access_size(mr, l, addr1);
1959 /* 32 bit read access */
1960 error |= io_mem_read(mr, addr1, &val, 4);
1962 } else if (l == 2) {
1963 /* 16 bit read access */
1964 error |= io_mem_read(mr, addr1, &val, 2);
1967 /* 8 bit read access */
1968 error |= io_mem_read(mr, addr1, &val, 1);
1973 ptr = qemu_get_ram_ptr(mr->ram_addr + addr1);
1974 memcpy(buf, ptr, l);
1985 bool address_space_write(AddressSpace *as, hwaddr addr,
1986 const uint8_t *buf, int len)
1988 return address_space_rw(as, addr, (uint8_t *)buf, len, true);
1991 bool address_space_read(AddressSpace *as, hwaddr addr, uint8_t *buf, int len)
1993 return address_space_rw(as, addr, buf, len, false);
1997 void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
1998 int len, int is_write)
2000 address_space_rw(&address_space_memory, addr, buf, len, is_write);
2003 /* used for ROM loading : can write in RAM and ROM */
2004 void cpu_physical_memory_write_rom(hwaddr addr,
2005 const uint8_t *buf, int len)
2014 mr = address_space_translate(&address_space_memory,
2015 addr, &addr1, &l, true);
2017 if (!(memory_region_is_ram(mr) ||
2018 memory_region_is_romd(mr))) {
2021 addr1 += memory_region_get_ram_addr(mr);
2023 ptr = qemu_get_ram_ptr(addr1);
2024 memcpy(ptr, buf, l);
2025 invalidate_and_set_dirty(addr1, l);
2040 static BounceBuffer bounce;
2042 typedef struct MapClient {
2044 void (*callback)(void *opaque);
2045 QLIST_ENTRY(MapClient) link;
2048 static QLIST_HEAD(map_client_list, MapClient) map_client_list
2049 = QLIST_HEAD_INITIALIZER(map_client_list);
2051 void *cpu_register_map_client(void *opaque, void (*callback)(void *opaque))
2053 MapClient *client = g_malloc(sizeof(*client));
2055 client->opaque = opaque;
2056 client->callback = callback;
2057 QLIST_INSERT_HEAD(&map_client_list, client, link);
2061 static void cpu_unregister_map_client(void *_client)
2063 MapClient *client = (MapClient *)_client;
2065 QLIST_REMOVE(client, link);
2069 static void cpu_notify_map_clients(void)
2073 while (!QLIST_EMPTY(&map_client_list)) {
2074 client = QLIST_FIRST(&map_client_list);
2075 client->callback(client->opaque);
2076 cpu_unregister_map_client(client);
2080 bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write)
2087 mr = address_space_translate(as, addr, &xlat, &l, is_write);
2088 if (!memory_access_is_direct(mr, is_write)) {
2089 l = memory_access_size(mr, l, addr);
2090 if (!memory_region_access_valid(mr, xlat, l, is_write)) {
2101 /* Map a physical memory region into a host virtual address.
2102 * May map a subset of the requested range, given by and returned in *plen.
2103 * May return NULL if resources needed to perform the mapping are exhausted.
2104 * Use only for reads OR writes - not for read-modify-write operations.
2105 * Use cpu_register_map_client() to know when retrying the map operation is
2106 * likely to succeed.
2108 void *address_space_map(AddressSpace *as,
2115 hwaddr l, xlat, base;
2116 MemoryRegion *mr, *this_mr;
2124 mr = address_space_translate(as, addr, &xlat, &l, is_write);
2125 if (!memory_access_is_direct(mr, is_write)) {
2126 if (bounce.buffer) {
2129 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, TARGET_PAGE_SIZE);
2133 memory_region_ref(mr);
2136 address_space_read(as, addr, bounce.buffer, l);
2140 return bounce.buffer;
2144 raddr = memory_region_get_ram_addr(mr);
2155 this_mr = address_space_translate(as, addr, &xlat, &l, is_write);
2156 if (this_mr != mr || xlat != base + done) {
2161 memory_region_ref(mr);
2163 return qemu_ram_ptr_length(raddr + base, plen);
2166 /* Unmaps a memory region previously mapped by address_space_map().
2167 * Will also mark the memory as dirty if is_write == 1. access_len gives
2168 * the amount of memory that was actually read or written by the caller.
2170 void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
2171 int is_write, hwaddr access_len)
2173 if (buffer != bounce.buffer) {
2177 mr = qemu_ram_addr_from_host(buffer, &addr1);
2180 while (access_len) {
2182 l = TARGET_PAGE_SIZE;
2185 invalidate_and_set_dirty(addr1, l);
2190 if (xen_enabled()) {
2191 xen_invalidate_map_cache_entry(buffer);
2193 memory_region_unref(mr);
2197 address_space_write(as, bounce.addr, bounce.buffer, access_len);
2199 qemu_vfree(bounce.buffer);
2200 bounce.buffer = NULL;
2201 memory_region_unref(bounce.mr);
2202 cpu_notify_map_clients();
2205 void *cpu_physical_memory_map(hwaddr addr,
2209 return address_space_map(&address_space_memory, addr, plen, is_write);
2212 void cpu_physical_memory_unmap(void *buffer, hwaddr len,
2213 int is_write, hwaddr access_len)
2215 return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len);
2218 /* warning: addr must be aligned */
2219 static inline uint32_t ldl_phys_internal(hwaddr addr,
2220 enum device_endian endian)
2228 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2230 if (l < 4 || !memory_access_is_direct(mr, false)) {
2232 io_mem_read(mr, addr1, &val, 4);
2233 #if defined(TARGET_WORDS_BIGENDIAN)
2234 if (endian == DEVICE_LITTLE_ENDIAN) {
2238 if (endian == DEVICE_BIG_ENDIAN) {
2244 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
2248 case DEVICE_LITTLE_ENDIAN:
2249 val = ldl_le_p(ptr);
2251 case DEVICE_BIG_ENDIAN:
2252 val = ldl_be_p(ptr);
2262 uint32_t ldl_phys(hwaddr addr)
2264 return ldl_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
2267 uint32_t ldl_le_phys(hwaddr addr)
2269 return ldl_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
2272 uint32_t ldl_be_phys(hwaddr addr)
2274 return ldl_phys_internal(addr, DEVICE_BIG_ENDIAN);
2277 /* warning: addr must be aligned */
2278 static inline uint64_t ldq_phys_internal(hwaddr addr,
2279 enum device_endian endian)
2287 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2289 if (l < 8 || !memory_access_is_direct(mr, false)) {
2291 io_mem_read(mr, addr1, &val, 8);
2292 #if defined(TARGET_WORDS_BIGENDIAN)
2293 if (endian == DEVICE_LITTLE_ENDIAN) {
2297 if (endian == DEVICE_BIG_ENDIAN) {
2303 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
2307 case DEVICE_LITTLE_ENDIAN:
2308 val = ldq_le_p(ptr);
2310 case DEVICE_BIG_ENDIAN:
2311 val = ldq_be_p(ptr);
2321 uint64_t ldq_phys(hwaddr addr)
2323 return ldq_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
2326 uint64_t ldq_le_phys(hwaddr addr)
2328 return ldq_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
2331 uint64_t ldq_be_phys(hwaddr addr)
2333 return ldq_phys_internal(addr, DEVICE_BIG_ENDIAN);
2337 uint32_t ldub_phys(hwaddr addr)
2340 cpu_physical_memory_read(addr, &val, 1);
2344 /* warning: addr must be aligned */
2345 static inline uint32_t lduw_phys_internal(hwaddr addr,
2346 enum device_endian endian)
2354 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2356 if (l < 2 || !memory_access_is_direct(mr, false)) {
2358 io_mem_read(mr, addr1, &val, 2);
2359 #if defined(TARGET_WORDS_BIGENDIAN)
2360 if (endian == DEVICE_LITTLE_ENDIAN) {
2364 if (endian == DEVICE_BIG_ENDIAN) {
2370 ptr = qemu_get_ram_ptr((memory_region_get_ram_addr(mr)
2374 case DEVICE_LITTLE_ENDIAN:
2375 val = lduw_le_p(ptr);
2377 case DEVICE_BIG_ENDIAN:
2378 val = lduw_be_p(ptr);
2388 uint32_t lduw_phys(hwaddr addr)
2390 return lduw_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
2393 uint32_t lduw_le_phys(hwaddr addr)
2395 return lduw_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
2398 uint32_t lduw_be_phys(hwaddr addr)
2400 return lduw_phys_internal(addr, DEVICE_BIG_ENDIAN);
2403 /* warning: addr must be aligned. The ram page is not masked as dirty
2404 and the code inside is not invalidated. It is useful if the dirty
2405 bits are used to track modified PTEs */
2406 void stl_phys_notdirty(hwaddr addr, uint32_t val)
2413 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2415 if (l < 4 || !memory_access_is_direct(mr, true)) {
2416 io_mem_write(mr, addr1, val, 4);
2418 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
2419 ptr = qemu_get_ram_ptr(addr1);
2422 if (unlikely(in_migration)) {
2423 if (!cpu_physical_memory_is_dirty(addr1)) {
2424 /* invalidate code */
2425 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
2427 cpu_physical_memory_set_dirty_flags(
2428 addr1, (0xff & ~CODE_DIRTY_FLAG));
2434 /* warning: addr must be aligned */
2435 static inline void stl_phys_internal(hwaddr addr, uint32_t val,
2436 enum device_endian endian)
2443 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2445 if (l < 4 || !memory_access_is_direct(mr, true)) {
2446 #if defined(TARGET_WORDS_BIGENDIAN)
2447 if (endian == DEVICE_LITTLE_ENDIAN) {
2451 if (endian == DEVICE_BIG_ENDIAN) {
2455 io_mem_write(mr, addr1, val, 4);
2458 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
2459 ptr = qemu_get_ram_ptr(addr1);
2461 case DEVICE_LITTLE_ENDIAN:
2464 case DEVICE_BIG_ENDIAN:
2471 invalidate_and_set_dirty(addr1, 4);
2475 void stl_phys(hwaddr addr, uint32_t val)
2477 stl_phys_internal(addr, val, DEVICE_NATIVE_ENDIAN);
2480 void stl_le_phys(hwaddr addr, uint32_t val)
2482 stl_phys_internal(addr, val, DEVICE_LITTLE_ENDIAN);
2485 void stl_be_phys(hwaddr addr, uint32_t val)
2487 stl_phys_internal(addr, val, DEVICE_BIG_ENDIAN);
2491 void stb_phys(hwaddr addr, uint32_t val)
2494 cpu_physical_memory_write(addr, &v, 1);
2497 /* warning: addr must be aligned */
2498 static inline void stw_phys_internal(hwaddr addr, uint32_t val,
2499 enum device_endian endian)
2506 mr = address_space_translate(&address_space_memory, addr, &addr1, &l,
2508 if (l < 2 || !memory_access_is_direct(mr, true)) {
2509 #if defined(TARGET_WORDS_BIGENDIAN)
2510 if (endian == DEVICE_LITTLE_ENDIAN) {
2514 if (endian == DEVICE_BIG_ENDIAN) {
2518 io_mem_write(mr, addr1, val, 2);
2521 addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK;
2522 ptr = qemu_get_ram_ptr(addr1);
2524 case DEVICE_LITTLE_ENDIAN:
2527 case DEVICE_BIG_ENDIAN:
2534 invalidate_and_set_dirty(addr1, 2);
2538 void stw_phys(hwaddr addr, uint32_t val)
2540 stw_phys_internal(addr, val, DEVICE_NATIVE_ENDIAN);
2543 void stw_le_phys(hwaddr addr, uint32_t val)
2545 stw_phys_internal(addr, val, DEVICE_LITTLE_ENDIAN);
2548 void stw_be_phys(hwaddr addr, uint32_t val)
2550 stw_phys_internal(addr, val, DEVICE_BIG_ENDIAN);
2554 void stq_phys(hwaddr addr, uint64_t val)
2557 cpu_physical_memory_write(addr, &val, 8);
2560 void stq_le_phys(hwaddr addr, uint64_t val)
2562 val = cpu_to_le64(val);
2563 cpu_physical_memory_write(addr, &val, 8);
2566 void stq_be_phys(hwaddr addr, uint64_t val)
2568 val = cpu_to_be64(val);
2569 cpu_physical_memory_write(addr, &val, 8);
2572 /* virtual memory access for debug (includes writing to ROM) */
2573 int cpu_memory_rw_debug(CPUArchState *env, target_ulong addr,
2574 uint8_t *buf, int len, int is_write)
2581 page = addr & TARGET_PAGE_MASK;
2582 phys_addr = cpu_get_phys_page_debug(env, page);
2583 /* if no physical page mapped, return an error */
2584 if (phys_addr == -1)
2586 l = (page + TARGET_PAGE_SIZE) - addr;
2589 phys_addr += (addr & ~TARGET_PAGE_MASK);
2591 cpu_physical_memory_write_rom(phys_addr, buf, l);
2593 cpu_physical_memory_rw(phys_addr, buf, l, is_write);
2602 #if !defined(CONFIG_USER_ONLY)
2605 * A helper function for the _utterly broken_ virtio device model to find out if
2606 * it's running on a big endian machine. Don't do this at home kids!
2608 bool virtio_is_big_endian(void);
2609 bool virtio_is_big_endian(void)
2611 #if defined(TARGET_WORDS_BIGENDIAN)
2620 #ifndef CONFIG_USER_ONLY
2621 bool cpu_physical_memory_is_io(hwaddr phys_addr)
2626 mr = address_space_translate(&address_space_memory,
2627 phys_addr, &phys_addr, &l, false);
2629 return !(memory_region_is_ram(mr) ||
2630 memory_region_is_romd(mr));
2633 void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
2637 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
2638 func(block->host, block->offset, block->length, opaque);