2 * virtual page mapping and translated block handling
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>
36 #include "qemu-common.h"
41 #include "qemu-timer.h"
42 #if defined(CONFIG_USER_ONLY)
45 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
46 #include <sys/param.h>
47 #if __FreeBSD_version >= 700104
48 #define HAVE_KINFO_GETVMMAP
49 #define sigqueue sigqueue_freebsd /* avoid redefinition */
52 #include <machine/profile.h>
62 //#define DEBUG_TB_INVALIDATE
65 //#define DEBUG_UNASSIGNED
67 /* make various TB consistency checks */
68 //#define DEBUG_TB_CHECK
69 //#define DEBUG_TLB_CHECK
71 //#define DEBUG_IOPORT
72 //#define DEBUG_SUBPAGE
74 #if !defined(CONFIG_USER_ONLY)
75 /* TB consistency checks only implemented for usermode emulation. */
79 #define SMC_BITMAP_USE_THRESHOLD 10
81 static TranslationBlock *tbs;
82 int code_gen_max_blocks;
83 TranslationBlock *tb_phys_hash[CODE_GEN_PHYS_HASH_SIZE];
85 /* any access to the tbs or the page table must use this lock */
86 spinlock_t tb_lock = SPIN_LOCK_UNLOCKED;
88 #if defined(__arm__) || defined(__sparc_v9__)
89 /* The prologue must be reachable with a direct jump. ARM and Sparc64
90 have limited branch ranges (possibly also PPC) so place it in a
91 section close to code segment. */
92 #define code_gen_section \
93 __attribute__((__section__(".gen_code"))) \
94 __attribute__((aligned (32)))
96 /* Maximum alignment for Win32 is 16. */
97 #define code_gen_section \
98 __attribute__((aligned (16)))
100 #define code_gen_section \
101 __attribute__((aligned (32)))
104 uint8_t code_gen_prologue[1024] code_gen_section;
105 static uint8_t *code_gen_buffer;
106 static unsigned long code_gen_buffer_size;
107 /* threshold to flush the translated code buffer */
108 static unsigned long code_gen_buffer_max_size;
109 uint8_t *code_gen_ptr;
111 #if !defined(CONFIG_USER_ONLY)
113 uint8_t *phys_ram_dirty;
114 static int in_migration;
116 typedef struct RAMBlock {
120 struct RAMBlock *next;
123 static RAMBlock *ram_blocks;
124 /* TODO: When we implement (and use) ram deallocation (e.g. for hotplug)
125 then we can no longer assume contiguous ram offsets, and external uses
126 of this variable will break. */
127 ram_addr_t last_ram_offset;
131 /* current CPU in the current thread. It is only valid inside
133 CPUState *cpu_single_env;
134 /* 0 = Do not count executed instructions.
135 1 = Precise instruction counting.
136 2 = Adaptive rate instruction counting. */
138 /* Current instruction counter. While executing translated code this may
139 include some instructions that have not yet been executed. */
142 typedef struct PageDesc {
143 /* list of TBs intersecting this ram page */
144 TranslationBlock *first_tb;
145 /* in order to optimize self modifying code, we count the number
146 of lookups we do to a given page to use a bitmap */
147 unsigned int code_write_count;
148 uint8_t *code_bitmap;
149 #if defined(CONFIG_USER_ONLY)
154 /* In system mode we want L1_MAP to be based on ram offsets,
155 while in user mode we want it to be based on virtual addresses. */
156 #if !defined(CONFIG_USER_ONLY)
157 #if HOST_LONG_BITS < TARGET_PHYS_ADDR_SPACE_BITS
158 # define L1_MAP_ADDR_SPACE_BITS HOST_LONG_BITS
160 # define L1_MAP_ADDR_SPACE_BITS TARGET_PHYS_ADDR_SPACE_BITS
163 # define L1_MAP_ADDR_SPACE_BITS TARGET_VIRT_ADDR_SPACE_BITS
166 /* Size of the L2 (and L3, etc) page tables. */
168 #define L2_SIZE (1 << L2_BITS)
170 /* The bits remaining after N lower levels of page tables. */
171 #define P_L1_BITS_REM \
172 ((TARGET_PHYS_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % L2_BITS)
173 #define V_L1_BITS_REM \
174 ((L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % L2_BITS)
176 /* Size of the L1 page table. Avoid silly small sizes. */
177 #if P_L1_BITS_REM < 4
178 #define P_L1_BITS (P_L1_BITS_REM + L2_BITS)
180 #define P_L1_BITS P_L1_BITS_REM
183 #if V_L1_BITS_REM < 4
184 #define V_L1_BITS (V_L1_BITS_REM + L2_BITS)
186 #define V_L1_BITS V_L1_BITS_REM
189 #define P_L1_SIZE ((target_phys_addr_t)1 << P_L1_BITS)
190 #define V_L1_SIZE ((target_ulong)1 << V_L1_BITS)
192 #define P_L1_SHIFT (TARGET_PHYS_ADDR_SPACE_BITS - TARGET_PAGE_BITS - P_L1_BITS)
193 #define V_L1_SHIFT (L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS - V_L1_BITS)
195 unsigned long qemu_real_host_page_size;
196 unsigned long qemu_host_page_bits;
197 unsigned long qemu_host_page_size;
198 unsigned long qemu_host_page_mask;
200 /* This is a multi-level map on the virtual address space.
201 The bottom level has pointers to PageDesc. */
202 static void *l1_map[V_L1_SIZE];
204 #if !defined(CONFIG_USER_ONLY)
205 typedef struct PhysPageDesc {
206 /* offset in host memory of the page + io_index in the low bits */
207 ram_addr_t phys_offset;
208 ram_addr_t region_offset;
211 /* This is a multi-level map on the physical address space.
212 The bottom level has pointers to PhysPageDesc. */
213 static void *l1_phys_map[P_L1_SIZE];
215 static void io_mem_init(void);
217 /* io memory support */
218 CPUWriteMemoryFunc *io_mem_write[IO_MEM_NB_ENTRIES][4];
219 CPUReadMemoryFunc *io_mem_read[IO_MEM_NB_ENTRIES][4];
220 void *io_mem_opaque[IO_MEM_NB_ENTRIES];
221 static char io_mem_used[IO_MEM_NB_ENTRIES];
222 static int io_mem_watch;
227 static const char *logfilename = "qemu.log";
229 static const char *logfilename = "/tmp/qemu.log";
233 static int log_append = 0;
236 #if !defined(CONFIG_USER_ONLY)
237 static int tlb_flush_count;
239 static int tb_flush_count;
240 static int tb_phys_invalidate_count;
243 static void map_exec(void *addr, long size)
246 VirtualProtect(addr, size,
247 PAGE_EXECUTE_READWRITE, &old_protect);
251 static void map_exec(void *addr, long size)
253 unsigned long start, end, page_size;
255 page_size = getpagesize();
256 start = (unsigned long)addr;
257 start &= ~(page_size - 1);
259 end = (unsigned long)addr + size;
260 end += page_size - 1;
261 end &= ~(page_size - 1);
263 mprotect((void *)start, end - start,
264 PROT_READ | PROT_WRITE | PROT_EXEC);
268 static void page_init(void)
270 /* NOTE: we can always suppose that qemu_host_page_size >=
274 SYSTEM_INFO system_info;
276 GetSystemInfo(&system_info);
277 qemu_real_host_page_size = system_info.dwPageSize;
280 qemu_real_host_page_size = getpagesize();
282 if (qemu_host_page_size == 0)
283 qemu_host_page_size = qemu_real_host_page_size;
284 if (qemu_host_page_size < TARGET_PAGE_SIZE)
285 qemu_host_page_size = TARGET_PAGE_SIZE;
286 qemu_host_page_bits = 0;
287 while ((1 << qemu_host_page_bits) < qemu_host_page_size)
288 qemu_host_page_bits++;
289 qemu_host_page_mask = ~(qemu_host_page_size - 1);
291 #if !defined(_WIN32) && defined(CONFIG_USER_ONLY)
293 #ifdef HAVE_KINFO_GETVMMAP
294 struct kinfo_vmentry *freep;
297 freep = kinfo_getvmmap(getpid(), &cnt);
300 for (i = 0; i < cnt; i++) {
301 unsigned long startaddr, endaddr;
303 startaddr = freep[i].kve_start;
304 endaddr = freep[i].kve_end;
305 if (h2g_valid(startaddr)) {
306 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
308 if (h2g_valid(endaddr)) {
309 endaddr = h2g(endaddr);
310 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
312 #if TARGET_ABI_BITS <= L1_MAP_ADDR_SPACE_BITS
314 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
325 last_brk = (unsigned long)sbrk(0);
327 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__)
328 f = fopen("/compat/linux/proc/self/maps", "r");
330 f = fopen("/proc/self/maps", "r");
336 unsigned long startaddr, endaddr;
339 n = fscanf (f, "%lx-%lx %*[^\n]\n", &startaddr, &endaddr);
341 if (n == 2 && h2g_valid(startaddr)) {
342 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
344 if (h2g_valid(endaddr)) {
345 endaddr = h2g(endaddr);
349 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
361 static PageDesc *page_find_alloc(tb_page_addr_t index, int alloc)
367 #if defined(CONFIG_USER_ONLY)
368 /* We can't use qemu_malloc because it may recurse into a locked mutex.
369 Neither can we record the new pages we reserve while allocating a
370 given page because that may recurse into an unallocated page table
371 entry. Stuff the allocations we do make into a queue and process
372 them after having completed one entire page table allocation. */
374 unsigned long reserve[2 * (V_L1_SHIFT / L2_BITS)];
377 # define ALLOC(P, SIZE) \
379 P = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, \
380 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); \
381 if (h2g_valid(P)) { \
382 reserve[reserve_idx] = h2g(P); \
383 reserve[reserve_idx + 1] = SIZE; \
388 # define ALLOC(P, SIZE) \
389 do { P = qemu_mallocz(SIZE); } while (0)
392 /* Level 1. Always allocated. */
393 lp = l1_map + ((index >> V_L1_SHIFT) & (V_L1_SIZE - 1));
396 for (i = V_L1_SHIFT / L2_BITS - 1; i > 0; i--) {
403 ALLOC(p, sizeof(void *) * L2_SIZE);
407 lp = p + ((index >> (i * L2_BITS)) & (L2_SIZE - 1));
415 ALLOC(pd, sizeof(PageDesc) * L2_SIZE);
420 #if defined(CONFIG_USER_ONLY)
421 for (i = 0; i < reserve_idx; i += 2) {
422 unsigned long addr = reserve[i];
423 unsigned long len = reserve[i + 1];
425 page_set_flags(addr & TARGET_PAGE_MASK,
426 TARGET_PAGE_ALIGN(addr + len),
431 return pd + (index & (L2_SIZE - 1));
434 static inline PageDesc *page_find(tb_page_addr_t index)
436 return page_find_alloc(index, 0);
439 #if !defined(CONFIG_USER_ONLY)
440 static PhysPageDesc *phys_page_find_alloc(target_phys_addr_t index, int alloc)
446 /* Level 1. Always allocated. */
447 lp = l1_phys_map + ((index >> P_L1_SHIFT) & (P_L1_SIZE - 1));
450 for (i = P_L1_SHIFT / L2_BITS - 1; i > 0; i--) {
456 *lp = p = qemu_mallocz(sizeof(void *) * L2_SIZE);
458 lp = p + ((index >> (i * L2_BITS)) & (L2_SIZE - 1));
469 *lp = pd = qemu_malloc(sizeof(PhysPageDesc) * L2_SIZE);
471 for (i = 0; i < L2_SIZE; i++) {
472 pd[i].phys_offset = IO_MEM_UNASSIGNED;
473 pd[i].region_offset = (index + i) << TARGET_PAGE_BITS;
477 return pd + (index & (L2_SIZE - 1));
480 static inline PhysPageDesc *phys_page_find(target_phys_addr_t index)
482 return phys_page_find_alloc(index, 0);
485 static void tlb_protect_code(ram_addr_t ram_addr);
486 static void tlb_unprotect_code_phys(CPUState *env, ram_addr_t ram_addr,
488 #define mmap_lock() do { } while(0)
489 #define mmap_unlock() do { } while(0)
492 #define DEFAULT_CODE_GEN_BUFFER_SIZE (32 * 1024 * 1024)
494 #if defined(CONFIG_USER_ONLY)
495 /* Currently it is not recommended to allocate big chunks of data in
496 user mode. It will change when a dedicated libc will be used */
497 #define USE_STATIC_CODE_GEN_BUFFER
500 #ifdef USE_STATIC_CODE_GEN_BUFFER
501 static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE];
504 static void code_gen_alloc(unsigned long tb_size)
506 #ifdef USE_STATIC_CODE_GEN_BUFFER
507 code_gen_buffer = static_code_gen_buffer;
508 code_gen_buffer_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
509 map_exec(code_gen_buffer, code_gen_buffer_size);
511 code_gen_buffer_size = tb_size;
512 if (code_gen_buffer_size == 0) {
513 #if defined(CONFIG_USER_ONLY)
514 /* in user mode, phys_ram_size is not meaningful */
515 code_gen_buffer_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
517 /* XXX: needs adjustments */
518 code_gen_buffer_size = (unsigned long)(ram_size / 4);
521 if (code_gen_buffer_size < MIN_CODE_GEN_BUFFER_SIZE)
522 code_gen_buffer_size = MIN_CODE_GEN_BUFFER_SIZE;
523 /* The code gen buffer location may have constraints depending on
524 the host cpu and OS */
525 #if defined(__linux__)
530 flags = MAP_PRIVATE | MAP_ANONYMOUS;
531 #if defined(__x86_64__)
533 /* Cannot map more than that */
534 if (code_gen_buffer_size > (800 * 1024 * 1024))
535 code_gen_buffer_size = (800 * 1024 * 1024);
536 #elif defined(__sparc_v9__)
537 // Map the buffer below 2G, so we can use direct calls and branches
539 start = (void *) 0x60000000UL;
540 if (code_gen_buffer_size > (512 * 1024 * 1024))
541 code_gen_buffer_size = (512 * 1024 * 1024);
542 #elif defined(__arm__)
543 /* Map the buffer below 32M, so we can use direct calls and branches */
545 start = (void *) 0x01000000UL;
546 if (code_gen_buffer_size > 16 * 1024 * 1024)
547 code_gen_buffer_size = 16 * 1024 * 1024;
549 code_gen_buffer = mmap(start, code_gen_buffer_size,
550 PROT_WRITE | PROT_READ | PROT_EXEC,
552 if (code_gen_buffer == MAP_FAILED) {
553 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
557 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__)
561 flags = MAP_PRIVATE | MAP_ANONYMOUS;
562 #if defined(__x86_64__)
563 /* FreeBSD doesn't have MAP_32BIT, use MAP_FIXED and assume
564 * 0x40000000 is free */
566 addr = (void *)0x40000000;
567 /* Cannot map more than that */
568 if (code_gen_buffer_size > (800 * 1024 * 1024))
569 code_gen_buffer_size = (800 * 1024 * 1024);
571 code_gen_buffer = mmap(addr, code_gen_buffer_size,
572 PROT_WRITE | PROT_READ | PROT_EXEC,
574 if (code_gen_buffer == MAP_FAILED) {
575 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
580 code_gen_buffer = qemu_malloc(code_gen_buffer_size);
581 map_exec(code_gen_buffer, code_gen_buffer_size);
583 #endif /* !USE_STATIC_CODE_GEN_BUFFER */
584 map_exec(code_gen_prologue, sizeof(code_gen_prologue));
585 code_gen_buffer_max_size = code_gen_buffer_size -
586 code_gen_max_block_size();
587 code_gen_max_blocks = code_gen_buffer_size / CODE_GEN_AVG_BLOCK_SIZE;
588 tbs = qemu_malloc(code_gen_max_blocks * sizeof(TranslationBlock));
591 /* Must be called before using the QEMU cpus. 'tb_size' is the size
592 (in bytes) allocated to the translation buffer. Zero means default
594 void cpu_exec_init_all(unsigned long tb_size)
597 code_gen_alloc(tb_size);
598 code_gen_ptr = code_gen_buffer;
600 #if !defined(CONFIG_USER_ONLY)
605 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
607 static int cpu_common_post_load(void *opaque, int version_id)
609 CPUState *env = opaque;
611 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
612 version_id is increased. */
613 env->interrupt_request &= ~0x01;
619 static const VMStateDescription vmstate_cpu_common = {
620 .name = "cpu_common",
622 .minimum_version_id = 1,
623 .minimum_version_id_old = 1,
624 .post_load = cpu_common_post_load,
625 .fields = (VMStateField []) {
626 VMSTATE_UINT32(halted, CPUState),
627 VMSTATE_UINT32(interrupt_request, CPUState),
628 VMSTATE_END_OF_LIST()
633 CPUState *qemu_get_cpu(int cpu)
635 CPUState *env = first_cpu;
638 if (env->cpu_index == cpu)
646 void cpu_exec_init(CPUState *env)
651 #if defined(CONFIG_USER_ONLY)
654 env->next_cpu = NULL;
657 while (*penv != NULL) {
658 penv = &(*penv)->next_cpu;
661 env->cpu_index = cpu_index;
663 QTAILQ_INIT(&env->breakpoints);
664 QTAILQ_INIT(&env->watchpoints);
666 #if defined(CONFIG_USER_ONLY)
669 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
670 vmstate_register(cpu_index, &vmstate_cpu_common, env);
671 register_savevm("cpu", cpu_index, CPU_SAVE_VERSION,
672 cpu_save, cpu_load, env);
676 static inline void invalidate_page_bitmap(PageDesc *p)
678 if (p->code_bitmap) {
679 qemu_free(p->code_bitmap);
680 p->code_bitmap = NULL;
682 p->code_write_count = 0;
685 /* Set to NULL all the 'first_tb' fields in all PageDescs. */
687 static void page_flush_tb_1 (int level, void **lp)
696 for (i = 0; i < L2_SIZE; ++i) {
697 pd[i].first_tb = NULL;
698 invalidate_page_bitmap(pd + i);
702 for (i = 0; i < L2_SIZE; ++i) {
703 page_flush_tb_1 (level - 1, pp + i);
708 static void page_flush_tb(void)
711 for (i = 0; i < V_L1_SIZE; i++) {
712 page_flush_tb_1(V_L1_SHIFT / L2_BITS - 1, l1_map + i);
716 /* flush all the translation blocks */
717 /* XXX: tb_flush is currently not thread safe */
718 void tb_flush(CPUState *env1)
721 #if defined(DEBUG_FLUSH)
722 printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n",
723 (unsigned long)(code_gen_ptr - code_gen_buffer),
725 ((unsigned long)(code_gen_ptr - code_gen_buffer)) / nb_tbs : 0);
727 if ((unsigned long)(code_gen_ptr - code_gen_buffer) > code_gen_buffer_size)
728 cpu_abort(env1, "Internal error: code buffer overflow\n");
732 for(env = first_cpu; env != NULL; env = env->next_cpu) {
733 memset (env->tb_jmp_cache, 0, TB_JMP_CACHE_SIZE * sizeof (void *));
736 memset (tb_phys_hash, 0, CODE_GEN_PHYS_HASH_SIZE * sizeof (void *));
739 code_gen_ptr = code_gen_buffer;
740 /* XXX: flush processor icache at this point if cache flush is
745 #ifdef DEBUG_TB_CHECK
747 static void tb_invalidate_check(target_ulong address)
749 TranslationBlock *tb;
751 address &= TARGET_PAGE_MASK;
752 for(i = 0;i < CODE_GEN_PHYS_HASH_SIZE; i++) {
753 for(tb = tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
754 if (!(address + TARGET_PAGE_SIZE <= tb->pc ||
755 address >= tb->pc + tb->size)) {
756 printf("ERROR invalidate: address=" TARGET_FMT_lx
757 " PC=%08lx size=%04x\n",
758 address, (long)tb->pc, tb->size);
764 /* verify that all the pages have correct rights for code */
765 static void tb_page_check(void)
767 TranslationBlock *tb;
768 int i, flags1, flags2;
770 for(i = 0;i < CODE_GEN_PHYS_HASH_SIZE; i++) {
771 for(tb = tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
772 flags1 = page_get_flags(tb->pc);
773 flags2 = page_get_flags(tb->pc + tb->size - 1);
774 if ((flags1 & PAGE_WRITE) || (flags2 & PAGE_WRITE)) {
775 printf("ERROR page flags: PC=%08lx size=%04x f1=%x f2=%x\n",
776 (long)tb->pc, tb->size, flags1, flags2);
784 /* invalidate one TB */
785 static inline void tb_remove(TranslationBlock **ptb, TranslationBlock *tb,
788 TranslationBlock *tb1;
792 *ptb = *(TranslationBlock **)((char *)tb1 + next_offset);
795 ptb = (TranslationBlock **)((char *)tb1 + next_offset);
799 static inline void tb_page_remove(TranslationBlock **ptb, TranslationBlock *tb)
801 TranslationBlock *tb1;
807 tb1 = (TranslationBlock *)((long)tb1 & ~3);
809 *ptb = tb1->page_next[n1];
812 ptb = &tb1->page_next[n1];
816 static inline void tb_jmp_remove(TranslationBlock *tb, int n)
818 TranslationBlock *tb1, **ptb;
821 ptb = &tb->jmp_next[n];
824 /* find tb(n) in circular list */
828 tb1 = (TranslationBlock *)((long)tb1 & ~3);
829 if (n1 == n && tb1 == tb)
832 ptb = &tb1->jmp_first;
834 ptb = &tb1->jmp_next[n1];
837 /* now we can suppress tb(n) from the list */
838 *ptb = tb->jmp_next[n];
840 tb->jmp_next[n] = NULL;
844 /* reset the jump entry 'n' of a TB so that it is not chained to
846 static inline void tb_reset_jump(TranslationBlock *tb, int n)
848 tb_set_jmp_target(tb, n, (unsigned long)(tb->tc_ptr + tb->tb_next_offset[n]));
851 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr)
856 tb_page_addr_t phys_pc;
857 TranslationBlock *tb1, *tb2;
859 /* remove the TB from the hash list */
860 phys_pc = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
861 h = tb_phys_hash_func(phys_pc);
862 tb_remove(&tb_phys_hash[h], tb,
863 offsetof(TranslationBlock, phys_hash_next));
865 /* remove the TB from the page list */
866 if (tb->page_addr[0] != page_addr) {
867 p = page_find(tb->page_addr[0] >> TARGET_PAGE_BITS);
868 tb_page_remove(&p->first_tb, tb);
869 invalidate_page_bitmap(p);
871 if (tb->page_addr[1] != -1 && tb->page_addr[1] != page_addr) {
872 p = page_find(tb->page_addr[1] >> TARGET_PAGE_BITS);
873 tb_page_remove(&p->first_tb, tb);
874 invalidate_page_bitmap(p);
877 tb_invalidated_flag = 1;
879 /* remove the TB from the hash list */
880 h = tb_jmp_cache_hash_func(tb->pc);
881 for(env = first_cpu; env != NULL; env = env->next_cpu) {
882 if (env->tb_jmp_cache[h] == tb)
883 env->tb_jmp_cache[h] = NULL;
886 /* suppress this TB from the two jump lists */
887 tb_jmp_remove(tb, 0);
888 tb_jmp_remove(tb, 1);
890 /* suppress any remaining jumps to this TB */
896 tb1 = (TranslationBlock *)((long)tb1 & ~3);
897 tb2 = tb1->jmp_next[n1];
898 tb_reset_jump(tb1, n1);
899 tb1->jmp_next[n1] = NULL;
902 tb->jmp_first = (TranslationBlock *)((long)tb | 2); /* fail safe */
904 tb_phys_invalidate_count++;
907 static inline void set_bits(uint8_t *tab, int start, int len)
913 mask = 0xff << (start & 7);
914 if ((start & ~7) == (end & ~7)) {
916 mask &= ~(0xff << (end & 7));
921 start = (start + 8) & ~7;
923 while (start < end1) {
928 mask = ~(0xff << (end & 7));
934 static void build_page_bitmap(PageDesc *p)
936 int n, tb_start, tb_end;
937 TranslationBlock *tb;
939 p->code_bitmap = qemu_mallocz(TARGET_PAGE_SIZE / 8);
944 tb = (TranslationBlock *)((long)tb & ~3);
945 /* NOTE: this is subtle as a TB may span two physical pages */
947 /* NOTE: tb_end may be after the end of the page, but
948 it is not a problem */
949 tb_start = tb->pc & ~TARGET_PAGE_MASK;
950 tb_end = tb_start + tb->size;
951 if (tb_end > TARGET_PAGE_SIZE)
952 tb_end = TARGET_PAGE_SIZE;
955 tb_end = ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
957 set_bits(p->code_bitmap, tb_start, tb_end - tb_start);
958 tb = tb->page_next[n];
962 TranslationBlock *tb_gen_code(CPUState *env,
963 target_ulong pc, target_ulong cs_base,
964 int flags, int cflags)
966 TranslationBlock *tb;
968 tb_page_addr_t phys_pc, phys_page2;
969 target_ulong virt_page2;
972 phys_pc = get_page_addr_code(env, pc);
975 /* flush must be done */
977 /* cannot fail at this point */
979 /* Don't forget to invalidate previous TB info. */
980 tb_invalidated_flag = 1;
982 tc_ptr = code_gen_ptr;
984 tb->cs_base = cs_base;
987 cpu_gen_code(env, tb, &code_gen_size);
988 code_gen_ptr = (void *)(((unsigned long)code_gen_ptr + code_gen_size + CODE_GEN_ALIGN - 1) & ~(CODE_GEN_ALIGN - 1));
990 /* check next page if needed */
991 virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
993 if ((pc & TARGET_PAGE_MASK) != virt_page2) {
994 phys_page2 = get_page_addr_code(env, virt_page2);
996 tb_link_page(tb, phys_pc, phys_page2);
1000 /* invalidate all TBs which intersect with the target physical page
1001 starting in range [start;end[. NOTE: start and end must refer to
1002 the same physical page. 'is_cpu_write_access' should be true if called
1003 from a real cpu write access: the virtual CPU will exit the current
1004 TB if code is modified inside this TB. */
1005 void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
1006 int is_cpu_write_access)
1008 TranslationBlock *tb, *tb_next, *saved_tb;
1009 CPUState *env = cpu_single_env;
1010 tb_page_addr_t tb_start, tb_end;
1013 #ifdef TARGET_HAS_PRECISE_SMC
1014 int current_tb_not_found = is_cpu_write_access;
1015 TranslationBlock *current_tb = NULL;
1016 int current_tb_modified = 0;
1017 target_ulong current_pc = 0;
1018 target_ulong current_cs_base = 0;
1019 int current_flags = 0;
1020 #endif /* TARGET_HAS_PRECISE_SMC */
1022 p = page_find(start >> TARGET_PAGE_BITS);
1025 if (!p->code_bitmap &&
1026 ++p->code_write_count >= SMC_BITMAP_USE_THRESHOLD &&
1027 is_cpu_write_access) {
1028 /* build code bitmap */
1029 build_page_bitmap(p);
1032 /* we remove all the TBs in the range [start, end[ */
1033 /* XXX: see if in some cases it could be faster to invalidate all the code */
1035 while (tb != NULL) {
1037 tb = (TranslationBlock *)((long)tb & ~3);
1038 tb_next = tb->page_next[n];
1039 /* NOTE: this is subtle as a TB may span two physical pages */
1041 /* NOTE: tb_end may be after the end of the page, but
1042 it is not a problem */
1043 tb_start = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
1044 tb_end = tb_start + tb->size;
1046 tb_start = tb->page_addr[1];
1047 tb_end = tb_start + ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
1049 if (!(tb_end <= start || tb_start >= end)) {
1050 #ifdef TARGET_HAS_PRECISE_SMC
1051 if (current_tb_not_found) {
1052 current_tb_not_found = 0;
1054 if (env->mem_io_pc) {
1055 /* now we have a real cpu fault */
1056 current_tb = tb_find_pc(env->mem_io_pc);
1059 if (current_tb == tb &&
1060 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1061 /* If we are modifying the current TB, we must stop
1062 its execution. We could be more precise by checking
1063 that the modification is after the current PC, but it
1064 would require a specialized function to partially
1065 restore the CPU state */
1067 current_tb_modified = 1;
1068 cpu_restore_state(current_tb, env,
1069 env->mem_io_pc, NULL);
1070 cpu_get_tb_cpu_state(env, ¤t_pc, ¤t_cs_base,
1073 #endif /* TARGET_HAS_PRECISE_SMC */
1074 /* we need to do that to handle the case where a signal
1075 occurs while doing tb_phys_invalidate() */
1078 saved_tb = env->current_tb;
1079 env->current_tb = NULL;
1081 tb_phys_invalidate(tb, -1);
1083 env->current_tb = saved_tb;
1084 if (env->interrupt_request && env->current_tb)
1085 cpu_interrupt(env, env->interrupt_request);
1090 #if !defined(CONFIG_USER_ONLY)
1091 /* if no code remaining, no need to continue to use slow writes */
1093 invalidate_page_bitmap(p);
1094 if (is_cpu_write_access) {
1095 tlb_unprotect_code_phys(env, start, env->mem_io_vaddr);
1099 #ifdef TARGET_HAS_PRECISE_SMC
1100 if (current_tb_modified) {
1101 /* we generate a block containing just the instruction
1102 modifying the memory. It will ensure that it cannot modify
1104 env->current_tb = NULL;
1105 tb_gen_code(env, current_pc, current_cs_base, current_flags, 1);
1106 cpu_resume_from_signal(env, NULL);
1111 /* len must be <= 8 and start must be a multiple of len */
1112 static inline void tb_invalidate_phys_page_fast(tb_page_addr_t start, int len)
1118 qemu_log("modifying code at 0x%x size=%d EIP=%x PC=%08x\n",
1119 cpu_single_env->mem_io_vaddr, len,
1120 cpu_single_env->eip,
1121 cpu_single_env->eip + (long)cpu_single_env->segs[R_CS].base);
1124 p = page_find(start >> TARGET_PAGE_BITS);
1127 if (p->code_bitmap) {
1128 offset = start & ~TARGET_PAGE_MASK;
1129 b = p->code_bitmap[offset >> 3] >> (offset & 7);
1130 if (b & ((1 << len) - 1))
1134 tb_invalidate_phys_page_range(start, start + len, 1);
1138 #if !defined(CONFIG_SOFTMMU)
1139 static void tb_invalidate_phys_page(tb_page_addr_t addr,
1140 unsigned long pc, void *puc)
1142 TranslationBlock *tb;
1145 #ifdef TARGET_HAS_PRECISE_SMC
1146 TranslationBlock *current_tb = NULL;
1147 CPUState *env = cpu_single_env;
1148 int current_tb_modified = 0;
1149 target_ulong current_pc = 0;
1150 target_ulong current_cs_base = 0;
1151 int current_flags = 0;
1154 addr &= TARGET_PAGE_MASK;
1155 p = page_find(addr >> TARGET_PAGE_BITS);
1159 #ifdef TARGET_HAS_PRECISE_SMC
1160 if (tb && pc != 0) {
1161 current_tb = tb_find_pc(pc);
1164 while (tb != NULL) {
1166 tb = (TranslationBlock *)((long)tb & ~3);
1167 #ifdef TARGET_HAS_PRECISE_SMC
1168 if (current_tb == tb &&
1169 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1170 /* If we are modifying the current TB, we must stop
1171 its execution. We could be more precise by checking
1172 that the modification is after the current PC, but it
1173 would require a specialized function to partially
1174 restore the CPU state */
1176 current_tb_modified = 1;
1177 cpu_restore_state(current_tb, env, pc, puc);
1178 cpu_get_tb_cpu_state(env, ¤t_pc, ¤t_cs_base,
1181 #endif /* TARGET_HAS_PRECISE_SMC */
1182 tb_phys_invalidate(tb, addr);
1183 tb = tb->page_next[n];
1186 #ifdef TARGET_HAS_PRECISE_SMC
1187 if (current_tb_modified) {
1188 /* we generate a block containing just the instruction
1189 modifying the memory. It will ensure that it cannot modify
1191 env->current_tb = NULL;
1192 tb_gen_code(env, current_pc, current_cs_base, current_flags, 1);
1193 cpu_resume_from_signal(env, puc);
1199 /* add the tb in the target page and protect it if necessary */
1200 static inline void tb_alloc_page(TranslationBlock *tb,
1201 unsigned int n, tb_page_addr_t page_addr)
1204 TranslationBlock *last_first_tb;
1206 tb->page_addr[n] = page_addr;
1207 p = page_find_alloc(page_addr >> TARGET_PAGE_BITS, 1);
1208 tb->page_next[n] = p->first_tb;
1209 last_first_tb = p->first_tb;
1210 p->first_tb = (TranslationBlock *)((long)tb | n);
1211 invalidate_page_bitmap(p);
1213 #if defined(TARGET_HAS_SMC) || 1
1215 #if defined(CONFIG_USER_ONLY)
1216 if (p->flags & PAGE_WRITE) {
1221 /* force the host page as non writable (writes will have a
1222 page fault + mprotect overhead) */
1223 page_addr &= qemu_host_page_mask;
1225 for(addr = page_addr; addr < page_addr + qemu_host_page_size;
1226 addr += TARGET_PAGE_SIZE) {
1228 p2 = page_find (addr >> TARGET_PAGE_BITS);
1232 p2->flags &= ~PAGE_WRITE;
1234 mprotect(g2h(page_addr), qemu_host_page_size,
1235 (prot & PAGE_BITS) & ~PAGE_WRITE);
1236 #ifdef DEBUG_TB_INVALIDATE
1237 printf("protecting code page: 0x" TARGET_FMT_lx "\n",
1242 /* if some code is already present, then the pages are already
1243 protected. So we handle the case where only the first TB is
1244 allocated in a physical page */
1245 if (!last_first_tb) {
1246 tlb_protect_code(page_addr);
1250 #endif /* TARGET_HAS_SMC */
1253 /* Allocate a new translation block. Flush the translation buffer if
1254 too many translation blocks or too much generated code. */
1255 TranslationBlock *tb_alloc(target_ulong pc)
1257 TranslationBlock *tb;
1259 if (nb_tbs >= code_gen_max_blocks ||
1260 (code_gen_ptr - code_gen_buffer) >= code_gen_buffer_max_size)
1262 tb = &tbs[nb_tbs++];
1268 void tb_free(TranslationBlock *tb)
1270 /* In practice this is mostly used for single use temporary TB
1271 Ignore the hard cases and just back up if this TB happens to
1272 be the last one generated. */
1273 if (nb_tbs > 0 && tb == &tbs[nb_tbs - 1]) {
1274 code_gen_ptr = tb->tc_ptr;
1279 /* add a new TB and link it to the physical page tables. phys_page2 is
1280 (-1) to indicate that only one page contains the TB. */
1281 void tb_link_page(TranslationBlock *tb,
1282 tb_page_addr_t phys_pc, tb_page_addr_t phys_page2)
1285 TranslationBlock **ptb;
1287 /* Grab the mmap lock to stop another thread invalidating this TB
1288 before we are done. */
1290 /* add in the physical hash table */
1291 h = tb_phys_hash_func(phys_pc);
1292 ptb = &tb_phys_hash[h];
1293 tb->phys_hash_next = *ptb;
1296 /* add in the page list */
1297 tb_alloc_page(tb, 0, phys_pc & TARGET_PAGE_MASK);
1298 if (phys_page2 != -1)
1299 tb_alloc_page(tb, 1, phys_page2);
1301 tb->page_addr[1] = -1;
1303 tb->jmp_first = (TranslationBlock *)((long)tb | 2);
1304 tb->jmp_next[0] = NULL;
1305 tb->jmp_next[1] = NULL;
1307 /* init original jump addresses */
1308 if (tb->tb_next_offset[0] != 0xffff)
1309 tb_reset_jump(tb, 0);
1310 if (tb->tb_next_offset[1] != 0xffff)
1311 tb_reset_jump(tb, 1);
1313 #ifdef DEBUG_TB_CHECK
1319 /* find the TB 'tb' such that tb[0].tc_ptr <= tc_ptr <
1320 tb[1].tc_ptr. Return NULL if not found */
1321 TranslationBlock *tb_find_pc(unsigned long tc_ptr)
1323 int m_min, m_max, m;
1325 TranslationBlock *tb;
1329 if (tc_ptr < (unsigned long)code_gen_buffer ||
1330 tc_ptr >= (unsigned long)code_gen_ptr)
1332 /* binary search (cf Knuth) */
1335 while (m_min <= m_max) {
1336 m = (m_min + m_max) >> 1;
1338 v = (unsigned long)tb->tc_ptr;
1341 else if (tc_ptr < v) {
1350 static void tb_reset_jump_recursive(TranslationBlock *tb);
1352 static inline void tb_reset_jump_recursive2(TranslationBlock *tb, int n)
1354 TranslationBlock *tb1, *tb_next, **ptb;
1357 tb1 = tb->jmp_next[n];
1359 /* find head of list */
1362 tb1 = (TranslationBlock *)((long)tb1 & ~3);
1365 tb1 = tb1->jmp_next[n1];
1367 /* we are now sure now that tb jumps to tb1 */
1370 /* remove tb from the jmp_first list */
1371 ptb = &tb_next->jmp_first;
1375 tb1 = (TranslationBlock *)((long)tb1 & ~3);
1376 if (n1 == n && tb1 == tb)
1378 ptb = &tb1->jmp_next[n1];
1380 *ptb = tb->jmp_next[n];
1381 tb->jmp_next[n] = NULL;
1383 /* suppress the jump to next tb in generated code */
1384 tb_reset_jump(tb, n);
1386 /* suppress jumps in the tb on which we could have jumped */
1387 tb_reset_jump_recursive(tb_next);
1391 static void tb_reset_jump_recursive(TranslationBlock *tb)
1393 tb_reset_jump_recursive2(tb, 0);
1394 tb_reset_jump_recursive2(tb, 1);
1397 #if defined(TARGET_HAS_ICE)
1398 #if defined(CONFIG_USER_ONLY)
1399 static void breakpoint_invalidate(CPUState *env, target_ulong pc)
1401 tb_invalidate_phys_page_range(pc, pc + 1, 0);
1404 static void breakpoint_invalidate(CPUState *env, target_ulong pc)
1406 target_phys_addr_t addr;
1408 ram_addr_t ram_addr;
1411 addr = cpu_get_phys_page_debug(env, pc);
1412 p = phys_page_find(addr >> TARGET_PAGE_BITS);
1414 pd = IO_MEM_UNASSIGNED;
1416 pd = p->phys_offset;
1418 ram_addr = (pd & TARGET_PAGE_MASK) | (pc & ~TARGET_PAGE_MASK);
1419 tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
1422 #endif /* TARGET_HAS_ICE */
1424 #if defined(CONFIG_USER_ONLY)
1425 void cpu_watchpoint_remove_all(CPUState *env, int mask)
1430 int cpu_watchpoint_insert(CPUState *env, target_ulong addr, target_ulong len,
1431 int flags, CPUWatchpoint **watchpoint)
1436 /* Add a watchpoint. */
1437 int cpu_watchpoint_insert(CPUState *env, target_ulong addr, target_ulong len,
1438 int flags, CPUWatchpoint **watchpoint)
1440 target_ulong len_mask = ~(len - 1);
1443 /* sanity checks: allow power-of-2 lengths, deny unaligned watchpoints */
1444 if ((len != 1 && len != 2 && len != 4 && len != 8) || (addr & ~len_mask)) {
1445 fprintf(stderr, "qemu: tried to set invalid watchpoint at "
1446 TARGET_FMT_lx ", len=" TARGET_FMT_lu "\n", addr, len);
1449 wp = qemu_malloc(sizeof(*wp));
1452 wp->len_mask = len_mask;
1455 /* keep all GDB-injected watchpoints in front */
1457 QTAILQ_INSERT_HEAD(&env->watchpoints, wp, entry);
1459 QTAILQ_INSERT_TAIL(&env->watchpoints, wp, entry);
1461 tlb_flush_page(env, addr);
1468 /* Remove a specific watchpoint. */
1469 int cpu_watchpoint_remove(CPUState *env, target_ulong addr, target_ulong len,
1472 target_ulong len_mask = ~(len - 1);
1475 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1476 if (addr == wp->vaddr && len_mask == wp->len_mask
1477 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
1478 cpu_watchpoint_remove_by_ref(env, wp);
1485 /* Remove a specific watchpoint by reference. */
1486 void cpu_watchpoint_remove_by_ref(CPUState *env, CPUWatchpoint *watchpoint)
1488 QTAILQ_REMOVE(&env->watchpoints, watchpoint, entry);
1490 tlb_flush_page(env, watchpoint->vaddr);
1492 qemu_free(watchpoint);
1495 /* Remove all matching watchpoints. */
1496 void cpu_watchpoint_remove_all(CPUState *env, int mask)
1498 CPUWatchpoint *wp, *next;
1500 QTAILQ_FOREACH_SAFE(wp, &env->watchpoints, entry, next) {
1501 if (wp->flags & mask)
1502 cpu_watchpoint_remove_by_ref(env, wp);
1507 /* Add a breakpoint. */
1508 int cpu_breakpoint_insert(CPUState *env, target_ulong pc, int flags,
1509 CPUBreakpoint **breakpoint)
1511 #if defined(TARGET_HAS_ICE)
1514 bp = qemu_malloc(sizeof(*bp));
1519 /* keep all GDB-injected breakpoints in front */
1521 QTAILQ_INSERT_HEAD(&env->breakpoints, bp, entry);
1523 QTAILQ_INSERT_TAIL(&env->breakpoints, bp, entry);
1525 breakpoint_invalidate(env, pc);
1535 /* Remove a specific breakpoint. */
1536 int cpu_breakpoint_remove(CPUState *env, target_ulong pc, int flags)
1538 #if defined(TARGET_HAS_ICE)
1541 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
1542 if (bp->pc == pc && bp->flags == flags) {
1543 cpu_breakpoint_remove_by_ref(env, bp);
1553 /* Remove a specific breakpoint by reference. */
1554 void cpu_breakpoint_remove_by_ref(CPUState *env, CPUBreakpoint *breakpoint)
1556 #if defined(TARGET_HAS_ICE)
1557 QTAILQ_REMOVE(&env->breakpoints, breakpoint, entry);
1559 breakpoint_invalidate(env, breakpoint->pc);
1561 qemu_free(breakpoint);
1565 /* Remove all matching breakpoints. */
1566 void cpu_breakpoint_remove_all(CPUState *env, int mask)
1568 #if defined(TARGET_HAS_ICE)
1569 CPUBreakpoint *bp, *next;
1571 QTAILQ_FOREACH_SAFE(bp, &env->breakpoints, entry, next) {
1572 if (bp->flags & mask)
1573 cpu_breakpoint_remove_by_ref(env, bp);
1578 /* enable or disable single step mode. EXCP_DEBUG is returned by the
1579 CPU loop after each instruction */
1580 void cpu_single_step(CPUState *env, int enabled)
1582 #if defined(TARGET_HAS_ICE)
1583 if (env->singlestep_enabled != enabled) {
1584 env->singlestep_enabled = enabled;
1586 kvm_update_guest_debug(env, 0);
1588 /* must flush all the translated code to avoid inconsistencies */
1589 /* XXX: only flush what is necessary */
1596 /* enable or disable low levels log */
1597 void cpu_set_log(int log_flags)
1599 loglevel = log_flags;
1600 if (loglevel && !logfile) {
1601 logfile = fopen(logfilename, log_append ? "a" : "w");
1603 perror(logfilename);
1606 #if !defined(CONFIG_SOFTMMU)
1607 /* must avoid mmap() usage of glibc by setting a buffer "by hand" */
1609 static char logfile_buf[4096];
1610 setvbuf(logfile, logfile_buf, _IOLBF, sizeof(logfile_buf));
1612 #elif !defined(_WIN32)
1613 /* Win32 doesn't support line-buffering and requires size >= 2 */
1614 setvbuf(logfile, NULL, _IOLBF, 0);
1618 if (!loglevel && logfile) {
1624 void cpu_set_log_filename(const char *filename)
1626 logfilename = strdup(filename);
1631 cpu_set_log(loglevel);
1634 static void cpu_unlink_tb(CPUState *env)
1636 /* FIXME: TB unchaining isn't SMP safe. For now just ignore the
1637 problem and hope the cpu will stop of its own accord. For userspace
1638 emulation this often isn't actually as bad as it sounds. Often
1639 signals are used primarily to interrupt blocking syscalls. */
1640 TranslationBlock *tb;
1641 static spinlock_t interrupt_lock = SPIN_LOCK_UNLOCKED;
1643 spin_lock(&interrupt_lock);
1644 tb = env->current_tb;
1645 /* if the cpu is currently executing code, we must unlink it and
1646 all the potentially executing TB */
1648 env->current_tb = NULL;
1649 tb_reset_jump_recursive(tb);
1651 spin_unlock(&interrupt_lock);
1654 /* mask must never be zero, except for A20 change call */
1655 void cpu_interrupt(CPUState *env, int mask)
1659 old_mask = env->interrupt_request;
1660 env->interrupt_request |= mask;
1662 #ifndef CONFIG_USER_ONLY
1664 * If called from iothread context, wake the target cpu in
1667 if (!qemu_cpu_self(env)) {
1674 env->icount_decr.u16.high = 0xffff;
1675 #ifndef CONFIG_USER_ONLY
1677 && (mask & ~old_mask) != 0) {
1678 cpu_abort(env, "Raised interrupt while not in I/O function");
1686 void cpu_reset_interrupt(CPUState *env, int mask)
1688 env->interrupt_request &= ~mask;
1691 void cpu_exit(CPUState *env)
1693 env->exit_request = 1;
1697 const CPULogItem cpu_log_items[] = {
1698 { CPU_LOG_TB_OUT_ASM, "out_asm",
1699 "show generated host assembly code for each compiled TB" },
1700 { CPU_LOG_TB_IN_ASM, "in_asm",
1701 "show target assembly code for each compiled TB" },
1702 { CPU_LOG_TB_OP, "op",
1703 "show micro ops for each compiled TB" },
1704 { CPU_LOG_TB_OP_OPT, "op_opt",
1707 "before eflags optimization and "
1709 "after liveness analysis" },
1710 { CPU_LOG_INT, "int",
1711 "show interrupts/exceptions in short format" },
1712 { CPU_LOG_EXEC, "exec",
1713 "show trace before each executed TB (lots of logs)" },
1714 { CPU_LOG_TB_CPU, "cpu",
1715 "show CPU state before block translation" },
1717 { CPU_LOG_PCALL, "pcall",
1718 "show protected mode far calls/returns/exceptions" },
1719 { CPU_LOG_RESET, "cpu_reset",
1720 "show CPU state before CPU resets" },
1723 { CPU_LOG_IOPORT, "ioport",
1724 "show all i/o ports accesses" },
1729 #ifndef CONFIG_USER_ONLY
1730 static QLIST_HEAD(memory_client_list, CPUPhysMemoryClient) memory_client_list
1731 = QLIST_HEAD_INITIALIZER(memory_client_list);
1733 static void cpu_notify_set_memory(target_phys_addr_t start_addr,
1735 ram_addr_t phys_offset)
1737 CPUPhysMemoryClient *client;
1738 QLIST_FOREACH(client, &memory_client_list, list) {
1739 client->set_memory(client, start_addr, size, phys_offset);
1743 static int cpu_notify_sync_dirty_bitmap(target_phys_addr_t start,
1744 target_phys_addr_t end)
1746 CPUPhysMemoryClient *client;
1747 QLIST_FOREACH(client, &memory_client_list, list) {
1748 int r = client->sync_dirty_bitmap(client, start, end);
1755 static int cpu_notify_migration_log(int enable)
1757 CPUPhysMemoryClient *client;
1758 QLIST_FOREACH(client, &memory_client_list, list) {
1759 int r = client->migration_log(client, enable);
1766 static void phys_page_for_each_1(CPUPhysMemoryClient *client,
1767 int level, void **lp)
1775 PhysPageDesc *pd = *lp;
1776 for (i = 0; i < L2_SIZE; ++i) {
1777 if (pd[i].phys_offset != IO_MEM_UNASSIGNED) {
1778 client->set_memory(client, pd[i].region_offset,
1779 TARGET_PAGE_SIZE, pd[i].phys_offset);
1784 for (i = 0; i < L2_SIZE; ++i) {
1785 phys_page_for_each_1(client, level - 1, pp + i);
1790 static void phys_page_for_each(CPUPhysMemoryClient *client)
1793 for (i = 0; i < P_L1_SIZE; ++i) {
1794 phys_page_for_each_1(client, P_L1_SHIFT / L2_BITS - 1,
1799 void cpu_register_phys_memory_client(CPUPhysMemoryClient *client)
1801 QLIST_INSERT_HEAD(&memory_client_list, client, list);
1802 phys_page_for_each(client);
1805 void cpu_unregister_phys_memory_client(CPUPhysMemoryClient *client)
1807 QLIST_REMOVE(client, list);
1811 static int cmp1(const char *s1, int n, const char *s2)
1813 if (strlen(s2) != n)
1815 return memcmp(s1, s2, n) == 0;
1818 /* takes a comma separated list of log masks. Return 0 if error. */
1819 int cpu_str_to_log_mask(const char *str)
1821 const CPULogItem *item;
1828 p1 = strchr(p, ',');
1831 if(cmp1(p,p1-p,"all")) {
1832 for(item = cpu_log_items; item->mask != 0; item++) {
1836 for(item = cpu_log_items; item->mask != 0; item++) {
1837 if (cmp1(p, p1 - p, item->name))
1851 void cpu_abort(CPUState *env, const char *fmt, ...)
1858 fprintf(stderr, "qemu: fatal: ");
1859 vfprintf(stderr, fmt, ap);
1860 fprintf(stderr, "\n");
1862 cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU | X86_DUMP_CCOP);
1864 cpu_dump_state(env, stderr, fprintf, 0);
1866 if (qemu_log_enabled()) {
1867 qemu_log("qemu: fatal: ");
1868 qemu_log_vprintf(fmt, ap2);
1871 log_cpu_state(env, X86_DUMP_FPU | X86_DUMP_CCOP);
1873 log_cpu_state(env, 0);
1880 #if defined(CONFIG_USER_ONLY)
1882 struct sigaction act;
1883 sigfillset(&act.sa_mask);
1884 act.sa_handler = SIG_DFL;
1885 sigaction(SIGABRT, &act, NULL);
1891 CPUState *cpu_copy(CPUState *env)
1893 CPUState *new_env = cpu_init(env->cpu_model_str);
1894 CPUState *next_cpu = new_env->next_cpu;
1895 int cpu_index = new_env->cpu_index;
1896 #if defined(TARGET_HAS_ICE)
1901 memcpy(new_env, env, sizeof(CPUState));
1903 /* Preserve chaining and index. */
1904 new_env->next_cpu = next_cpu;
1905 new_env->cpu_index = cpu_index;
1907 /* Clone all break/watchpoints.
1908 Note: Once we support ptrace with hw-debug register access, make sure
1909 BP_CPU break/watchpoints are handled correctly on clone. */
1910 QTAILQ_INIT(&env->breakpoints);
1911 QTAILQ_INIT(&env->watchpoints);
1912 #if defined(TARGET_HAS_ICE)
1913 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
1914 cpu_breakpoint_insert(new_env, bp->pc, bp->flags, NULL);
1916 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1917 cpu_watchpoint_insert(new_env, wp->vaddr, (~wp->len_mask) + 1,
1925 #if !defined(CONFIG_USER_ONLY)
1927 static inline void tlb_flush_jmp_cache(CPUState *env, target_ulong addr)
1931 /* Discard jump cache entries for any tb which might potentially
1932 overlap the flushed page. */
1933 i = tb_jmp_cache_hash_page(addr - TARGET_PAGE_SIZE);
1934 memset (&env->tb_jmp_cache[i], 0,
1935 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1937 i = tb_jmp_cache_hash_page(addr);
1938 memset (&env->tb_jmp_cache[i], 0,
1939 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1942 static CPUTLBEntry s_cputlb_empty_entry = {
1949 /* NOTE: if flush_global is true, also flush global entries (not
1951 void tlb_flush(CPUState *env, int flush_global)
1955 #if defined(DEBUG_TLB)
1956 printf("tlb_flush:\n");
1958 /* must reset current TB so that interrupts cannot modify the
1959 links while we are modifying them */
1960 env->current_tb = NULL;
1962 for(i = 0; i < CPU_TLB_SIZE; i++) {
1964 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
1965 env->tlb_table[mmu_idx][i] = s_cputlb_empty_entry;
1969 memset (env->tb_jmp_cache, 0, TB_JMP_CACHE_SIZE * sizeof (void *));
1971 env->tlb_flush_addr = -1;
1972 env->tlb_flush_mask = 0;
1976 static inline void tlb_flush_entry(CPUTLBEntry *tlb_entry, target_ulong addr)
1978 if (addr == (tlb_entry->addr_read &
1979 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) ||
1980 addr == (tlb_entry->addr_write &
1981 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) ||
1982 addr == (tlb_entry->addr_code &
1983 (TARGET_PAGE_MASK | TLB_INVALID_MASK))) {
1984 *tlb_entry = s_cputlb_empty_entry;
1988 void tlb_flush_page(CPUState *env, target_ulong addr)
1993 #if defined(DEBUG_TLB)
1994 printf("tlb_flush_page: " TARGET_FMT_lx "\n", addr);
1996 /* Check if we need to flush due to large pages. */
1997 if ((addr & env->tlb_flush_mask) == env->tlb_flush_addr) {
1998 #if defined(DEBUG_TLB)
1999 printf("tlb_flush_page: forced full flush ("
2000 TARGET_FMT_lx "/" TARGET_FMT_lx ")\n",
2001 env->tlb_flush_addr, env->tlb_flush_mask);
2006 /* must reset current TB so that interrupts cannot modify the
2007 links while we are modifying them */
2008 env->current_tb = NULL;
2010 addr &= TARGET_PAGE_MASK;
2011 i = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2012 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++)
2013 tlb_flush_entry(&env->tlb_table[mmu_idx][i], addr);
2015 tlb_flush_jmp_cache(env, addr);
2018 /* update the TLBs so that writes to code in the virtual page 'addr'
2020 static void tlb_protect_code(ram_addr_t ram_addr)
2022 cpu_physical_memory_reset_dirty(ram_addr,
2023 ram_addr + TARGET_PAGE_SIZE,
2027 /* update the TLB so that writes in physical page 'phys_addr' are no longer
2028 tested for self modifying code */
2029 static void tlb_unprotect_code_phys(CPUState *env, ram_addr_t ram_addr,
2032 phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] |= CODE_DIRTY_FLAG;
2035 static inline void tlb_reset_dirty_range(CPUTLBEntry *tlb_entry,
2036 unsigned long start, unsigned long length)
2039 if ((tlb_entry->addr_write & ~TARGET_PAGE_MASK) == IO_MEM_RAM) {
2040 addr = (tlb_entry->addr_write & TARGET_PAGE_MASK) + tlb_entry->addend;
2041 if ((addr - start) < length) {
2042 tlb_entry->addr_write = (tlb_entry->addr_write & TARGET_PAGE_MASK) | TLB_NOTDIRTY;
2047 /* Note: start and end must be within the same ram block. */
2048 void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t end,
2052 unsigned long length, start1;
2056 start &= TARGET_PAGE_MASK;
2057 end = TARGET_PAGE_ALIGN(end);
2059 length = end - start;
2062 len = length >> TARGET_PAGE_BITS;
2063 mask = ~dirty_flags;
2064 p = phys_ram_dirty + (start >> TARGET_PAGE_BITS);
2065 for(i = 0; i < len; i++)
2068 /* we modify the TLB cache so that the dirty bit will be set again
2069 when accessing the range */
2070 start1 = (unsigned long)qemu_get_ram_ptr(start);
2071 /* Chek that we don't span multiple blocks - this breaks the
2072 address comparisons below. */
2073 if ((unsigned long)qemu_get_ram_ptr(end - 1) - start1
2074 != (end - 1) - start) {
2078 for(env = first_cpu; env != NULL; env = env->next_cpu) {
2080 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
2081 for(i = 0; i < CPU_TLB_SIZE; i++)
2082 tlb_reset_dirty_range(&env->tlb_table[mmu_idx][i],
2088 int cpu_physical_memory_set_dirty_tracking(int enable)
2091 in_migration = enable;
2092 ret = cpu_notify_migration_log(!!enable);
2096 int cpu_physical_memory_get_dirty_tracking(void)
2098 return in_migration;
2101 int cpu_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
2102 target_phys_addr_t end_addr)
2106 ret = cpu_notify_sync_dirty_bitmap(start_addr, end_addr);
2110 static inline void tlb_update_dirty(CPUTLBEntry *tlb_entry)
2112 ram_addr_t ram_addr;
2115 if ((tlb_entry->addr_write & ~TARGET_PAGE_MASK) == IO_MEM_RAM) {
2116 p = (void *)(unsigned long)((tlb_entry->addr_write & TARGET_PAGE_MASK)
2117 + tlb_entry->addend);
2118 ram_addr = qemu_ram_addr_from_host(p);
2119 if (!cpu_physical_memory_is_dirty(ram_addr)) {
2120 tlb_entry->addr_write |= TLB_NOTDIRTY;
2125 /* update the TLB according to the current state of the dirty bits */
2126 void cpu_tlb_update_dirty(CPUState *env)
2130 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
2131 for(i = 0; i < CPU_TLB_SIZE; i++)
2132 tlb_update_dirty(&env->tlb_table[mmu_idx][i]);
2136 static inline void tlb_set_dirty1(CPUTLBEntry *tlb_entry, target_ulong vaddr)
2138 if (tlb_entry->addr_write == (vaddr | TLB_NOTDIRTY))
2139 tlb_entry->addr_write = vaddr;
2142 /* update the TLB corresponding to virtual page vaddr
2143 so that it is no longer dirty */
2144 static inline void tlb_set_dirty(CPUState *env, target_ulong vaddr)
2149 vaddr &= TARGET_PAGE_MASK;
2150 i = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2151 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++)
2152 tlb_set_dirty1(&env->tlb_table[mmu_idx][i], vaddr);
2155 /* Our TLB does not support large pages, so remember the area covered by
2156 large pages and trigger a full TLB flush if these are invalidated. */
2157 static void tlb_add_large_page(CPUState *env, target_ulong vaddr,
2160 target_ulong mask = ~(size - 1);
2162 if (env->tlb_flush_addr == (target_ulong)-1) {
2163 env->tlb_flush_addr = vaddr & mask;
2164 env->tlb_flush_mask = mask;
2167 /* Extend the existing region to include the new page.
2168 This is a compromise between unnecessary flushes and the cost
2169 of maintaining a full variable size TLB. */
2170 mask &= env->tlb_flush_mask;
2171 while (((env->tlb_flush_addr ^ vaddr) & mask) != 0) {
2174 env->tlb_flush_addr &= mask;
2175 env->tlb_flush_mask = mask;
2178 /* Add a new TLB entry. At most one entry for a given virtual address
2179 is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
2180 supplied size is only used by tlb_flush_page. */
2181 void tlb_set_page(CPUState *env, target_ulong vaddr,
2182 target_phys_addr_t paddr, int prot,
2183 int mmu_idx, target_ulong size)
2188 target_ulong address;
2189 target_ulong code_address;
2190 target_phys_addr_t addend;
2193 target_phys_addr_t iotlb;
2195 assert(size >= TARGET_PAGE_SIZE);
2196 if (size != TARGET_PAGE_SIZE) {
2197 tlb_add_large_page(env, vaddr, size);
2199 p = phys_page_find(paddr >> TARGET_PAGE_BITS);
2201 pd = IO_MEM_UNASSIGNED;
2203 pd = p->phys_offset;
2205 #if defined(DEBUG_TLB)
2206 printf("tlb_set_page: vaddr=" TARGET_FMT_lx " paddr=0x%08x prot=%x idx=%d smmu=%d pd=0x%08lx\n",
2207 vaddr, (int)paddr, prot, mmu_idx, is_softmmu, pd);
2211 if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM && !(pd & IO_MEM_ROMD)) {
2212 /* IO memory case (romd handled later) */
2213 address |= TLB_MMIO;
2215 addend = (unsigned long)qemu_get_ram_ptr(pd & TARGET_PAGE_MASK);
2216 if ((pd & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
2218 iotlb = pd & TARGET_PAGE_MASK;
2219 if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_RAM)
2220 iotlb |= IO_MEM_NOTDIRTY;
2222 iotlb |= IO_MEM_ROM;
2224 /* IO handlers are currently passed a physical address.
2225 It would be nice to pass an offset from the base address
2226 of that region. This would avoid having to special case RAM,
2227 and avoid full address decoding in every device.
2228 We can't use the high bits of pd for this because
2229 IO_MEM_ROMD uses these as a ram address. */
2230 iotlb = (pd & ~TARGET_PAGE_MASK);
2232 iotlb += p->region_offset;
2238 code_address = address;
2239 /* Make accesses to pages with watchpoints go via the
2240 watchpoint trap routines. */
2241 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
2242 if (vaddr == (wp->vaddr & TARGET_PAGE_MASK)) {
2243 iotlb = io_mem_watch + paddr;
2244 /* TODO: The memory case can be optimized by not trapping
2245 reads of pages with a write breakpoint. */
2246 address |= TLB_MMIO;
2250 index = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2251 env->iotlb[mmu_idx][index] = iotlb - vaddr;
2252 te = &env->tlb_table[mmu_idx][index];
2253 te->addend = addend - vaddr;
2254 if (prot & PAGE_READ) {
2255 te->addr_read = address;
2260 if (prot & PAGE_EXEC) {
2261 te->addr_code = code_address;
2265 if (prot & PAGE_WRITE) {
2266 if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_ROM ||
2267 (pd & IO_MEM_ROMD)) {
2268 /* Write access calls the I/O callback. */
2269 te->addr_write = address | TLB_MMIO;
2270 } else if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_RAM &&
2271 !cpu_physical_memory_is_dirty(pd)) {
2272 te->addr_write = address | TLB_NOTDIRTY;
2274 te->addr_write = address;
2277 te->addr_write = -1;
2283 void tlb_flush(CPUState *env, int flush_global)
2287 void tlb_flush_page(CPUState *env, target_ulong addr)
2292 * Walks guest process memory "regions" one by one
2293 * and calls callback function 'fn' for each region.
2296 struct walk_memory_regions_data
2298 walk_memory_regions_fn fn;
2300 unsigned long start;
2304 static int walk_memory_regions_end(struct walk_memory_regions_data *data,
2305 abi_ulong end, int new_prot)
2307 if (data->start != -1ul) {
2308 int rc = data->fn(data->priv, data->start, end, data->prot);
2314 data->start = (new_prot ? end : -1ul);
2315 data->prot = new_prot;
2320 static int walk_memory_regions_1(struct walk_memory_regions_data *data,
2321 abi_ulong base, int level, void **lp)
2327 return walk_memory_regions_end(data, base, 0);
2332 for (i = 0; i < L2_SIZE; ++i) {
2333 int prot = pd[i].flags;
2335 pa = base | (i << TARGET_PAGE_BITS);
2336 if (prot != data->prot) {
2337 rc = walk_memory_regions_end(data, pa, prot);
2345 for (i = 0; i < L2_SIZE; ++i) {
2346 pa = base | ((abi_ulong)i <<
2347 (TARGET_PAGE_BITS + L2_BITS * level));
2348 rc = walk_memory_regions_1(data, pa, level - 1, pp + i);
2358 int walk_memory_regions(void *priv, walk_memory_regions_fn fn)
2360 struct walk_memory_regions_data data;
2368 for (i = 0; i < V_L1_SIZE; i++) {
2369 int rc = walk_memory_regions_1(&data, (abi_ulong)i << V_L1_SHIFT,
2370 V_L1_SHIFT / L2_BITS - 1, l1_map + i);
2376 return walk_memory_regions_end(&data, 0, 0);
2379 static int dump_region(void *priv, abi_ulong start,
2380 abi_ulong end, unsigned long prot)
2382 FILE *f = (FILE *)priv;
2384 (void) fprintf(f, TARGET_ABI_FMT_lx"-"TARGET_ABI_FMT_lx
2385 " "TARGET_ABI_FMT_lx" %c%c%c\n",
2386 start, end, end - start,
2387 ((prot & PAGE_READ) ? 'r' : '-'),
2388 ((prot & PAGE_WRITE) ? 'w' : '-'),
2389 ((prot & PAGE_EXEC) ? 'x' : '-'));
2394 /* dump memory mappings */
2395 void page_dump(FILE *f)
2397 (void) fprintf(f, "%-8s %-8s %-8s %s\n",
2398 "start", "end", "size", "prot");
2399 walk_memory_regions(f, dump_region);
2402 int page_get_flags(target_ulong address)
2406 p = page_find(address >> TARGET_PAGE_BITS);
2412 /* Modify the flags of a page and invalidate the code if necessary.
2413 The flag PAGE_WRITE_ORG is positioned automatically depending
2414 on PAGE_WRITE. The mmap_lock should already be held. */
2415 void page_set_flags(target_ulong start, target_ulong end, int flags)
2417 target_ulong addr, len;
2419 /* This function should never be called with addresses outside the
2420 guest address space. If this assert fires, it probably indicates
2421 a missing call to h2g_valid. */
2422 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
2423 assert(end < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
2425 assert(start < end);
2427 start = start & TARGET_PAGE_MASK;
2428 end = TARGET_PAGE_ALIGN(end);
2430 if (flags & PAGE_WRITE) {
2431 flags |= PAGE_WRITE_ORG;
2434 for (addr = start, len = end - start;
2436 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
2437 PageDesc *p = page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
2439 /* If the write protection bit is set, then we invalidate
2441 if (!(p->flags & PAGE_WRITE) &&
2442 (flags & PAGE_WRITE) &&
2444 tb_invalidate_phys_page(addr, 0, NULL);
2450 int page_check_range(target_ulong start, target_ulong len, int flags)
2456 /* This function should never be called with addresses outside the
2457 guest address space. If this assert fires, it probably indicates
2458 a missing call to h2g_valid. */
2459 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
2460 assert(start < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
2463 if (start + len - 1 < start) {
2464 /* We've wrapped around. */
2468 end = TARGET_PAGE_ALIGN(start+len); /* must do before we loose bits in the next step */
2469 start = start & TARGET_PAGE_MASK;
2471 for (addr = start, len = end - start;
2473 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
2474 p = page_find(addr >> TARGET_PAGE_BITS);
2477 if( !(p->flags & PAGE_VALID) )
2480 if ((flags & PAGE_READ) && !(p->flags & PAGE_READ))
2482 if (flags & PAGE_WRITE) {
2483 if (!(p->flags & PAGE_WRITE_ORG))
2485 /* unprotect the page if it was put read-only because it
2486 contains translated code */
2487 if (!(p->flags & PAGE_WRITE)) {
2488 if (!page_unprotect(addr, 0, NULL))
2497 /* called from signal handler: invalidate the code and unprotect the
2498 page. Return TRUE if the fault was successfully handled. */
2499 int page_unprotect(target_ulong address, unsigned long pc, void *puc)
2501 unsigned int page_index, prot, pindex;
2503 target_ulong host_start, host_end, addr;
2505 /* Technically this isn't safe inside a signal handler. However we
2506 know this only ever happens in a synchronous SEGV handler, so in
2507 practice it seems to be ok. */
2510 host_start = address & qemu_host_page_mask;
2511 page_index = host_start >> TARGET_PAGE_BITS;
2512 p1 = page_find(page_index);
2517 host_end = host_start + qemu_host_page_size;
2520 for(addr = host_start;addr < host_end; addr += TARGET_PAGE_SIZE) {
2524 /* if the page was really writable, then we change its
2525 protection back to writable */
2526 if (prot & PAGE_WRITE_ORG) {
2527 pindex = (address - host_start) >> TARGET_PAGE_BITS;
2528 if (!(p1[pindex].flags & PAGE_WRITE)) {
2529 mprotect((void *)g2h(host_start), qemu_host_page_size,
2530 (prot & PAGE_BITS) | PAGE_WRITE);
2531 p1[pindex].flags |= PAGE_WRITE;
2532 /* and since the content will be modified, we must invalidate
2533 the corresponding translated code. */
2534 tb_invalidate_phys_page(address, pc, puc);
2535 #ifdef DEBUG_TB_CHECK
2536 tb_invalidate_check(address);
2546 static inline void tlb_set_dirty(CPUState *env,
2547 unsigned long addr, target_ulong vaddr)
2550 #endif /* defined(CONFIG_USER_ONLY) */
2552 #if !defined(CONFIG_USER_ONLY)
2554 #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
2555 typedef struct subpage_t {
2556 target_phys_addr_t base;
2557 CPUReadMemoryFunc * const *mem_read[TARGET_PAGE_SIZE][4];
2558 CPUWriteMemoryFunc * const *mem_write[TARGET_PAGE_SIZE][4];
2559 void *opaque[TARGET_PAGE_SIZE][2][4];
2560 ram_addr_t region_offset[TARGET_PAGE_SIZE][2][4];
2563 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
2564 ram_addr_t memory, ram_addr_t region_offset);
2565 static void *subpage_init (target_phys_addr_t base, ram_addr_t *phys,
2566 ram_addr_t orig_memory, ram_addr_t region_offset);
2567 #define CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr, end_addr2, \
2570 if (addr > start_addr) \
2573 start_addr2 = start_addr & ~TARGET_PAGE_MASK; \
2574 if (start_addr2 > 0) \
2578 if ((start_addr + orig_size) - addr >= TARGET_PAGE_SIZE) \
2579 end_addr2 = TARGET_PAGE_SIZE - 1; \
2581 end_addr2 = (start_addr + orig_size - 1) & ~TARGET_PAGE_MASK; \
2582 if (end_addr2 < TARGET_PAGE_SIZE - 1) \
2587 /* register physical memory.
2588 For RAM, 'size' must be a multiple of the target page size.
2589 If (phys_offset & ~TARGET_PAGE_MASK) != 0, then it is an
2590 io memory page. The address used when calling the IO function is
2591 the offset from the start of the region, plus region_offset. Both
2592 start_addr and region_offset are rounded down to a page boundary
2593 before calculating this offset. This should not be a problem unless
2594 the low bits of start_addr and region_offset differ. */
2595 void cpu_register_physical_memory_offset(target_phys_addr_t start_addr,
2597 ram_addr_t phys_offset,
2598 ram_addr_t region_offset)
2600 target_phys_addr_t addr, end_addr;
2603 ram_addr_t orig_size = size;
2606 cpu_notify_set_memory(start_addr, size, phys_offset);
2608 if (phys_offset == IO_MEM_UNASSIGNED) {
2609 region_offset = start_addr;
2611 region_offset &= TARGET_PAGE_MASK;
2612 size = (size + TARGET_PAGE_SIZE - 1) & TARGET_PAGE_MASK;
2613 end_addr = start_addr + (target_phys_addr_t)size;
2614 for(addr = start_addr; addr != end_addr; addr += TARGET_PAGE_SIZE) {
2615 p = phys_page_find(addr >> TARGET_PAGE_BITS);
2616 if (p && p->phys_offset != IO_MEM_UNASSIGNED) {
2617 ram_addr_t orig_memory = p->phys_offset;
2618 target_phys_addr_t start_addr2, end_addr2;
2619 int need_subpage = 0;
2621 CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr, end_addr2,
2623 if (need_subpage || phys_offset & IO_MEM_SUBWIDTH) {
2624 if (!(orig_memory & IO_MEM_SUBPAGE)) {
2625 subpage = subpage_init((addr & TARGET_PAGE_MASK),
2626 &p->phys_offset, orig_memory,
2629 subpage = io_mem_opaque[(orig_memory & ~TARGET_PAGE_MASK)
2632 subpage_register(subpage, start_addr2, end_addr2, phys_offset,
2634 p->region_offset = 0;
2636 p->phys_offset = phys_offset;
2637 if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM ||
2638 (phys_offset & IO_MEM_ROMD))
2639 phys_offset += TARGET_PAGE_SIZE;
2642 p = phys_page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
2643 p->phys_offset = phys_offset;
2644 p->region_offset = region_offset;
2645 if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM ||
2646 (phys_offset & IO_MEM_ROMD)) {
2647 phys_offset += TARGET_PAGE_SIZE;
2649 target_phys_addr_t start_addr2, end_addr2;
2650 int need_subpage = 0;
2652 CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr,
2653 end_addr2, need_subpage);
2655 if (need_subpage || phys_offset & IO_MEM_SUBWIDTH) {
2656 subpage = subpage_init((addr & TARGET_PAGE_MASK),
2657 &p->phys_offset, IO_MEM_UNASSIGNED,
2658 addr & TARGET_PAGE_MASK);
2659 subpage_register(subpage, start_addr2, end_addr2,
2660 phys_offset, region_offset);
2661 p->region_offset = 0;
2665 region_offset += TARGET_PAGE_SIZE;
2668 /* since each CPU stores ram addresses in its TLB cache, we must
2669 reset the modified entries */
2671 for(env = first_cpu; env != NULL; env = env->next_cpu) {
2676 /* XXX: temporary until new memory mapping API */
2677 ram_addr_t cpu_get_physical_page_desc(target_phys_addr_t addr)
2681 p = phys_page_find(addr >> TARGET_PAGE_BITS);
2683 return IO_MEM_UNASSIGNED;
2684 return p->phys_offset;
2687 void qemu_register_coalesced_mmio(target_phys_addr_t addr, ram_addr_t size)
2690 kvm_coalesce_mmio_region(addr, size);
2693 void qemu_unregister_coalesced_mmio(target_phys_addr_t addr, ram_addr_t size)
2696 kvm_uncoalesce_mmio_region(addr, size);
2699 void qemu_flush_coalesced_mmio_buffer(void)
2702 kvm_flush_coalesced_mmio_buffer();
2705 #if defined(__linux__) && !defined(TARGET_S390X)
2707 #include <sys/vfs.h>
2709 #define HUGETLBFS_MAGIC 0x958458f6
2711 static long gethugepagesize(const char *path)
2717 ret = statfs(path, &fs);
2718 } while (ret != 0 && errno == EINTR);
2725 if (fs.f_type != HUGETLBFS_MAGIC)
2726 fprintf(stderr, "Warning: path not on HugeTLBFS: %s\n", path);
2731 static void *file_ram_alloc(ram_addr_t memory, const char *path)
2739 unsigned long hpagesize;
2741 hpagesize = gethugepagesize(path);
2746 if (memory < hpagesize) {
2750 if (kvm_enabled() && !kvm_has_sync_mmu()) {
2751 fprintf(stderr, "host lacks kvm mmu notifiers, -mem-path unsupported\n");
2755 if (asprintf(&filename, "%s/qemu_back_mem.XXXXXX", path) == -1) {
2759 fd = mkstemp(filename);
2761 perror("unable to create backing store for hugepages");
2768 memory = (memory+hpagesize-1) & ~(hpagesize-1);
2771 * ftruncate is not supported by hugetlbfs in older
2772 * hosts, so don't bother bailing out on errors.
2773 * If anything goes wrong with it under other filesystems,
2776 if (ftruncate(fd, memory))
2777 perror("ftruncate");
2780 /* NB: MAP_POPULATE won't exhaustively alloc all phys pages in the case
2781 * MAP_PRIVATE is requested. For mem_prealloc we mmap as MAP_SHARED
2782 * to sidestep this quirk.
2784 flags = mem_prealloc ? MAP_POPULATE | MAP_SHARED : MAP_PRIVATE;
2785 area = mmap(0, memory, PROT_READ | PROT_WRITE, flags, fd, 0);
2787 area = mmap(0, memory, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
2789 if (area == MAP_FAILED) {
2790 perror("file_ram_alloc: can't mmap RAM pages");
2798 ram_addr_t qemu_ram_alloc(ram_addr_t size)
2800 RAMBlock *new_block;
2802 size = TARGET_PAGE_ALIGN(size);
2803 new_block = qemu_malloc(sizeof(*new_block));
2806 #if defined (__linux__) && !defined(TARGET_S390X)
2807 new_block->host = file_ram_alloc(size, mem_path);
2808 if (!new_block->host)
2811 fprintf(stderr, "-mem-path option unsupported\n");
2815 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2816 /* XXX S390 KVM requires the topmost vma of the RAM to be < 256GB */
2817 new_block->host = mmap((void*)0x1000000, size,
2818 PROT_EXEC|PROT_READ|PROT_WRITE,
2819 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
2821 new_block->host = qemu_vmalloc(size);
2823 #ifdef MADV_MERGEABLE
2824 madvise(new_block->host, size, MADV_MERGEABLE);
2827 new_block->offset = last_ram_offset;
2828 new_block->length = size;
2830 new_block->next = ram_blocks;
2831 ram_blocks = new_block;
2833 phys_ram_dirty = qemu_realloc(phys_ram_dirty,
2834 (last_ram_offset + size) >> TARGET_PAGE_BITS);
2835 memset(phys_ram_dirty + (last_ram_offset >> TARGET_PAGE_BITS),
2836 0xff, size >> TARGET_PAGE_BITS);
2838 last_ram_offset += size;
2841 kvm_setup_guest_memory(new_block->host, size);
2843 return new_block->offset;
2846 void qemu_ram_free(ram_addr_t addr)
2848 /* TODO: implement this. */
2851 /* Return a host pointer to ram allocated with qemu_ram_alloc.
2852 With the exception of the softmmu code in this file, this should
2853 only be used for local memory (e.g. video ram) that the device owns,
2854 and knows it isn't going to access beyond the end of the block.
2856 It should not be used for general purpose DMA.
2857 Use cpu_physical_memory_map/cpu_physical_memory_rw instead.
2859 void *qemu_get_ram_ptr(ram_addr_t addr)
2866 prevp = &ram_blocks;
2868 while (block && (block->offset > addr
2869 || block->offset + block->length <= addr)) {
2871 prevp = &prev->next;
2873 block = block->next;
2876 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
2879 /* Move this entry to to start of the list. */
2881 prev->next = block->next;
2882 block->next = *prevp;
2885 return block->host + (addr - block->offset);
2888 /* Some of the softmmu routines need to translate from a host pointer
2889 (typically a TLB entry) back to a ram offset. */
2890 ram_addr_t qemu_ram_addr_from_host(void *ptr)
2894 uint8_t *host = ptr;
2898 while (block && (block->host > host
2899 || block->host + block->length <= host)) {
2901 block = block->next;
2904 fprintf(stderr, "Bad ram pointer %p\n", ptr);
2907 return block->offset + (host - block->host);
2910 static uint32_t unassigned_mem_readb(void *opaque, target_phys_addr_t addr)
2912 #ifdef DEBUG_UNASSIGNED
2913 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
2915 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
2916 do_unassigned_access(addr, 0, 0, 0, 1);
2921 static uint32_t unassigned_mem_readw(void *opaque, target_phys_addr_t addr)
2923 #ifdef DEBUG_UNASSIGNED
2924 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
2926 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
2927 do_unassigned_access(addr, 0, 0, 0, 2);
2932 static uint32_t unassigned_mem_readl(void *opaque, target_phys_addr_t addr)
2934 #ifdef DEBUG_UNASSIGNED
2935 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
2937 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
2938 do_unassigned_access(addr, 0, 0, 0, 4);
2943 static void unassigned_mem_writeb(void *opaque, target_phys_addr_t addr, uint32_t val)
2945 #ifdef DEBUG_UNASSIGNED
2946 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%x\n", addr, val);
2948 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
2949 do_unassigned_access(addr, 1, 0, 0, 1);
2953 static void unassigned_mem_writew(void *opaque, target_phys_addr_t addr, uint32_t val)
2955 #ifdef DEBUG_UNASSIGNED
2956 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%x\n", addr, val);
2958 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
2959 do_unassigned_access(addr, 1, 0, 0, 2);
2963 static void unassigned_mem_writel(void *opaque, target_phys_addr_t addr, uint32_t val)
2965 #ifdef DEBUG_UNASSIGNED
2966 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%x\n", addr, val);
2968 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
2969 do_unassigned_access(addr, 1, 0, 0, 4);
2973 static CPUReadMemoryFunc * const unassigned_mem_read[3] = {
2974 unassigned_mem_readb,
2975 unassigned_mem_readw,
2976 unassigned_mem_readl,
2979 static CPUWriteMemoryFunc * const unassigned_mem_write[3] = {
2980 unassigned_mem_writeb,
2981 unassigned_mem_writew,
2982 unassigned_mem_writel,
2985 static void notdirty_mem_writeb(void *opaque, target_phys_addr_t ram_addr,
2989 dirty_flags = phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS];
2990 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
2991 #if !defined(CONFIG_USER_ONLY)
2992 tb_invalidate_phys_page_fast(ram_addr, 1);
2993 dirty_flags = phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS];
2996 stb_p(qemu_get_ram_ptr(ram_addr), val);
2997 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
2998 phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] = dirty_flags;
2999 /* we remove the notdirty callback only if the code has been
3001 if (dirty_flags == 0xff)
3002 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3005 static void notdirty_mem_writew(void *opaque, target_phys_addr_t ram_addr,
3009 dirty_flags = phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS];
3010 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
3011 #if !defined(CONFIG_USER_ONLY)
3012 tb_invalidate_phys_page_fast(ram_addr, 2);
3013 dirty_flags = phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS];
3016 stw_p(qemu_get_ram_ptr(ram_addr), val);
3017 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
3018 phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] = dirty_flags;
3019 /* we remove the notdirty callback only if the code has been
3021 if (dirty_flags == 0xff)
3022 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3025 static void notdirty_mem_writel(void *opaque, target_phys_addr_t ram_addr,
3029 dirty_flags = phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS];
3030 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
3031 #if !defined(CONFIG_USER_ONLY)
3032 tb_invalidate_phys_page_fast(ram_addr, 4);
3033 dirty_flags = phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS];
3036 stl_p(qemu_get_ram_ptr(ram_addr), val);
3037 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
3038 phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] = dirty_flags;
3039 /* we remove the notdirty callback only if the code has been
3041 if (dirty_flags == 0xff)
3042 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3045 static CPUReadMemoryFunc * const error_mem_read[3] = {
3046 NULL, /* never used */
3047 NULL, /* never used */
3048 NULL, /* never used */
3051 static CPUWriteMemoryFunc * const notdirty_mem_write[3] = {
3052 notdirty_mem_writeb,
3053 notdirty_mem_writew,
3054 notdirty_mem_writel,
3057 /* Generate a debug exception if a watchpoint has been hit. */
3058 static void check_watchpoint(int offset, int len_mask, int flags)
3060 CPUState *env = cpu_single_env;
3061 target_ulong pc, cs_base;
3062 TranslationBlock *tb;
3067 if (env->watchpoint_hit) {
3068 /* We re-entered the check after replacing the TB. Now raise
3069 * the debug interrupt so that is will trigger after the
3070 * current instruction. */
3071 cpu_interrupt(env, CPU_INTERRUPT_DEBUG);
3074 vaddr = (env->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
3075 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
3076 if ((vaddr == (wp->vaddr & len_mask) ||
3077 (vaddr & wp->len_mask) == wp->vaddr) && (wp->flags & flags)) {
3078 wp->flags |= BP_WATCHPOINT_HIT;
3079 if (!env->watchpoint_hit) {
3080 env->watchpoint_hit = wp;
3081 tb = tb_find_pc(env->mem_io_pc);
3083 cpu_abort(env, "check_watchpoint: could not find TB for "
3084 "pc=%p", (void *)env->mem_io_pc);
3086 cpu_restore_state(tb, env, env->mem_io_pc, NULL);
3087 tb_phys_invalidate(tb, -1);
3088 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
3089 env->exception_index = EXCP_DEBUG;
3091 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
3092 tb_gen_code(env, pc, cs_base, cpu_flags, 1);
3094 cpu_resume_from_signal(env, NULL);
3097 wp->flags &= ~BP_WATCHPOINT_HIT;
3102 /* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
3103 so these check for a hit then pass through to the normal out-of-line
3105 static uint32_t watch_mem_readb(void *opaque, target_phys_addr_t addr)
3107 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x0, BP_MEM_READ);
3108 return ldub_phys(addr);
3111 static uint32_t watch_mem_readw(void *opaque, target_phys_addr_t addr)
3113 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x1, BP_MEM_READ);
3114 return lduw_phys(addr);
3117 static uint32_t watch_mem_readl(void *opaque, target_phys_addr_t addr)
3119 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x3, BP_MEM_READ);
3120 return ldl_phys(addr);
3123 static void watch_mem_writeb(void *opaque, target_phys_addr_t addr,
3126 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x0, BP_MEM_WRITE);
3127 stb_phys(addr, val);
3130 static void watch_mem_writew(void *opaque, target_phys_addr_t addr,
3133 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x1, BP_MEM_WRITE);
3134 stw_phys(addr, val);
3137 static void watch_mem_writel(void *opaque, target_phys_addr_t addr,
3140 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x3, BP_MEM_WRITE);
3141 stl_phys(addr, val);
3144 static CPUReadMemoryFunc * const watch_mem_read[3] = {
3150 static CPUWriteMemoryFunc * const watch_mem_write[3] = {
3156 static inline uint32_t subpage_readlen (subpage_t *mmio, target_phys_addr_t addr,
3162 idx = SUBPAGE_IDX(addr);
3163 #if defined(DEBUG_SUBPAGE)
3164 printf("%s: subpage %p len %d addr " TARGET_FMT_plx " idx %d\n", __func__,
3165 mmio, len, addr, idx);
3167 ret = (**mmio->mem_read[idx][len])(mmio->opaque[idx][0][len],
3168 addr + mmio->region_offset[idx][0][len]);
3173 static inline void subpage_writelen (subpage_t *mmio, target_phys_addr_t addr,
3174 uint32_t value, unsigned int len)
3178 idx = SUBPAGE_IDX(addr);
3179 #if defined(DEBUG_SUBPAGE)
3180 printf("%s: subpage %p len %d addr " TARGET_FMT_plx " idx %d value %08x\n", __func__,
3181 mmio, len, addr, idx, value);
3183 (**mmio->mem_write[idx][len])(mmio->opaque[idx][1][len],
3184 addr + mmio->region_offset[idx][1][len],
3188 static uint32_t subpage_readb (void *opaque, target_phys_addr_t addr)
3190 #if defined(DEBUG_SUBPAGE)
3191 printf("%s: addr " TARGET_FMT_plx "\n", __func__, addr);
3194 return subpage_readlen(opaque, addr, 0);
3197 static void subpage_writeb (void *opaque, target_phys_addr_t addr,
3200 #if defined(DEBUG_SUBPAGE)
3201 printf("%s: addr " TARGET_FMT_plx " val %08x\n", __func__, addr, value);
3203 subpage_writelen(opaque, addr, value, 0);
3206 static uint32_t subpage_readw (void *opaque, target_phys_addr_t addr)
3208 #if defined(DEBUG_SUBPAGE)
3209 printf("%s: addr " TARGET_FMT_plx "\n", __func__, addr);
3212 return subpage_readlen(opaque, addr, 1);
3215 static void subpage_writew (void *opaque, target_phys_addr_t addr,
3218 #if defined(DEBUG_SUBPAGE)
3219 printf("%s: addr " TARGET_FMT_plx " val %08x\n", __func__, addr, value);
3221 subpage_writelen(opaque, addr, value, 1);
3224 static uint32_t subpage_readl (void *opaque, target_phys_addr_t addr)
3226 #if defined(DEBUG_SUBPAGE)
3227 printf("%s: addr " TARGET_FMT_plx "\n", __func__, addr);
3230 return subpage_readlen(opaque, addr, 2);
3233 static void subpage_writel (void *opaque,
3234 target_phys_addr_t addr, uint32_t value)
3236 #if defined(DEBUG_SUBPAGE)
3237 printf("%s: addr " TARGET_FMT_plx " val %08x\n", __func__, addr, value);
3239 subpage_writelen(opaque, addr, value, 2);
3242 static CPUReadMemoryFunc * const subpage_read[] = {
3248 static CPUWriteMemoryFunc * const subpage_write[] = {
3254 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
3255 ram_addr_t memory, ram_addr_t region_offset)
3260 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
3262 idx = SUBPAGE_IDX(start);
3263 eidx = SUBPAGE_IDX(end);
3264 #if defined(DEBUG_SUBPAGE)
3265 printf("%s: %p start %08x end %08x idx %08x eidx %08x mem %ld\n", __func__,
3266 mmio, start, end, idx, eidx, memory);
3268 memory >>= IO_MEM_SHIFT;
3269 for (; idx <= eidx; idx++) {
3270 for (i = 0; i < 4; i++) {
3271 if (io_mem_read[memory][i]) {
3272 mmio->mem_read[idx][i] = &io_mem_read[memory][i];
3273 mmio->opaque[idx][0][i] = io_mem_opaque[memory];
3274 mmio->region_offset[idx][0][i] = region_offset;
3276 if (io_mem_write[memory][i]) {
3277 mmio->mem_write[idx][i] = &io_mem_write[memory][i];
3278 mmio->opaque[idx][1][i] = io_mem_opaque[memory];
3279 mmio->region_offset[idx][1][i] = region_offset;
3287 static void *subpage_init (target_phys_addr_t base, ram_addr_t *phys,
3288 ram_addr_t orig_memory, ram_addr_t region_offset)
3293 mmio = qemu_mallocz(sizeof(subpage_t));
3296 subpage_memory = cpu_register_io_memory(subpage_read, subpage_write, mmio);
3297 #if defined(DEBUG_SUBPAGE)
3298 printf("%s: %p base " TARGET_FMT_plx " len %08x %d\n", __func__,
3299 mmio, base, TARGET_PAGE_SIZE, subpage_memory);
3301 *phys = subpage_memory | IO_MEM_SUBPAGE;
3302 subpage_register(mmio, 0, TARGET_PAGE_SIZE - 1, orig_memory,
3308 static int get_free_io_mem_idx(void)
3312 for (i = 0; i<IO_MEM_NB_ENTRIES; i++)
3313 if (!io_mem_used[i]) {
3317 fprintf(stderr, "RAN out out io_mem_idx, max %d !\n", IO_MEM_NB_ENTRIES);
3321 /* mem_read and mem_write are arrays of functions containing the
3322 function to access byte (index 0), word (index 1) and dword (index
3323 2). Functions can be omitted with a NULL function pointer.
3324 If io_index is non zero, the corresponding io zone is
3325 modified. If it is zero, a new io zone is allocated. The return
3326 value can be used with cpu_register_physical_memory(). (-1) is
3327 returned if error. */
3328 static int cpu_register_io_memory_fixed(int io_index,
3329 CPUReadMemoryFunc * const *mem_read,
3330 CPUWriteMemoryFunc * const *mem_write,
3333 int i, subwidth = 0;
3335 if (io_index <= 0) {
3336 io_index = get_free_io_mem_idx();
3340 io_index >>= IO_MEM_SHIFT;
3341 if (io_index >= IO_MEM_NB_ENTRIES)
3345 for(i = 0;i < 3; i++) {
3346 if (!mem_read[i] || !mem_write[i])
3347 subwidth = IO_MEM_SUBWIDTH;
3348 io_mem_read[io_index][i] = mem_read[i];
3349 io_mem_write[io_index][i] = mem_write[i];
3351 io_mem_opaque[io_index] = opaque;
3352 return (io_index << IO_MEM_SHIFT) | subwidth;
3355 int cpu_register_io_memory(CPUReadMemoryFunc * const *mem_read,
3356 CPUWriteMemoryFunc * const *mem_write,
3359 return cpu_register_io_memory_fixed(0, mem_read, mem_write, opaque);
3362 void cpu_unregister_io_memory(int io_table_address)
3365 int io_index = io_table_address >> IO_MEM_SHIFT;
3367 for (i=0;i < 3; i++) {
3368 io_mem_read[io_index][i] = unassigned_mem_read[i];
3369 io_mem_write[io_index][i] = unassigned_mem_write[i];
3371 io_mem_opaque[io_index] = NULL;
3372 io_mem_used[io_index] = 0;
3375 static void io_mem_init(void)
3379 cpu_register_io_memory_fixed(IO_MEM_ROM, error_mem_read, unassigned_mem_write, NULL);
3380 cpu_register_io_memory_fixed(IO_MEM_UNASSIGNED, unassigned_mem_read, unassigned_mem_write, NULL);
3381 cpu_register_io_memory_fixed(IO_MEM_NOTDIRTY, error_mem_read, notdirty_mem_write, NULL);
3385 io_mem_watch = cpu_register_io_memory(watch_mem_read,
3386 watch_mem_write, NULL);
3389 #endif /* !defined(CONFIG_USER_ONLY) */
3391 /* physical memory access (slow version, mainly for debug) */
3392 #if defined(CONFIG_USER_ONLY)
3393 int cpu_memory_rw_debug(CPUState *env, target_ulong addr,
3394 uint8_t *buf, int len, int is_write)
3401 page = addr & TARGET_PAGE_MASK;
3402 l = (page + TARGET_PAGE_SIZE) - addr;
3405 flags = page_get_flags(page);
3406 if (!(flags & PAGE_VALID))
3409 if (!(flags & PAGE_WRITE))
3411 /* XXX: this code should not depend on lock_user */
3412 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
3415 unlock_user(p, addr, l);
3417 if (!(flags & PAGE_READ))
3419 /* XXX: this code should not depend on lock_user */
3420 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
3423 unlock_user(p, addr, 0);
3433 void cpu_physical_memory_rw(target_phys_addr_t addr, uint8_t *buf,
3434 int len, int is_write)
3439 target_phys_addr_t page;
3444 page = addr & TARGET_PAGE_MASK;
3445 l = (page + TARGET_PAGE_SIZE) - addr;
3448 p = phys_page_find(page >> TARGET_PAGE_BITS);
3450 pd = IO_MEM_UNASSIGNED;
3452 pd = p->phys_offset;
3456 if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
3457 target_phys_addr_t addr1 = addr;
3458 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3460 addr1 = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3461 /* XXX: could force cpu_single_env to NULL to avoid
3463 if (l >= 4 && ((addr1 & 3) == 0)) {
3464 /* 32 bit write access */
3466 io_mem_write[io_index][2](io_mem_opaque[io_index], addr1, val);
3468 } else if (l >= 2 && ((addr1 & 1) == 0)) {
3469 /* 16 bit write access */
3471 io_mem_write[io_index][1](io_mem_opaque[io_index], addr1, val);
3474 /* 8 bit write access */
3476 io_mem_write[io_index][0](io_mem_opaque[io_index], addr1, val);
3480 unsigned long addr1;
3481 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3483 ptr = qemu_get_ram_ptr(addr1);
3484 memcpy(ptr, buf, l);
3485 if (!cpu_physical_memory_is_dirty(addr1)) {
3486 /* invalidate code */
3487 tb_invalidate_phys_page_range(addr1, addr1 + l, 0);
3489 phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] |=
3490 (0xff & ~CODE_DIRTY_FLAG);
3494 if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM &&
3495 !(pd & IO_MEM_ROMD)) {
3496 target_phys_addr_t addr1 = addr;
3498 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3500 addr1 = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3501 if (l >= 4 && ((addr1 & 3) == 0)) {
3502 /* 32 bit read access */
3503 val = io_mem_read[io_index][2](io_mem_opaque[io_index], addr1);
3506 } else if (l >= 2 && ((addr1 & 1) == 0)) {
3507 /* 16 bit read access */
3508 val = io_mem_read[io_index][1](io_mem_opaque[io_index], addr1);
3512 /* 8 bit read access */
3513 val = io_mem_read[io_index][0](io_mem_opaque[io_index], addr1);
3519 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
3520 (addr & ~TARGET_PAGE_MASK);
3521 memcpy(buf, ptr, l);
3530 /* used for ROM loading : can write in RAM and ROM */
3531 void cpu_physical_memory_write_rom(target_phys_addr_t addr,
3532 const uint8_t *buf, int len)
3536 target_phys_addr_t page;
3541 page = addr & TARGET_PAGE_MASK;
3542 l = (page + TARGET_PAGE_SIZE) - addr;
3545 p = phys_page_find(page >> TARGET_PAGE_BITS);
3547 pd = IO_MEM_UNASSIGNED;
3549 pd = p->phys_offset;
3552 if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM &&
3553 (pd & ~TARGET_PAGE_MASK) != IO_MEM_ROM &&
3554 !(pd & IO_MEM_ROMD)) {
3557 unsigned long addr1;
3558 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3560 ptr = qemu_get_ram_ptr(addr1);
3561 memcpy(ptr, buf, l);
3571 target_phys_addr_t addr;
3572 target_phys_addr_t len;
3575 static BounceBuffer bounce;
3577 typedef struct MapClient {
3579 void (*callback)(void *opaque);
3580 QLIST_ENTRY(MapClient) link;
3583 static QLIST_HEAD(map_client_list, MapClient) map_client_list
3584 = QLIST_HEAD_INITIALIZER(map_client_list);
3586 void *cpu_register_map_client(void *opaque, void (*callback)(void *opaque))
3588 MapClient *client = qemu_malloc(sizeof(*client));
3590 client->opaque = opaque;
3591 client->callback = callback;
3592 QLIST_INSERT_HEAD(&map_client_list, client, link);
3596 void cpu_unregister_map_client(void *_client)
3598 MapClient *client = (MapClient *)_client;
3600 QLIST_REMOVE(client, link);
3604 static void cpu_notify_map_clients(void)
3608 while (!QLIST_EMPTY(&map_client_list)) {
3609 client = QLIST_FIRST(&map_client_list);
3610 client->callback(client->opaque);
3611 cpu_unregister_map_client(client);
3615 /* Map a physical memory region into a host virtual address.
3616 * May map a subset of the requested range, given by and returned in *plen.
3617 * May return NULL if resources needed to perform the mapping are exhausted.
3618 * Use only for reads OR writes - not for read-modify-write operations.
3619 * Use cpu_register_map_client() to know when retrying the map operation is
3620 * likely to succeed.
3622 void *cpu_physical_memory_map(target_phys_addr_t addr,
3623 target_phys_addr_t *plen,
3626 target_phys_addr_t len = *plen;
3627 target_phys_addr_t done = 0;
3629 uint8_t *ret = NULL;
3631 target_phys_addr_t page;
3634 unsigned long addr1;
3637 page = addr & TARGET_PAGE_MASK;
3638 l = (page + TARGET_PAGE_SIZE) - addr;
3641 p = phys_page_find(page >> TARGET_PAGE_BITS);
3643 pd = IO_MEM_UNASSIGNED;
3645 pd = p->phys_offset;
3648 if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
3649 if (done || bounce.buffer) {
3652 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, TARGET_PAGE_SIZE);
3656 cpu_physical_memory_rw(addr, bounce.buffer, l, 0);
3658 ptr = bounce.buffer;
3660 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3661 ptr = qemu_get_ram_ptr(addr1);
3665 } else if (ret + done != ptr) {
3677 /* Unmaps a memory region previously mapped by cpu_physical_memory_map().
3678 * Will also mark the memory as dirty if is_write == 1. access_len gives
3679 * the amount of memory that was actually read or written by the caller.
3681 void cpu_physical_memory_unmap(void *buffer, target_phys_addr_t len,
3682 int is_write, target_phys_addr_t access_len)
3684 if (buffer != bounce.buffer) {
3686 ram_addr_t addr1 = qemu_ram_addr_from_host(buffer);
3687 while (access_len) {
3689 l = TARGET_PAGE_SIZE;
3692 if (!cpu_physical_memory_is_dirty(addr1)) {
3693 /* invalidate code */
3694 tb_invalidate_phys_page_range(addr1, addr1 + l, 0);
3696 phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] |=
3697 (0xff & ~CODE_DIRTY_FLAG);
3706 cpu_physical_memory_write(bounce.addr, bounce.buffer, access_len);
3708 qemu_vfree(bounce.buffer);
3709 bounce.buffer = NULL;
3710 cpu_notify_map_clients();
3713 /* warning: addr must be aligned */
3714 uint32_t ldl_phys(target_phys_addr_t addr)
3722 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3724 pd = IO_MEM_UNASSIGNED;
3726 pd = p->phys_offset;
3729 if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM &&
3730 !(pd & IO_MEM_ROMD)) {
3732 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3734 addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3735 val = io_mem_read[io_index][2](io_mem_opaque[io_index], addr);
3738 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
3739 (addr & ~TARGET_PAGE_MASK);
3745 /* warning: addr must be aligned */
3746 uint64_t ldq_phys(target_phys_addr_t addr)
3754 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3756 pd = IO_MEM_UNASSIGNED;
3758 pd = p->phys_offset;
3761 if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM &&
3762 !(pd & IO_MEM_ROMD)) {
3764 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3766 addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3767 #ifdef TARGET_WORDS_BIGENDIAN
3768 val = (uint64_t)io_mem_read[io_index][2](io_mem_opaque[io_index], addr) << 32;
3769 val |= io_mem_read[io_index][2](io_mem_opaque[io_index], addr + 4);
3771 val = io_mem_read[io_index][2](io_mem_opaque[io_index], addr);
3772 val |= (uint64_t)io_mem_read[io_index][2](io_mem_opaque[io_index], addr + 4) << 32;
3776 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
3777 (addr & ~TARGET_PAGE_MASK);
3784 uint32_t ldub_phys(target_phys_addr_t addr)
3787 cpu_physical_memory_read(addr, &val, 1);
3792 uint32_t lduw_phys(target_phys_addr_t addr)
3795 cpu_physical_memory_read(addr, (uint8_t *)&val, 2);
3796 return tswap16(val);
3799 /* warning: addr must be aligned. The ram page is not masked as dirty
3800 and the code inside is not invalidated. It is useful if the dirty
3801 bits are used to track modified PTEs */
3802 void stl_phys_notdirty(target_phys_addr_t addr, uint32_t val)
3809 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3811 pd = IO_MEM_UNASSIGNED;
3813 pd = p->phys_offset;
3816 if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
3817 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3819 addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3820 io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val);
3822 unsigned long addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3823 ptr = qemu_get_ram_ptr(addr1);
3826 if (unlikely(in_migration)) {
3827 if (!cpu_physical_memory_is_dirty(addr1)) {
3828 /* invalidate code */
3829 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
3831 phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] |=
3832 (0xff & ~CODE_DIRTY_FLAG);
3838 void stq_phys_notdirty(target_phys_addr_t addr, uint64_t val)
3845 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3847 pd = IO_MEM_UNASSIGNED;
3849 pd = p->phys_offset;
3852 if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
3853 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3855 addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3856 #ifdef TARGET_WORDS_BIGENDIAN
3857 io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val >> 32);
3858 io_mem_write[io_index][2](io_mem_opaque[io_index], addr + 4, val);
3860 io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val);
3861 io_mem_write[io_index][2](io_mem_opaque[io_index], addr + 4, val >> 32);
3864 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
3865 (addr & ~TARGET_PAGE_MASK);
3870 /* warning: addr must be aligned */
3871 void stl_phys(target_phys_addr_t addr, uint32_t val)
3878 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3880 pd = IO_MEM_UNASSIGNED;
3882 pd = p->phys_offset;
3885 if ((pd & ~TARGET_PAGE_MASK) != IO_MEM_RAM) {
3886 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3888 addr = (addr & ~TARGET_PAGE_MASK) + p->region_offset;
3889 io_mem_write[io_index][2](io_mem_opaque[io_index], addr, val);
3891 unsigned long addr1;
3892 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3894 ptr = qemu_get_ram_ptr(addr1);
3896 if (!cpu_physical_memory_is_dirty(addr1)) {
3897 /* invalidate code */
3898 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
3900 phys_ram_dirty[addr1 >> TARGET_PAGE_BITS] |=
3901 (0xff & ~CODE_DIRTY_FLAG);
3907 void stb_phys(target_phys_addr_t addr, uint32_t val)
3910 cpu_physical_memory_write(addr, &v, 1);
3914 void stw_phys(target_phys_addr_t addr, uint32_t val)
3916 uint16_t v = tswap16(val);
3917 cpu_physical_memory_write(addr, (const uint8_t *)&v, 2);
3921 void stq_phys(target_phys_addr_t addr, uint64_t val)
3924 cpu_physical_memory_write(addr, (const uint8_t *)&val, 8);
3927 /* virtual memory access for debug (includes writing to ROM) */
3928 int cpu_memory_rw_debug(CPUState *env, target_ulong addr,
3929 uint8_t *buf, int len, int is_write)
3932 target_phys_addr_t phys_addr;
3936 page = addr & TARGET_PAGE_MASK;
3937 phys_addr = cpu_get_phys_page_debug(env, page);
3938 /* if no physical page mapped, return an error */
3939 if (phys_addr == -1)
3941 l = (page + TARGET_PAGE_SIZE) - addr;
3944 phys_addr += (addr & ~TARGET_PAGE_MASK);
3946 cpu_physical_memory_write_rom(phys_addr, buf, l);
3948 cpu_physical_memory_rw(phys_addr, buf, l, is_write);
3957 /* in deterministic execution mode, instructions doing device I/Os
3958 must be at the end of the TB */
3959 void cpu_io_recompile(CPUState *env, void *retaddr)
3961 TranslationBlock *tb;
3963 target_ulong pc, cs_base;
3966 tb = tb_find_pc((unsigned long)retaddr);
3968 cpu_abort(env, "cpu_io_recompile: could not find TB for pc=%p",
3971 n = env->icount_decr.u16.low + tb->icount;
3972 cpu_restore_state(tb, env, (unsigned long)retaddr, NULL);
3973 /* Calculate how many instructions had been executed before the fault
3975 n = n - env->icount_decr.u16.low;
3976 /* Generate a new TB ending on the I/O insn. */
3978 /* On MIPS and SH, delay slot instructions can only be restarted if
3979 they were already the first instruction in the TB. If this is not
3980 the first instruction in a TB then re-execute the preceding
3982 #if defined(TARGET_MIPS)
3983 if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
3984 env->active_tc.PC -= 4;
3985 env->icount_decr.u16.low++;
3986 env->hflags &= ~MIPS_HFLAG_BMASK;
3988 #elif defined(TARGET_SH4)
3989 if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
3992 env->icount_decr.u16.low++;
3993 env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
3996 /* This should never happen. */
3997 if (n > CF_COUNT_MASK)
3998 cpu_abort(env, "TB too big during recompile");
4000 cflags = n | CF_LAST_IO;
4002 cs_base = tb->cs_base;
4004 tb_phys_invalidate(tb, -1);
4005 /* FIXME: In theory this could raise an exception. In practice
4006 we have already translated the block once so it's probably ok. */
4007 tb_gen_code(env, pc, cs_base, flags, cflags);
4008 /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
4009 the first in the TB) then we end up generating a whole new TB and
4010 repeating the fault, which is horribly inefficient.
4011 Better would be to execute just this insn uncached, or generate a
4013 cpu_resume_from_signal(env, NULL);
4016 #if !defined(CONFIG_USER_ONLY)
4018 void dump_exec_info(FILE *f,
4019 int (*cpu_fprintf)(FILE *f, const char *fmt, ...))
4021 int i, target_code_size, max_target_code_size;
4022 int direct_jmp_count, direct_jmp2_count, cross_page;
4023 TranslationBlock *tb;
4025 target_code_size = 0;
4026 max_target_code_size = 0;
4028 direct_jmp_count = 0;
4029 direct_jmp2_count = 0;
4030 for(i = 0; i < nb_tbs; i++) {
4032 target_code_size += tb->size;
4033 if (tb->size > max_target_code_size)
4034 max_target_code_size = tb->size;
4035 if (tb->page_addr[1] != -1)
4037 if (tb->tb_next_offset[0] != 0xffff) {
4039 if (tb->tb_next_offset[1] != 0xffff) {
4040 direct_jmp2_count++;
4044 /* XXX: avoid using doubles ? */
4045 cpu_fprintf(f, "Translation buffer state:\n");
4046 cpu_fprintf(f, "gen code size %ld/%ld\n",
4047 code_gen_ptr - code_gen_buffer, code_gen_buffer_max_size);
4048 cpu_fprintf(f, "TB count %d/%d\n",
4049 nb_tbs, code_gen_max_blocks);
4050 cpu_fprintf(f, "TB avg target size %d max=%d bytes\n",
4051 nb_tbs ? target_code_size / nb_tbs : 0,
4052 max_target_code_size);
4053 cpu_fprintf(f, "TB avg host size %d bytes (expansion ratio: %0.1f)\n",
4054 nb_tbs ? (code_gen_ptr - code_gen_buffer) / nb_tbs : 0,
4055 target_code_size ? (double) (code_gen_ptr - code_gen_buffer) / target_code_size : 0);
4056 cpu_fprintf(f, "cross page TB count %d (%d%%)\n",
4058 nb_tbs ? (cross_page * 100) / nb_tbs : 0);
4059 cpu_fprintf(f, "direct jump count %d (%d%%) (2 jumps=%d %d%%)\n",
4061 nb_tbs ? (direct_jmp_count * 100) / nb_tbs : 0,
4063 nb_tbs ? (direct_jmp2_count * 100) / nb_tbs : 0);
4064 cpu_fprintf(f, "\nStatistics:\n");
4065 cpu_fprintf(f, "TB flush count %d\n", tb_flush_count);
4066 cpu_fprintf(f, "TB invalidate count %d\n", tb_phys_invalidate_count);
4067 cpu_fprintf(f, "TLB flush count %d\n", tlb_flush_count);
4068 tcg_dump_info(f, cpu_fprintf);
4071 #define MMUSUFFIX _cmmu
4072 #define GETPC() NULL
4073 #define env cpu_single_env
4074 #define SOFTMMU_CODE_ACCESS
4077 #include "softmmu_template.h"
4080 #include "softmmu_template.h"
4083 #include "softmmu_template.h"
4086 #include "softmmu_template.h"