CPUPhysMemoryClient: Fix typo in phys memory client registration
[qemu.git] / exec.c
1 /*
2 * virtual page mapping and translated block handling
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18 */
19 #include "config.h"
20 #ifdef _WIN32
21 #include <windows.h>
22 #else
23 #include <sys/types.h>
24 #include <sys/mman.h>
25 #endif
26
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "exec-all.h"
30 #include "tcg.h"
31 #include "hw/hw.h"
32 #include "hw/qdev.h"
33 #include "osdep.h"
34 #include "kvm.h"
35 #include "qemu-timer.h"
36 #if defined(CONFIG_USER_ONLY)
37 #include <qemu.h>
38 #include <signal.h>
39 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
40 #include <sys/param.h>
41 #if __FreeBSD_version >= 700104
42 #define HAVE_KINFO_GETVMMAP
43 #define sigqueue sigqueue_freebsd /* avoid redefinition */
44 #include <sys/time.h>
45 #include <sys/proc.h>
46 #include <machine/profile.h>
47 #define _KERNEL
48 #include <sys/user.h>
49 #undef _KERNEL
50 #undef sigqueue
51 #include <libutil.h>
52 #endif
53 #endif
54 #endif
55
56 //#define DEBUG_TB_INVALIDATE
57 //#define DEBUG_FLUSH
58 //#define DEBUG_TLB
59 //#define DEBUG_UNASSIGNED
60
61 /* make various TB consistency checks */
62 //#define DEBUG_TB_CHECK
63 //#define DEBUG_TLB_CHECK
64
65 //#define DEBUG_IOPORT
66 //#define DEBUG_SUBPAGE
67
68 #if !defined(CONFIG_USER_ONLY)
69 /* TB consistency checks only implemented for usermode emulation. */
70 #undef DEBUG_TB_CHECK
71 #endif
72
73 #define SMC_BITMAP_USE_THRESHOLD 10
74
75 static TranslationBlock *tbs;
76 static int code_gen_max_blocks;
77 TranslationBlock *tb_phys_hash[CODE_GEN_PHYS_HASH_SIZE];
78 static int nb_tbs;
79 /* any access to the tbs or the page table must use this lock */
80 spinlock_t tb_lock = SPIN_LOCK_UNLOCKED;
81
82 #if defined(__arm__) || defined(__sparc_v9__)
83 /* The prologue must be reachable with a direct jump. ARM and Sparc64
84 have limited branch ranges (possibly also PPC) so place it in a
85 section close to code segment. */
86 #define code_gen_section \
87 __attribute__((__section__(".gen_code"))) \
88 __attribute__((aligned (32)))
89 #elif defined(_WIN32)
90 /* Maximum alignment for Win32 is 16. */
91 #define code_gen_section \
92 __attribute__((aligned (16)))
93 #else
94 #define code_gen_section \
95 __attribute__((aligned (32)))
96 #endif
97
98 uint8_t code_gen_prologue[1024] code_gen_section;
99 static uint8_t *code_gen_buffer;
100 static unsigned long code_gen_buffer_size;
101 /* threshold to flush the translated code buffer */
102 static unsigned long code_gen_buffer_max_size;
103 static uint8_t *code_gen_ptr;
104
105 #if !defined(CONFIG_USER_ONLY)
106 int phys_ram_fd;
107 static int in_migration;
108
109 RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list) };
110 #endif
111
112 CPUState *first_cpu;
113 /* current CPU in the current thread. It is only valid inside
114 cpu_exec() */
115 CPUState *cpu_single_env;
116 /* 0 = Do not count executed instructions.
117 1 = Precise instruction counting.
118 2 = Adaptive rate instruction counting. */
119 int use_icount = 0;
120 /* Current instruction counter. While executing translated code this may
121 include some instructions that have not yet been executed. */
122 int64_t qemu_icount;
123
124 typedef struct PageDesc {
125 /* list of TBs intersecting this ram page */
126 TranslationBlock *first_tb;
127 /* in order to optimize self modifying code, we count the number
128 of lookups we do to a given page to use a bitmap */
129 unsigned int code_write_count;
130 uint8_t *code_bitmap;
131 #if defined(CONFIG_USER_ONLY)
132 unsigned long flags;
133 #endif
134 } PageDesc;
135
136 /* In system mode we want L1_MAP to be based on ram offsets,
137 while in user mode we want it to be based on virtual addresses. */
138 #if !defined(CONFIG_USER_ONLY)
139 #if HOST_LONG_BITS < TARGET_PHYS_ADDR_SPACE_BITS
140 # define L1_MAP_ADDR_SPACE_BITS HOST_LONG_BITS
141 #else
142 # define L1_MAP_ADDR_SPACE_BITS TARGET_PHYS_ADDR_SPACE_BITS
143 #endif
144 #else
145 # define L1_MAP_ADDR_SPACE_BITS TARGET_VIRT_ADDR_SPACE_BITS
146 #endif
147
148 /* Size of the L2 (and L3, etc) page tables. */
149 #define L2_BITS 10
150 #define L2_SIZE (1 << L2_BITS)
151
152 /* The bits remaining after N lower levels of page tables. */
153 #define P_L1_BITS_REM \
154 ((TARGET_PHYS_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % L2_BITS)
155 #define V_L1_BITS_REM \
156 ((L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % L2_BITS)
157
158 /* Size of the L1 page table. Avoid silly small sizes. */
159 #if P_L1_BITS_REM < 4
160 #define P_L1_BITS (P_L1_BITS_REM + L2_BITS)
161 #else
162 #define P_L1_BITS P_L1_BITS_REM
163 #endif
164
165 #if V_L1_BITS_REM < 4
166 #define V_L1_BITS (V_L1_BITS_REM + L2_BITS)
167 #else
168 #define V_L1_BITS V_L1_BITS_REM
169 #endif
170
171 #define P_L1_SIZE ((target_phys_addr_t)1 << P_L1_BITS)
172 #define V_L1_SIZE ((target_ulong)1 << V_L1_BITS)
173
174 #define P_L1_SHIFT (TARGET_PHYS_ADDR_SPACE_BITS - TARGET_PAGE_BITS - P_L1_BITS)
175 #define V_L1_SHIFT (L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS - V_L1_BITS)
176
177 unsigned long qemu_real_host_page_size;
178 unsigned long qemu_host_page_bits;
179 unsigned long qemu_host_page_size;
180 unsigned long qemu_host_page_mask;
181
182 /* This is a multi-level map on the virtual address space.
183 The bottom level has pointers to PageDesc. */
184 static void *l1_map[V_L1_SIZE];
185
186 #if !defined(CONFIG_USER_ONLY)
187 typedef struct PhysPageDesc {
188 /* offset in host memory of the page + io_index in the low bits */
189 ram_addr_t phys_offset;
190 ram_addr_t region_offset;
191 } PhysPageDesc;
192
193 /* This is a multi-level map on the physical address space.
194 The bottom level has pointers to PhysPageDesc. */
195 static void *l1_phys_map[P_L1_SIZE];
196
197 static void io_mem_init(void);
198
199 /* io memory support */
200 CPUWriteMemoryFunc *io_mem_write[IO_MEM_NB_ENTRIES][4];
201 CPUReadMemoryFunc *io_mem_read[IO_MEM_NB_ENTRIES][4];
202 void *io_mem_opaque[IO_MEM_NB_ENTRIES];
203 static char io_mem_used[IO_MEM_NB_ENTRIES];
204 static int io_mem_watch;
205 #endif
206
207 /* log support */
208 #ifdef WIN32
209 static const char *logfilename = "qemu.log";
210 #else
211 static const char *logfilename = "/tmp/qemu.log";
212 #endif
213 FILE *logfile;
214 int loglevel;
215 static int log_append = 0;
216
217 /* statistics */
218 #if !defined(CONFIG_USER_ONLY)
219 static int tlb_flush_count;
220 #endif
221 static int tb_flush_count;
222 static int tb_phys_invalidate_count;
223
224 #ifdef _WIN32
225 static void map_exec(void *addr, long size)
226 {
227 DWORD old_protect;
228 VirtualProtect(addr, size,
229 PAGE_EXECUTE_READWRITE, &old_protect);
230
231 }
232 #else
233 static void map_exec(void *addr, long size)
234 {
235 unsigned long start, end, page_size;
236
237 page_size = getpagesize();
238 start = (unsigned long)addr;
239 start &= ~(page_size - 1);
240
241 end = (unsigned long)addr + size;
242 end += page_size - 1;
243 end &= ~(page_size - 1);
244
245 mprotect((void *)start, end - start,
246 PROT_READ | PROT_WRITE | PROT_EXEC);
247 }
248 #endif
249
250 static void page_init(void)
251 {
252 /* NOTE: we can always suppose that qemu_host_page_size >=
253 TARGET_PAGE_SIZE */
254 #ifdef _WIN32
255 {
256 SYSTEM_INFO system_info;
257
258 GetSystemInfo(&system_info);
259 qemu_real_host_page_size = system_info.dwPageSize;
260 }
261 #else
262 qemu_real_host_page_size = getpagesize();
263 #endif
264 if (qemu_host_page_size == 0)
265 qemu_host_page_size = qemu_real_host_page_size;
266 if (qemu_host_page_size < TARGET_PAGE_SIZE)
267 qemu_host_page_size = TARGET_PAGE_SIZE;
268 qemu_host_page_bits = 0;
269 while ((1 << qemu_host_page_bits) < qemu_host_page_size)
270 qemu_host_page_bits++;
271 qemu_host_page_mask = ~(qemu_host_page_size - 1);
272
273 #if defined(CONFIG_BSD) && defined(CONFIG_USER_ONLY)
274 {
275 #ifdef HAVE_KINFO_GETVMMAP
276 struct kinfo_vmentry *freep;
277 int i, cnt;
278
279 freep = kinfo_getvmmap(getpid(), &cnt);
280 if (freep) {
281 mmap_lock();
282 for (i = 0; i < cnt; i++) {
283 unsigned long startaddr, endaddr;
284
285 startaddr = freep[i].kve_start;
286 endaddr = freep[i].kve_end;
287 if (h2g_valid(startaddr)) {
288 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
289
290 if (h2g_valid(endaddr)) {
291 endaddr = h2g(endaddr);
292 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
293 } else {
294 #if TARGET_ABI_BITS <= L1_MAP_ADDR_SPACE_BITS
295 endaddr = ~0ul;
296 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
297 #endif
298 }
299 }
300 }
301 free(freep);
302 mmap_unlock();
303 }
304 #else
305 FILE *f;
306
307 last_brk = (unsigned long)sbrk(0);
308
309 f = fopen("/compat/linux/proc/self/maps", "r");
310 if (f) {
311 mmap_lock();
312
313 do {
314 unsigned long startaddr, endaddr;
315 int n;
316
317 n = fscanf (f, "%lx-%lx %*[^\n]\n", &startaddr, &endaddr);
318
319 if (n == 2 && h2g_valid(startaddr)) {
320 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
321
322 if (h2g_valid(endaddr)) {
323 endaddr = h2g(endaddr);
324 } else {
325 endaddr = ~0ul;
326 }
327 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
328 }
329 } while (!feof(f));
330
331 fclose(f);
332 mmap_unlock();
333 }
334 #endif
335 }
336 #endif
337 }
338
339 static PageDesc *page_find_alloc(tb_page_addr_t index, int alloc)
340 {
341 PageDesc *pd;
342 void **lp;
343 int i;
344
345 #if defined(CONFIG_USER_ONLY)
346 /* We can't use qemu_malloc because it may recurse into a locked mutex. */
347 # define ALLOC(P, SIZE) \
348 do { \
349 P = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, \
350 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); \
351 } while (0)
352 #else
353 # define ALLOC(P, SIZE) \
354 do { P = qemu_mallocz(SIZE); } while (0)
355 #endif
356
357 /* Level 1. Always allocated. */
358 lp = l1_map + ((index >> V_L1_SHIFT) & (V_L1_SIZE - 1));
359
360 /* Level 2..N-1. */
361 for (i = V_L1_SHIFT / L2_BITS - 1; i > 0; i--) {
362 void **p = *lp;
363
364 if (p == NULL) {
365 if (!alloc) {
366 return NULL;
367 }
368 ALLOC(p, sizeof(void *) * L2_SIZE);
369 *lp = p;
370 }
371
372 lp = p + ((index >> (i * L2_BITS)) & (L2_SIZE - 1));
373 }
374
375 pd = *lp;
376 if (pd == NULL) {
377 if (!alloc) {
378 return NULL;
379 }
380 ALLOC(pd, sizeof(PageDesc) * L2_SIZE);
381 *lp = pd;
382 }
383
384 #undef ALLOC
385
386 return pd + (index & (L2_SIZE - 1));
387 }
388
389 static inline PageDesc *page_find(tb_page_addr_t index)
390 {
391 return page_find_alloc(index, 0);
392 }
393
394 #if !defined(CONFIG_USER_ONLY)
395 static PhysPageDesc *phys_page_find_alloc(target_phys_addr_t index, int alloc)
396 {
397 PhysPageDesc *pd;
398 void **lp;
399 int i;
400
401 /* Level 1. Always allocated. */
402 lp = l1_phys_map + ((index >> P_L1_SHIFT) & (P_L1_SIZE - 1));
403
404 /* Level 2..N-1. */
405 for (i = P_L1_SHIFT / L2_BITS - 1; i > 0; i--) {
406 void **p = *lp;
407 if (p == NULL) {
408 if (!alloc) {
409 return NULL;
410 }
411 *lp = p = qemu_mallocz(sizeof(void *) * L2_SIZE);
412 }
413 lp = p + ((index >> (i * L2_BITS)) & (L2_SIZE - 1));
414 }
415
416 pd = *lp;
417 if (pd == NULL) {
418 int i;
419
420 if (!alloc) {
421 return NULL;
422 }
423
424 *lp = pd = qemu_malloc(sizeof(PhysPageDesc) * L2_SIZE);
425
426 for (i = 0; i < L2_SIZE; i++) {
427 pd[i].phys_offset = IO_MEM_UNASSIGNED;
428 pd[i].region_offset = (index + i) << TARGET_PAGE_BITS;
429 }
430 }
431
432 return pd + (index & (L2_SIZE - 1));
433 }
434
435 static inline PhysPageDesc *phys_page_find(target_phys_addr_t index)
436 {
437 return phys_page_find_alloc(index, 0);
438 }
439
440 static void tlb_protect_code(ram_addr_t ram_addr);
441 static void tlb_unprotect_code_phys(CPUState *env, ram_addr_t ram_addr,
442 target_ulong vaddr);
443 #define mmap_lock() do { } while(0)
444 #define mmap_unlock() do { } while(0)
445 #endif
446
447 #define DEFAULT_CODE_GEN_BUFFER_SIZE (32 * 1024 * 1024)
448
449 #if defined(CONFIG_USER_ONLY)
450 /* Currently it is not recommended to allocate big chunks of data in
451 user mode. It will change when a dedicated libc will be used */
452 #define USE_STATIC_CODE_GEN_BUFFER
453 #endif
454
455 #ifdef USE_STATIC_CODE_GEN_BUFFER
456 static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE]
457 __attribute__((aligned (CODE_GEN_ALIGN)));
458 #endif
459
460 static void code_gen_alloc(unsigned long tb_size)
461 {
462 #ifdef USE_STATIC_CODE_GEN_BUFFER
463 code_gen_buffer = static_code_gen_buffer;
464 code_gen_buffer_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
465 map_exec(code_gen_buffer, code_gen_buffer_size);
466 #else
467 code_gen_buffer_size = tb_size;
468 if (code_gen_buffer_size == 0) {
469 #if defined(CONFIG_USER_ONLY)
470 /* in user mode, phys_ram_size is not meaningful */
471 code_gen_buffer_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
472 #else
473 /* XXX: needs adjustments */
474 code_gen_buffer_size = (unsigned long)(ram_size / 4);
475 #endif
476 }
477 if (code_gen_buffer_size < MIN_CODE_GEN_BUFFER_SIZE)
478 code_gen_buffer_size = MIN_CODE_GEN_BUFFER_SIZE;
479 /* The code gen buffer location may have constraints depending on
480 the host cpu and OS */
481 #if defined(__linux__)
482 {
483 int flags;
484 void *start = NULL;
485
486 flags = MAP_PRIVATE | MAP_ANONYMOUS;
487 #if defined(__x86_64__)
488 flags |= MAP_32BIT;
489 /* Cannot map more than that */
490 if (code_gen_buffer_size > (800 * 1024 * 1024))
491 code_gen_buffer_size = (800 * 1024 * 1024);
492 #elif defined(__sparc_v9__)
493 // Map the buffer below 2G, so we can use direct calls and branches
494 flags |= MAP_FIXED;
495 start = (void *) 0x60000000UL;
496 if (code_gen_buffer_size > (512 * 1024 * 1024))
497 code_gen_buffer_size = (512 * 1024 * 1024);
498 #elif defined(__arm__)
499 /* Map the buffer below 32M, so we can use direct calls and branches */
500 flags |= MAP_FIXED;
501 start = (void *) 0x01000000UL;
502 if (code_gen_buffer_size > 16 * 1024 * 1024)
503 code_gen_buffer_size = 16 * 1024 * 1024;
504 #elif defined(__s390x__)
505 /* Map the buffer so that we can use direct calls and branches. */
506 /* We have a +- 4GB range on the branches; leave some slop. */
507 if (code_gen_buffer_size > (3ul * 1024 * 1024 * 1024)) {
508 code_gen_buffer_size = 3ul * 1024 * 1024 * 1024;
509 }
510 start = (void *)0x90000000UL;
511 #endif
512 code_gen_buffer = mmap(start, code_gen_buffer_size,
513 PROT_WRITE | PROT_READ | PROT_EXEC,
514 flags, -1, 0);
515 if (code_gen_buffer == MAP_FAILED) {
516 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
517 exit(1);
518 }
519 }
520 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) \
521 || defined(__DragonFly__) || defined(__OpenBSD__)
522 {
523 int flags;
524 void *addr = NULL;
525 flags = MAP_PRIVATE | MAP_ANONYMOUS;
526 #if defined(__x86_64__)
527 /* FreeBSD doesn't have MAP_32BIT, use MAP_FIXED and assume
528 * 0x40000000 is free */
529 flags |= MAP_FIXED;
530 addr = (void *)0x40000000;
531 /* Cannot map more than that */
532 if (code_gen_buffer_size > (800 * 1024 * 1024))
533 code_gen_buffer_size = (800 * 1024 * 1024);
534 #elif defined(__sparc_v9__)
535 // Map the buffer below 2G, so we can use direct calls and branches
536 flags |= MAP_FIXED;
537 addr = (void *) 0x60000000UL;
538 if (code_gen_buffer_size > (512 * 1024 * 1024)) {
539 code_gen_buffer_size = (512 * 1024 * 1024);
540 }
541 #endif
542 code_gen_buffer = mmap(addr, code_gen_buffer_size,
543 PROT_WRITE | PROT_READ | PROT_EXEC,
544 flags, -1, 0);
545 if (code_gen_buffer == MAP_FAILED) {
546 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
547 exit(1);
548 }
549 }
550 #else
551 code_gen_buffer = qemu_malloc(code_gen_buffer_size);
552 map_exec(code_gen_buffer, code_gen_buffer_size);
553 #endif
554 #endif /* !USE_STATIC_CODE_GEN_BUFFER */
555 map_exec(code_gen_prologue, sizeof(code_gen_prologue));
556 code_gen_buffer_max_size = code_gen_buffer_size -
557 (TCG_MAX_OP_SIZE * OPC_MAX_SIZE);
558 code_gen_max_blocks = code_gen_buffer_size / CODE_GEN_AVG_BLOCK_SIZE;
559 tbs = qemu_malloc(code_gen_max_blocks * sizeof(TranslationBlock));
560 }
561
562 /* Must be called before using the QEMU cpus. 'tb_size' is the size
563 (in bytes) allocated to the translation buffer. Zero means default
564 size. */
565 void cpu_exec_init_all(unsigned long tb_size)
566 {
567 cpu_gen_init();
568 code_gen_alloc(tb_size);
569 code_gen_ptr = code_gen_buffer;
570 page_init();
571 #if !defined(CONFIG_USER_ONLY)
572 io_mem_init();
573 #endif
574 #if !defined(CONFIG_USER_ONLY) || !defined(CONFIG_USE_GUEST_BASE)
575 /* There's no guest base to take into account, so go ahead and
576 initialize the prologue now. */
577 tcg_prologue_init(&tcg_ctx);
578 #endif
579 }
580
581 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
582
583 static int cpu_common_post_load(void *opaque, int version_id)
584 {
585 CPUState *env = opaque;
586
587 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
588 version_id is increased. */
589 env->interrupt_request &= ~0x01;
590 tlb_flush(env, 1);
591
592 return 0;
593 }
594
595 static const VMStateDescription vmstate_cpu_common = {
596 .name = "cpu_common",
597 .version_id = 1,
598 .minimum_version_id = 1,
599 .minimum_version_id_old = 1,
600 .post_load = cpu_common_post_load,
601 .fields = (VMStateField []) {
602 VMSTATE_UINT32(halted, CPUState),
603 VMSTATE_UINT32(interrupt_request, CPUState),
604 VMSTATE_END_OF_LIST()
605 }
606 };
607 #endif
608
609 CPUState *qemu_get_cpu(int cpu)
610 {
611 CPUState *env = first_cpu;
612
613 while (env) {
614 if (env->cpu_index == cpu)
615 break;
616 env = env->next_cpu;
617 }
618
619 return env;
620 }
621
622 void cpu_exec_init(CPUState *env)
623 {
624 CPUState **penv;
625 int cpu_index;
626
627 #if defined(CONFIG_USER_ONLY)
628 cpu_list_lock();
629 #endif
630 env->next_cpu = NULL;
631 penv = &first_cpu;
632 cpu_index = 0;
633 while (*penv != NULL) {
634 penv = &(*penv)->next_cpu;
635 cpu_index++;
636 }
637 env->cpu_index = cpu_index;
638 env->numa_node = 0;
639 QTAILQ_INIT(&env->breakpoints);
640 QTAILQ_INIT(&env->watchpoints);
641 #ifndef CONFIG_USER_ONLY
642 env->thread_id = qemu_get_thread_id();
643 #endif
644 *penv = env;
645 #if defined(CONFIG_USER_ONLY)
646 cpu_list_unlock();
647 #endif
648 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
649 vmstate_register(NULL, cpu_index, &vmstate_cpu_common, env);
650 register_savevm(NULL, "cpu", cpu_index, CPU_SAVE_VERSION,
651 cpu_save, cpu_load, env);
652 #endif
653 }
654
655 /* Allocate a new translation block. Flush the translation buffer if
656 too many translation blocks or too much generated code. */
657 static TranslationBlock *tb_alloc(target_ulong pc)
658 {
659 TranslationBlock *tb;
660
661 if (nb_tbs >= code_gen_max_blocks ||
662 (code_gen_ptr - code_gen_buffer) >= code_gen_buffer_max_size)
663 return NULL;
664 tb = &tbs[nb_tbs++];
665 tb->pc = pc;
666 tb->cflags = 0;
667 return tb;
668 }
669
670 void tb_free(TranslationBlock *tb)
671 {
672 /* In practice this is mostly used for single use temporary TB
673 Ignore the hard cases and just back up if this TB happens to
674 be the last one generated. */
675 if (nb_tbs > 0 && tb == &tbs[nb_tbs - 1]) {
676 code_gen_ptr = tb->tc_ptr;
677 nb_tbs--;
678 }
679 }
680
681 static inline void invalidate_page_bitmap(PageDesc *p)
682 {
683 if (p->code_bitmap) {
684 qemu_free(p->code_bitmap);
685 p->code_bitmap = NULL;
686 }
687 p->code_write_count = 0;
688 }
689
690 /* Set to NULL all the 'first_tb' fields in all PageDescs. */
691
692 static void page_flush_tb_1 (int level, void **lp)
693 {
694 int i;
695
696 if (*lp == NULL) {
697 return;
698 }
699 if (level == 0) {
700 PageDesc *pd = *lp;
701 for (i = 0; i < L2_SIZE; ++i) {
702 pd[i].first_tb = NULL;
703 invalidate_page_bitmap(pd + i);
704 }
705 } else {
706 void **pp = *lp;
707 for (i = 0; i < L2_SIZE; ++i) {
708 page_flush_tb_1 (level - 1, pp + i);
709 }
710 }
711 }
712
713 static void page_flush_tb(void)
714 {
715 int i;
716 for (i = 0; i < V_L1_SIZE; i++) {
717 page_flush_tb_1(V_L1_SHIFT / L2_BITS - 1, l1_map + i);
718 }
719 }
720
721 /* flush all the translation blocks */
722 /* XXX: tb_flush is currently not thread safe */
723 void tb_flush(CPUState *env1)
724 {
725 CPUState *env;
726 #if defined(DEBUG_FLUSH)
727 printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n",
728 (unsigned long)(code_gen_ptr - code_gen_buffer),
729 nb_tbs, nb_tbs > 0 ?
730 ((unsigned long)(code_gen_ptr - code_gen_buffer)) / nb_tbs : 0);
731 #endif
732 if ((unsigned long)(code_gen_ptr - code_gen_buffer) > code_gen_buffer_size)
733 cpu_abort(env1, "Internal error: code buffer overflow\n");
734
735 nb_tbs = 0;
736
737 for(env = first_cpu; env != NULL; env = env->next_cpu) {
738 memset (env->tb_jmp_cache, 0, TB_JMP_CACHE_SIZE * sizeof (void *));
739 }
740
741 memset (tb_phys_hash, 0, CODE_GEN_PHYS_HASH_SIZE * sizeof (void *));
742 page_flush_tb();
743
744 code_gen_ptr = code_gen_buffer;
745 /* XXX: flush processor icache at this point if cache flush is
746 expensive */
747 tb_flush_count++;
748 }
749
750 #ifdef DEBUG_TB_CHECK
751
752 static void tb_invalidate_check(target_ulong address)
753 {
754 TranslationBlock *tb;
755 int i;
756 address &= TARGET_PAGE_MASK;
757 for(i = 0;i < CODE_GEN_PHYS_HASH_SIZE; i++) {
758 for(tb = tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
759 if (!(address + TARGET_PAGE_SIZE <= tb->pc ||
760 address >= tb->pc + tb->size)) {
761 printf("ERROR invalidate: address=" TARGET_FMT_lx
762 " PC=%08lx size=%04x\n",
763 address, (long)tb->pc, tb->size);
764 }
765 }
766 }
767 }
768
769 /* verify that all the pages have correct rights for code */
770 static void tb_page_check(void)
771 {
772 TranslationBlock *tb;
773 int i, flags1, flags2;
774
775 for(i = 0;i < CODE_GEN_PHYS_HASH_SIZE; i++) {
776 for(tb = tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
777 flags1 = page_get_flags(tb->pc);
778 flags2 = page_get_flags(tb->pc + tb->size - 1);
779 if ((flags1 & PAGE_WRITE) || (flags2 & PAGE_WRITE)) {
780 printf("ERROR page flags: PC=%08lx size=%04x f1=%x f2=%x\n",
781 (long)tb->pc, tb->size, flags1, flags2);
782 }
783 }
784 }
785 }
786
787 #endif
788
789 /* invalidate one TB */
790 static inline void tb_remove(TranslationBlock **ptb, TranslationBlock *tb,
791 int next_offset)
792 {
793 TranslationBlock *tb1;
794 for(;;) {
795 tb1 = *ptb;
796 if (tb1 == tb) {
797 *ptb = *(TranslationBlock **)((char *)tb1 + next_offset);
798 break;
799 }
800 ptb = (TranslationBlock **)((char *)tb1 + next_offset);
801 }
802 }
803
804 static inline void tb_page_remove(TranslationBlock **ptb, TranslationBlock *tb)
805 {
806 TranslationBlock *tb1;
807 unsigned int n1;
808
809 for(;;) {
810 tb1 = *ptb;
811 n1 = (long)tb1 & 3;
812 tb1 = (TranslationBlock *)((long)tb1 & ~3);
813 if (tb1 == tb) {
814 *ptb = tb1->page_next[n1];
815 break;
816 }
817 ptb = &tb1->page_next[n1];
818 }
819 }
820
821 static inline void tb_jmp_remove(TranslationBlock *tb, int n)
822 {
823 TranslationBlock *tb1, **ptb;
824 unsigned int n1;
825
826 ptb = &tb->jmp_next[n];
827 tb1 = *ptb;
828 if (tb1) {
829 /* find tb(n) in circular list */
830 for(;;) {
831 tb1 = *ptb;
832 n1 = (long)tb1 & 3;
833 tb1 = (TranslationBlock *)((long)tb1 & ~3);
834 if (n1 == n && tb1 == tb)
835 break;
836 if (n1 == 2) {
837 ptb = &tb1->jmp_first;
838 } else {
839 ptb = &tb1->jmp_next[n1];
840 }
841 }
842 /* now we can suppress tb(n) from the list */
843 *ptb = tb->jmp_next[n];
844
845 tb->jmp_next[n] = NULL;
846 }
847 }
848
849 /* reset the jump entry 'n' of a TB so that it is not chained to
850 another TB */
851 static inline void tb_reset_jump(TranslationBlock *tb, int n)
852 {
853 tb_set_jmp_target(tb, n, (unsigned long)(tb->tc_ptr + tb->tb_next_offset[n]));
854 }
855
856 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr)
857 {
858 CPUState *env;
859 PageDesc *p;
860 unsigned int h, n1;
861 tb_page_addr_t phys_pc;
862 TranslationBlock *tb1, *tb2;
863
864 /* remove the TB from the hash list */
865 phys_pc = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
866 h = tb_phys_hash_func(phys_pc);
867 tb_remove(&tb_phys_hash[h], tb,
868 offsetof(TranslationBlock, phys_hash_next));
869
870 /* remove the TB from the page list */
871 if (tb->page_addr[0] != page_addr) {
872 p = page_find(tb->page_addr[0] >> TARGET_PAGE_BITS);
873 tb_page_remove(&p->first_tb, tb);
874 invalidate_page_bitmap(p);
875 }
876 if (tb->page_addr[1] != -1 && tb->page_addr[1] != page_addr) {
877 p = page_find(tb->page_addr[1] >> TARGET_PAGE_BITS);
878 tb_page_remove(&p->first_tb, tb);
879 invalidate_page_bitmap(p);
880 }
881
882 tb_invalidated_flag = 1;
883
884 /* remove the TB from the hash list */
885 h = tb_jmp_cache_hash_func(tb->pc);
886 for(env = first_cpu; env != NULL; env = env->next_cpu) {
887 if (env->tb_jmp_cache[h] == tb)
888 env->tb_jmp_cache[h] = NULL;
889 }
890
891 /* suppress this TB from the two jump lists */
892 tb_jmp_remove(tb, 0);
893 tb_jmp_remove(tb, 1);
894
895 /* suppress any remaining jumps to this TB */
896 tb1 = tb->jmp_first;
897 for(;;) {
898 n1 = (long)tb1 & 3;
899 if (n1 == 2)
900 break;
901 tb1 = (TranslationBlock *)((long)tb1 & ~3);
902 tb2 = tb1->jmp_next[n1];
903 tb_reset_jump(tb1, n1);
904 tb1->jmp_next[n1] = NULL;
905 tb1 = tb2;
906 }
907 tb->jmp_first = (TranslationBlock *)((long)tb | 2); /* fail safe */
908
909 tb_phys_invalidate_count++;
910 }
911
912 static inline void set_bits(uint8_t *tab, int start, int len)
913 {
914 int end, mask, end1;
915
916 end = start + len;
917 tab += start >> 3;
918 mask = 0xff << (start & 7);
919 if ((start & ~7) == (end & ~7)) {
920 if (start < end) {
921 mask &= ~(0xff << (end & 7));
922 *tab |= mask;
923 }
924 } else {
925 *tab++ |= mask;
926 start = (start + 8) & ~7;
927 end1 = end & ~7;
928 while (start < end1) {
929 *tab++ = 0xff;
930 start += 8;
931 }
932 if (start < end) {
933 mask = ~(0xff << (end & 7));
934 *tab |= mask;
935 }
936 }
937 }
938
939 static void build_page_bitmap(PageDesc *p)
940 {
941 int n, tb_start, tb_end;
942 TranslationBlock *tb;
943
944 p->code_bitmap = qemu_mallocz(TARGET_PAGE_SIZE / 8);
945
946 tb = p->first_tb;
947 while (tb != NULL) {
948 n = (long)tb & 3;
949 tb = (TranslationBlock *)((long)tb & ~3);
950 /* NOTE: this is subtle as a TB may span two physical pages */
951 if (n == 0) {
952 /* NOTE: tb_end may be after the end of the page, but
953 it is not a problem */
954 tb_start = tb->pc & ~TARGET_PAGE_MASK;
955 tb_end = tb_start + tb->size;
956 if (tb_end > TARGET_PAGE_SIZE)
957 tb_end = TARGET_PAGE_SIZE;
958 } else {
959 tb_start = 0;
960 tb_end = ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
961 }
962 set_bits(p->code_bitmap, tb_start, tb_end - tb_start);
963 tb = tb->page_next[n];
964 }
965 }
966
967 TranslationBlock *tb_gen_code(CPUState *env,
968 target_ulong pc, target_ulong cs_base,
969 int flags, int cflags)
970 {
971 TranslationBlock *tb;
972 uint8_t *tc_ptr;
973 tb_page_addr_t phys_pc, phys_page2;
974 target_ulong virt_page2;
975 int code_gen_size;
976
977 phys_pc = get_page_addr_code(env, pc);
978 tb = tb_alloc(pc);
979 if (!tb) {
980 /* flush must be done */
981 tb_flush(env);
982 /* cannot fail at this point */
983 tb = tb_alloc(pc);
984 /* Don't forget to invalidate previous TB info. */
985 tb_invalidated_flag = 1;
986 }
987 tc_ptr = code_gen_ptr;
988 tb->tc_ptr = tc_ptr;
989 tb->cs_base = cs_base;
990 tb->flags = flags;
991 tb->cflags = cflags;
992 cpu_gen_code(env, tb, &code_gen_size);
993 code_gen_ptr = (void *)(((unsigned long)code_gen_ptr + code_gen_size + CODE_GEN_ALIGN - 1) & ~(CODE_GEN_ALIGN - 1));
994
995 /* check next page if needed */
996 virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
997 phys_page2 = -1;
998 if ((pc & TARGET_PAGE_MASK) != virt_page2) {
999 phys_page2 = get_page_addr_code(env, virt_page2);
1000 }
1001 tb_link_page(tb, phys_pc, phys_page2);
1002 return tb;
1003 }
1004
1005 /* invalidate all TBs which intersect with the target physical page
1006 starting in range [start;end[. NOTE: start and end must refer to
1007 the same physical page. 'is_cpu_write_access' should be true if called
1008 from a real cpu write access: the virtual CPU will exit the current
1009 TB if code is modified inside this TB. */
1010 void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
1011 int is_cpu_write_access)
1012 {
1013 TranslationBlock *tb, *tb_next, *saved_tb;
1014 CPUState *env = cpu_single_env;
1015 tb_page_addr_t tb_start, tb_end;
1016 PageDesc *p;
1017 int n;
1018 #ifdef TARGET_HAS_PRECISE_SMC
1019 int current_tb_not_found = is_cpu_write_access;
1020 TranslationBlock *current_tb = NULL;
1021 int current_tb_modified = 0;
1022 target_ulong current_pc = 0;
1023 target_ulong current_cs_base = 0;
1024 int current_flags = 0;
1025 #endif /* TARGET_HAS_PRECISE_SMC */
1026
1027 p = page_find(start >> TARGET_PAGE_BITS);
1028 if (!p)
1029 return;
1030 if (!p->code_bitmap &&
1031 ++p->code_write_count >= SMC_BITMAP_USE_THRESHOLD &&
1032 is_cpu_write_access) {
1033 /* build code bitmap */
1034 build_page_bitmap(p);
1035 }
1036
1037 /* we remove all the TBs in the range [start, end[ */
1038 /* XXX: see if in some cases it could be faster to invalidate all the code */
1039 tb = p->first_tb;
1040 while (tb != NULL) {
1041 n = (long)tb & 3;
1042 tb = (TranslationBlock *)((long)tb & ~3);
1043 tb_next = tb->page_next[n];
1044 /* NOTE: this is subtle as a TB may span two physical pages */
1045 if (n == 0) {
1046 /* NOTE: tb_end may be after the end of the page, but
1047 it is not a problem */
1048 tb_start = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
1049 tb_end = tb_start + tb->size;
1050 } else {
1051 tb_start = tb->page_addr[1];
1052 tb_end = tb_start + ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
1053 }
1054 if (!(tb_end <= start || tb_start >= end)) {
1055 #ifdef TARGET_HAS_PRECISE_SMC
1056 if (current_tb_not_found) {
1057 current_tb_not_found = 0;
1058 current_tb = NULL;
1059 if (env->mem_io_pc) {
1060 /* now we have a real cpu fault */
1061 current_tb = tb_find_pc(env->mem_io_pc);
1062 }
1063 }
1064 if (current_tb == tb &&
1065 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1066 /* If we are modifying the current TB, we must stop
1067 its execution. We could be more precise by checking
1068 that the modification is after the current PC, but it
1069 would require a specialized function to partially
1070 restore the CPU state */
1071
1072 current_tb_modified = 1;
1073 cpu_restore_state(current_tb, env,
1074 env->mem_io_pc, NULL);
1075 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1076 &current_flags);
1077 }
1078 #endif /* TARGET_HAS_PRECISE_SMC */
1079 /* we need to do that to handle the case where a signal
1080 occurs while doing tb_phys_invalidate() */
1081 saved_tb = NULL;
1082 if (env) {
1083 saved_tb = env->current_tb;
1084 env->current_tb = NULL;
1085 }
1086 tb_phys_invalidate(tb, -1);
1087 if (env) {
1088 env->current_tb = saved_tb;
1089 if (env->interrupt_request && env->current_tb)
1090 cpu_interrupt(env, env->interrupt_request);
1091 }
1092 }
1093 tb = tb_next;
1094 }
1095 #if !defined(CONFIG_USER_ONLY)
1096 /* if no code remaining, no need to continue to use slow writes */
1097 if (!p->first_tb) {
1098 invalidate_page_bitmap(p);
1099 if (is_cpu_write_access) {
1100 tlb_unprotect_code_phys(env, start, env->mem_io_vaddr);
1101 }
1102 }
1103 #endif
1104 #ifdef TARGET_HAS_PRECISE_SMC
1105 if (current_tb_modified) {
1106 /* we generate a block containing just the instruction
1107 modifying the memory. It will ensure that it cannot modify
1108 itself */
1109 env->current_tb = NULL;
1110 tb_gen_code(env, current_pc, current_cs_base, current_flags, 1);
1111 cpu_resume_from_signal(env, NULL);
1112 }
1113 #endif
1114 }
1115
1116 /* len must be <= 8 and start must be a multiple of len */
1117 static inline void tb_invalidate_phys_page_fast(tb_page_addr_t start, int len)
1118 {
1119 PageDesc *p;
1120 int offset, b;
1121 #if 0
1122 if (1) {
1123 qemu_log("modifying code at 0x%x size=%d EIP=%x PC=%08x\n",
1124 cpu_single_env->mem_io_vaddr, len,
1125 cpu_single_env->eip,
1126 cpu_single_env->eip + (long)cpu_single_env->segs[R_CS].base);
1127 }
1128 #endif
1129 p = page_find(start >> TARGET_PAGE_BITS);
1130 if (!p)
1131 return;
1132 if (p->code_bitmap) {
1133 offset = start & ~TARGET_PAGE_MASK;
1134 b = p->code_bitmap[offset >> 3] >> (offset & 7);
1135 if (b & ((1 << len) - 1))
1136 goto do_invalidate;
1137 } else {
1138 do_invalidate:
1139 tb_invalidate_phys_page_range(start, start + len, 1);
1140 }
1141 }
1142
1143 #if !defined(CONFIG_SOFTMMU)
1144 static void tb_invalidate_phys_page(tb_page_addr_t addr,
1145 unsigned long pc, void *puc)
1146 {
1147 TranslationBlock *tb;
1148 PageDesc *p;
1149 int n;
1150 #ifdef TARGET_HAS_PRECISE_SMC
1151 TranslationBlock *current_tb = NULL;
1152 CPUState *env = cpu_single_env;
1153 int current_tb_modified = 0;
1154 target_ulong current_pc = 0;
1155 target_ulong current_cs_base = 0;
1156 int current_flags = 0;
1157 #endif
1158
1159 addr &= TARGET_PAGE_MASK;
1160 p = page_find(addr >> TARGET_PAGE_BITS);
1161 if (!p)
1162 return;
1163 tb = p->first_tb;
1164 #ifdef TARGET_HAS_PRECISE_SMC
1165 if (tb && pc != 0) {
1166 current_tb = tb_find_pc(pc);
1167 }
1168 #endif
1169 while (tb != NULL) {
1170 n = (long)tb & 3;
1171 tb = (TranslationBlock *)((long)tb & ~3);
1172 #ifdef TARGET_HAS_PRECISE_SMC
1173 if (current_tb == tb &&
1174 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1175 /* If we are modifying the current TB, we must stop
1176 its execution. We could be more precise by checking
1177 that the modification is after the current PC, but it
1178 would require a specialized function to partially
1179 restore the CPU state */
1180
1181 current_tb_modified = 1;
1182 cpu_restore_state(current_tb, env, pc, puc);
1183 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1184 &current_flags);
1185 }
1186 #endif /* TARGET_HAS_PRECISE_SMC */
1187 tb_phys_invalidate(tb, addr);
1188 tb = tb->page_next[n];
1189 }
1190 p->first_tb = NULL;
1191 #ifdef TARGET_HAS_PRECISE_SMC
1192 if (current_tb_modified) {
1193 /* we generate a block containing just the instruction
1194 modifying the memory. It will ensure that it cannot modify
1195 itself */
1196 env->current_tb = NULL;
1197 tb_gen_code(env, current_pc, current_cs_base, current_flags, 1);
1198 cpu_resume_from_signal(env, puc);
1199 }
1200 #endif
1201 }
1202 #endif
1203
1204 /* add the tb in the target page and protect it if necessary */
1205 static inline void tb_alloc_page(TranslationBlock *tb,
1206 unsigned int n, tb_page_addr_t page_addr)
1207 {
1208 PageDesc *p;
1209 TranslationBlock *last_first_tb;
1210
1211 tb->page_addr[n] = page_addr;
1212 p = page_find_alloc(page_addr >> TARGET_PAGE_BITS, 1);
1213 tb->page_next[n] = p->first_tb;
1214 last_first_tb = p->first_tb;
1215 p->first_tb = (TranslationBlock *)((long)tb | n);
1216 invalidate_page_bitmap(p);
1217
1218 #if defined(TARGET_HAS_SMC) || 1
1219
1220 #if defined(CONFIG_USER_ONLY)
1221 if (p->flags & PAGE_WRITE) {
1222 target_ulong addr;
1223 PageDesc *p2;
1224 int prot;
1225
1226 /* force the host page as non writable (writes will have a
1227 page fault + mprotect overhead) */
1228 page_addr &= qemu_host_page_mask;
1229 prot = 0;
1230 for(addr = page_addr; addr < page_addr + qemu_host_page_size;
1231 addr += TARGET_PAGE_SIZE) {
1232
1233 p2 = page_find (addr >> TARGET_PAGE_BITS);
1234 if (!p2)
1235 continue;
1236 prot |= p2->flags;
1237 p2->flags &= ~PAGE_WRITE;
1238 }
1239 mprotect(g2h(page_addr), qemu_host_page_size,
1240 (prot & PAGE_BITS) & ~PAGE_WRITE);
1241 #ifdef DEBUG_TB_INVALIDATE
1242 printf("protecting code page: 0x" TARGET_FMT_lx "\n",
1243 page_addr);
1244 #endif
1245 }
1246 #else
1247 /* if some code is already present, then the pages are already
1248 protected. So we handle the case where only the first TB is
1249 allocated in a physical page */
1250 if (!last_first_tb) {
1251 tlb_protect_code(page_addr);
1252 }
1253 #endif
1254
1255 #endif /* TARGET_HAS_SMC */
1256 }
1257
1258 /* add a new TB and link it to the physical page tables. phys_page2 is
1259 (-1) to indicate that only one page contains the TB. */
1260 void tb_link_page(TranslationBlock *tb,
1261 tb_page_addr_t phys_pc, tb_page_addr_t phys_page2)
1262 {
1263 unsigned int h;
1264 TranslationBlock **ptb;
1265
1266 /* Grab the mmap lock to stop another thread invalidating this TB
1267 before we are done. */
1268 mmap_lock();
1269 /* add in the physical hash table */
1270 h = tb_phys_hash_func(phys_pc);
1271 ptb = &tb_phys_hash[h];
1272 tb->phys_hash_next = *ptb;
1273 *ptb = tb;
1274
1275 /* add in the page list */
1276 tb_alloc_page(tb, 0, phys_pc & TARGET_PAGE_MASK);
1277 if (phys_page2 != -1)
1278 tb_alloc_page(tb, 1, phys_page2);
1279 else
1280 tb->page_addr[1] = -1;
1281
1282 tb->jmp_first = (TranslationBlock *)((long)tb | 2);
1283 tb->jmp_next[0] = NULL;
1284 tb->jmp_next[1] = NULL;
1285
1286 /* init original jump addresses */
1287 if (tb->tb_next_offset[0] != 0xffff)
1288 tb_reset_jump(tb, 0);
1289 if (tb->tb_next_offset[1] != 0xffff)
1290 tb_reset_jump(tb, 1);
1291
1292 #ifdef DEBUG_TB_CHECK
1293 tb_page_check();
1294 #endif
1295 mmap_unlock();
1296 }
1297
1298 /* find the TB 'tb' such that tb[0].tc_ptr <= tc_ptr <
1299 tb[1].tc_ptr. Return NULL if not found */
1300 TranslationBlock *tb_find_pc(unsigned long tc_ptr)
1301 {
1302 int m_min, m_max, m;
1303 unsigned long v;
1304 TranslationBlock *tb;
1305
1306 if (nb_tbs <= 0)
1307 return NULL;
1308 if (tc_ptr < (unsigned long)code_gen_buffer ||
1309 tc_ptr >= (unsigned long)code_gen_ptr)
1310 return NULL;
1311 /* binary search (cf Knuth) */
1312 m_min = 0;
1313 m_max = nb_tbs - 1;
1314 while (m_min <= m_max) {
1315 m = (m_min + m_max) >> 1;
1316 tb = &tbs[m];
1317 v = (unsigned long)tb->tc_ptr;
1318 if (v == tc_ptr)
1319 return tb;
1320 else if (tc_ptr < v) {
1321 m_max = m - 1;
1322 } else {
1323 m_min = m + 1;
1324 }
1325 }
1326 return &tbs[m_max];
1327 }
1328
1329 static void tb_reset_jump_recursive(TranslationBlock *tb);
1330
1331 static inline void tb_reset_jump_recursive2(TranslationBlock *tb, int n)
1332 {
1333 TranslationBlock *tb1, *tb_next, **ptb;
1334 unsigned int n1;
1335
1336 tb1 = tb->jmp_next[n];
1337 if (tb1 != NULL) {
1338 /* find head of list */
1339 for(;;) {
1340 n1 = (long)tb1 & 3;
1341 tb1 = (TranslationBlock *)((long)tb1 & ~3);
1342 if (n1 == 2)
1343 break;
1344 tb1 = tb1->jmp_next[n1];
1345 }
1346 /* we are now sure now that tb jumps to tb1 */
1347 tb_next = tb1;
1348
1349 /* remove tb from the jmp_first list */
1350 ptb = &tb_next->jmp_first;
1351 for(;;) {
1352 tb1 = *ptb;
1353 n1 = (long)tb1 & 3;
1354 tb1 = (TranslationBlock *)((long)tb1 & ~3);
1355 if (n1 == n && tb1 == tb)
1356 break;
1357 ptb = &tb1->jmp_next[n1];
1358 }
1359 *ptb = tb->jmp_next[n];
1360 tb->jmp_next[n] = NULL;
1361
1362 /* suppress the jump to next tb in generated code */
1363 tb_reset_jump(tb, n);
1364
1365 /* suppress jumps in the tb on which we could have jumped */
1366 tb_reset_jump_recursive(tb_next);
1367 }
1368 }
1369
1370 static void tb_reset_jump_recursive(TranslationBlock *tb)
1371 {
1372 tb_reset_jump_recursive2(tb, 0);
1373 tb_reset_jump_recursive2(tb, 1);
1374 }
1375
1376 #if defined(TARGET_HAS_ICE)
1377 #if defined(CONFIG_USER_ONLY)
1378 static void breakpoint_invalidate(CPUState *env, target_ulong pc)
1379 {
1380 tb_invalidate_phys_page_range(pc, pc + 1, 0);
1381 }
1382 #else
1383 static void breakpoint_invalidate(CPUState *env, target_ulong pc)
1384 {
1385 target_phys_addr_t addr;
1386 target_ulong pd;
1387 ram_addr_t ram_addr;
1388 PhysPageDesc *p;
1389
1390 addr = cpu_get_phys_page_debug(env, pc);
1391 p = phys_page_find(addr >> TARGET_PAGE_BITS);
1392 if (!p) {
1393 pd = IO_MEM_UNASSIGNED;
1394 } else {
1395 pd = p->phys_offset;
1396 }
1397 ram_addr = (pd & TARGET_PAGE_MASK) | (pc & ~TARGET_PAGE_MASK);
1398 tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
1399 }
1400 #endif
1401 #endif /* TARGET_HAS_ICE */
1402
1403 #if defined(CONFIG_USER_ONLY)
1404 void cpu_watchpoint_remove_all(CPUState *env, int mask)
1405
1406 {
1407 }
1408
1409 int cpu_watchpoint_insert(CPUState *env, target_ulong addr, target_ulong len,
1410 int flags, CPUWatchpoint **watchpoint)
1411 {
1412 return -ENOSYS;
1413 }
1414 #else
1415 /* Add a watchpoint. */
1416 int cpu_watchpoint_insert(CPUState *env, target_ulong addr, target_ulong len,
1417 int flags, CPUWatchpoint **watchpoint)
1418 {
1419 target_ulong len_mask = ~(len - 1);
1420 CPUWatchpoint *wp;
1421
1422 /* sanity checks: allow power-of-2 lengths, deny unaligned watchpoints */
1423 if ((len != 1 && len != 2 && len != 4 && len != 8) || (addr & ~len_mask)) {
1424 fprintf(stderr, "qemu: tried to set invalid watchpoint at "
1425 TARGET_FMT_lx ", len=" TARGET_FMT_lu "\n", addr, len);
1426 return -EINVAL;
1427 }
1428 wp = qemu_malloc(sizeof(*wp));
1429
1430 wp->vaddr = addr;
1431 wp->len_mask = len_mask;
1432 wp->flags = flags;
1433
1434 /* keep all GDB-injected watchpoints in front */
1435 if (flags & BP_GDB)
1436 QTAILQ_INSERT_HEAD(&env->watchpoints, wp, entry);
1437 else
1438 QTAILQ_INSERT_TAIL(&env->watchpoints, wp, entry);
1439
1440 tlb_flush_page(env, addr);
1441
1442 if (watchpoint)
1443 *watchpoint = wp;
1444 return 0;
1445 }
1446
1447 /* Remove a specific watchpoint. */
1448 int cpu_watchpoint_remove(CPUState *env, target_ulong addr, target_ulong len,
1449 int flags)
1450 {
1451 target_ulong len_mask = ~(len - 1);
1452 CPUWatchpoint *wp;
1453
1454 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1455 if (addr == wp->vaddr && len_mask == wp->len_mask
1456 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
1457 cpu_watchpoint_remove_by_ref(env, wp);
1458 return 0;
1459 }
1460 }
1461 return -ENOENT;
1462 }
1463
1464 /* Remove a specific watchpoint by reference. */
1465 void cpu_watchpoint_remove_by_ref(CPUState *env, CPUWatchpoint *watchpoint)
1466 {
1467 QTAILQ_REMOVE(&env->watchpoints, watchpoint, entry);
1468
1469 tlb_flush_page(env, watchpoint->vaddr);
1470
1471 qemu_free(watchpoint);
1472 }
1473
1474 /* Remove all matching watchpoints. */
1475 void cpu_watchpoint_remove_all(CPUState *env, int mask)
1476 {
1477 CPUWatchpoint *wp, *next;
1478
1479 QTAILQ_FOREACH_SAFE(wp, &env->watchpoints, entry, next) {
1480 if (wp->flags & mask)
1481 cpu_watchpoint_remove_by_ref(env, wp);
1482 }
1483 }
1484 #endif
1485
1486 /* Add a breakpoint. */
1487 int cpu_breakpoint_insert(CPUState *env, target_ulong pc, int flags,
1488 CPUBreakpoint **breakpoint)
1489 {
1490 #if defined(TARGET_HAS_ICE)
1491 CPUBreakpoint *bp;
1492
1493 bp = qemu_malloc(sizeof(*bp));
1494
1495 bp->pc = pc;
1496 bp->flags = flags;
1497
1498 /* keep all GDB-injected breakpoints in front */
1499 if (flags & BP_GDB)
1500 QTAILQ_INSERT_HEAD(&env->breakpoints, bp, entry);
1501 else
1502 QTAILQ_INSERT_TAIL(&env->breakpoints, bp, entry);
1503
1504 breakpoint_invalidate(env, pc);
1505
1506 if (breakpoint)
1507 *breakpoint = bp;
1508 return 0;
1509 #else
1510 return -ENOSYS;
1511 #endif
1512 }
1513
1514 /* Remove a specific breakpoint. */
1515 int cpu_breakpoint_remove(CPUState *env, target_ulong pc, int flags)
1516 {
1517 #if defined(TARGET_HAS_ICE)
1518 CPUBreakpoint *bp;
1519
1520 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
1521 if (bp->pc == pc && bp->flags == flags) {
1522 cpu_breakpoint_remove_by_ref(env, bp);
1523 return 0;
1524 }
1525 }
1526 return -ENOENT;
1527 #else
1528 return -ENOSYS;
1529 #endif
1530 }
1531
1532 /* Remove a specific breakpoint by reference. */
1533 void cpu_breakpoint_remove_by_ref(CPUState *env, CPUBreakpoint *breakpoint)
1534 {
1535 #if defined(TARGET_HAS_ICE)
1536 QTAILQ_REMOVE(&env->breakpoints, breakpoint, entry);
1537
1538 breakpoint_invalidate(env, breakpoint->pc);
1539
1540 qemu_free(breakpoint);
1541 #endif
1542 }
1543
1544 /* Remove all matching breakpoints. */
1545 void cpu_breakpoint_remove_all(CPUState *env, int mask)
1546 {
1547 #if defined(TARGET_HAS_ICE)
1548 CPUBreakpoint *bp, *next;
1549
1550 QTAILQ_FOREACH_SAFE(bp, &env->breakpoints, entry, next) {
1551 if (bp->flags & mask)
1552 cpu_breakpoint_remove_by_ref(env, bp);
1553 }
1554 #endif
1555 }
1556
1557 /* enable or disable single step mode. EXCP_DEBUG is returned by the
1558 CPU loop after each instruction */
1559 void cpu_single_step(CPUState *env, int enabled)
1560 {
1561 #if defined(TARGET_HAS_ICE)
1562 if (env->singlestep_enabled != enabled) {
1563 env->singlestep_enabled = enabled;
1564 if (kvm_enabled())
1565 kvm_update_guest_debug(env, 0);
1566 else {
1567 /* must flush all the translated code to avoid inconsistencies */
1568 /* XXX: only flush what is necessary */
1569 tb_flush(env);
1570 }
1571 }
1572 #endif
1573 }
1574
1575 /* enable or disable low levels log */
1576 void cpu_set_log(int log_flags)
1577 {
1578 loglevel = log_flags;
1579 if (loglevel && !logfile) {
1580 logfile = fopen(logfilename, log_append ? "a" : "w");
1581 if (!logfile) {
1582 perror(logfilename);
1583 _exit(1);
1584 }
1585 #if !defined(CONFIG_SOFTMMU)
1586 /* must avoid mmap() usage of glibc by setting a buffer "by hand" */
1587 {
1588 static char logfile_buf[4096];
1589 setvbuf(logfile, logfile_buf, _IOLBF, sizeof(logfile_buf));
1590 }
1591 #elif !defined(_WIN32)
1592 /* Win32 doesn't support line-buffering and requires size >= 2 */
1593 setvbuf(logfile, NULL, _IOLBF, 0);
1594 #endif
1595 log_append = 1;
1596 }
1597 if (!loglevel && logfile) {
1598 fclose(logfile);
1599 logfile = NULL;
1600 }
1601 }
1602
1603 void cpu_set_log_filename(const char *filename)
1604 {
1605 logfilename = strdup(filename);
1606 if (logfile) {
1607 fclose(logfile);
1608 logfile = NULL;
1609 }
1610 cpu_set_log(loglevel);
1611 }
1612
1613 static void cpu_unlink_tb(CPUState *env)
1614 {
1615 /* FIXME: TB unchaining isn't SMP safe. For now just ignore the
1616 problem and hope the cpu will stop of its own accord. For userspace
1617 emulation this often isn't actually as bad as it sounds. Often
1618 signals are used primarily to interrupt blocking syscalls. */
1619 TranslationBlock *tb;
1620 static spinlock_t interrupt_lock = SPIN_LOCK_UNLOCKED;
1621
1622 spin_lock(&interrupt_lock);
1623 tb = env->current_tb;
1624 /* if the cpu is currently executing code, we must unlink it and
1625 all the potentially executing TB */
1626 if (tb) {
1627 env->current_tb = NULL;
1628 tb_reset_jump_recursive(tb);
1629 }
1630 spin_unlock(&interrupt_lock);
1631 }
1632
1633 /* mask must never be zero, except for A20 change call */
1634 void cpu_interrupt(CPUState *env, int mask)
1635 {
1636 int old_mask;
1637
1638 old_mask = env->interrupt_request;
1639 env->interrupt_request |= mask;
1640
1641 #ifndef CONFIG_USER_ONLY
1642 /*
1643 * If called from iothread context, wake the target cpu in
1644 * case its halted.
1645 */
1646 if (!qemu_cpu_is_self(env)) {
1647 qemu_cpu_kick(env);
1648 return;
1649 }
1650 #endif
1651
1652 if (use_icount) {
1653 env->icount_decr.u16.high = 0xffff;
1654 #ifndef CONFIG_USER_ONLY
1655 if (!can_do_io(env)
1656 && (mask & ~old_mask) != 0) {
1657 cpu_abort(env, "Raised interrupt while not in I/O function");
1658 }
1659 #endif
1660 } else {
1661 cpu_unlink_tb(env);
1662 }
1663 }
1664
1665 void cpu_reset_interrupt(CPUState *env, int mask)
1666 {
1667 env->interrupt_request &= ~mask;
1668 }
1669
1670 void cpu_exit(CPUState *env)
1671 {
1672 env->exit_request = 1;
1673 cpu_unlink_tb(env);
1674 }
1675
1676 const CPULogItem cpu_log_items[] = {
1677 { CPU_LOG_TB_OUT_ASM, "out_asm",
1678 "show generated host assembly code for each compiled TB" },
1679 { CPU_LOG_TB_IN_ASM, "in_asm",
1680 "show target assembly code for each compiled TB" },
1681 { CPU_LOG_TB_OP, "op",
1682 "show micro ops for each compiled TB" },
1683 { CPU_LOG_TB_OP_OPT, "op_opt",
1684 "show micro ops "
1685 #ifdef TARGET_I386
1686 "before eflags optimization and "
1687 #endif
1688 "after liveness analysis" },
1689 { CPU_LOG_INT, "int",
1690 "show interrupts/exceptions in short format" },
1691 { CPU_LOG_EXEC, "exec",
1692 "show trace before each executed TB (lots of logs)" },
1693 { CPU_LOG_TB_CPU, "cpu",
1694 "show CPU state before block translation" },
1695 #ifdef TARGET_I386
1696 { CPU_LOG_PCALL, "pcall",
1697 "show protected mode far calls/returns/exceptions" },
1698 { CPU_LOG_RESET, "cpu_reset",
1699 "show CPU state before CPU resets" },
1700 #endif
1701 #ifdef DEBUG_IOPORT
1702 { CPU_LOG_IOPORT, "ioport",
1703 "show all i/o ports accesses" },
1704 #endif
1705 { 0, NULL, NULL },
1706 };
1707
1708 #ifndef CONFIG_USER_ONLY
1709 static QLIST_HEAD(memory_client_list, CPUPhysMemoryClient) memory_client_list
1710 = QLIST_HEAD_INITIALIZER(memory_client_list);
1711
1712 static void cpu_notify_set_memory(target_phys_addr_t start_addr,
1713 ram_addr_t size,
1714 ram_addr_t phys_offset,
1715 bool log_dirty)
1716 {
1717 CPUPhysMemoryClient *client;
1718 QLIST_FOREACH(client, &memory_client_list, list) {
1719 client->set_memory(client, start_addr, size, phys_offset, log_dirty);
1720 }
1721 }
1722
1723 static int cpu_notify_sync_dirty_bitmap(target_phys_addr_t start,
1724 target_phys_addr_t end)
1725 {
1726 CPUPhysMemoryClient *client;
1727 QLIST_FOREACH(client, &memory_client_list, list) {
1728 int r = client->sync_dirty_bitmap(client, start, end);
1729 if (r < 0)
1730 return r;
1731 }
1732 return 0;
1733 }
1734
1735 static int cpu_notify_migration_log(int enable)
1736 {
1737 CPUPhysMemoryClient *client;
1738 QLIST_FOREACH(client, &memory_client_list, list) {
1739 int r = client->migration_log(client, enable);
1740 if (r < 0)
1741 return r;
1742 }
1743 return 0;
1744 }
1745
1746 static void phys_page_for_each_1(CPUPhysMemoryClient *client,
1747 int level, void **lp)
1748 {
1749 int i;
1750
1751 if (*lp == NULL) {
1752 return;
1753 }
1754 if (level == 0) {
1755 PhysPageDesc *pd = *lp;
1756 for (i = 0; i < L2_SIZE; ++i) {
1757 if (pd[i].phys_offset != IO_MEM_UNASSIGNED) {
1758 client->set_memory(client, pd[i].region_offset,
1759 TARGET_PAGE_SIZE, pd[i].phys_offset, false);
1760 }
1761 }
1762 } else {
1763 void **pp = *lp;
1764 for (i = 0; i < L2_SIZE; ++i) {
1765 phys_page_for_each_1(client, level - 1, pp + i);
1766 }
1767 }
1768 }
1769
1770 static void phys_page_for_each(CPUPhysMemoryClient *client)
1771 {
1772 int i;
1773 for (i = 0; i < P_L1_SIZE; ++i) {
1774 phys_page_for_each_1(client, P_L1_SHIFT / L2_BITS - 1,
1775 l1_phys_map + i);
1776 }
1777 }
1778
1779 void cpu_register_phys_memory_client(CPUPhysMemoryClient *client)
1780 {
1781 QLIST_INSERT_HEAD(&memory_client_list, client, list);
1782 phys_page_for_each(client);
1783 }
1784
1785 void cpu_unregister_phys_memory_client(CPUPhysMemoryClient *client)
1786 {
1787 QLIST_REMOVE(client, list);
1788 }
1789 #endif
1790
1791 static int cmp1(const char *s1, int n, const char *s2)
1792 {
1793 if (strlen(s2) != n)
1794 return 0;
1795 return memcmp(s1, s2, n) == 0;
1796 }
1797
1798 /* takes a comma separated list of log masks. Return 0 if error. */
1799 int cpu_str_to_log_mask(const char *str)
1800 {
1801 const CPULogItem *item;
1802 int mask;
1803 const char *p, *p1;
1804
1805 p = str;
1806 mask = 0;
1807 for(;;) {
1808 p1 = strchr(p, ',');
1809 if (!p1)
1810 p1 = p + strlen(p);
1811 if(cmp1(p,p1-p,"all")) {
1812 for(item = cpu_log_items; item->mask != 0; item++) {
1813 mask |= item->mask;
1814 }
1815 } else {
1816 for(item = cpu_log_items; item->mask != 0; item++) {
1817 if (cmp1(p, p1 - p, item->name))
1818 goto found;
1819 }
1820 return 0;
1821 }
1822 found:
1823 mask |= item->mask;
1824 if (*p1 != ',')
1825 break;
1826 p = p1 + 1;
1827 }
1828 return mask;
1829 }
1830
1831 void cpu_abort(CPUState *env, const char *fmt, ...)
1832 {
1833 va_list ap;
1834 va_list ap2;
1835
1836 va_start(ap, fmt);
1837 va_copy(ap2, ap);
1838 fprintf(stderr, "qemu: fatal: ");
1839 vfprintf(stderr, fmt, ap);
1840 fprintf(stderr, "\n");
1841 #ifdef TARGET_I386
1842 cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU | X86_DUMP_CCOP);
1843 #else
1844 cpu_dump_state(env, stderr, fprintf, 0);
1845 #endif
1846 if (qemu_log_enabled()) {
1847 qemu_log("qemu: fatal: ");
1848 qemu_log_vprintf(fmt, ap2);
1849 qemu_log("\n");
1850 #ifdef TARGET_I386
1851 log_cpu_state(env, X86_DUMP_FPU | X86_DUMP_CCOP);
1852 #else
1853 log_cpu_state(env, 0);
1854 #endif
1855 qemu_log_flush();
1856 qemu_log_close();
1857 }
1858 va_end(ap2);
1859 va_end(ap);
1860 #if defined(CONFIG_USER_ONLY)
1861 {
1862 struct sigaction act;
1863 sigfillset(&act.sa_mask);
1864 act.sa_handler = SIG_DFL;
1865 sigaction(SIGABRT, &act, NULL);
1866 }
1867 #endif
1868 abort();
1869 }
1870
1871 CPUState *cpu_copy(CPUState *env)
1872 {
1873 CPUState *new_env = cpu_init(env->cpu_model_str);
1874 CPUState *next_cpu = new_env->next_cpu;
1875 int cpu_index = new_env->cpu_index;
1876 #if defined(TARGET_HAS_ICE)
1877 CPUBreakpoint *bp;
1878 CPUWatchpoint *wp;
1879 #endif
1880
1881 memcpy(new_env, env, sizeof(CPUState));
1882
1883 /* Preserve chaining and index. */
1884 new_env->next_cpu = next_cpu;
1885 new_env->cpu_index = cpu_index;
1886
1887 /* Clone all break/watchpoints.
1888 Note: Once we support ptrace with hw-debug register access, make sure
1889 BP_CPU break/watchpoints are handled correctly on clone. */
1890 QTAILQ_INIT(&env->breakpoints);
1891 QTAILQ_INIT(&env->watchpoints);
1892 #if defined(TARGET_HAS_ICE)
1893 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
1894 cpu_breakpoint_insert(new_env, bp->pc, bp->flags, NULL);
1895 }
1896 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1897 cpu_watchpoint_insert(new_env, wp->vaddr, (~wp->len_mask) + 1,
1898 wp->flags, NULL);
1899 }
1900 #endif
1901
1902 return new_env;
1903 }
1904
1905 #if !defined(CONFIG_USER_ONLY)
1906
1907 static inline void tlb_flush_jmp_cache(CPUState *env, target_ulong addr)
1908 {
1909 unsigned int i;
1910
1911 /* Discard jump cache entries for any tb which might potentially
1912 overlap the flushed page. */
1913 i = tb_jmp_cache_hash_page(addr - TARGET_PAGE_SIZE);
1914 memset (&env->tb_jmp_cache[i], 0,
1915 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1916
1917 i = tb_jmp_cache_hash_page(addr);
1918 memset (&env->tb_jmp_cache[i], 0,
1919 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1920 }
1921
1922 static CPUTLBEntry s_cputlb_empty_entry = {
1923 .addr_read = -1,
1924 .addr_write = -1,
1925 .addr_code = -1,
1926 .addend = -1,
1927 };
1928
1929 /* NOTE: if flush_global is true, also flush global entries (not
1930 implemented yet) */
1931 void tlb_flush(CPUState *env, int flush_global)
1932 {
1933 int i;
1934
1935 #if defined(DEBUG_TLB)
1936 printf("tlb_flush:\n");
1937 #endif
1938 /* must reset current TB so that interrupts cannot modify the
1939 links while we are modifying them */
1940 env->current_tb = NULL;
1941
1942 for(i = 0; i < CPU_TLB_SIZE; i++) {
1943 int mmu_idx;
1944 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
1945 env->tlb_table[mmu_idx][i] = s_cputlb_empty_entry;
1946 }
1947 }
1948
1949 memset (env->tb_jmp_cache, 0, TB_JMP_CACHE_SIZE * sizeof (void *));
1950
1951 env->tlb_flush_addr = -1;
1952 env->tlb_flush_mask = 0;
1953 tlb_flush_count++;
1954 }
1955
1956 static inline void tlb_flush_entry(CPUTLBEntry *tlb_entry, target_ulong addr)
1957 {
1958 if (addr == (tlb_entry->addr_read &
1959 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) ||
1960 addr == (tlb_entry->addr_write &
1961 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) ||
1962 addr == (tlb_entry->addr_code &
1963 (TARGET_PAGE_MASK | TLB_INVALID_MASK))) {
1964 *tlb_entry = s_cputlb_empty_entry;
1965 }
1966 }
1967
1968 void tlb_flush_page(CPUState *env, target_ulong addr)
1969 {
1970 int i;
1971 int mmu_idx;
1972
1973 #if defined(DEBUG_TLB)
1974 printf("tlb_flush_page: " TARGET_FMT_lx "\n", addr);
1975 #endif
1976 /* Check if we need to flush due to large pages. */
1977 if ((addr & env->tlb_flush_mask) == env->tlb_flush_addr) {
1978 #if defined(DEBUG_TLB)
1979 printf("tlb_flush_page: forced full flush ("
1980 TARGET_FMT_lx "/" TARGET_FMT_lx ")\n",
1981 env->tlb_flush_addr, env->tlb_flush_mask);
1982 #endif
1983 tlb_flush(env, 1);
1984 return;
1985 }
1986 /* must reset current TB so that interrupts cannot modify the
1987 links while we are modifying them */
1988 env->current_tb = NULL;
1989
1990 addr &= TARGET_PAGE_MASK;
1991 i = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
1992 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++)
1993 tlb_flush_entry(&env->tlb_table[mmu_idx][i], addr);
1994
1995 tlb_flush_jmp_cache(env, addr);
1996 }
1997
1998 /* update the TLBs so that writes to code in the virtual page 'addr'
1999 can be detected */
2000 static void tlb_protect_code(ram_addr_t ram_addr)
2001 {
2002 cpu_physical_memory_reset_dirty(ram_addr,
2003 ram_addr + TARGET_PAGE_SIZE,
2004 CODE_DIRTY_FLAG);
2005 }
2006
2007 /* update the TLB so that writes in physical page 'phys_addr' are no longer
2008 tested for self modifying code */
2009 static void tlb_unprotect_code_phys(CPUState *env, ram_addr_t ram_addr,
2010 target_ulong vaddr)
2011 {
2012 cpu_physical_memory_set_dirty_flags(ram_addr, CODE_DIRTY_FLAG);
2013 }
2014
2015 static inline void tlb_reset_dirty_range(CPUTLBEntry *tlb_entry,
2016 unsigned long start, unsigned long length)
2017 {
2018 unsigned long addr;
2019 if ((tlb_entry->addr_write & ~TARGET_PAGE_MASK) == IO_MEM_RAM) {
2020 addr = (tlb_entry->addr_write & TARGET_PAGE_MASK) + tlb_entry->addend;
2021 if ((addr - start) < length) {
2022 tlb_entry->addr_write = (tlb_entry->addr_write & TARGET_PAGE_MASK) | TLB_NOTDIRTY;
2023 }
2024 }
2025 }
2026
2027 /* Note: start and end must be within the same ram block. */
2028 void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t end,
2029 int dirty_flags)
2030 {
2031 CPUState *env;
2032 unsigned long length, start1;
2033 int i;
2034
2035 start &= TARGET_PAGE_MASK;
2036 end = TARGET_PAGE_ALIGN(end);
2037
2038 length = end - start;
2039 if (length == 0)
2040 return;
2041 cpu_physical_memory_mask_dirty_range(start, length, dirty_flags);
2042
2043 /* we modify the TLB cache so that the dirty bit will be set again
2044 when accessing the range */
2045 start1 = (unsigned long)qemu_safe_ram_ptr(start);
2046 /* Chek that we don't span multiple blocks - this breaks the
2047 address comparisons below. */
2048 if ((unsigned long)qemu_safe_ram_ptr(end - 1) - start1
2049 != (end - 1) - start) {
2050 abort();
2051 }
2052
2053 for(env = first_cpu; env != NULL; env = env->next_cpu) {
2054 int mmu_idx;
2055 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
2056 for(i = 0; i < CPU_TLB_SIZE; i++)
2057 tlb_reset_dirty_range(&env->tlb_table[mmu_idx][i],
2058 start1, length);
2059 }
2060 }
2061 }
2062
2063 int cpu_physical_memory_set_dirty_tracking(int enable)
2064 {
2065 int ret = 0;
2066 in_migration = enable;
2067 ret = cpu_notify_migration_log(!!enable);
2068 return ret;
2069 }
2070
2071 int cpu_physical_memory_get_dirty_tracking(void)
2072 {
2073 return in_migration;
2074 }
2075
2076 int cpu_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
2077 target_phys_addr_t end_addr)
2078 {
2079 int ret;
2080
2081 ret = cpu_notify_sync_dirty_bitmap(start_addr, end_addr);
2082 return ret;
2083 }
2084
2085 int cpu_physical_log_start(target_phys_addr_t start_addr,
2086 ram_addr_t size)
2087 {
2088 CPUPhysMemoryClient *client;
2089 QLIST_FOREACH(client, &memory_client_list, list) {
2090 if (client->log_start) {
2091 int r = client->log_start(client, start_addr, size);
2092 if (r < 0) {
2093 return r;
2094 }
2095 }
2096 }
2097 return 0;
2098 }
2099
2100 int cpu_physical_log_stop(target_phys_addr_t start_addr,
2101 ram_addr_t size)
2102 {
2103 CPUPhysMemoryClient *client;
2104 QLIST_FOREACH(client, &memory_client_list, list) {
2105 if (client->log_stop) {
2106 int r = client->log_stop(client, start_addr, size);
2107 if (r < 0) {
2108 return r;
2109 }
2110 }
2111 }
2112 return 0;
2113 }
2114
2115 static inline void tlb_update_dirty(CPUTLBEntry *tlb_entry)
2116 {
2117 ram_addr_t ram_addr;
2118 void *p;
2119
2120 if ((tlb_entry->addr_write & ~TARGET_PAGE_MASK) == IO_MEM_RAM) {
2121 p = (void *)(unsigned long)((tlb_entry->addr_write & TARGET_PAGE_MASK)
2122 + tlb_entry->addend);
2123 ram_addr = qemu_ram_addr_from_host_nofail(p);
2124 if (!cpu_physical_memory_is_dirty(ram_addr)) {
2125 tlb_entry->addr_write |= TLB_NOTDIRTY;
2126 }
2127 }
2128 }
2129
2130 /* update the TLB according to the current state of the dirty bits */
2131 void cpu_tlb_update_dirty(CPUState *env)
2132 {
2133 int i;
2134 int mmu_idx;
2135 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
2136 for(i = 0; i < CPU_TLB_SIZE; i++)
2137 tlb_update_dirty(&env->tlb_table[mmu_idx][i]);
2138 }
2139 }
2140
2141 static inline void tlb_set_dirty1(CPUTLBEntry *tlb_entry, target_ulong vaddr)
2142 {
2143 if (tlb_entry->addr_write == (vaddr | TLB_NOTDIRTY))
2144 tlb_entry->addr_write = vaddr;
2145 }
2146
2147 /* update the TLB corresponding to virtual page vaddr
2148 so that it is no longer dirty */
2149 static inline void tlb_set_dirty(CPUState *env, target_ulong vaddr)
2150 {
2151 int i;
2152 int mmu_idx;
2153
2154 vaddr &= TARGET_PAGE_MASK;
2155 i = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2156 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++)
2157 tlb_set_dirty1(&env->tlb_table[mmu_idx][i], vaddr);
2158 }
2159
2160 /* Our TLB does not support large pages, so remember the area covered by
2161 large pages and trigger a full TLB flush if these are invalidated. */
2162 static void tlb_add_large_page(CPUState *env, target_ulong vaddr,
2163 target_ulong size)
2164 {
2165 target_ulong mask = ~(size - 1);
2166
2167 if (env->tlb_flush_addr == (target_ulong)-1) {
2168 env->tlb_flush_addr = vaddr & mask;
2169 env->tlb_flush_mask = mask;
2170 return;
2171 }
2172 /* Extend the existing region to include the new page.
2173 This is a compromise between unnecessary flushes and the cost
2174 of maintaining a full variable size TLB. */
2175 mask &= env->tlb_flush_mask;
2176 while (((env->tlb_flush_addr ^ vaddr) & mask) != 0) {
2177 mask <<= 1;
2178 }
2179 env->tlb_flush_addr &= mask;
2180 env->tlb_flush_mask = mask;
2181 }
2182
2183 /* Add a new TLB entry. At most one entry for a given virtual address
2184 is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
2185 supplied size is only used by tlb_flush_page. */
2186 void tlb_set_page(CPUState *env, target_ulong vaddr,
2187 target_phys_addr_t paddr, int prot,
2188 int mmu_idx, target_ulong size)
2189 {
2190 PhysPageDesc *p;
2191 unsigned long pd;
2192 unsigned int index;
2193 target_ulong address;
2194 target_ulong code_address;
2195 unsigned long addend;
2196 CPUTLBEntry *te;
2197 CPUWatchpoint *wp;
2198 target_phys_addr_t iotlb;
2199
2200 assert(size >= TARGET_PAGE_SIZE);
2201 if (size != TARGET_PAGE_SIZE) {
2202 tlb_add_large_page(env, vaddr, size);
2203 }
2204 p = phys_page_find(paddr >> TARGET_PAGE_BITS);
2205 if (!p) {
2206 pd = IO_MEM_UNASSIGNED;
2207 } else {
2208 pd = p->phys_offset;
2209 }
2210 #if defined(DEBUG_TLB)
2211 printf("tlb_set_page: vaddr=" TARGET_FMT_lx " paddr=0x" TARGET_FMT_plx
2212 " prot=%x idx=%d pd=0x%08lx\n",
2213 vaddr, paddr, prot, mmu_idx, pd);
2214 #endif
2215
2216 address = vaddr;
2217 if ((pd & ~TARGET_PAGE_MASK) > IO_MEM_ROM && !(pd & IO_MEM_ROMD)) {
2218 /* IO memory case (romd handled later) */
2219 address |= TLB_MMIO;
2220 }
2221 addend = (unsigned long)qemu_get_ram_ptr(pd & TARGET_PAGE_MASK);
2222 if ((pd & ~TARGET_PAGE_MASK) <= IO_MEM_ROM) {
2223 /* Normal RAM. */
2224 iotlb = pd & TARGET_PAGE_MASK;
2225 if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_RAM)
2226 iotlb |= IO_MEM_NOTDIRTY;
2227 else
2228 iotlb |= IO_MEM_ROM;
2229 } else {
2230 /* IO handlers are currently passed a physical address.
2231 It would be nice to pass an offset from the base address
2232 of that region. This would avoid having to special case RAM,
2233 and avoid full address decoding in every device.
2234 We can't use the high bits of pd for this because
2235 IO_MEM_ROMD uses these as a ram address. */
2236 iotlb = (pd & ~TARGET_PAGE_MASK);
2237 if (p) {
2238 iotlb += p->region_offset;
2239 } else {
2240 iotlb += paddr;
2241 }
2242 }
2243
2244 code_address = address;
2245 /* Make accesses to pages with watchpoints go via the
2246 watchpoint trap routines. */
2247 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
2248 if (vaddr == (wp->vaddr & TARGET_PAGE_MASK)) {
2249 /* Avoid trapping reads of pages with a write breakpoint. */
2250 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
2251 iotlb = io_mem_watch + paddr;
2252 address |= TLB_MMIO;
2253 break;
2254 }
2255 }
2256 }
2257
2258 index = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2259 env->iotlb[mmu_idx][index] = iotlb - vaddr;
2260 te = &env->tlb_table[mmu_idx][index];
2261 te->addend = addend - vaddr;
2262 if (prot & PAGE_READ) {
2263 te->addr_read = address;
2264 } else {
2265 te->addr_read = -1;
2266 }
2267
2268 if (prot & PAGE_EXEC) {
2269 te->addr_code = code_address;
2270 } else {
2271 te->addr_code = -1;
2272 }
2273 if (prot & PAGE_WRITE) {
2274 if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_ROM ||
2275 (pd & IO_MEM_ROMD)) {
2276 /* Write access calls the I/O callback. */
2277 te->addr_write = address | TLB_MMIO;
2278 } else if ((pd & ~TARGET_PAGE_MASK) == IO_MEM_RAM &&
2279 !cpu_physical_memory_is_dirty(pd)) {
2280 te->addr_write = address | TLB_NOTDIRTY;
2281 } else {
2282 te->addr_write = address;
2283 }
2284 } else {
2285 te->addr_write = -1;
2286 }
2287 }
2288
2289 #else
2290
2291 void tlb_flush(CPUState *env, int flush_global)
2292 {
2293 }
2294
2295 void tlb_flush_page(CPUState *env, target_ulong addr)
2296 {
2297 }
2298
2299 /*
2300 * Walks guest process memory "regions" one by one
2301 * and calls callback function 'fn' for each region.
2302 */
2303
2304 struct walk_memory_regions_data
2305 {
2306 walk_memory_regions_fn fn;
2307 void *priv;
2308 unsigned long start;
2309 int prot;
2310 };
2311
2312 static int walk_memory_regions_end(struct walk_memory_regions_data *data,
2313 abi_ulong end, int new_prot)
2314 {
2315 if (data->start != -1ul) {
2316 int rc = data->fn(data->priv, data->start, end, data->prot);
2317 if (rc != 0) {
2318 return rc;
2319 }
2320 }
2321
2322 data->start = (new_prot ? end : -1ul);
2323 data->prot = new_prot;
2324
2325 return 0;
2326 }
2327
2328 static int walk_memory_regions_1(struct walk_memory_regions_data *data,
2329 abi_ulong base, int level, void **lp)
2330 {
2331 abi_ulong pa;
2332 int i, rc;
2333
2334 if (*lp == NULL) {
2335 return walk_memory_regions_end(data, base, 0);
2336 }
2337
2338 if (level == 0) {
2339 PageDesc *pd = *lp;
2340 for (i = 0; i < L2_SIZE; ++i) {
2341 int prot = pd[i].flags;
2342
2343 pa = base | (i << TARGET_PAGE_BITS);
2344 if (prot != data->prot) {
2345 rc = walk_memory_regions_end(data, pa, prot);
2346 if (rc != 0) {
2347 return rc;
2348 }
2349 }
2350 }
2351 } else {
2352 void **pp = *lp;
2353 for (i = 0; i < L2_SIZE; ++i) {
2354 pa = base | ((abi_ulong)i <<
2355 (TARGET_PAGE_BITS + L2_BITS * level));
2356 rc = walk_memory_regions_1(data, pa, level - 1, pp + i);
2357 if (rc != 0) {
2358 return rc;
2359 }
2360 }
2361 }
2362
2363 return 0;
2364 }
2365
2366 int walk_memory_regions(void *priv, walk_memory_regions_fn fn)
2367 {
2368 struct walk_memory_regions_data data;
2369 unsigned long i;
2370
2371 data.fn = fn;
2372 data.priv = priv;
2373 data.start = -1ul;
2374 data.prot = 0;
2375
2376 for (i = 0; i < V_L1_SIZE; i++) {
2377 int rc = walk_memory_regions_1(&data, (abi_ulong)i << V_L1_SHIFT,
2378 V_L1_SHIFT / L2_BITS - 1, l1_map + i);
2379 if (rc != 0) {
2380 return rc;
2381 }
2382 }
2383
2384 return walk_memory_regions_end(&data, 0, 0);
2385 }
2386
2387 static int dump_region(void *priv, abi_ulong start,
2388 abi_ulong end, unsigned long prot)
2389 {
2390 FILE *f = (FILE *)priv;
2391
2392 (void) fprintf(f, TARGET_ABI_FMT_lx"-"TARGET_ABI_FMT_lx
2393 " "TARGET_ABI_FMT_lx" %c%c%c\n",
2394 start, end, end - start,
2395 ((prot & PAGE_READ) ? 'r' : '-'),
2396 ((prot & PAGE_WRITE) ? 'w' : '-'),
2397 ((prot & PAGE_EXEC) ? 'x' : '-'));
2398
2399 return (0);
2400 }
2401
2402 /* dump memory mappings */
2403 void page_dump(FILE *f)
2404 {
2405 (void) fprintf(f, "%-8s %-8s %-8s %s\n",
2406 "start", "end", "size", "prot");
2407 walk_memory_regions(f, dump_region);
2408 }
2409
2410 int page_get_flags(target_ulong address)
2411 {
2412 PageDesc *p;
2413
2414 p = page_find(address >> TARGET_PAGE_BITS);
2415 if (!p)
2416 return 0;
2417 return p->flags;
2418 }
2419
2420 /* Modify the flags of a page and invalidate the code if necessary.
2421 The flag PAGE_WRITE_ORG is positioned automatically depending
2422 on PAGE_WRITE. The mmap_lock should already be held. */
2423 void page_set_flags(target_ulong start, target_ulong end, int flags)
2424 {
2425 target_ulong addr, len;
2426
2427 /* This function should never be called with addresses outside the
2428 guest address space. If this assert fires, it probably indicates
2429 a missing call to h2g_valid. */
2430 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
2431 assert(end < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
2432 #endif
2433 assert(start < end);
2434
2435 start = start & TARGET_PAGE_MASK;
2436 end = TARGET_PAGE_ALIGN(end);
2437
2438 if (flags & PAGE_WRITE) {
2439 flags |= PAGE_WRITE_ORG;
2440 }
2441
2442 for (addr = start, len = end - start;
2443 len != 0;
2444 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
2445 PageDesc *p = page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
2446
2447 /* If the write protection bit is set, then we invalidate
2448 the code inside. */
2449 if (!(p->flags & PAGE_WRITE) &&
2450 (flags & PAGE_WRITE) &&
2451 p->first_tb) {
2452 tb_invalidate_phys_page(addr, 0, NULL);
2453 }
2454 p->flags = flags;
2455 }
2456 }
2457
2458 int page_check_range(target_ulong start, target_ulong len, int flags)
2459 {
2460 PageDesc *p;
2461 target_ulong end;
2462 target_ulong addr;
2463
2464 /* This function should never be called with addresses outside the
2465 guest address space. If this assert fires, it probably indicates
2466 a missing call to h2g_valid. */
2467 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
2468 assert(start < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
2469 #endif
2470
2471 if (len == 0) {
2472 return 0;
2473 }
2474 if (start + len - 1 < start) {
2475 /* We've wrapped around. */
2476 return -1;
2477 }
2478
2479 end = TARGET_PAGE_ALIGN(start+len); /* must do before we loose bits in the next step */
2480 start = start & TARGET_PAGE_MASK;
2481
2482 for (addr = start, len = end - start;
2483 len != 0;
2484 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
2485 p = page_find(addr >> TARGET_PAGE_BITS);
2486 if( !p )
2487 return -1;
2488 if( !(p->flags & PAGE_VALID) )
2489 return -1;
2490
2491 if ((flags & PAGE_READ) && !(p->flags & PAGE_READ))
2492 return -1;
2493 if (flags & PAGE_WRITE) {
2494 if (!(p->flags & PAGE_WRITE_ORG))
2495 return -1;
2496 /* unprotect the page if it was put read-only because it
2497 contains translated code */
2498 if (!(p->flags & PAGE_WRITE)) {
2499 if (!page_unprotect(addr, 0, NULL))
2500 return -1;
2501 }
2502 return 0;
2503 }
2504 }
2505 return 0;
2506 }
2507
2508 /* called from signal handler: invalidate the code and unprotect the
2509 page. Return TRUE if the fault was successfully handled. */
2510 int page_unprotect(target_ulong address, unsigned long pc, void *puc)
2511 {
2512 unsigned int prot;
2513 PageDesc *p;
2514 target_ulong host_start, host_end, addr;
2515
2516 /* Technically this isn't safe inside a signal handler. However we
2517 know this only ever happens in a synchronous SEGV handler, so in
2518 practice it seems to be ok. */
2519 mmap_lock();
2520
2521 p = page_find(address >> TARGET_PAGE_BITS);
2522 if (!p) {
2523 mmap_unlock();
2524 return 0;
2525 }
2526
2527 /* if the page was really writable, then we change its
2528 protection back to writable */
2529 if ((p->flags & PAGE_WRITE_ORG) && !(p->flags & PAGE_WRITE)) {
2530 host_start = address & qemu_host_page_mask;
2531 host_end = host_start + qemu_host_page_size;
2532
2533 prot = 0;
2534 for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) {
2535 p = page_find(addr >> TARGET_PAGE_BITS);
2536 p->flags |= PAGE_WRITE;
2537 prot |= p->flags;
2538
2539 /* and since the content will be modified, we must invalidate
2540 the corresponding translated code. */
2541 tb_invalidate_phys_page(addr, pc, puc);
2542 #ifdef DEBUG_TB_CHECK
2543 tb_invalidate_check(addr);
2544 #endif
2545 }
2546 mprotect((void *)g2h(host_start), qemu_host_page_size,
2547 prot & PAGE_BITS);
2548
2549 mmap_unlock();
2550 return 1;
2551 }
2552 mmap_unlock();
2553 return 0;
2554 }
2555
2556 static inline void tlb_set_dirty(CPUState *env,
2557 unsigned long addr, target_ulong vaddr)
2558 {
2559 }
2560 #endif /* defined(CONFIG_USER_ONLY) */
2561
2562 #if !defined(CONFIG_USER_ONLY)
2563
2564 #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
2565 typedef struct subpage_t {
2566 target_phys_addr_t base;
2567 ram_addr_t sub_io_index[TARGET_PAGE_SIZE];
2568 ram_addr_t region_offset[TARGET_PAGE_SIZE];
2569 } subpage_t;
2570
2571 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
2572 ram_addr_t memory, ram_addr_t region_offset);
2573 static subpage_t *subpage_init (target_phys_addr_t base, ram_addr_t *phys,
2574 ram_addr_t orig_memory,
2575 ram_addr_t region_offset);
2576 #define CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr, end_addr2, \
2577 need_subpage) \
2578 do { \
2579 if (addr > start_addr) \
2580 start_addr2 = 0; \
2581 else { \
2582 start_addr2 = start_addr & ~TARGET_PAGE_MASK; \
2583 if (start_addr2 > 0) \
2584 need_subpage = 1; \
2585 } \
2586 \
2587 if ((start_addr + orig_size) - addr >= TARGET_PAGE_SIZE) \
2588 end_addr2 = TARGET_PAGE_SIZE - 1; \
2589 else { \
2590 end_addr2 = (start_addr + orig_size - 1) & ~TARGET_PAGE_MASK; \
2591 if (end_addr2 < TARGET_PAGE_SIZE - 1) \
2592 need_subpage = 1; \
2593 } \
2594 } while (0)
2595
2596 /* register physical memory.
2597 For RAM, 'size' must be a multiple of the target page size.
2598 If (phys_offset & ~TARGET_PAGE_MASK) != 0, then it is an
2599 io memory page. The address used when calling the IO function is
2600 the offset from the start of the region, plus region_offset. Both
2601 start_addr and region_offset are rounded down to a page boundary
2602 before calculating this offset. This should not be a problem unless
2603 the low bits of start_addr and region_offset differ. */
2604 void cpu_register_physical_memory_log(target_phys_addr_t start_addr,
2605 ram_addr_t size,
2606 ram_addr_t phys_offset,
2607 ram_addr_t region_offset,
2608 bool log_dirty)
2609 {
2610 target_phys_addr_t addr, end_addr;
2611 PhysPageDesc *p;
2612 CPUState *env;
2613 ram_addr_t orig_size = size;
2614 subpage_t *subpage;
2615
2616 cpu_notify_set_memory(start_addr, size, phys_offset, log_dirty);
2617
2618 if (phys_offset == IO_MEM_UNASSIGNED) {
2619 region_offset = start_addr;
2620 }
2621 region_offset &= TARGET_PAGE_MASK;
2622 size = (size + TARGET_PAGE_SIZE - 1) & TARGET_PAGE_MASK;
2623 end_addr = start_addr + (target_phys_addr_t)size;
2624 for(addr = start_addr; addr != end_addr; addr += TARGET_PAGE_SIZE) {
2625 p = phys_page_find(addr >> TARGET_PAGE_BITS);
2626 if (p && p->phys_offset != IO_MEM_UNASSIGNED) {
2627 ram_addr_t orig_memory = p->phys_offset;
2628 target_phys_addr_t start_addr2, end_addr2;
2629 int need_subpage = 0;
2630
2631 CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr, end_addr2,
2632 need_subpage);
2633 if (need_subpage) {
2634 if (!(orig_memory & IO_MEM_SUBPAGE)) {
2635 subpage = subpage_init((addr & TARGET_PAGE_MASK),
2636 &p->phys_offset, orig_memory,
2637 p->region_offset);
2638 } else {
2639 subpage = io_mem_opaque[(orig_memory & ~TARGET_PAGE_MASK)
2640 >> IO_MEM_SHIFT];
2641 }
2642 subpage_register(subpage, start_addr2, end_addr2, phys_offset,
2643 region_offset);
2644 p->region_offset = 0;
2645 } else {
2646 p->phys_offset = phys_offset;
2647 if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM ||
2648 (phys_offset & IO_MEM_ROMD))
2649 phys_offset += TARGET_PAGE_SIZE;
2650 }
2651 } else {
2652 p = phys_page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
2653 p->phys_offset = phys_offset;
2654 p->region_offset = region_offset;
2655 if ((phys_offset & ~TARGET_PAGE_MASK) <= IO_MEM_ROM ||
2656 (phys_offset & IO_MEM_ROMD)) {
2657 phys_offset += TARGET_PAGE_SIZE;
2658 } else {
2659 target_phys_addr_t start_addr2, end_addr2;
2660 int need_subpage = 0;
2661
2662 CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr,
2663 end_addr2, need_subpage);
2664
2665 if (need_subpage) {
2666 subpage = subpage_init((addr & TARGET_PAGE_MASK),
2667 &p->phys_offset, IO_MEM_UNASSIGNED,
2668 addr & TARGET_PAGE_MASK);
2669 subpage_register(subpage, start_addr2, end_addr2,
2670 phys_offset, region_offset);
2671 p->region_offset = 0;
2672 }
2673 }
2674 }
2675 region_offset += TARGET_PAGE_SIZE;
2676 }
2677
2678 /* since each CPU stores ram addresses in its TLB cache, we must
2679 reset the modified entries */
2680 /* XXX: slow ! */
2681 for(env = first_cpu; env != NULL; env = env->next_cpu) {
2682 tlb_flush(env, 1);
2683 }
2684 }
2685
2686 /* XXX: temporary until new memory mapping API */
2687 ram_addr_t cpu_get_physical_page_desc(target_phys_addr_t addr)
2688 {
2689 PhysPageDesc *p;
2690
2691 p = phys_page_find(addr >> TARGET_PAGE_BITS);
2692 if (!p)
2693 return IO_MEM_UNASSIGNED;
2694 return p->phys_offset;
2695 }
2696
2697 void qemu_register_coalesced_mmio(target_phys_addr_t addr, ram_addr_t size)
2698 {
2699 if (kvm_enabled())
2700 kvm_coalesce_mmio_region(addr, size);
2701 }
2702
2703 void qemu_unregister_coalesced_mmio(target_phys_addr_t addr, ram_addr_t size)
2704 {
2705 if (kvm_enabled())
2706 kvm_uncoalesce_mmio_region(addr, size);
2707 }
2708
2709 void qemu_flush_coalesced_mmio_buffer(void)
2710 {
2711 if (kvm_enabled())
2712 kvm_flush_coalesced_mmio_buffer();
2713 }
2714
2715 #if defined(__linux__) && !defined(TARGET_S390X)
2716
2717 #include <sys/vfs.h>
2718
2719 #define HUGETLBFS_MAGIC 0x958458f6
2720
2721 static long gethugepagesize(const char *path)
2722 {
2723 struct statfs fs;
2724 int ret;
2725
2726 do {
2727 ret = statfs(path, &fs);
2728 } while (ret != 0 && errno == EINTR);
2729
2730 if (ret != 0) {
2731 perror(path);
2732 return 0;
2733 }
2734
2735 if (fs.f_type != HUGETLBFS_MAGIC)
2736 fprintf(stderr, "Warning: path not on HugeTLBFS: %s\n", path);
2737
2738 return fs.f_bsize;
2739 }
2740
2741 static void *file_ram_alloc(RAMBlock *block,
2742 ram_addr_t memory,
2743 const char *path)
2744 {
2745 char *filename;
2746 void *area;
2747 int fd;
2748 #ifdef MAP_POPULATE
2749 int flags;
2750 #endif
2751 unsigned long hpagesize;
2752
2753 hpagesize = gethugepagesize(path);
2754 if (!hpagesize) {
2755 return NULL;
2756 }
2757
2758 if (memory < hpagesize) {
2759 return NULL;
2760 }
2761
2762 if (kvm_enabled() && !kvm_has_sync_mmu()) {
2763 fprintf(stderr, "host lacks kvm mmu notifiers, -mem-path unsupported\n");
2764 return NULL;
2765 }
2766
2767 if (asprintf(&filename, "%s/qemu_back_mem.XXXXXX", path) == -1) {
2768 return NULL;
2769 }
2770
2771 fd = mkstemp(filename);
2772 if (fd < 0) {
2773 perror("unable to create backing store for hugepages");
2774 free(filename);
2775 return NULL;
2776 }
2777 unlink(filename);
2778 free(filename);
2779
2780 memory = (memory+hpagesize-1) & ~(hpagesize-1);
2781
2782 /*
2783 * ftruncate is not supported by hugetlbfs in older
2784 * hosts, so don't bother bailing out on errors.
2785 * If anything goes wrong with it under other filesystems,
2786 * mmap will fail.
2787 */
2788 if (ftruncate(fd, memory))
2789 perror("ftruncate");
2790
2791 #ifdef MAP_POPULATE
2792 /* NB: MAP_POPULATE won't exhaustively alloc all phys pages in the case
2793 * MAP_PRIVATE is requested. For mem_prealloc we mmap as MAP_SHARED
2794 * to sidestep this quirk.
2795 */
2796 flags = mem_prealloc ? MAP_POPULATE | MAP_SHARED : MAP_PRIVATE;
2797 area = mmap(0, memory, PROT_READ | PROT_WRITE, flags, fd, 0);
2798 #else
2799 area = mmap(0, memory, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
2800 #endif
2801 if (area == MAP_FAILED) {
2802 perror("file_ram_alloc: can't mmap RAM pages");
2803 close(fd);
2804 return (NULL);
2805 }
2806 block->fd = fd;
2807 return area;
2808 }
2809 #endif
2810
2811 static ram_addr_t find_ram_offset(ram_addr_t size)
2812 {
2813 RAMBlock *block, *next_block;
2814 ram_addr_t offset = 0, mingap = ULONG_MAX;
2815
2816 if (QLIST_EMPTY(&ram_list.blocks))
2817 return 0;
2818
2819 QLIST_FOREACH(block, &ram_list.blocks, next) {
2820 ram_addr_t end, next = ULONG_MAX;
2821
2822 end = block->offset + block->length;
2823
2824 QLIST_FOREACH(next_block, &ram_list.blocks, next) {
2825 if (next_block->offset >= end) {
2826 next = MIN(next, next_block->offset);
2827 }
2828 }
2829 if (next - end >= size && next - end < mingap) {
2830 offset = end;
2831 mingap = next - end;
2832 }
2833 }
2834 return offset;
2835 }
2836
2837 static ram_addr_t last_ram_offset(void)
2838 {
2839 RAMBlock *block;
2840 ram_addr_t last = 0;
2841
2842 QLIST_FOREACH(block, &ram_list.blocks, next)
2843 last = MAX(last, block->offset + block->length);
2844
2845 return last;
2846 }
2847
2848 ram_addr_t qemu_ram_alloc_from_ptr(DeviceState *dev, const char *name,
2849 ram_addr_t size, void *host)
2850 {
2851 RAMBlock *new_block, *block;
2852
2853 size = TARGET_PAGE_ALIGN(size);
2854 new_block = qemu_mallocz(sizeof(*new_block));
2855
2856 if (dev && dev->parent_bus && dev->parent_bus->info->get_dev_path) {
2857 char *id = dev->parent_bus->info->get_dev_path(dev);
2858 if (id) {
2859 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
2860 qemu_free(id);
2861 }
2862 }
2863 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
2864
2865 QLIST_FOREACH(block, &ram_list.blocks, next) {
2866 if (!strcmp(block->idstr, new_block->idstr)) {
2867 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
2868 new_block->idstr);
2869 abort();
2870 }
2871 }
2872
2873 if (host) {
2874 new_block->host = host;
2875 new_block->flags |= RAM_PREALLOC_MASK;
2876 } else {
2877 if (mem_path) {
2878 #if defined (__linux__) && !defined(TARGET_S390X)
2879 new_block->host = file_ram_alloc(new_block, size, mem_path);
2880 if (!new_block->host) {
2881 new_block->host = qemu_vmalloc(size);
2882 qemu_madvise(new_block->host, size, QEMU_MADV_MERGEABLE);
2883 }
2884 #else
2885 fprintf(stderr, "-mem-path option unsupported\n");
2886 exit(1);
2887 #endif
2888 } else {
2889 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2890 /* XXX S390 KVM requires the topmost vma of the RAM to be < 256GB */
2891 new_block->host = mmap((void*)0x1000000, size,
2892 PROT_EXEC|PROT_READ|PROT_WRITE,
2893 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
2894 #else
2895 new_block->host = qemu_vmalloc(size);
2896 #endif
2897 qemu_madvise(new_block->host, size, QEMU_MADV_MERGEABLE);
2898 }
2899 }
2900
2901 new_block->offset = find_ram_offset(size);
2902 new_block->length = size;
2903
2904 QLIST_INSERT_HEAD(&ram_list.blocks, new_block, next);
2905
2906 ram_list.phys_dirty = qemu_realloc(ram_list.phys_dirty,
2907 last_ram_offset() >> TARGET_PAGE_BITS);
2908 memset(ram_list.phys_dirty + (new_block->offset >> TARGET_PAGE_BITS),
2909 0xff, size >> TARGET_PAGE_BITS);
2910
2911 if (kvm_enabled())
2912 kvm_setup_guest_memory(new_block->host, size);
2913
2914 return new_block->offset;
2915 }
2916
2917 ram_addr_t qemu_ram_alloc(DeviceState *dev, const char *name, ram_addr_t size)
2918 {
2919 return qemu_ram_alloc_from_ptr(dev, name, size, NULL);
2920 }
2921
2922 void qemu_ram_free(ram_addr_t addr)
2923 {
2924 RAMBlock *block;
2925
2926 QLIST_FOREACH(block, &ram_list.blocks, next) {
2927 if (addr == block->offset) {
2928 QLIST_REMOVE(block, next);
2929 if (block->flags & RAM_PREALLOC_MASK) {
2930 ;
2931 } else if (mem_path) {
2932 #if defined (__linux__) && !defined(TARGET_S390X)
2933 if (block->fd) {
2934 munmap(block->host, block->length);
2935 close(block->fd);
2936 } else {
2937 qemu_vfree(block->host);
2938 }
2939 #else
2940 abort();
2941 #endif
2942 } else {
2943 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2944 munmap(block->host, block->length);
2945 #else
2946 qemu_vfree(block->host);
2947 #endif
2948 }
2949 qemu_free(block);
2950 return;
2951 }
2952 }
2953
2954 }
2955
2956 #ifndef _WIN32
2957 void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
2958 {
2959 RAMBlock *block;
2960 ram_addr_t offset;
2961 int flags;
2962 void *area, *vaddr;
2963
2964 QLIST_FOREACH(block, &ram_list.blocks, next) {
2965 offset = addr - block->offset;
2966 if (offset < block->length) {
2967 vaddr = block->host + offset;
2968 if (block->flags & RAM_PREALLOC_MASK) {
2969 ;
2970 } else {
2971 flags = MAP_FIXED;
2972 munmap(vaddr, length);
2973 if (mem_path) {
2974 #if defined(__linux__) && !defined(TARGET_S390X)
2975 if (block->fd) {
2976 #ifdef MAP_POPULATE
2977 flags |= mem_prealloc ? MAP_POPULATE | MAP_SHARED :
2978 MAP_PRIVATE;
2979 #else
2980 flags |= MAP_PRIVATE;
2981 #endif
2982 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2983 flags, block->fd, offset);
2984 } else {
2985 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2986 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2987 flags, -1, 0);
2988 }
2989 #else
2990 abort();
2991 #endif
2992 } else {
2993 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2994 flags |= MAP_SHARED | MAP_ANONYMOUS;
2995 area = mmap(vaddr, length, PROT_EXEC|PROT_READ|PROT_WRITE,
2996 flags, -1, 0);
2997 #else
2998 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2999 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
3000 flags, -1, 0);
3001 #endif
3002 }
3003 if (area != vaddr) {
3004 fprintf(stderr, "Could not remap addr: %lx@%lx\n",
3005 length, addr);
3006 exit(1);
3007 }
3008 qemu_madvise(vaddr, length, QEMU_MADV_MERGEABLE);
3009 }
3010 return;
3011 }
3012 }
3013 }
3014 #endif /* !_WIN32 */
3015
3016 /* Return a host pointer to ram allocated with qemu_ram_alloc.
3017 With the exception of the softmmu code in this file, this should
3018 only be used for local memory (e.g. video ram) that the device owns,
3019 and knows it isn't going to access beyond the end of the block.
3020
3021 It should not be used for general purpose DMA.
3022 Use cpu_physical_memory_map/cpu_physical_memory_rw instead.
3023 */
3024 void *qemu_get_ram_ptr(ram_addr_t addr)
3025 {
3026 RAMBlock *block;
3027
3028 QLIST_FOREACH(block, &ram_list.blocks, next) {
3029 if (addr - block->offset < block->length) {
3030 /* Move this entry to to start of the list. */
3031 if (block != QLIST_FIRST(&ram_list.blocks)) {
3032 QLIST_REMOVE(block, next);
3033 QLIST_INSERT_HEAD(&ram_list.blocks, block, next);
3034 }
3035 return block->host + (addr - block->offset);
3036 }
3037 }
3038
3039 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
3040 abort();
3041
3042 return NULL;
3043 }
3044
3045 /* Return a host pointer to ram allocated with qemu_ram_alloc.
3046 * Same as qemu_get_ram_ptr but avoid reordering ramblocks.
3047 */
3048 void *qemu_safe_ram_ptr(ram_addr_t addr)
3049 {
3050 RAMBlock *block;
3051
3052 QLIST_FOREACH(block, &ram_list.blocks, next) {
3053 if (addr - block->offset < block->length) {
3054 return block->host + (addr - block->offset);
3055 }
3056 }
3057
3058 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
3059 abort();
3060
3061 return NULL;
3062 }
3063
3064 int qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
3065 {
3066 RAMBlock *block;
3067 uint8_t *host = ptr;
3068
3069 QLIST_FOREACH(block, &ram_list.blocks, next) {
3070 if (host - block->host < block->length) {
3071 *ram_addr = block->offset + (host - block->host);
3072 return 0;
3073 }
3074 }
3075 return -1;
3076 }
3077
3078 /* Some of the softmmu routines need to translate from a host pointer
3079 (typically a TLB entry) back to a ram offset. */
3080 ram_addr_t qemu_ram_addr_from_host_nofail(void *ptr)
3081 {
3082 ram_addr_t ram_addr;
3083
3084 if (qemu_ram_addr_from_host(ptr, &ram_addr)) {
3085 fprintf(stderr, "Bad ram pointer %p\n", ptr);
3086 abort();
3087 }
3088 return ram_addr;
3089 }
3090
3091 static uint32_t unassigned_mem_readb(void *opaque, target_phys_addr_t addr)
3092 {
3093 #ifdef DEBUG_UNASSIGNED
3094 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
3095 #endif
3096 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3097 do_unassigned_access(addr, 0, 0, 0, 1);
3098 #endif
3099 return 0;
3100 }
3101
3102 static uint32_t unassigned_mem_readw(void *opaque, target_phys_addr_t addr)
3103 {
3104 #ifdef DEBUG_UNASSIGNED
3105 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
3106 #endif
3107 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3108 do_unassigned_access(addr, 0, 0, 0, 2);
3109 #endif
3110 return 0;
3111 }
3112
3113 static uint32_t unassigned_mem_readl(void *opaque, target_phys_addr_t addr)
3114 {
3115 #ifdef DEBUG_UNASSIGNED
3116 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
3117 #endif
3118 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3119 do_unassigned_access(addr, 0, 0, 0, 4);
3120 #endif
3121 return 0;
3122 }
3123
3124 static void unassigned_mem_writeb(void *opaque, target_phys_addr_t addr, uint32_t val)
3125 {
3126 #ifdef DEBUG_UNASSIGNED
3127 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%x\n", addr, val);
3128 #endif
3129 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3130 do_unassigned_access(addr, 1, 0, 0, 1);
3131 #endif
3132 }
3133
3134 static void unassigned_mem_writew(void *opaque, target_phys_addr_t addr, uint32_t val)
3135 {
3136 #ifdef DEBUG_UNASSIGNED
3137 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%x\n", addr, val);
3138 #endif
3139 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3140 do_unassigned_access(addr, 1, 0, 0, 2);
3141 #endif
3142 }
3143
3144 static void unassigned_mem_writel(void *opaque, target_phys_addr_t addr, uint32_t val)
3145 {
3146 #ifdef DEBUG_UNASSIGNED
3147 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%x\n", addr, val);
3148 #endif
3149 #if defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3150 do_unassigned_access(addr, 1, 0, 0, 4);
3151 #endif
3152 }
3153
3154 static CPUReadMemoryFunc * const unassigned_mem_read[3] = {
3155 unassigned_mem_readb,
3156 unassigned_mem_readw,
3157 unassigned_mem_readl,
3158 };
3159
3160 static CPUWriteMemoryFunc * const unassigned_mem_write[3] = {
3161 unassigned_mem_writeb,
3162 unassigned_mem_writew,
3163 unassigned_mem_writel,
3164 };
3165
3166 static void notdirty_mem_writeb(void *opaque, target_phys_addr_t ram_addr,
3167 uint32_t val)
3168 {
3169 int dirty_flags;
3170 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3171 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
3172 #if !defined(CONFIG_USER_ONLY)
3173 tb_invalidate_phys_page_fast(ram_addr, 1);
3174 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3175 #endif
3176 }
3177 stb_p(qemu_get_ram_ptr(ram_addr), val);
3178 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
3179 cpu_physical_memory_set_dirty_flags(ram_addr, dirty_flags);
3180 /* we remove the notdirty callback only if the code has been
3181 flushed */
3182 if (dirty_flags == 0xff)
3183 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3184 }
3185
3186 static void notdirty_mem_writew(void *opaque, target_phys_addr_t ram_addr,
3187 uint32_t val)
3188 {
3189 int dirty_flags;
3190 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3191 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
3192 #if !defined(CONFIG_USER_ONLY)
3193 tb_invalidate_phys_page_fast(ram_addr, 2);
3194 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3195 #endif
3196 }
3197 stw_p(qemu_get_ram_ptr(ram_addr), val);
3198 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
3199 cpu_physical_memory_set_dirty_flags(ram_addr, dirty_flags);
3200 /* we remove the notdirty callback only if the code has been
3201 flushed */
3202 if (dirty_flags == 0xff)
3203 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3204 }
3205
3206 static void notdirty_mem_writel(void *opaque, target_phys_addr_t ram_addr,
3207 uint32_t val)
3208 {
3209 int dirty_flags;
3210 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3211 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
3212 #if !defined(CONFIG_USER_ONLY)
3213 tb_invalidate_phys_page_fast(ram_addr, 4);
3214 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3215 #endif
3216 }
3217 stl_p(qemu_get_ram_ptr(ram_addr), val);
3218 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
3219 cpu_physical_memory_set_dirty_flags(ram_addr, dirty_flags);
3220 /* we remove the notdirty callback only if the code has been
3221 flushed */
3222 if (dirty_flags == 0xff)
3223 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3224 }
3225
3226 static CPUReadMemoryFunc * const error_mem_read[3] = {
3227 NULL, /* never used */
3228 NULL, /* never used */
3229 NULL, /* never used */
3230 };
3231
3232 static CPUWriteMemoryFunc * const notdirty_mem_write[3] = {
3233 notdirty_mem_writeb,
3234 notdirty_mem_writew,
3235 notdirty_mem_writel,
3236 };
3237
3238 /* Generate a debug exception if a watchpoint has been hit. */
3239 static void check_watchpoint(int offset, int len_mask, int flags)
3240 {
3241 CPUState *env = cpu_single_env;
3242 target_ulong pc, cs_base;
3243 TranslationBlock *tb;
3244 target_ulong vaddr;
3245 CPUWatchpoint *wp;
3246 int cpu_flags;
3247
3248 if (env->watchpoint_hit) {
3249 /* We re-entered the check after replacing the TB. Now raise
3250 * the debug interrupt so that is will trigger after the
3251 * current instruction. */
3252 cpu_interrupt(env, CPU_INTERRUPT_DEBUG);
3253 return;
3254 }
3255 vaddr = (env->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
3256 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
3257 if ((vaddr == (wp->vaddr & len_mask) ||
3258 (vaddr & wp->len_mask) == wp->vaddr) && (wp->flags & flags)) {
3259 wp->flags |= BP_WATCHPOINT_HIT;
3260 if (!env->watchpoint_hit) {
3261 env->watchpoint_hit = wp;
3262 tb = tb_find_pc(env->mem_io_pc);
3263 if (!tb) {
3264 cpu_abort(env, "check_watchpoint: could not find TB for "
3265 "pc=%p", (void *)env->mem_io_pc);
3266 }
3267 cpu_restore_state(tb, env, env->mem_io_pc, NULL);
3268 tb_phys_invalidate(tb, -1);
3269 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
3270 env->exception_index = EXCP_DEBUG;
3271 } else {
3272 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
3273 tb_gen_code(env, pc, cs_base, cpu_flags, 1);
3274 }
3275 cpu_resume_from_signal(env, NULL);
3276 }
3277 } else {
3278 wp->flags &= ~BP_WATCHPOINT_HIT;
3279 }
3280 }
3281 }
3282
3283 /* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
3284 so these check for a hit then pass through to the normal out-of-line
3285 phys routines. */
3286 static uint32_t watch_mem_readb(void *opaque, target_phys_addr_t addr)
3287 {
3288 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x0, BP_MEM_READ);
3289 return ldub_phys(addr);
3290 }
3291
3292 static uint32_t watch_mem_readw(void *opaque, target_phys_addr_t addr)
3293 {
3294 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x1, BP_MEM_READ);
3295 return lduw_phys(addr);
3296 }
3297
3298 static uint32_t watch_mem_readl(void *opaque, target_phys_addr_t addr)
3299 {
3300 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x3, BP_MEM_READ);
3301 return ldl_phys(addr);
3302 }
3303
3304 static void watch_mem_writeb(void *opaque, target_phys_addr_t addr,
3305 uint32_t val)
3306 {
3307 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x0, BP_MEM_WRITE);
3308 stb_phys(addr, val);
3309 }
3310
3311 static void watch_mem_writew(void *opaque, target_phys_addr_t addr,
3312 uint32_t val)
3313 {
3314 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x1, BP_MEM_WRITE);
3315 stw_phys(addr, val);
3316 }
3317
3318 static void watch_mem_writel(void *opaque, target_phys_addr_t addr,
3319 uint32_t val)
3320 {
3321 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x3, BP_MEM_WRITE);
3322 stl_phys(addr, val);
3323 }
3324
3325 static CPUReadMemoryFunc * const watch_mem_read[3] = {
3326 watch_mem_readb,
3327 watch_mem_readw,
3328 watch_mem_readl,
3329 };
3330
3331 static CPUWriteMemoryFunc * const watch_mem_write[3] = {
3332 watch_mem_writeb,
3333 watch_mem_writew,
3334 watch_mem_writel,
3335 };
3336
3337 static inline uint32_t subpage_readlen (subpage_t *mmio,
3338 target_phys_addr_t addr,
3339 unsigned int len)
3340 {
3341 unsigned int idx = SUBPAGE_IDX(addr);
3342 #if defined(DEBUG_SUBPAGE)
3343 printf("%s: subpage %p len %d addr " TARGET_FMT_plx " idx %d\n", __func__,
3344 mmio, len, addr, idx);
3345 #endif
3346
3347 addr += mmio->region_offset[idx];
3348 idx = mmio->sub_io_index[idx];
3349 return io_mem_read[idx][len](io_mem_opaque[idx], addr);
3350 }
3351
3352 static inline void subpage_writelen (subpage_t *mmio, target_phys_addr_t addr,
3353 uint32_t value, unsigned int len)
3354 {
3355 unsigned int idx = SUBPAGE_IDX(addr);
3356 #if defined(DEBUG_SUBPAGE)
3357 printf("%s: subpage %p len %d addr " TARGET_FMT_plx " idx %d value %08x\n",
3358 __func__, mmio, len, addr, idx, value);
3359 #endif
3360
3361 addr += mmio->region_offset[idx];
3362 idx = mmio->sub_io_index[idx];
3363 io_mem_write[idx][len](io_mem_opaque[idx], addr, value);
3364 }
3365
3366 static uint32_t subpage_readb (void *opaque, target_phys_addr_t addr)
3367 {
3368 return subpage_readlen(opaque, addr, 0);
3369 }
3370
3371 static void subpage_writeb (void *opaque, target_phys_addr_t addr,
3372 uint32_t value)
3373 {
3374 subpage_writelen(opaque, addr, value, 0);
3375 }
3376
3377 static uint32_t subpage_readw (void *opaque, target_phys_addr_t addr)
3378 {
3379 return subpage_readlen(opaque, addr, 1);
3380 }
3381
3382 static void subpage_writew (void *opaque, target_phys_addr_t addr,
3383 uint32_t value)
3384 {
3385 subpage_writelen(opaque, addr, value, 1);
3386 }
3387
3388 static uint32_t subpage_readl (void *opaque, target_phys_addr_t addr)
3389 {
3390 return subpage_readlen(opaque, addr, 2);
3391 }
3392
3393 static void subpage_writel (void *opaque, target_phys_addr_t addr,
3394 uint32_t value)
3395 {
3396 subpage_writelen(opaque, addr, value, 2);
3397 }
3398
3399 static CPUReadMemoryFunc * const subpage_read[] = {
3400 &subpage_readb,
3401 &subpage_readw,
3402 &subpage_readl,
3403 };
3404
3405 static CPUWriteMemoryFunc * const subpage_write[] = {
3406 &subpage_writeb,
3407 &subpage_writew,
3408 &subpage_writel,
3409 };
3410
3411 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
3412 ram_addr_t memory, ram_addr_t region_offset)
3413 {
3414 int idx, eidx;
3415
3416 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
3417 return -1;
3418 idx = SUBPAGE_IDX(start);
3419 eidx = SUBPAGE_IDX(end);
3420 #if defined(DEBUG_SUBPAGE)
3421 printf("%s: %p start %08x end %08x idx %08x eidx %08x mem %ld\n", __func__,
3422 mmio, start, end, idx, eidx, memory);
3423 #endif
3424 if ((memory & ~TARGET_PAGE_MASK) == IO_MEM_RAM)
3425 memory = IO_MEM_UNASSIGNED;
3426 memory = (memory >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3427 for (; idx <= eidx; idx++) {
3428 mmio->sub_io_index[idx] = memory;
3429 mmio->region_offset[idx] = region_offset;
3430 }
3431
3432 return 0;
3433 }
3434
3435 static subpage_t *subpage_init (target_phys_addr_t base, ram_addr_t *phys,
3436 ram_addr_t orig_memory,
3437 ram_addr_t region_offset)
3438 {
3439 subpage_t *mmio;
3440 int subpage_memory;
3441
3442 mmio = qemu_mallocz(sizeof(subpage_t));
3443
3444 mmio->base = base;
3445 subpage_memory = cpu_register_io_memory(subpage_read, subpage_write, mmio,
3446 DEVICE_NATIVE_ENDIAN);
3447 #if defined(DEBUG_SUBPAGE)
3448 printf("%s: %p base " TARGET_FMT_plx " len %08x %d\n", __func__,
3449 mmio, base, TARGET_PAGE_SIZE, subpage_memory);
3450 #endif
3451 *phys = subpage_memory | IO_MEM_SUBPAGE;
3452 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, orig_memory, region_offset);
3453
3454 return mmio;
3455 }
3456
3457 static int get_free_io_mem_idx(void)
3458 {
3459 int i;
3460
3461 for (i = 0; i<IO_MEM_NB_ENTRIES; i++)
3462 if (!io_mem_used[i]) {
3463 io_mem_used[i] = 1;
3464 return i;
3465 }
3466 fprintf(stderr, "RAN out out io_mem_idx, max %d !\n", IO_MEM_NB_ENTRIES);
3467 return -1;
3468 }
3469
3470 /*
3471 * Usually, devices operate in little endian mode. There are devices out
3472 * there that operate in big endian too. Each device gets byte swapped
3473 * mmio if plugged onto a CPU that does the other endianness.
3474 *
3475 * CPU Device swap?
3476 *
3477 * little little no
3478 * little big yes
3479 * big little yes
3480 * big big no
3481 */
3482
3483 typedef struct SwapEndianContainer {
3484 CPUReadMemoryFunc *read[3];
3485 CPUWriteMemoryFunc *write[3];
3486 void *opaque;
3487 } SwapEndianContainer;
3488
3489 static uint32_t swapendian_mem_readb (void *opaque, target_phys_addr_t addr)
3490 {
3491 uint32_t val;
3492 SwapEndianContainer *c = opaque;
3493 val = c->read[0](c->opaque, addr);
3494 return val;
3495 }
3496
3497 static uint32_t swapendian_mem_readw(void *opaque, target_phys_addr_t addr)
3498 {
3499 uint32_t val;
3500 SwapEndianContainer *c = opaque;
3501 val = bswap16(c->read[1](c->opaque, addr));
3502 return val;
3503 }
3504
3505 static uint32_t swapendian_mem_readl(void *opaque, target_phys_addr_t addr)
3506 {
3507 uint32_t val;
3508 SwapEndianContainer *c = opaque;
3509 val = bswap32(c->read[2](c->opaque, addr));
3510 return val;
3511 }
3512
3513 static CPUReadMemoryFunc * const swapendian_readfn[3]={
3514 swapendian_mem_readb,
3515 swapendian_mem_readw,
3516 swapendian_mem_readl
3517 };
3518
3519 static void swapendian_mem_writeb(void *opaque, target_phys_addr_t addr,
3520 uint32_t val)
3521 {
3522 SwapEndianContainer *c = opaque;
3523 c->write[0](c->opaque, addr, val);
3524 }
3525
3526 static void swapendian_mem_writew(void *opaque, target_phys_addr_t addr,
3527 uint32_t val)
3528 {
3529 SwapEndianContainer *c = opaque;
3530 c->write[1](c->opaque, addr, bswap16(val));
3531 }
3532
3533 static void swapendian_mem_writel(void *opaque, target_phys_addr_t addr,
3534 uint32_t val)
3535 {
3536 SwapEndianContainer *c = opaque;
3537 c->write[2](c->opaque, addr, bswap32(val));
3538 }
3539
3540 static CPUWriteMemoryFunc * const swapendian_writefn[3]={
3541 swapendian_mem_writeb,
3542 swapendian_mem_writew,
3543 swapendian_mem_writel
3544 };
3545
3546 static void swapendian_init(int io_index)
3547 {
3548 SwapEndianContainer *c = qemu_malloc(sizeof(SwapEndianContainer));
3549 int i;
3550
3551 /* Swap mmio for big endian targets */
3552 c->opaque = io_mem_opaque[io_index];
3553 for (i = 0; i < 3; i++) {
3554 c