]> git.proxmox.com Git - mirror_qemu.git/blob - exec.c
Convert the subpage wrapper to be a MemoryRegion
[mirror_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 "tcg.h"
30 #include "hw/hw.h"
31 #include "hw/qdev.h"
32 #include "osdep.h"
33 #include "kvm.h"
34 #include "hw/xen.h"
35 #include "qemu-timer.h"
36 #include "memory.h"
37 #include "exec-memory.h"
38 #if defined(CONFIG_USER_ONLY)
39 #include <qemu.h>
40 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
41 #include <sys/param.h>
42 #if __FreeBSD_version >= 700104
43 #define HAVE_KINFO_GETVMMAP
44 #define sigqueue sigqueue_freebsd /* avoid redefinition */
45 #include <sys/time.h>
46 #include <sys/proc.h>
47 #include <machine/profile.h>
48 #define _KERNEL
49 #include <sys/user.h>
50 #undef _KERNEL
51 #undef sigqueue
52 #include <libutil.h>
53 #endif
54 #endif
55 #else /* !CONFIG_USER_ONLY */
56 #include "xen-mapcache.h"
57 #include "trace.h"
58 #endif
59
60 #define WANT_EXEC_OBSOLETE
61 #include "exec-obsolete.h"
62
63 //#define DEBUG_TB_INVALIDATE
64 //#define DEBUG_FLUSH
65 //#define DEBUG_TLB
66 //#define DEBUG_UNASSIGNED
67
68 /* make various TB consistency checks */
69 //#define DEBUG_TB_CHECK
70 //#define DEBUG_TLB_CHECK
71
72 //#define DEBUG_IOPORT
73 //#define DEBUG_SUBPAGE
74
75 #if !defined(CONFIG_USER_ONLY)
76 /* TB consistency checks only implemented for usermode emulation. */
77 #undef DEBUG_TB_CHECK
78 #endif
79
80 #define SMC_BITMAP_USE_THRESHOLD 10
81
82 static TranslationBlock *tbs;
83 static int code_gen_max_blocks;
84 TranslationBlock *tb_phys_hash[CODE_GEN_PHYS_HASH_SIZE];
85 static int nb_tbs;
86 /* any access to the tbs or the page table must use this lock */
87 spinlock_t tb_lock = SPIN_LOCK_UNLOCKED;
88
89 #if defined(__arm__) || defined(__sparc_v9__)
90 /* The prologue must be reachable with a direct jump. ARM and Sparc64
91 have limited branch ranges (possibly also PPC) so place it in a
92 section close to code segment. */
93 #define code_gen_section \
94 __attribute__((__section__(".gen_code"))) \
95 __attribute__((aligned (32)))
96 #elif defined(_WIN32)
97 /* Maximum alignment for Win32 is 16. */
98 #define code_gen_section \
99 __attribute__((aligned (16)))
100 #else
101 #define code_gen_section \
102 __attribute__((aligned (32)))
103 #endif
104
105 uint8_t code_gen_prologue[1024] code_gen_section;
106 static uint8_t *code_gen_buffer;
107 static unsigned long code_gen_buffer_size;
108 /* threshold to flush the translated code buffer */
109 static unsigned long code_gen_buffer_max_size;
110 static uint8_t *code_gen_ptr;
111
112 #if !defined(CONFIG_USER_ONLY)
113 int phys_ram_fd;
114 static int in_migration;
115
116 RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) };
117
118 static MemoryRegion *system_memory;
119 static MemoryRegion *system_io;
120
121 MemoryRegion io_mem_ram, io_mem_rom, io_mem_unassigned, io_mem_notdirty;
122
123 #endif
124
125 CPUState *first_cpu;
126 /* current CPU in the current thread. It is only valid inside
127 cpu_exec() */
128 DEFINE_TLS(CPUState *,cpu_single_env);
129 /* 0 = Do not count executed instructions.
130 1 = Precise instruction counting.
131 2 = Adaptive rate instruction counting. */
132 int use_icount = 0;
133
134 typedef struct PageDesc {
135 /* list of TBs intersecting this ram page */
136 TranslationBlock *first_tb;
137 /* in order to optimize self modifying code, we count the number
138 of lookups we do to a given page to use a bitmap */
139 unsigned int code_write_count;
140 uint8_t *code_bitmap;
141 #if defined(CONFIG_USER_ONLY)
142 unsigned long flags;
143 #endif
144 } PageDesc;
145
146 /* In system mode we want L1_MAP to be based on ram offsets,
147 while in user mode we want it to be based on virtual addresses. */
148 #if !defined(CONFIG_USER_ONLY)
149 #if HOST_LONG_BITS < TARGET_PHYS_ADDR_SPACE_BITS
150 # define L1_MAP_ADDR_SPACE_BITS HOST_LONG_BITS
151 #else
152 # define L1_MAP_ADDR_SPACE_BITS TARGET_PHYS_ADDR_SPACE_BITS
153 #endif
154 #else
155 # define L1_MAP_ADDR_SPACE_BITS TARGET_VIRT_ADDR_SPACE_BITS
156 #endif
157
158 /* Size of the L2 (and L3, etc) page tables. */
159 #define L2_BITS 10
160 #define L2_SIZE (1 << L2_BITS)
161
162 /* The bits remaining after N lower levels of page tables. */
163 #define P_L1_BITS_REM \
164 ((TARGET_PHYS_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % L2_BITS)
165 #define V_L1_BITS_REM \
166 ((L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % L2_BITS)
167
168 /* Size of the L1 page table. Avoid silly small sizes. */
169 #if P_L1_BITS_REM < 4
170 #define P_L1_BITS (P_L1_BITS_REM + L2_BITS)
171 #else
172 #define P_L1_BITS P_L1_BITS_REM
173 #endif
174
175 #if V_L1_BITS_REM < 4
176 #define V_L1_BITS (V_L1_BITS_REM + L2_BITS)
177 #else
178 #define V_L1_BITS V_L1_BITS_REM
179 #endif
180
181 #define P_L1_SIZE ((target_phys_addr_t)1 << P_L1_BITS)
182 #define V_L1_SIZE ((target_ulong)1 << V_L1_BITS)
183
184 #define P_L1_SHIFT (TARGET_PHYS_ADDR_SPACE_BITS - TARGET_PAGE_BITS - P_L1_BITS)
185 #define V_L1_SHIFT (L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS - V_L1_BITS)
186
187 unsigned long qemu_real_host_page_size;
188 unsigned long qemu_host_page_size;
189 unsigned long qemu_host_page_mask;
190
191 /* This is a multi-level map on the virtual address space.
192 The bottom level has pointers to PageDesc. */
193 static void *l1_map[V_L1_SIZE];
194
195 #if !defined(CONFIG_USER_ONLY)
196 typedef struct PhysPageDesc {
197 /* offset in host memory of the page + io_index in the low bits */
198 ram_addr_t phys_offset;
199 ram_addr_t region_offset;
200 } PhysPageDesc;
201
202 /* This is a multi-level map on the physical address space.
203 The bottom level has pointers to PhysPageDesc. */
204 static void *l1_phys_map[P_L1_SIZE];
205
206 static void io_mem_init(void);
207 static void memory_map_init(void);
208
209 /* io memory support */
210 CPUWriteMemoryFunc *_io_mem_write[IO_MEM_NB_ENTRIES][4];
211 CPUReadMemoryFunc *_io_mem_read[IO_MEM_NB_ENTRIES][4];
212 void *io_mem_opaque[IO_MEM_NB_ENTRIES];
213 static char io_mem_used[IO_MEM_NB_ENTRIES];
214 static int io_mem_watch;
215 #endif
216
217 /* log support */
218 #ifdef WIN32
219 static const char *logfilename = "qemu.log";
220 #else
221 static const char *logfilename = "/tmp/qemu.log";
222 #endif
223 FILE *logfile;
224 int loglevel;
225 static int log_append = 0;
226
227 /* statistics */
228 #if !defined(CONFIG_USER_ONLY)
229 static int tlb_flush_count;
230 #endif
231 static int tb_flush_count;
232 static int tb_phys_invalidate_count;
233
234 #ifdef _WIN32
235 static void map_exec(void *addr, long size)
236 {
237 DWORD old_protect;
238 VirtualProtect(addr, size,
239 PAGE_EXECUTE_READWRITE, &old_protect);
240
241 }
242 #else
243 static void map_exec(void *addr, long size)
244 {
245 unsigned long start, end, page_size;
246
247 page_size = getpagesize();
248 start = (unsigned long)addr;
249 start &= ~(page_size - 1);
250
251 end = (unsigned long)addr + size;
252 end += page_size - 1;
253 end &= ~(page_size - 1);
254
255 mprotect((void *)start, end - start,
256 PROT_READ | PROT_WRITE | PROT_EXEC);
257 }
258 #endif
259
260 static void page_init(void)
261 {
262 /* NOTE: we can always suppose that qemu_host_page_size >=
263 TARGET_PAGE_SIZE */
264 #ifdef _WIN32
265 {
266 SYSTEM_INFO system_info;
267
268 GetSystemInfo(&system_info);
269 qemu_real_host_page_size = system_info.dwPageSize;
270 }
271 #else
272 qemu_real_host_page_size = getpagesize();
273 #endif
274 if (qemu_host_page_size == 0)
275 qemu_host_page_size = qemu_real_host_page_size;
276 if (qemu_host_page_size < TARGET_PAGE_SIZE)
277 qemu_host_page_size = TARGET_PAGE_SIZE;
278 qemu_host_page_mask = ~(qemu_host_page_size - 1);
279
280 #if defined(CONFIG_BSD) && defined(CONFIG_USER_ONLY)
281 {
282 #ifdef HAVE_KINFO_GETVMMAP
283 struct kinfo_vmentry *freep;
284 int i, cnt;
285
286 freep = kinfo_getvmmap(getpid(), &cnt);
287 if (freep) {
288 mmap_lock();
289 for (i = 0; i < cnt; i++) {
290 unsigned long startaddr, endaddr;
291
292 startaddr = freep[i].kve_start;
293 endaddr = freep[i].kve_end;
294 if (h2g_valid(startaddr)) {
295 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
296
297 if (h2g_valid(endaddr)) {
298 endaddr = h2g(endaddr);
299 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
300 } else {
301 #if TARGET_ABI_BITS <= L1_MAP_ADDR_SPACE_BITS
302 endaddr = ~0ul;
303 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
304 #endif
305 }
306 }
307 }
308 free(freep);
309 mmap_unlock();
310 }
311 #else
312 FILE *f;
313
314 last_brk = (unsigned long)sbrk(0);
315
316 f = fopen("/compat/linux/proc/self/maps", "r");
317 if (f) {
318 mmap_lock();
319
320 do {
321 unsigned long startaddr, endaddr;
322 int n;
323
324 n = fscanf (f, "%lx-%lx %*[^\n]\n", &startaddr, &endaddr);
325
326 if (n == 2 && h2g_valid(startaddr)) {
327 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
328
329 if (h2g_valid(endaddr)) {
330 endaddr = h2g(endaddr);
331 } else {
332 endaddr = ~0ul;
333 }
334 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
335 }
336 } while (!feof(f));
337
338 fclose(f);
339 mmap_unlock();
340 }
341 #endif
342 }
343 #endif
344 }
345
346 static PageDesc *page_find_alloc(tb_page_addr_t index, int alloc)
347 {
348 PageDesc *pd;
349 void **lp;
350 int i;
351
352 #if defined(CONFIG_USER_ONLY)
353 /* We can't use g_malloc because it may recurse into a locked mutex. */
354 # define ALLOC(P, SIZE) \
355 do { \
356 P = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, \
357 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); \
358 } while (0)
359 #else
360 # define ALLOC(P, SIZE) \
361 do { P = g_malloc0(SIZE); } while (0)
362 #endif
363
364 /* Level 1. Always allocated. */
365 lp = l1_map + ((index >> V_L1_SHIFT) & (V_L1_SIZE - 1));
366
367 /* Level 2..N-1. */
368 for (i = V_L1_SHIFT / L2_BITS - 1; i > 0; i--) {
369 void **p = *lp;
370
371 if (p == NULL) {
372 if (!alloc) {
373 return NULL;
374 }
375 ALLOC(p, sizeof(void *) * L2_SIZE);
376 *lp = p;
377 }
378
379 lp = p + ((index >> (i * L2_BITS)) & (L2_SIZE - 1));
380 }
381
382 pd = *lp;
383 if (pd == NULL) {
384 if (!alloc) {
385 return NULL;
386 }
387 ALLOC(pd, sizeof(PageDesc) * L2_SIZE);
388 *lp = pd;
389 }
390
391 #undef ALLOC
392
393 return pd + (index & (L2_SIZE - 1));
394 }
395
396 static inline PageDesc *page_find(tb_page_addr_t index)
397 {
398 return page_find_alloc(index, 0);
399 }
400
401 #if !defined(CONFIG_USER_ONLY)
402 static PhysPageDesc *phys_page_find_alloc(target_phys_addr_t index, int alloc)
403 {
404 PhysPageDesc *pd;
405 void **lp;
406 int i;
407
408 /* Level 1. Always allocated. */
409 lp = l1_phys_map + ((index >> P_L1_SHIFT) & (P_L1_SIZE - 1));
410
411 /* Level 2..N-1. */
412 for (i = P_L1_SHIFT / L2_BITS - 1; i > 0; i--) {
413 void **p = *lp;
414 if (p == NULL) {
415 if (!alloc) {
416 return NULL;
417 }
418 *lp = p = g_malloc0(sizeof(void *) * L2_SIZE);
419 }
420 lp = p + ((index >> (i * L2_BITS)) & (L2_SIZE - 1));
421 }
422
423 pd = *lp;
424 if (pd == NULL) {
425 int i;
426 int first_index = index & ~(L2_SIZE - 1);
427
428 if (!alloc) {
429 return NULL;
430 }
431
432 *lp = pd = g_malloc(sizeof(PhysPageDesc) * L2_SIZE);
433
434 for (i = 0; i < L2_SIZE; i++) {
435 pd[i].phys_offset = io_mem_unassigned.ram_addr;
436 pd[i].region_offset = (first_index + i) << TARGET_PAGE_BITS;
437 }
438 }
439
440 return pd + (index & (L2_SIZE - 1));
441 }
442
443 static inline PhysPageDesc phys_page_find(target_phys_addr_t index)
444 {
445 PhysPageDesc *p = phys_page_find_alloc(index, 0);
446
447 if (p) {
448 return *p;
449 } else {
450 return (PhysPageDesc) {
451 .phys_offset = io_mem_unassigned.ram_addr,
452 .region_offset = index << TARGET_PAGE_BITS,
453 };
454 }
455 }
456
457 static void tlb_protect_code(ram_addr_t ram_addr);
458 static void tlb_unprotect_code_phys(CPUState *env, ram_addr_t ram_addr,
459 target_ulong vaddr);
460 #define mmap_lock() do { } while(0)
461 #define mmap_unlock() do { } while(0)
462 #endif
463
464 #define DEFAULT_CODE_GEN_BUFFER_SIZE (32 * 1024 * 1024)
465
466 #if defined(CONFIG_USER_ONLY)
467 /* Currently it is not recommended to allocate big chunks of data in
468 user mode. It will change when a dedicated libc will be used */
469 #define USE_STATIC_CODE_GEN_BUFFER
470 #endif
471
472 #ifdef USE_STATIC_CODE_GEN_BUFFER
473 static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE]
474 __attribute__((aligned (CODE_GEN_ALIGN)));
475 #endif
476
477 static void code_gen_alloc(unsigned long tb_size)
478 {
479 #ifdef USE_STATIC_CODE_GEN_BUFFER
480 code_gen_buffer = static_code_gen_buffer;
481 code_gen_buffer_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
482 map_exec(code_gen_buffer, code_gen_buffer_size);
483 #else
484 code_gen_buffer_size = tb_size;
485 if (code_gen_buffer_size == 0) {
486 #if defined(CONFIG_USER_ONLY)
487 code_gen_buffer_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
488 #else
489 /* XXX: needs adjustments */
490 code_gen_buffer_size = (unsigned long)(ram_size / 4);
491 #endif
492 }
493 if (code_gen_buffer_size < MIN_CODE_GEN_BUFFER_SIZE)
494 code_gen_buffer_size = MIN_CODE_GEN_BUFFER_SIZE;
495 /* The code gen buffer location may have constraints depending on
496 the host cpu and OS */
497 #if defined(__linux__)
498 {
499 int flags;
500 void *start = NULL;
501
502 flags = MAP_PRIVATE | MAP_ANONYMOUS;
503 #if defined(__x86_64__)
504 flags |= MAP_32BIT;
505 /* Cannot map more than that */
506 if (code_gen_buffer_size > (800 * 1024 * 1024))
507 code_gen_buffer_size = (800 * 1024 * 1024);
508 #elif defined(__sparc_v9__)
509 // Map the buffer below 2G, so we can use direct calls and branches
510 flags |= MAP_FIXED;
511 start = (void *) 0x60000000UL;
512 if (code_gen_buffer_size > (512 * 1024 * 1024))
513 code_gen_buffer_size = (512 * 1024 * 1024);
514 #elif defined(__arm__)
515 /* Keep the buffer no bigger than 16GB to branch between blocks */
516 if (code_gen_buffer_size > 16 * 1024 * 1024)
517 code_gen_buffer_size = 16 * 1024 * 1024;
518 #elif defined(__s390x__)
519 /* Map the buffer so that we can use direct calls and branches. */
520 /* We have a +- 4GB range on the branches; leave some slop. */
521 if (code_gen_buffer_size > (3ul * 1024 * 1024 * 1024)) {
522 code_gen_buffer_size = 3ul * 1024 * 1024 * 1024;
523 }
524 start = (void *)0x90000000UL;
525 #endif
526 code_gen_buffer = mmap(start, code_gen_buffer_size,
527 PROT_WRITE | PROT_READ | PROT_EXEC,
528 flags, -1, 0);
529 if (code_gen_buffer == MAP_FAILED) {
530 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
531 exit(1);
532 }
533 }
534 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) \
535 || defined(__DragonFly__) || defined(__OpenBSD__) \
536 || defined(__NetBSD__)
537 {
538 int flags;
539 void *addr = NULL;
540 flags = MAP_PRIVATE | MAP_ANONYMOUS;
541 #if defined(__x86_64__)
542 /* FreeBSD doesn't have MAP_32BIT, use MAP_FIXED and assume
543 * 0x40000000 is free */
544 flags |= MAP_FIXED;
545 addr = (void *)0x40000000;
546 /* Cannot map more than that */
547 if (code_gen_buffer_size > (800 * 1024 * 1024))
548 code_gen_buffer_size = (800 * 1024 * 1024);
549 #elif defined(__sparc_v9__)
550 // Map the buffer below 2G, so we can use direct calls and branches
551 flags |= MAP_FIXED;
552 addr = (void *) 0x60000000UL;
553 if (code_gen_buffer_size > (512 * 1024 * 1024)) {
554 code_gen_buffer_size = (512 * 1024 * 1024);
555 }
556 #endif
557 code_gen_buffer = mmap(addr, code_gen_buffer_size,
558 PROT_WRITE | PROT_READ | PROT_EXEC,
559 flags, -1, 0);
560 if (code_gen_buffer == MAP_FAILED) {
561 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
562 exit(1);
563 }
564 }
565 #else
566 code_gen_buffer = g_malloc(code_gen_buffer_size);
567 map_exec(code_gen_buffer, code_gen_buffer_size);
568 #endif
569 #endif /* !USE_STATIC_CODE_GEN_BUFFER */
570 map_exec(code_gen_prologue, sizeof(code_gen_prologue));
571 code_gen_buffer_max_size = code_gen_buffer_size -
572 (TCG_MAX_OP_SIZE * OPC_BUF_SIZE);
573 code_gen_max_blocks = code_gen_buffer_size / CODE_GEN_AVG_BLOCK_SIZE;
574 tbs = g_malloc(code_gen_max_blocks * sizeof(TranslationBlock));
575 }
576
577 /* Must be called before using the QEMU cpus. 'tb_size' is the size
578 (in bytes) allocated to the translation buffer. Zero means default
579 size. */
580 void tcg_exec_init(unsigned long tb_size)
581 {
582 cpu_gen_init();
583 code_gen_alloc(tb_size);
584 code_gen_ptr = code_gen_buffer;
585 page_init();
586 #if !defined(CONFIG_USER_ONLY) || !defined(CONFIG_USE_GUEST_BASE)
587 /* There's no guest base to take into account, so go ahead and
588 initialize the prologue now. */
589 tcg_prologue_init(&tcg_ctx);
590 #endif
591 }
592
593 bool tcg_enabled(void)
594 {
595 return code_gen_buffer != NULL;
596 }
597
598 void cpu_exec_init_all(void)
599 {
600 #if !defined(CONFIG_USER_ONLY)
601 memory_map_init();
602 io_mem_init();
603 #endif
604 }
605
606 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
607
608 static int cpu_common_post_load(void *opaque, int version_id)
609 {
610 CPUState *env = opaque;
611
612 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
613 version_id is increased. */
614 env->interrupt_request &= ~0x01;
615 tlb_flush(env, 1);
616
617 return 0;
618 }
619
620 static const VMStateDescription vmstate_cpu_common = {
621 .name = "cpu_common",
622 .version_id = 1,
623 .minimum_version_id = 1,
624 .minimum_version_id_old = 1,
625 .post_load = cpu_common_post_load,
626 .fields = (VMStateField []) {
627 VMSTATE_UINT32(halted, CPUState),
628 VMSTATE_UINT32(interrupt_request, CPUState),
629 VMSTATE_END_OF_LIST()
630 }
631 };
632 #endif
633
634 CPUState *qemu_get_cpu(int cpu)
635 {
636 CPUState *env = first_cpu;
637
638 while (env) {
639 if (env->cpu_index == cpu)
640 break;
641 env = env->next_cpu;
642 }
643
644 return env;
645 }
646
647 void cpu_exec_init(CPUState *env)
648 {
649 CPUState **penv;
650 int cpu_index;
651
652 #if defined(CONFIG_USER_ONLY)
653 cpu_list_lock();
654 #endif
655 env->next_cpu = NULL;
656 penv = &first_cpu;
657 cpu_index = 0;
658 while (*penv != NULL) {
659 penv = &(*penv)->next_cpu;
660 cpu_index++;
661 }
662 env->cpu_index = cpu_index;
663 env->numa_node = 0;
664 QTAILQ_INIT(&env->breakpoints);
665 QTAILQ_INIT(&env->watchpoints);
666 #ifndef CONFIG_USER_ONLY
667 env->thread_id = qemu_get_thread_id();
668 #endif
669 *penv = env;
670 #if defined(CONFIG_USER_ONLY)
671 cpu_list_unlock();
672 #endif
673 #if defined(CPU_SAVE_VERSION) && !defined(CONFIG_USER_ONLY)
674 vmstate_register(NULL, cpu_index, &vmstate_cpu_common, env);
675 register_savevm(NULL, "cpu", cpu_index, CPU_SAVE_VERSION,
676 cpu_save, cpu_load, env);
677 #endif
678 }
679
680 /* Allocate a new translation block. Flush the translation buffer if
681 too many translation blocks or too much generated code. */
682 static TranslationBlock *tb_alloc(target_ulong pc)
683 {
684 TranslationBlock *tb;
685
686 if (nb_tbs >= code_gen_max_blocks ||
687 (code_gen_ptr - code_gen_buffer) >= code_gen_buffer_max_size)
688 return NULL;
689 tb = &tbs[nb_tbs++];
690 tb->pc = pc;
691 tb->cflags = 0;
692 return tb;
693 }
694
695 void tb_free(TranslationBlock *tb)
696 {
697 /* In practice this is mostly used for single use temporary TB
698 Ignore the hard cases and just back up if this TB happens to
699 be the last one generated. */
700 if (nb_tbs > 0 && tb == &tbs[nb_tbs - 1]) {
701 code_gen_ptr = tb->tc_ptr;
702 nb_tbs--;
703 }
704 }
705
706 static inline void invalidate_page_bitmap(PageDesc *p)
707 {
708 if (p->code_bitmap) {
709 g_free(p->code_bitmap);
710 p->code_bitmap = NULL;
711 }
712 p->code_write_count = 0;
713 }
714
715 /* Set to NULL all the 'first_tb' fields in all PageDescs. */
716
717 static void page_flush_tb_1 (int level, void **lp)
718 {
719 int i;
720
721 if (*lp == NULL) {
722 return;
723 }
724 if (level == 0) {
725 PageDesc *pd = *lp;
726 for (i = 0; i < L2_SIZE; ++i) {
727 pd[i].first_tb = NULL;
728 invalidate_page_bitmap(pd + i);
729 }
730 } else {
731 void **pp = *lp;
732 for (i = 0; i < L2_SIZE; ++i) {
733 page_flush_tb_1 (level - 1, pp + i);
734 }
735 }
736 }
737
738 static void page_flush_tb(void)
739 {
740 int i;
741 for (i = 0; i < V_L1_SIZE; i++) {
742 page_flush_tb_1(V_L1_SHIFT / L2_BITS - 1, l1_map + i);
743 }
744 }
745
746 /* flush all the translation blocks */
747 /* XXX: tb_flush is currently not thread safe */
748 void tb_flush(CPUState *env1)
749 {
750 CPUState *env;
751 #if defined(DEBUG_FLUSH)
752 printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n",
753 (unsigned long)(code_gen_ptr - code_gen_buffer),
754 nb_tbs, nb_tbs > 0 ?
755 ((unsigned long)(code_gen_ptr - code_gen_buffer)) / nb_tbs : 0);
756 #endif
757 if ((unsigned long)(code_gen_ptr - code_gen_buffer) > code_gen_buffer_size)
758 cpu_abort(env1, "Internal error: code buffer overflow\n");
759
760 nb_tbs = 0;
761
762 for(env = first_cpu; env != NULL; env = env->next_cpu) {
763 memset (env->tb_jmp_cache, 0, TB_JMP_CACHE_SIZE * sizeof (void *));
764 }
765
766 memset (tb_phys_hash, 0, CODE_GEN_PHYS_HASH_SIZE * sizeof (void *));
767 page_flush_tb();
768
769 code_gen_ptr = code_gen_buffer;
770 /* XXX: flush processor icache at this point if cache flush is
771 expensive */
772 tb_flush_count++;
773 }
774
775 #ifdef DEBUG_TB_CHECK
776
777 static void tb_invalidate_check(target_ulong address)
778 {
779 TranslationBlock *tb;
780 int i;
781 address &= TARGET_PAGE_MASK;
782 for(i = 0;i < CODE_GEN_PHYS_HASH_SIZE; i++) {
783 for(tb = tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
784 if (!(address + TARGET_PAGE_SIZE <= tb->pc ||
785 address >= tb->pc + tb->size)) {
786 printf("ERROR invalidate: address=" TARGET_FMT_lx
787 " PC=%08lx size=%04x\n",
788 address, (long)tb->pc, tb->size);
789 }
790 }
791 }
792 }
793
794 /* verify that all the pages have correct rights for code */
795 static void tb_page_check(void)
796 {
797 TranslationBlock *tb;
798 int i, flags1, flags2;
799
800 for(i = 0;i < CODE_GEN_PHYS_HASH_SIZE; i++) {
801 for(tb = tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
802 flags1 = page_get_flags(tb->pc);
803 flags2 = page_get_flags(tb->pc + tb->size - 1);
804 if ((flags1 & PAGE_WRITE) || (flags2 & PAGE_WRITE)) {
805 printf("ERROR page flags: PC=%08lx size=%04x f1=%x f2=%x\n",
806 (long)tb->pc, tb->size, flags1, flags2);
807 }
808 }
809 }
810 }
811
812 #endif
813
814 /* invalidate one TB */
815 static inline void tb_remove(TranslationBlock **ptb, TranslationBlock *tb,
816 int next_offset)
817 {
818 TranslationBlock *tb1;
819 for(;;) {
820 tb1 = *ptb;
821 if (tb1 == tb) {
822 *ptb = *(TranslationBlock **)((char *)tb1 + next_offset);
823 break;
824 }
825 ptb = (TranslationBlock **)((char *)tb1 + next_offset);
826 }
827 }
828
829 static inline void tb_page_remove(TranslationBlock **ptb, TranslationBlock *tb)
830 {
831 TranslationBlock *tb1;
832 unsigned int n1;
833
834 for(;;) {
835 tb1 = *ptb;
836 n1 = (long)tb1 & 3;
837 tb1 = (TranslationBlock *)((long)tb1 & ~3);
838 if (tb1 == tb) {
839 *ptb = tb1->page_next[n1];
840 break;
841 }
842 ptb = &tb1->page_next[n1];
843 }
844 }
845
846 static inline void tb_jmp_remove(TranslationBlock *tb, int n)
847 {
848 TranslationBlock *tb1, **ptb;
849 unsigned int n1;
850
851 ptb = &tb->jmp_next[n];
852 tb1 = *ptb;
853 if (tb1) {
854 /* find tb(n) in circular list */
855 for(;;) {
856 tb1 = *ptb;
857 n1 = (long)tb1 & 3;
858 tb1 = (TranslationBlock *)((long)tb1 & ~3);
859 if (n1 == n && tb1 == tb)
860 break;
861 if (n1 == 2) {
862 ptb = &tb1->jmp_first;
863 } else {
864 ptb = &tb1->jmp_next[n1];
865 }
866 }
867 /* now we can suppress tb(n) from the list */
868 *ptb = tb->jmp_next[n];
869
870 tb->jmp_next[n] = NULL;
871 }
872 }
873
874 /* reset the jump entry 'n' of a TB so that it is not chained to
875 another TB */
876 static inline void tb_reset_jump(TranslationBlock *tb, int n)
877 {
878 tb_set_jmp_target(tb, n, (unsigned long)(tb->tc_ptr + tb->tb_next_offset[n]));
879 }
880
881 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr)
882 {
883 CPUState *env;
884 PageDesc *p;
885 unsigned int h, n1;
886 tb_page_addr_t phys_pc;
887 TranslationBlock *tb1, *tb2;
888
889 /* remove the TB from the hash list */
890 phys_pc = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
891 h = tb_phys_hash_func(phys_pc);
892 tb_remove(&tb_phys_hash[h], tb,
893 offsetof(TranslationBlock, phys_hash_next));
894
895 /* remove the TB from the page list */
896 if (tb->page_addr[0] != page_addr) {
897 p = page_find(tb->page_addr[0] >> TARGET_PAGE_BITS);
898 tb_page_remove(&p->first_tb, tb);
899 invalidate_page_bitmap(p);
900 }
901 if (tb->page_addr[1] != -1 && tb->page_addr[1] != page_addr) {
902 p = page_find(tb->page_addr[1] >> TARGET_PAGE_BITS);
903 tb_page_remove(&p->first_tb, tb);
904 invalidate_page_bitmap(p);
905 }
906
907 tb_invalidated_flag = 1;
908
909 /* remove the TB from the hash list */
910 h = tb_jmp_cache_hash_func(tb->pc);
911 for(env = first_cpu; env != NULL; env = env->next_cpu) {
912 if (env->tb_jmp_cache[h] == tb)
913 env->tb_jmp_cache[h] = NULL;
914 }
915
916 /* suppress this TB from the two jump lists */
917 tb_jmp_remove(tb, 0);
918 tb_jmp_remove(tb, 1);
919
920 /* suppress any remaining jumps to this TB */
921 tb1 = tb->jmp_first;
922 for(;;) {
923 n1 = (long)tb1 & 3;
924 if (n1 == 2)
925 break;
926 tb1 = (TranslationBlock *)((long)tb1 & ~3);
927 tb2 = tb1->jmp_next[n1];
928 tb_reset_jump(tb1, n1);
929 tb1->jmp_next[n1] = NULL;
930 tb1 = tb2;
931 }
932 tb->jmp_first = (TranslationBlock *)((long)tb | 2); /* fail safe */
933
934 tb_phys_invalidate_count++;
935 }
936
937 static inline void set_bits(uint8_t *tab, int start, int len)
938 {
939 int end, mask, end1;
940
941 end = start + len;
942 tab += start >> 3;
943 mask = 0xff << (start & 7);
944 if ((start & ~7) == (end & ~7)) {
945 if (start < end) {
946 mask &= ~(0xff << (end & 7));
947 *tab |= mask;
948 }
949 } else {
950 *tab++ |= mask;
951 start = (start + 8) & ~7;
952 end1 = end & ~7;
953 while (start < end1) {
954 *tab++ = 0xff;
955 start += 8;
956 }
957 if (start < end) {
958 mask = ~(0xff << (end & 7));
959 *tab |= mask;
960 }
961 }
962 }
963
964 static void build_page_bitmap(PageDesc *p)
965 {
966 int n, tb_start, tb_end;
967 TranslationBlock *tb;
968
969 p->code_bitmap = g_malloc0(TARGET_PAGE_SIZE / 8);
970
971 tb = p->first_tb;
972 while (tb != NULL) {
973 n = (long)tb & 3;
974 tb = (TranslationBlock *)((long)tb & ~3);
975 /* NOTE: this is subtle as a TB may span two physical pages */
976 if (n == 0) {
977 /* NOTE: tb_end may be after the end of the page, but
978 it is not a problem */
979 tb_start = tb->pc & ~TARGET_PAGE_MASK;
980 tb_end = tb_start + tb->size;
981 if (tb_end > TARGET_PAGE_SIZE)
982 tb_end = TARGET_PAGE_SIZE;
983 } else {
984 tb_start = 0;
985 tb_end = ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
986 }
987 set_bits(p->code_bitmap, tb_start, tb_end - tb_start);
988 tb = tb->page_next[n];
989 }
990 }
991
992 TranslationBlock *tb_gen_code(CPUState *env,
993 target_ulong pc, target_ulong cs_base,
994 int flags, int cflags)
995 {
996 TranslationBlock *tb;
997 uint8_t *tc_ptr;
998 tb_page_addr_t phys_pc, phys_page2;
999 target_ulong virt_page2;
1000 int code_gen_size;
1001
1002 phys_pc = get_page_addr_code(env, pc);
1003 tb = tb_alloc(pc);
1004 if (!tb) {
1005 /* flush must be done */
1006 tb_flush(env);
1007 /* cannot fail at this point */
1008 tb = tb_alloc(pc);
1009 /* Don't forget to invalidate previous TB info. */
1010 tb_invalidated_flag = 1;
1011 }
1012 tc_ptr = code_gen_ptr;
1013 tb->tc_ptr = tc_ptr;
1014 tb->cs_base = cs_base;
1015 tb->flags = flags;
1016 tb->cflags = cflags;
1017 cpu_gen_code(env, tb, &code_gen_size);
1018 code_gen_ptr = (void *)(((unsigned long)code_gen_ptr + code_gen_size + CODE_GEN_ALIGN - 1) & ~(CODE_GEN_ALIGN - 1));
1019
1020 /* check next page if needed */
1021 virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
1022 phys_page2 = -1;
1023 if ((pc & TARGET_PAGE_MASK) != virt_page2) {
1024 phys_page2 = get_page_addr_code(env, virt_page2);
1025 }
1026 tb_link_page(tb, phys_pc, phys_page2);
1027 return tb;
1028 }
1029
1030 /* invalidate all TBs which intersect with the target physical page
1031 starting in range [start;end[. NOTE: start and end must refer to
1032 the same physical page. 'is_cpu_write_access' should be true if called
1033 from a real cpu write access: the virtual CPU will exit the current
1034 TB if code is modified inside this TB. */
1035 void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
1036 int is_cpu_write_access)
1037 {
1038 TranslationBlock *tb, *tb_next, *saved_tb;
1039 CPUState *env = cpu_single_env;
1040 tb_page_addr_t tb_start, tb_end;
1041 PageDesc *p;
1042 int n;
1043 #ifdef TARGET_HAS_PRECISE_SMC
1044 int current_tb_not_found = is_cpu_write_access;
1045 TranslationBlock *current_tb = NULL;
1046 int current_tb_modified = 0;
1047 target_ulong current_pc = 0;
1048 target_ulong current_cs_base = 0;
1049 int current_flags = 0;
1050 #endif /* TARGET_HAS_PRECISE_SMC */
1051
1052 p = page_find(start >> TARGET_PAGE_BITS);
1053 if (!p)
1054 return;
1055 if (!p->code_bitmap &&
1056 ++p->code_write_count >= SMC_BITMAP_USE_THRESHOLD &&
1057 is_cpu_write_access) {
1058 /* build code bitmap */
1059 build_page_bitmap(p);
1060 }
1061
1062 /* we remove all the TBs in the range [start, end[ */
1063 /* XXX: see if in some cases it could be faster to invalidate all the code */
1064 tb = p->first_tb;
1065 while (tb != NULL) {
1066 n = (long)tb & 3;
1067 tb = (TranslationBlock *)((long)tb & ~3);
1068 tb_next = tb->page_next[n];
1069 /* NOTE: this is subtle as a TB may span two physical pages */
1070 if (n == 0) {
1071 /* NOTE: tb_end may be after the end of the page, but
1072 it is not a problem */
1073 tb_start = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
1074 tb_end = tb_start + tb->size;
1075 } else {
1076 tb_start = tb->page_addr[1];
1077 tb_end = tb_start + ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
1078 }
1079 if (!(tb_end <= start || tb_start >= end)) {
1080 #ifdef TARGET_HAS_PRECISE_SMC
1081 if (current_tb_not_found) {
1082 current_tb_not_found = 0;
1083 current_tb = NULL;
1084 if (env->mem_io_pc) {
1085 /* now we have a real cpu fault */
1086 current_tb = tb_find_pc(env->mem_io_pc);
1087 }
1088 }
1089 if (current_tb == tb &&
1090 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1091 /* If we are modifying the current TB, we must stop
1092 its execution. We could be more precise by checking
1093 that the modification is after the current PC, but it
1094 would require a specialized function to partially
1095 restore the CPU state */
1096
1097 current_tb_modified = 1;
1098 cpu_restore_state(current_tb, env, env->mem_io_pc);
1099 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1100 &current_flags);
1101 }
1102 #endif /* TARGET_HAS_PRECISE_SMC */
1103 /* we need to do that to handle the case where a signal
1104 occurs while doing tb_phys_invalidate() */
1105 saved_tb = NULL;
1106 if (env) {
1107 saved_tb = env->current_tb;
1108 env->current_tb = NULL;
1109 }
1110 tb_phys_invalidate(tb, -1);
1111 if (env) {
1112 env->current_tb = saved_tb;
1113 if (env->interrupt_request && env->current_tb)
1114 cpu_interrupt(env, env->interrupt_request);
1115 }
1116 }
1117 tb = tb_next;
1118 }
1119 #if !defined(CONFIG_USER_ONLY)
1120 /* if no code remaining, no need to continue to use slow writes */
1121 if (!p->first_tb) {
1122 invalidate_page_bitmap(p);
1123 if (is_cpu_write_access) {
1124 tlb_unprotect_code_phys(env, start, env->mem_io_vaddr);
1125 }
1126 }
1127 #endif
1128 #ifdef TARGET_HAS_PRECISE_SMC
1129 if (current_tb_modified) {
1130 /* we generate a block containing just the instruction
1131 modifying the memory. It will ensure that it cannot modify
1132 itself */
1133 env->current_tb = NULL;
1134 tb_gen_code(env, current_pc, current_cs_base, current_flags, 1);
1135 cpu_resume_from_signal(env, NULL);
1136 }
1137 #endif
1138 }
1139
1140 /* len must be <= 8 and start must be a multiple of len */
1141 static inline void tb_invalidate_phys_page_fast(tb_page_addr_t start, int len)
1142 {
1143 PageDesc *p;
1144 int offset, b;
1145 #if 0
1146 if (1) {
1147 qemu_log("modifying code at 0x%x size=%d EIP=%x PC=%08x\n",
1148 cpu_single_env->mem_io_vaddr, len,
1149 cpu_single_env->eip,
1150 cpu_single_env->eip + (long)cpu_single_env->segs[R_CS].base);
1151 }
1152 #endif
1153 p = page_find(start >> TARGET_PAGE_BITS);
1154 if (!p)
1155 return;
1156 if (p->code_bitmap) {
1157 offset = start & ~TARGET_PAGE_MASK;
1158 b = p->code_bitmap[offset >> 3] >> (offset & 7);
1159 if (b & ((1 << len) - 1))
1160 goto do_invalidate;
1161 } else {
1162 do_invalidate:
1163 tb_invalidate_phys_page_range(start, start + len, 1);
1164 }
1165 }
1166
1167 #if !defined(CONFIG_SOFTMMU)
1168 static void tb_invalidate_phys_page(tb_page_addr_t addr,
1169 unsigned long pc, void *puc)
1170 {
1171 TranslationBlock *tb;
1172 PageDesc *p;
1173 int n;
1174 #ifdef TARGET_HAS_PRECISE_SMC
1175 TranslationBlock *current_tb = NULL;
1176 CPUState *env = cpu_single_env;
1177 int current_tb_modified = 0;
1178 target_ulong current_pc = 0;
1179 target_ulong current_cs_base = 0;
1180 int current_flags = 0;
1181 #endif
1182
1183 addr &= TARGET_PAGE_MASK;
1184 p = page_find(addr >> TARGET_PAGE_BITS);
1185 if (!p)
1186 return;
1187 tb = p->first_tb;
1188 #ifdef TARGET_HAS_PRECISE_SMC
1189 if (tb && pc != 0) {
1190 current_tb = tb_find_pc(pc);
1191 }
1192 #endif
1193 while (tb != NULL) {
1194 n = (long)tb & 3;
1195 tb = (TranslationBlock *)((long)tb & ~3);
1196 #ifdef TARGET_HAS_PRECISE_SMC
1197 if (current_tb == tb &&
1198 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1199 /* If we are modifying the current TB, we must stop
1200 its execution. We could be more precise by checking
1201 that the modification is after the current PC, but it
1202 would require a specialized function to partially
1203 restore the CPU state */
1204
1205 current_tb_modified = 1;
1206 cpu_restore_state(current_tb, env, pc);
1207 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1208 &current_flags);
1209 }
1210 #endif /* TARGET_HAS_PRECISE_SMC */
1211 tb_phys_invalidate(tb, addr);
1212 tb = tb->page_next[n];
1213 }
1214 p->first_tb = NULL;
1215 #ifdef TARGET_HAS_PRECISE_SMC
1216 if (current_tb_modified) {
1217 /* we generate a block containing just the instruction
1218 modifying the memory. It will ensure that it cannot modify
1219 itself */
1220 env->current_tb = NULL;
1221 tb_gen_code(env, current_pc, current_cs_base, current_flags, 1);
1222 cpu_resume_from_signal(env, puc);
1223 }
1224 #endif
1225 }
1226 #endif
1227
1228 /* add the tb in the target page and protect it if necessary */
1229 static inline void tb_alloc_page(TranslationBlock *tb,
1230 unsigned int n, tb_page_addr_t page_addr)
1231 {
1232 PageDesc *p;
1233 #ifndef CONFIG_USER_ONLY
1234 bool page_already_protected;
1235 #endif
1236
1237 tb->page_addr[n] = page_addr;
1238 p = page_find_alloc(page_addr >> TARGET_PAGE_BITS, 1);
1239 tb->page_next[n] = p->first_tb;
1240 #ifndef CONFIG_USER_ONLY
1241 page_already_protected = p->first_tb != NULL;
1242 #endif
1243 p->first_tb = (TranslationBlock *)((long)tb | n);
1244 invalidate_page_bitmap(p);
1245
1246 #if defined(TARGET_HAS_SMC) || 1
1247
1248 #if defined(CONFIG_USER_ONLY)
1249 if (p->flags & PAGE_WRITE) {
1250 target_ulong addr;
1251 PageDesc *p2;
1252 int prot;
1253
1254 /* force the host page as non writable (writes will have a
1255 page fault + mprotect overhead) */
1256 page_addr &= qemu_host_page_mask;
1257 prot = 0;
1258 for(addr = page_addr; addr < page_addr + qemu_host_page_size;
1259 addr += TARGET_PAGE_SIZE) {
1260
1261 p2 = page_find (addr >> TARGET_PAGE_BITS);
1262 if (!p2)
1263 continue;
1264 prot |= p2->flags;
1265 p2->flags &= ~PAGE_WRITE;
1266 }
1267 mprotect(g2h(page_addr), qemu_host_page_size,
1268 (prot & PAGE_BITS) & ~PAGE_WRITE);
1269 #ifdef DEBUG_TB_INVALIDATE
1270 printf("protecting code page: 0x" TARGET_FMT_lx "\n",
1271 page_addr);
1272 #endif
1273 }
1274 #else
1275 /* if some code is already present, then the pages are already
1276 protected. So we handle the case where only the first TB is
1277 allocated in a physical page */
1278 if (!page_already_protected) {
1279 tlb_protect_code(page_addr);
1280 }
1281 #endif
1282
1283 #endif /* TARGET_HAS_SMC */
1284 }
1285
1286 /* add a new TB and link it to the physical page tables. phys_page2 is
1287 (-1) to indicate that only one page contains the TB. */
1288 void tb_link_page(TranslationBlock *tb,
1289 tb_page_addr_t phys_pc, tb_page_addr_t phys_page2)
1290 {
1291 unsigned int h;
1292 TranslationBlock **ptb;
1293
1294 /* Grab the mmap lock to stop another thread invalidating this TB
1295 before we are done. */
1296 mmap_lock();
1297 /* add in the physical hash table */
1298 h = tb_phys_hash_func(phys_pc);
1299 ptb = &tb_phys_hash[h];
1300 tb->phys_hash_next = *ptb;
1301 *ptb = tb;
1302
1303 /* add in the page list */
1304 tb_alloc_page(tb, 0, phys_pc & TARGET_PAGE_MASK);
1305 if (phys_page2 != -1)
1306 tb_alloc_page(tb, 1, phys_page2);
1307 else
1308 tb->page_addr[1] = -1;
1309
1310 tb->jmp_first = (TranslationBlock *)((long)tb | 2);
1311 tb->jmp_next[0] = NULL;
1312 tb->jmp_next[1] = NULL;
1313
1314 /* init original jump addresses */
1315 if (tb->tb_next_offset[0] != 0xffff)
1316 tb_reset_jump(tb, 0);
1317 if (tb->tb_next_offset[1] != 0xffff)
1318 tb_reset_jump(tb, 1);
1319
1320 #ifdef DEBUG_TB_CHECK
1321 tb_page_check();
1322 #endif
1323 mmap_unlock();
1324 }
1325
1326 /* find the TB 'tb' such that tb[0].tc_ptr <= tc_ptr <
1327 tb[1].tc_ptr. Return NULL if not found */
1328 TranslationBlock *tb_find_pc(unsigned long tc_ptr)
1329 {
1330 int m_min, m_max, m;
1331 unsigned long v;
1332 TranslationBlock *tb;
1333
1334 if (nb_tbs <= 0)
1335 return NULL;
1336 if (tc_ptr < (unsigned long)code_gen_buffer ||
1337 tc_ptr >= (unsigned long)code_gen_ptr)
1338 return NULL;
1339 /* binary search (cf Knuth) */
1340 m_min = 0;
1341 m_max = nb_tbs - 1;
1342 while (m_min <= m_max) {
1343 m = (m_min + m_max) >> 1;
1344 tb = &tbs[m];
1345 v = (unsigned long)tb->tc_ptr;
1346 if (v == tc_ptr)
1347 return tb;
1348 else if (tc_ptr < v) {
1349 m_max = m - 1;
1350 } else {
1351 m_min = m + 1;
1352 }
1353 }
1354 return &tbs[m_max];
1355 }
1356
1357 static void tb_reset_jump_recursive(TranslationBlock *tb);
1358
1359 static inline void tb_reset_jump_recursive2(TranslationBlock *tb, int n)
1360 {
1361 TranslationBlock *tb1, *tb_next, **ptb;
1362 unsigned int n1;
1363
1364 tb1 = tb->jmp_next[n];
1365 if (tb1 != NULL) {
1366 /* find head of list */
1367 for(;;) {
1368 n1 = (long)tb1 & 3;
1369 tb1 = (TranslationBlock *)((long)tb1 & ~3);
1370 if (n1 == 2)
1371 break;
1372 tb1 = tb1->jmp_next[n1];
1373 }
1374 /* we are now sure now that tb jumps to tb1 */
1375 tb_next = tb1;
1376
1377 /* remove tb from the jmp_first list */
1378 ptb = &tb_next->jmp_first;
1379 for(;;) {
1380 tb1 = *ptb;
1381 n1 = (long)tb1 & 3;
1382 tb1 = (TranslationBlock *)((long)tb1 & ~3);
1383 if (n1 == n && tb1 == tb)
1384 break;
1385 ptb = &tb1->jmp_next[n1];
1386 }
1387 *ptb = tb->jmp_next[n];
1388 tb->jmp_next[n] = NULL;
1389
1390 /* suppress the jump to next tb in generated code */
1391 tb_reset_jump(tb, n);
1392
1393 /* suppress jumps in the tb on which we could have jumped */
1394 tb_reset_jump_recursive(tb_next);
1395 }
1396 }
1397
1398 static void tb_reset_jump_recursive(TranslationBlock *tb)
1399 {
1400 tb_reset_jump_recursive2(tb, 0);
1401 tb_reset_jump_recursive2(tb, 1);
1402 }
1403
1404 #if defined(TARGET_HAS_ICE)
1405 #if defined(CONFIG_USER_ONLY)
1406 static void breakpoint_invalidate(CPUState *env, target_ulong pc)
1407 {
1408 tb_invalidate_phys_page_range(pc, pc + 1, 0);
1409 }
1410 #else
1411 static void breakpoint_invalidate(CPUState *env, target_ulong pc)
1412 {
1413 target_phys_addr_t addr;
1414 target_ulong pd;
1415 ram_addr_t ram_addr;
1416 PhysPageDesc p;
1417
1418 addr = cpu_get_phys_page_debug(env, pc);
1419 p = phys_page_find(addr >> TARGET_PAGE_BITS);
1420 pd = p.phys_offset;
1421 ram_addr = (pd & TARGET_PAGE_MASK) | (pc & ~TARGET_PAGE_MASK);
1422 tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
1423 }
1424 #endif
1425 #endif /* TARGET_HAS_ICE */
1426
1427 #if defined(CONFIG_USER_ONLY)
1428 void cpu_watchpoint_remove_all(CPUState *env, int mask)
1429
1430 {
1431 }
1432
1433 int cpu_watchpoint_insert(CPUState *env, target_ulong addr, target_ulong len,
1434 int flags, CPUWatchpoint **watchpoint)
1435 {
1436 return -ENOSYS;
1437 }
1438 #else
1439 /* Add a watchpoint. */
1440 int cpu_watchpoint_insert(CPUState *env, target_ulong addr, target_ulong len,
1441 int flags, CPUWatchpoint **watchpoint)
1442 {
1443 target_ulong len_mask = ~(len - 1);
1444 CPUWatchpoint *wp;
1445
1446 /* sanity checks: allow power-of-2 lengths, deny unaligned watchpoints */
1447 if ((len != 1 && len != 2 && len != 4 && len != 8) || (addr & ~len_mask)) {
1448 fprintf(stderr, "qemu: tried to set invalid watchpoint at "
1449 TARGET_FMT_lx ", len=" TARGET_FMT_lu "\n", addr, len);
1450 return -EINVAL;
1451 }
1452 wp = g_malloc(sizeof(*wp));
1453
1454 wp->vaddr = addr;
1455 wp->len_mask = len_mask;
1456 wp->flags = flags;
1457
1458 /* keep all GDB-injected watchpoints in front */
1459 if (flags & BP_GDB)
1460 QTAILQ_INSERT_HEAD(&env->watchpoints, wp, entry);
1461 else
1462 QTAILQ_INSERT_TAIL(&env->watchpoints, wp, entry);
1463
1464 tlb_flush_page(env, addr);
1465
1466 if (watchpoint)
1467 *watchpoint = wp;
1468 return 0;
1469 }
1470
1471 /* Remove a specific watchpoint. */
1472 int cpu_watchpoint_remove(CPUState *env, target_ulong addr, target_ulong len,
1473 int flags)
1474 {
1475 target_ulong len_mask = ~(len - 1);
1476 CPUWatchpoint *wp;
1477
1478 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1479 if (addr == wp->vaddr && len_mask == wp->len_mask
1480 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
1481 cpu_watchpoint_remove_by_ref(env, wp);
1482 return 0;
1483 }
1484 }
1485 return -ENOENT;
1486 }
1487
1488 /* Remove a specific watchpoint by reference. */
1489 void cpu_watchpoint_remove_by_ref(CPUState *env, CPUWatchpoint *watchpoint)
1490 {
1491 QTAILQ_REMOVE(&env->watchpoints, watchpoint, entry);
1492
1493 tlb_flush_page(env, watchpoint->vaddr);
1494
1495 g_free(watchpoint);
1496 }
1497
1498 /* Remove all matching watchpoints. */
1499 void cpu_watchpoint_remove_all(CPUState *env, int mask)
1500 {
1501 CPUWatchpoint *wp, *next;
1502
1503 QTAILQ_FOREACH_SAFE(wp, &env->watchpoints, entry, next) {
1504 if (wp->flags & mask)
1505 cpu_watchpoint_remove_by_ref(env, wp);
1506 }
1507 }
1508 #endif
1509
1510 /* Add a breakpoint. */
1511 int cpu_breakpoint_insert(CPUState *env, target_ulong pc, int flags,
1512 CPUBreakpoint **breakpoint)
1513 {
1514 #if defined(TARGET_HAS_ICE)
1515 CPUBreakpoint *bp;
1516
1517 bp = g_malloc(sizeof(*bp));
1518
1519 bp->pc = pc;
1520 bp->flags = flags;
1521
1522 /* keep all GDB-injected breakpoints in front */
1523 if (flags & BP_GDB)
1524 QTAILQ_INSERT_HEAD(&env->breakpoints, bp, entry);
1525 else
1526 QTAILQ_INSERT_TAIL(&env->breakpoints, bp, entry);
1527
1528 breakpoint_invalidate(env, pc);
1529
1530 if (breakpoint)
1531 *breakpoint = bp;
1532 return 0;
1533 #else
1534 return -ENOSYS;
1535 #endif
1536 }
1537
1538 /* Remove a specific breakpoint. */
1539 int cpu_breakpoint_remove(CPUState *env, target_ulong pc, int flags)
1540 {
1541 #if defined(TARGET_HAS_ICE)
1542 CPUBreakpoint *bp;
1543
1544 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
1545 if (bp->pc == pc && bp->flags == flags) {
1546 cpu_breakpoint_remove_by_ref(env, bp);
1547 return 0;
1548 }
1549 }
1550 return -ENOENT;
1551 #else
1552 return -ENOSYS;
1553 #endif
1554 }
1555
1556 /* Remove a specific breakpoint by reference. */
1557 void cpu_breakpoint_remove_by_ref(CPUState *env, CPUBreakpoint *breakpoint)
1558 {
1559 #if defined(TARGET_HAS_ICE)
1560 QTAILQ_REMOVE(&env->breakpoints, breakpoint, entry);
1561
1562 breakpoint_invalidate(env, breakpoint->pc);
1563
1564 g_free(breakpoint);
1565 #endif
1566 }
1567
1568 /* Remove all matching breakpoints. */
1569 void cpu_breakpoint_remove_all(CPUState *env, int mask)
1570 {
1571 #if defined(TARGET_HAS_ICE)
1572 CPUBreakpoint *bp, *next;
1573
1574 QTAILQ_FOREACH_SAFE(bp, &env->breakpoints, entry, next) {
1575 if (bp->flags & mask)
1576 cpu_breakpoint_remove_by_ref(env, bp);
1577 }
1578 #endif
1579 }
1580
1581 /* enable or disable single step mode. EXCP_DEBUG is returned by the
1582 CPU loop after each instruction */
1583 void cpu_single_step(CPUState *env, int enabled)
1584 {
1585 #if defined(TARGET_HAS_ICE)
1586 if (env->singlestep_enabled != enabled) {
1587 env->singlestep_enabled = enabled;
1588 if (kvm_enabled())
1589 kvm_update_guest_debug(env, 0);
1590 else {
1591 /* must flush all the translated code to avoid inconsistencies */
1592 /* XXX: only flush what is necessary */
1593 tb_flush(env);
1594 }
1595 }
1596 #endif
1597 }
1598
1599 /* enable or disable low levels log */
1600 void cpu_set_log(int log_flags)
1601 {
1602 loglevel = log_flags;
1603 if (loglevel && !logfile) {
1604 logfile = fopen(logfilename, log_append ? "a" : "w");
1605 if (!logfile) {
1606 perror(logfilename);
1607 _exit(1);
1608 }
1609 #if !defined(CONFIG_SOFTMMU)
1610 /* must avoid mmap() usage of glibc by setting a buffer "by hand" */
1611 {
1612 static char logfile_buf[4096];
1613 setvbuf(logfile, logfile_buf, _IOLBF, sizeof(logfile_buf));
1614 }
1615 #elif defined(_WIN32)
1616 /* Win32 doesn't support line-buffering, so use unbuffered output. */
1617 setvbuf(logfile, NULL, _IONBF, 0);
1618 #else
1619 setvbuf(logfile, NULL, _IOLBF, 0);
1620 #endif
1621 log_append = 1;
1622 }
1623 if (!loglevel && logfile) {
1624 fclose(logfile);
1625 logfile = NULL;
1626 }
1627 }
1628
1629 void cpu_set_log_filename(const char *filename)
1630 {
1631 logfilename = strdup(filename);
1632 if (logfile) {
1633 fclose(logfile);
1634 logfile = NULL;
1635 }
1636 cpu_set_log(loglevel);
1637 }
1638
1639 static void cpu_unlink_tb(CPUState *env)
1640 {
1641 /* FIXME: TB unchaining isn't SMP safe. For now just ignore the
1642 problem and hope the cpu will stop of its own accord. For userspace
1643 emulation this often isn't actually as bad as it sounds. Often
1644 signals are used primarily to interrupt blocking syscalls. */
1645 TranslationBlock *tb;
1646 static spinlock_t interrupt_lock = SPIN_LOCK_UNLOCKED;
1647
1648 spin_lock(&interrupt_lock);
1649 tb = env->current_tb;
1650 /* if the cpu is currently executing code, we must unlink it and
1651 all the potentially executing TB */
1652 if (tb) {
1653 env->current_tb = NULL;
1654 tb_reset_jump_recursive(tb);
1655 }
1656 spin_unlock(&interrupt_lock);
1657 }
1658
1659 #ifndef CONFIG_USER_ONLY
1660 /* mask must never be zero, except for A20 change call */
1661 static void tcg_handle_interrupt(CPUState *env, int mask)
1662 {
1663 int old_mask;
1664
1665 old_mask = env->interrupt_request;
1666 env->interrupt_request |= mask;
1667
1668 /*
1669 * If called from iothread context, wake the target cpu in
1670 * case its halted.
1671 */
1672 if (!qemu_cpu_is_self(env)) {
1673 qemu_cpu_kick(env);
1674 return;
1675 }
1676
1677 if (use_icount) {
1678 env->icount_decr.u16.high = 0xffff;
1679 if (!can_do_io(env)
1680 && (mask & ~old_mask) != 0) {
1681 cpu_abort(env, "Raised interrupt while not in I/O function");
1682 }
1683 } else {
1684 cpu_unlink_tb(env);
1685 }
1686 }
1687
1688 CPUInterruptHandler cpu_interrupt_handler = tcg_handle_interrupt;
1689
1690 #else /* CONFIG_USER_ONLY */
1691
1692 void cpu_interrupt(CPUState *env, int mask)
1693 {
1694 env->interrupt_request |= mask;
1695 cpu_unlink_tb(env);
1696 }
1697 #endif /* CONFIG_USER_ONLY */
1698
1699 void cpu_reset_interrupt(CPUState *env, int mask)
1700 {
1701 env->interrupt_request &= ~mask;
1702 }
1703
1704 void cpu_exit(CPUState *env)
1705 {
1706 env->exit_request = 1;
1707 cpu_unlink_tb(env);
1708 }
1709
1710 const CPULogItem cpu_log_items[] = {
1711 { CPU_LOG_TB_OUT_ASM, "out_asm",
1712 "show generated host assembly code for each compiled TB" },
1713 { CPU_LOG_TB_IN_ASM, "in_asm",
1714 "show target assembly code for each compiled TB" },
1715 { CPU_LOG_TB_OP, "op",
1716 "show micro ops for each compiled TB" },
1717 { CPU_LOG_TB_OP_OPT, "op_opt",
1718 "show micro ops "
1719 #ifdef TARGET_I386
1720 "before eflags optimization and "
1721 #endif
1722 "after liveness analysis" },
1723 { CPU_LOG_INT, "int",
1724 "show interrupts/exceptions in short format" },
1725 { CPU_LOG_EXEC, "exec",
1726 "show trace before each executed TB (lots of logs)" },
1727 { CPU_LOG_TB_CPU, "cpu",
1728 "show CPU state before block translation" },
1729 #ifdef TARGET_I386
1730 { CPU_LOG_PCALL, "pcall",
1731 "show protected mode far calls/returns/exceptions" },
1732 { CPU_LOG_RESET, "cpu_reset",
1733 "show CPU state before CPU resets" },
1734 #endif
1735 #ifdef DEBUG_IOPORT
1736 { CPU_LOG_IOPORT, "ioport",
1737 "show all i/o ports accesses" },
1738 #endif
1739 { 0, NULL, NULL },
1740 };
1741
1742 static int cmp1(const char *s1, int n, const char *s2)
1743 {
1744 if (strlen(s2) != n)
1745 return 0;
1746 return memcmp(s1, s2, n) == 0;
1747 }
1748
1749 /* takes a comma separated list of log masks. Return 0 if error. */
1750 int cpu_str_to_log_mask(const char *str)
1751 {
1752 const CPULogItem *item;
1753 int mask;
1754 const char *p, *p1;
1755
1756 p = str;
1757 mask = 0;
1758 for(;;) {
1759 p1 = strchr(p, ',');
1760 if (!p1)
1761 p1 = p + strlen(p);
1762 if(cmp1(p,p1-p,"all")) {
1763 for(item = cpu_log_items; item->mask != 0; item++) {
1764 mask |= item->mask;
1765 }
1766 } else {
1767 for(item = cpu_log_items; item->mask != 0; item++) {
1768 if (cmp1(p, p1 - p, item->name))
1769 goto found;
1770 }
1771 return 0;
1772 }
1773 found:
1774 mask |= item->mask;
1775 if (*p1 != ',')
1776 break;
1777 p = p1 + 1;
1778 }
1779 return mask;
1780 }
1781
1782 void cpu_abort(CPUState *env, const char *fmt, ...)
1783 {
1784 va_list ap;
1785 va_list ap2;
1786
1787 va_start(ap, fmt);
1788 va_copy(ap2, ap);
1789 fprintf(stderr, "qemu: fatal: ");
1790 vfprintf(stderr, fmt, ap);
1791 fprintf(stderr, "\n");
1792 #ifdef TARGET_I386
1793 cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU | X86_DUMP_CCOP);
1794 #else
1795 cpu_dump_state(env, stderr, fprintf, 0);
1796 #endif
1797 if (qemu_log_enabled()) {
1798 qemu_log("qemu: fatal: ");
1799 qemu_log_vprintf(fmt, ap2);
1800 qemu_log("\n");
1801 #ifdef TARGET_I386
1802 log_cpu_state(env, X86_DUMP_FPU | X86_DUMP_CCOP);
1803 #else
1804 log_cpu_state(env, 0);
1805 #endif
1806 qemu_log_flush();
1807 qemu_log_close();
1808 }
1809 va_end(ap2);
1810 va_end(ap);
1811 #if defined(CONFIG_USER_ONLY)
1812 {
1813 struct sigaction act;
1814 sigfillset(&act.sa_mask);
1815 act.sa_handler = SIG_DFL;
1816 sigaction(SIGABRT, &act, NULL);
1817 }
1818 #endif
1819 abort();
1820 }
1821
1822 CPUState *cpu_copy(CPUState *env)
1823 {
1824 CPUState *new_env = cpu_init(env->cpu_model_str);
1825 CPUState *next_cpu = new_env->next_cpu;
1826 int cpu_index = new_env->cpu_index;
1827 #if defined(TARGET_HAS_ICE)
1828 CPUBreakpoint *bp;
1829 CPUWatchpoint *wp;
1830 #endif
1831
1832 memcpy(new_env, env, sizeof(CPUState));
1833
1834 /* Preserve chaining and index. */
1835 new_env->next_cpu = next_cpu;
1836 new_env->cpu_index = cpu_index;
1837
1838 /* Clone all break/watchpoints.
1839 Note: Once we support ptrace with hw-debug register access, make sure
1840 BP_CPU break/watchpoints are handled correctly on clone. */
1841 QTAILQ_INIT(&env->breakpoints);
1842 QTAILQ_INIT(&env->watchpoints);
1843 #if defined(TARGET_HAS_ICE)
1844 QTAILQ_FOREACH(bp, &env->breakpoints, entry) {
1845 cpu_breakpoint_insert(new_env, bp->pc, bp->flags, NULL);
1846 }
1847 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
1848 cpu_watchpoint_insert(new_env, wp->vaddr, (~wp->len_mask) + 1,
1849 wp->flags, NULL);
1850 }
1851 #endif
1852
1853 return new_env;
1854 }
1855
1856 #if !defined(CONFIG_USER_ONLY)
1857
1858 static inline void tlb_flush_jmp_cache(CPUState *env, target_ulong addr)
1859 {
1860 unsigned int i;
1861
1862 /* Discard jump cache entries for any tb which might potentially
1863 overlap the flushed page. */
1864 i = tb_jmp_cache_hash_page(addr - TARGET_PAGE_SIZE);
1865 memset (&env->tb_jmp_cache[i], 0,
1866 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1867
1868 i = tb_jmp_cache_hash_page(addr);
1869 memset (&env->tb_jmp_cache[i], 0,
1870 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1871 }
1872
1873 static CPUTLBEntry s_cputlb_empty_entry = {
1874 .addr_read = -1,
1875 .addr_write = -1,
1876 .addr_code = -1,
1877 .addend = -1,
1878 };
1879
1880 /* NOTE: if flush_global is true, also flush global entries (not
1881 implemented yet) */
1882 void tlb_flush(CPUState *env, int flush_global)
1883 {
1884 int i;
1885
1886 #if defined(DEBUG_TLB)
1887 printf("tlb_flush:\n");
1888 #endif
1889 /* must reset current TB so that interrupts cannot modify the
1890 links while we are modifying them */
1891 env->current_tb = NULL;
1892
1893 for(i = 0; i < CPU_TLB_SIZE; i++) {
1894 int mmu_idx;
1895 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
1896 env->tlb_table[mmu_idx][i] = s_cputlb_empty_entry;
1897 }
1898 }
1899
1900 memset (env->tb_jmp_cache, 0, TB_JMP_CACHE_SIZE * sizeof (void *));
1901
1902 env->tlb_flush_addr = -1;
1903 env->tlb_flush_mask = 0;
1904 tlb_flush_count++;
1905 }
1906
1907 static inline void tlb_flush_entry(CPUTLBEntry *tlb_entry, target_ulong addr)
1908 {
1909 if (addr == (tlb_entry->addr_read &
1910 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) ||
1911 addr == (tlb_entry->addr_write &
1912 (TARGET_PAGE_MASK | TLB_INVALID_MASK)) ||
1913 addr == (tlb_entry->addr_code &
1914 (TARGET_PAGE_MASK | TLB_INVALID_MASK))) {
1915 *tlb_entry = s_cputlb_empty_entry;
1916 }
1917 }
1918
1919 void tlb_flush_page(CPUState *env, target_ulong addr)
1920 {
1921 int i;
1922 int mmu_idx;
1923
1924 #if defined(DEBUG_TLB)
1925 printf("tlb_flush_page: " TARGET_FMT_lx "\n", addr);
1926 #endif
1927 /* Check if we need to flush due to large pages. */
1928 if ((addr & env->tlb_flush_mask) == env->tlb_flush_addr) {
1929 #if defined(DEBUG_TLB)
1930 printf("tlb_flush_page: forced full flush ("
1931 TARGET_FMT_lx "/" TARGET_FMT_lx ")\n",
1932 env->tlb_flush_addr, env->tlb_flush_mask);
1933 #endif
1934 tlb_flush(env, 1);
1935 return;
1936 }
1937 /* must reset current TB so that interrupts cannot modify the
1938 links while we are modifying them */
1939 env->current_tb = NULL;
1940
1941 addr &= TARGET_PAGE_MASK;
1942 i = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
1943 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++)
1944 tlb_flush_entry(&env->tlb_table[mmu_idx][i], addr);
1945
1946 tlb_flush_jmp_cache(env, addr);
1947 }
1948
1949 /* update the TLBs so that writes to code in the virtual page 'addr'
1950 can be detected */
1951 static void tlb_protect_code(ram_addr_t ram_addr)
1952 {
1953 cpu_physical_memory_reset_dirty(ram_addr,
1954 ram_addr + TARGET_PAGE_SIZE,
1955 CODE_DIRTY_FLAG);
1956 }
1957
1958 /* update the TLB so that writes in physical page 'phys_addr' are no longer
1959 tested for self modifying code */
1960 static void tlb_unprotect_code_phys(CPUState *env, ram_addr_t ram_addr,
1961 target_ulong vaddr)
1962 {
1963 cpu_physical_memory_set_dirty_flags(ram_addr, CODE_DIRTY_FLAG);
1964 }
1965
1966 static inline void tlb_reset_dirty_range(CPUTLBEntry *tlb_entry,
1967 unsigned long start, unsigned long length)
1968 {
1969 unsigned long addr;
1970 if ((tlb_entry->addr_write & ~TARGET_PAGE_MASK) == io_mem_ram.ram_addr) {
1971 addr = (tlb_entry->addr_write & TARGET_PAGE_MASK) + tlb_entry->addend;
1972 if ((addr - start) < length) {
1973 tlb_entry->addr_write = (tlb_entry->addr_write & TARGET_PAGE_MASK) | TLB_NOTDIRTY;
1974 }
1975 }
1976 }
1977
1978 /* Note: start and end must be within the same ram block. */
1979 void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t end,
1980 int dirty_flags)
1981 {
1982 CPUState *env;
1983 unsigned long length, start1;
1984 int i;
1985
1986 start &= TARGET_PAGE_MASK;
1987 end = TARGET_PAGE_ALIGN(end);
1988
1989 length = end - start;
1990 if (length == 0)
1991 return;
1992 cpu_physical_memory_mask_dirty_range(start, length, dirty_flags);
1993
1994 /* we modify the TLB cache so that the dirty bit will be set again
1995 when accessing the range */
1996 start1 = (unsigned long)qemu_safe_ram_ptr(start);
1997 /* Check that we don't span multiple blocks - this breaks the
1998 address comparisons below. */
1999 if ((unsigned long)qemu_safe_ram_ptr(end - 1) - start1
2000 != (end - 1) - start) {
2001 abort();
2002 }
2003
2004 for(env = first_cpu; env != NULL; env = env->next_cpu) {
2005 int mmu_idx;
2006 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
2007 for(i = 0; i < CPU_TLB_SIZE; i++)
2008 tlb_reset_dirty_range(&env->tlb_table[mmu_idx][i],
2009 start1, length);
2010 }
2011 }
2012 }
2013
2014 int cpu_physical_memory_set_dirty_tracking(int enable)
2015 {
2016 int ret = 0;
2017 in_migration = enable;
2018 return ret;
2019 }
2020
2021 static inline void tlb_update_dirty(CPUTLBEntry *tlb_entry)
2022 {
2023 ram_addr_t ram_addr;
2024 void *p;
2025
2026 if ((tlb_entry->addr_write & ~TARGET_PAGE_MASK) == io_mem_ram.ram_addr) {
2027 p = (void *)(unsigned long)((tlb_entry->addr_write & TARGET_PAGE_MASK)
2028 + tlb_entry->addend);
2029 ram_addr = qemu_ram_addr_from_host_nofail(p);
2030 if (!cpu_physical_memory_is_dirty(ram_addr)) {
2031 tlb_entry->addr_write |= TLB_NOTDIRTY;
2032 }
2033 }
2034 }
2035
2036 /* update the TLB according to the current state of the dirty bits */
2037 void cpu_tlb_update_dirty(CPUState *env)
2038 {
2039 int i;
2040 int mmu_idx;
2041 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
2042 for(i = 0; i < CPU_TLB_SIZE; i++)
2043 tlb_update_dirty(&env->tlb_table[mmu_idx][i]);
2044 }
2045 }
2046
2047 static inline void tlb_set_dirty1(CPUTLBEntry *tlb_entry, target_ulong vaddr)
2048 {
2049 if (tlb_entry->addr_write == (vaddr | TLB_NOTDIRTY))
2050 tlb_entry->addr_write = vaddr;
2051 }
2052
2053 /* update the TLB corresponding to virtual page vaddr
2054 so that it is no longer dirty */
2055 static inline void tlb_set_dirty(CPUState *env, target_ulong vaddr)
2056 {
2057 int i;
2058 int mmu_idx;
2059
2060 vaddr &= TARGET_PAGE_MASK;
2061 i = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2062 for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++)
2063 tlb_set_dirty1(&env->tlb_table[mmu_idx][i], vaddr);
2064 }
2065
2066 /* Our TLB does not support large pages, so remember the area covered by
2067 large pages and trigger a full TLB flush if these are invalidated. */
2068 static void tlb_add_large_page(CPUState *env, target_ulong vaddr,
2069 target_ulong size)
2070 {
2071 target_ulong mask = ~(size - 1);
2072
2073 if (env->tlb_flush_addr == (target_ulong)-1) {
2074 env->tlb_flush_addr = vaddr & mask;
2075 env->tlb_flush_mask = mask;
2076 return;
2077 }
2078 /* Extend the existing region to include the new page.
2079 This is a compromise between unnecessary flushes and the cost
2080 of maintaining a full variable size TLB. */
2081 mask &= env->tlb_flush_mask;
2082 while (((env->tlb_flush_addr ^ vaddr) & mask) != 0) {
2083 mask <<= 1;
2084 }
2085 env->tlb_flush_addr &= mask;
2086 env->tlb_flush_mask = mask;
2087 }
2088
2089 static bool is_ram_rom(ram_addr_t pd)
2090 {
2091 pd &= ~TARGET_PAGE_MASK;
2092 return pd == io_mem_ram.ram_addr || pd == io_mem_rom.ram_addr;
2093 }
2094
2095 static bool is_ram_rom_romd(ram_addr_t pd)
2096 {
2097 return is_ram_rom(pd) || (pd & IO_MEM_ROMD);
2098 }
2099
2100 /* Add a new TLB entry. At most one entry for a given virtual address
2101 is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
2102 supplied size is only used by tlb_flush_page. */
2103 void tlb_set_page(CPUState *env, target_ulong vaddr,
2104 target_phys_addr_t paddr, int prot,
2105 int mmu_idx, target_ulong size)
2106 {
2107 PhysPageDesc p;
2108 unsigned long pd;
2109 unsigned int index;
2110 target_ulong address;
2111 target_ulong code_address;
2112 unsigned long addend;
2113 CPUTLBEntry *te;
2114 CPUWatchpoint *wp;
2115 target_phys_addr_t iotlb;
2116
2117 assert(size >= TARGET_PAGE_SIZE);
2118 if (size != TARGET_PAGE_SIZE) {
2119 tlb_add_large_page(env, vaddr, size);
2120 }
2121 p = phys_page_find(paddr >> TARGET_PAGE_BITS);
2122 pd = p.phys_offset;
2123 #if defined(DEBUG_TLB)
2124 printf("tlb_set_page: vaddr=" TARGET_FMT_lx " paddr=0x" TARGET_FMT_plx
2125 " prot=%x idx=%d pd=0x%08lx\n",
2126 vaddr, paddr, prot, mmu_idx, pd);
2127 #endif
2128
2129 address = vaddr;
2130 if (!is_ram_rom_romd(pd)) {
2131 /* IO memory case (romd handled later) */
2132 address |= TLB_MMIO;
2133 }
2134 addend = (unsigned long)qemu_get_ram_ptr(pd & TARGET_PAGE_MASK);
2135 if (is_ram_rom(pd)) {
2136 /* Normal RAM. */
2137 iotlb = pd & TARGET_PAGE_MASK;
2138 if ((pd & ~TARGET_PAGE_MASK) == io_mem_ram.ram_addr)
2139 iotlb |= io_mem_notdirty.ram_addr;
2140 else
2141 iotlb |= io_mem_rom.ram_addr;
2142 } else {
2143 /* IO handlers are currently passed a physical address.
2144 It would be nice to pass an offset from the base address
2145 of that region. This would avoid having to special case RAM,
2146 and avoid full address decoding in every device.
2147 We can't use the high bits of pd for this because
2148 IO_MEM_ROMD uses these as a ram address. */
2149 iotlb = (pd & ~TARGET_PAGE_MASK);
2150 iotlb += p.region_offset;
2151 }
2152
2153 code_address = address;
2154 /* Make accesses to pages with watchpoints go via the
2155 watchpoint trap routines. */
2156 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
2157 if (vaddr == (wp->vaddr & TARGET_PAGE_MASK)) {
2158 /* Avoid trapping reads of pages with a write breakpoint. */
2159 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
2160 iotlb = io_mem_watch + paddr;
2161 address |= TLB_MMIO;
2162 break;
2163 }
2164 }
2165 }
2166
2167 index = (vaddr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
2168 env->iotlb[mmu_idx][index] = iotlb - vaddr;
2169 te = &env->tlb_table[mmu_idx][index];
2170 te->addend = addend - vaddr;
2171 if (prot & PAGE_READ) {
2172 te->addr_read = address;
2173 } else {
2174 te->addr_read = -1;
2175 }
2176
2177 if (prot & PAGE_EXEC) {
2178 te->addr_code = code_address;
2179 } else {
2180 te->addr_code = -1;
2181 }
2182 if (prot & PAGE_WRITE) {
2183 if ((pd & ~TARGET_PAGE_MASK) == io_mem_rom.ram_addr ||
2184 (pd & IO_MEM_ROMD)) {
2185 /* Write access calls the I/O callback. */
2186 te->addr_write = address | TLB_MMIO;
2187 } else if ((pd & ~TARGET_PAGE_MASK) == io_mem_ram.ram_addr &&
2188 !cpu_physical_memory_is_dirty(pd)) {
2189 te->addr_write = address | TLB_NOTDIRTY;
2190 } else {
2191 te->addr_write = address;
2192 }
2193 } else {
2194 te->addr_write = -1;
2195 }
2196 }
2197
2198 #else
2199
2200 void tlb_flush(CPUState *env, int flush_global)
2201 {
2202 }
2203
2204 void tlb_flush_page(CPUState *env, target_ulong addr)
2205 {
2206 }
2207
2208 /*
2209 * Walks guest process memory "regions" one by one
2210 * and calls callback function 'fn' for each region.
2211 */
2212
2213 struct walk_memory_regions_data
2214 {
2215 walk_memory_regions_fn fn;
2216 void *priv;
2217 unsigned long start;
2218 int prot;
2219 };
2220
2221 static int walk_memory_regions_end(struct walk_memory_regions_data *data,
2222 abi_ulong end, int new_prot)
2223 {
2224 if (data->start != -1ul) {
2225 int rc = data->fn(data->priv, data->start, end, data->prot);
2226 if (rc != 0) {
2227 return rc;
2228 }
2229 }
2230
2231 data->start = (new_prot ? end : -1ul);
2232 data->prot = new_prot;
2233
2234 return 0;
2235 }
2236
2237 static int walk_memory_regions_1(struct walk_memory_regions_data *data,
2238 abi_ulong base, int level, void **lp)
2239 {
2240 abi_ulong pa;
2241 int i, rc;
2242
2243 if (*lp == NULL) {
2244 return walk_memory_regions_end(data, base, 0);
2245 }
2246
2247 if (level == 0) {
2248 PageDesc *pd = *lp;
2249 for (i = 0; i < L2_SIZE; ++i) {
2250 int prot = pd[i].flags;
2251
2252 pa = base | (i << TARGET_PAGE_BITS);
2253 if (prot != data->prot) {
2254 rc = walk_memory_regions_end(data, pa, prot);
2255 if (rc != 0) {
2256 return rc;
2257 }
2258 }
2259 }
2260 } else {
2261 void **pp = *lp;
2262 for (i = 0; i < L2_SIZE; ++i) {
2263 pa = base | ((abi_ulong)i <<
2264 (TARGET_PAGE_BITS + L2_BITS * level));
2265 rc = walk_memory_regions_1(data, pa, level - 1, pp + i);
2266 if (rc != 0) {
2267 return rc;
2268 }
2269 }
2270 }
2271
2272 return 0;
2273 }
2274
2275 int walk_memory_regions(void *priv, walk_memory_regions_fn fn)
2276 {
2277 struct walk_memory_regions_data data;
2278 unsigned long i;
2279
2280 data.fn = fn;
2281 data.priv = priv;
2282 data.start = -1ul;
2283 data.prot = 0;
2284
2285 for (i = 0; i < V_L1_SIZE; i++) {
2286 int rc = walk_memory_regions_1(&data, (abi_ulong)i << V_L1_SHIFT,
2287 V_L1_SHIFT / L2_BITS - 1, l1_map + i);
2288 if (rc != 0) {
2289 return rc;
2290 }
2291 }
2292
2293 return walk_memory_regions_end(&data, 0, 0);
2294 }
2295
2296 static int dump_region(void *priv, abi_ulong start,
2297 abi_ulong end, unsigned long prot)
2298 {
2299 FILE *f = (FILE *)priv;
2300
2301 (void) fprintf(f, TARGET_ABI_FMT_lx"-"TARGET_ABI_FMT_lx
2302 " "TARGET_ABI_FMT_lx" %c%c%c\n",
2303 start, end, end - start,
2304 ((prot & PAGE_READ) ? 'r' : '-'),
2305 ((prot & PAGE_WRITE) ? 'w' : '-'),
2306 ((prot & PAGE_EXEC) ? 'x' : '-'));
2307
2308 return (0);
2309 }
2310
2311 /* dump memory mappings */
2312 void page_dump(FILE *f)
2313 {
2314 (void) fprintf(f, "%-8s %-8s %-8s %s\n",
2315 "start", "end", "size", "prot");
2316 walk_memory_regions(f, dump_region);
2317 }
2318
2319 int page_get_flags(target_ulong address)
2320 {
2321 PageDesc *p;
2322
2323 p = page_find(address >> TARGET_PAGE_BITS);
2324 if (!p)
2325 return 0;
2326 return p->flags;
2327 }
2328
2329 /* Modify the flags of a page and invalidate the code if necessary.
2330 The flag PAGE_WRITE_ORG is positioned automatically depending
2331 on PAGE_WRITE. The mmap_lock should already be held. */
2332 void page_set_flags(target_ulong start, target_ulong end, int flags)
2333 {
2334 target_ulong addr, len;
2335
2336 /* This function should never be called with addresses outside the
2337 guest address space. If this assert fires, it probably indicates
2338 a missing call to h2g_valid. */
2339 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
2340 assert(end < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
2341 #endif
2342 assert(start < end);
2343
2344 start = start & TARGET_PAGE_MASK;
2345 end = TARGET_PAGE_ALIGN(end);
2346
2347 if (flags & PAGE_WRITE) {
2348 flags |= PAGE_WRITE_ORG;
2349 }
2350
2351 for (addr = start, len = end - start;
2352 len != 0;
2353 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
2354 PageDesc *p = page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
2355
2356 /* If the write protection bit is set, then we invalidate
2357 the code inside. */
2358 if (!(p->flags & PAGE_WRITE) &&
2359 (flags & PAGE_WRITE) &&
2360 p->first_tb) {
2361 tb_invalidate_phys_page(addr, 0, NULL);
2362 }
2363 p->flags = flags;
2364 }
2365 }
2366
2367 int page_check_range(target_ulong start, target_ulong len, int flags)
2368 {
2369 PageDesc *p;
2370 target_ulong end;
2371 target_ulong addr;
2372
2373 /* This function should never be called with addresses outside the
2374 guest address space. If this assert fires, it probably indicates
2375 a missing call to h2g_valid. */
2376 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
2377 assert(start < ((abi_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
2378 #endif
2379
2380 if (len == 0) {
2381 return 0;
2382 }
2383 if (start + len - 1 < start) {
2384 /* We've wrapped around. */
2385 return -1;
2386 }
2387
2388 end = TARGET_PAGE_ALIGN(start+len); /* must do before we loose bits in the next step */
2389 start = start & TARGET_PAGE_MASK;
2390
2391 for (addr = start, len = end - start;
2392 len != 0;
2393 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
2394 p = page_find(addr >> TARGET_PAGE_BITS);
2395 if( !p )
2396 return -1;
2397 if( !(p->flags & PAGE_VALID) )
2398 return -1;
2399
2400 if ((flags & PAGE_READ) && !(p->flags & PAGE_READ))
2401 return -1;
2402 if (flags & PAGE_WRITE) {
2403 if (!(p->flags & PAGE_WRITE_ORG))
2404 return -1;
2405 /* unprotect the page if it was put read-only because it
2406 contains translated code */
2407 if (!(p->flags & PAGE_WRITE)) {
2408 if (!page_unprotect(addr, 0, NULL))
2409 return -1;
2410 }
2411 return 0;
2412 }
2413 }
2414 return 0;
2415 }
2416
2417 /* called from signal handler: invalidate the code and unprotect the
2418 page. Return TRUE if the fault was successfully handled. */
2419 int page_unprotect(target_ulong address, unsigned long pc, void *puc)
2420 {
2421 unsigned int prot;
2422 PageDesc *p;
2423 target_ulong host_start, host_end, addr;
2424
2425 /* Technically this isn't safe inside a signal handler. However we
2426 know this only ever happens in a synchronous SEGV handler, so in
2427 practice it seems to be ok. */
2428 mmap_lock();
2429
2430 p = page_find(address >> TARGET_PAGE_BITS);
2431 if (!p) {
2432 mmap_unlock();
2433 return 0;
2434 }
2435
2436 /* if the page was really writable, then we change its
2437 protection back to writable */
2438 if ((p->flags & PAGE_WRITE_ORG) && !(p->flags & PAGE_WRITE)) {
2439 host_start = address & qemu_host_page_mask;
2440 host_end = host_start + qemu_host_page_size;
2441
2442 prot = 0;
2443 for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) {
2444 p = page_find(addr >> TARGET_PAGE_BITS);
2445 p->flags |= PAGE_WRITE;
2446 prot |= p->flags;
2447
2448 /* and since the content will be modified, we must invalidate
2449 the corresponding translated code. */
2450 tb_invalidate_phys_page(addr, pc, puc);
2451 #ifdef DEBUG_TB_CHECK
2452 tb_invalidate_check(addr);
2453 #endif
2454 }
2455 mprotect((void *)g2h(host_start), qemu_host_page_size,
2456 prot & PAGE_BITS);
2457
2458 mmap_unlock();
2459 return 1;
2460 }
2461 mmap_unlock();
2462 return 0;
2463 }
2464
2465 static inline void tlb_set_dirty(CPUState *env,
2466 unsigned long addr, target_ulong vaddr)
2467 {
2468 }
2469 #endif /* defined(CONFIG_USER_ONLY) */
2470
2471 #if !defined(CONFIG_USER_ONLY)
2472
2473 #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
2474 typedef struct subpage_t {
2475 MemoryRegion iomem;
2476 target_phys_addr_t base;
2477 ram_addr_t sub_io_index[TARGET_PAGE_SIZE];
2478 ram_addr_t region_offset[TARGET_PAGE_SIZE];
2479 } subpage_t;
2480
2481 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
2482 ram_addr_t memory, ram_addr_t region_offset);
2483 static subpage_t *subpage_init (target_phys_addr_t base, ram_addr_t *phys,
2484 ram_addr_t orig_memory,
2485 ram_addr_t region_offset);
2486 #define CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr, end_addr2, \
2487 need_subpage) \
2488 do { \
2489 if (addr > start_addr) \
2490 start_addr2 = 0; \
2491 else { \
2492 start_addr2 = start_addr & ~TARGET_PAGE_MASK; \
2493 if (start_addr2 > 0) \
2494 need_subpage = 1; \
2495 } \
2496 \
2497 if ((start_addr + orig_size) - addr >= TARGET_PAGE_SIZE) \
2498 end_addr2 = TARGET_PAGE_SIZE - 1; \
2499 else { \
2500 end_addr2 = (start_addr + orig_size - 1) & ~TARGET_PAGE_MASK; \
2501 if (end_addr2 < TARGET_PAGE_SIZE - 1) \
2502 need_subpage = 1; \
2503 } \
2504 } while (0)
2505
2506 /* register physical memory.
2507 For RAM, 'size' must be a multiple of the target page size.
2508 If (phys_offset & ~TARGET_PAGE_MASK) != 0, then it is an
2509 io memory page. The address used when calling the IO function is
2510 the offset from the start of the region, plus region_offset. Both
2511 start_addr and region_offset are rounded down to a page boundary
2512 before calculating this offset. This should not be a problem unless
2513 the low bits of start_addr and region_offset differ. */
2514 void cpu_register_physical_memory_log(MemoryRegionSection *section,
2515 bool readable, bool readonly)
2516 {
2517 target_phys_addr_t start_addr = section->offset_within_address_space;
2518 ram_addr_t size = section->size;
2519 ram_addr_t phys_offset = section->mr->ram_addr;
2520 ram_addr_t region_offset = section->offset_within_region;
2521 target_phys_addr_t addr, end_addr;
2522 PhysPageDesc *p;
2523 CPUState *env;
2524 ram_addr_t orig_size = size;
2525 subpage_t *subpage;
2526
2527 if (memory_region_is_ram(section->mr)) {
2528 phys_offset += region_offset;
2529 region_offset = 0;
2530 }
2531
2532 if (!readable) {
2533 phys_offset &= ~TARGET_PAGE_MASK & ~IO_MEM_ROMD;
2534 }
2535
2536 if (readonly) {
2537 phys_offset |= io_mem_rom.ram_addr;
2538 }
2539
2540 assert(size);
2541
2542 if (phys_offset == io_mem_unassigned.ram_addr) {
2543 region_offset = start_addr;
2544 }
2545 region_offset &= TARGET_PAGE_MASK;
2546 size = (size + TARGET_PAGE_SIZE - 1) & TARGET_PAGE_MASK;
2547 end_addr = start_addr + (target_phys_addr_t)size;
2548
2549 addr = start_addr;
2550 do {
2551 p = phys_page_find_alloc(addr >> TARGET_PAGE_BITS, 0);
2552 if (p && p->phys_offset != io_mem_unassigned.ram_addr) {
2553 ram_addr_t orig_memory = p->phys_offset;
2554 target_phys_addr_t start_addr2, end_addr2;
2555 int need_subpage = 0;
2556
2557 CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr, end_addr2,
2558 need_subpage);
2559 if (need_subpage) {
2560 if (!(orig_memory & IO_MEM_SUBPAGE)) {
2561 subpage = subpage_init((addr & TARGET_PAGE_MASK),
2562 &p->phys_offset, orig_memory,
2563 p->region_offset);
2564 } else {
2565 subpage = io_mem_opaque[(orig_memory & ~TARGET_PAGE_MASK)
2566 >> IO_MEM_SHIFT];
2567 }
2568 subpage_register(subpage, start_addr2, end_addr2, phys_offset,
2569 region_offset);
2570 p->region_offset = 0;
2571 } else {
2572 p->phys_offset = phys_offset;
2573 p->region_offset = region_offset;
2574 if (is_ram_rom_romd(phys_offset))
2575 phys_offset += TARGET_PAGE_SIZE;
2576 }
2577 } else {
2578 p = phys_page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
2579 p->phys_offset = phys_offset;
2580 p->region_offset = region_offset;
2581 if (is_ram_rom_romd(phys_offset)) {
2582 phys_offset += TARGET_PAGE_SIZE;
2583 } else {
2584 target_phys_addr_t start_addr2, end_addr2;
2585 int need_subpage = 0;
2586
2587 CHECK_SUBPAGE(addr, start_addr, start_addr2, end_addr,
2588 end_addr2, need_subpage);
2589
2590 if (need_subpage) {
2591 subpage = subpage_init((addr & TARGET_PAGE_MASK),
2592 &p->phys_offset,
2593 io_mem_unassigned.ram_addr,
2594 addr & TARGET_PAGE_MASK);
2595 subpage_register(subpage, start_addr2, end_addr2,
2596 phys_offset, region_offset);
2597 p->region_offset = 0;
2598 }
2599 }
2600 }
2601 region_offset += TARGET_PAGE_SIZE;
2602 addr += TARGET_PAGE_SIZE;
2603 } while (addr != end_addr);
2604
2605 /* since each CPU stores ram addresses in its TLB cache, we must
2606 reset the modified entries */
2607 /* XXX: slow ! */
2608 for(env = first_cpu; env != NULL; env = env->next_cpu) {
2609 tlb_flush(env, 1);
2610 }
2611 }
2612
2613 void qemu_register_coalesced_mmio(target_phys_addr_t addr, ram_addr_t size)
2614 {
2615 if (kvm_enabled())
2616 kvm_coalesce_mmio_region(addr, size);
2617 }
2618
2619 void qemu_unregister_coalesced_mmio(target_phys_addr_t addr, ram_addr_t size)
2620 {
2621 if (kvm_enabled())
2622 kvm_uncoalesce_mmio_region(addr, size);
2623 }
2624
2625 void qemu_flush_coalesced_mmio_buffer(void)
2626 {
2627 if (kvm_enabled())
2628 kvm_flush_coalesced_mmio_buffer();
2629 }
2630
2631 #if defined(__linux__) && !defined(TARGET_S390X)
2632
2633 #include <sys/vfs.h>
2634
2635 #define HUGETLBFS_MAGIC 0x958458f6
2636
2637 static long gethugepagesize(const char *path)
2638 {
2639 struct statfs fs;
2640 int ret;
2641
2642 do {
2643 ret = statfs(path, &fs);
2644 } while (ret != 0 && errno == EINTR);
2645
2646 if (ret != 0) {
2647 perror(path);
2648 return 0;
2649 }
2650
2651 if (fs.f_type != HUGETLBFS_MAGIC)
2652 fprintf(stderr, "Warning: path not on HugeTLBFS: %s\n", path);
2653
2654 return fs.f_bsize;
2655 }
2656
2657 static void *file_ram_alloc(RAMBlock *block,
2658 ram_addr_t memory,
2659 const char *path)
2660 {
2661 char *filename;
2662 void *area;
2663 int fd;
2664 #ifdef MAP_POPULATE
2665 int flags;
2666 #endif
2667 unsigned long hpagesize;
2668
2669 hpagesize = gethugepagesize(path);
2670 if (!hpagesize) {
2671 return NULL;
2672 }
2673
2674 if (memory < hpagesize) {
2675 return NULL;
2676 }
2677
2678 if (kvm_enabled() && !kvm_has_sync_mmu()) {
2679 fprintf(stderr, "host lacks kvm mmu notifiers, -mem-path unsupported\n");
2680 return NULL;
2681 }
2682
2683 if (asprintf(&filename, "%s/qemu_back_mem.XXXXXX", path) == -1) {
2684 return NULL;
2685 }
2686
2687 fd = mkstemp(filename);
2688 if (fd < 0) {
2689 perror("unable to create backing store for hugepages");
2690 free(filename);
2691 return NULL;
2692 }
2693 unlink(filename);
2694 free(filename);
2695
2696 memory = (memory+hpagesize-1) & ~(hpagesize-1);
2697
2698 /*
2699 * ftruncate is not supported by hugetlbfs in older
2700 * hosts, so don't bother bailing out on errors.
2701 * If anything goes wrong with it under other filesystems,
2702 * mmap will fail.
2703 */
2704 if (ftruncate(fd, memory))
2705 perror("ftruncate");
2706
2707 #ifdef MAP_POPULATE
2708 /* NB: MAP_POPULATE won't exhaustively alloc all phys pages in the case
2709 * MAP_PRIVATE is requested. For mem_prealloc we mmap as MAP_SHARED
2710 * to sidestep this quirk.
2711 */
2712 flags = mem_prealloc ? MAP_POPULATE | MAP_SHARED : MAP_PRIVATE;
2713 area = mmap(0, memory, PROT_READ | PROT_WRITE, flags, fd, 0);
2714 #else
2715 area = mmap(0, memory, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
2716 #endif
2717 if (area == MAP_FAILED) {
2718 perror("file_ram_alloc: can't mmap RAM pages");
2719 close(fd);
2720 return (NULL);
2721 }
2722 block->fd = fd;
2723 return area;
2724 }
2725 #endif
2726
2727 static ram_addr_t find_ram_offset(ram_addr_t size)
2728 {
2729 RAMBlock *block, *next_block;
2730 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
2731
2732 if (QLIST_EMPTY(&ram_list.blocks))
2733 return 0;
2734
2735 QLIST_FOREACH(block, &ram_list.blocks, next) {
2736 ram_addr_t end, next = RAM_ADDR_MAX;
2737
2738 end = block->offset + block->length;
2739
2740 QLIST_FOREACH(next_block, &ram_list.blocks, next) {
2741 if (next_block->offset >= end) {
2742 next = MIN(next, next_block->offset);
2743 }
2744 }
2745 if (next - end >= size && next - end < mingap) {
2746 offset = end;
2747 mingap = next - end;
2748 }
2749 }
2750
2751 if (offset == RAM_ADDR_MAX) {
2752 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
2753 (uint64_t)size);
2754 abort();
2755 }
2756
2757 return offset;
2758 }
2759
2760 static ram_addr_t last_ram_offset(void)
2761 {
2762 RAMBlock *block;
2763 ram_addr_t last = 0;
2764
2765 QLIST_FOREACH(block, &ram_list.blocks, next)
2766 last = MAX(last, block->offset + block->length);
2767
2768 return last;
2769 }
2770
2771 void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
2772 {
2773 RAMBlock *new_block, *block;
2774
2775 new_block = NULL;
2776 QLIST_FOREACH(block, &ram_list.blocks, next) {
2777 if (block->offset == addr) {
2778 new_block = block;
2779 break;
2780 }
2781 }
2782 assert(new_block);
2783 assert(!new_block->idstr[0]);
2784
2785 if (dev && dev->parent_bus && dev->parent_bus->info->get_dev_path) {
2786 char *id = dev->parent_bus->info->get_dev_path(dev);
2787 if (id) {
2788 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
2789 g_free(id);
2790 }
2791 }
2792 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
2793
2794 QLIST_FOREACH(block, &ram_list.blocks, next) {
2795 if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
2796 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
2797 new_block->idstr);
2798 abort();
2799 }
2800 }
2801 }
2802
2803 ram_addr_t qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
2804 MemoryRegion *mr)
2805 {
2806 RAMBlock *new_block;
2807
2808 size = TARGET_PAGE_ALIGN(size);
2809 new_block = g_malloc0(sizeof(*new_block));
2810
2811 new_block->mr = mr;
2812 new_block->offset = find_ram_offset(size);
2813 if (host) {
2814 new_block->host = host;
2815 new_block->flags |= RAM_PREALLOC_MASK;
2816 } else {
2817 if (mem_path) {
2818 #if defined (__linux__) && !defined(TARGET_S390X)
2819 new_block->host = file_ram_alloc(new_block, size, mem_path);
2820 if (!new_block->host) {
2821 new_block->host = qemu_vmalloc(size);
2822 qemu_madvise(new_block->host, size, QEMU_MADV_MERGEABLE);
2823 }
2824 #else
2825 fprintf(stderr, "-mem-path option unsupported\n");
2826 exit(1);
2827 #endif
2828 } else {
2829 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2830 /* S390 KVM requires the topmost vma of the RAM to be smaller than
2831 an system defined value, which is at least 256GB. Larger systems
2832 have larger values. We put the guest between the end of data
2833 segment (system break) and this value. We use 32GB as a base to
2834 have enough room for the system break to grow. */
2835 new_block->host = mmap((void*)0x800000000, size,
2836 PROT_EXEC|PROT_READ|PROT_WRITE,
2837 MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
2838 if (new_block->host == MAP_FAILED) {
2839 fprintf(stderr, "Allocating RAM failed\n");
2840 abort();
2841 }
2842 #else
2843 if (xen_enabled()) {
2844 xen_ram_alloc(new_block->offset, size, mr);
2845 } else {
2846 new_block->host = qemu_vmalloc(size);
2847 }
2848 #endif
2849 qemu_madvise(new_block->host, size, QEMU_MADV_MERGEABLE);
2850 }
2851 }
2852 new_block->length = size;
2853
2854 QLIST_INSERT_HEAD(&ram_list.blocks, new_block, next);
2855
2856 ram_list.phys_dirty = g_realloc(ram_list.phys_dirty,
2857 last_ram_offset() >> TARGET_PAGE_BITS);
2858 memset(ram_list.phys_dirty + (new_block->offset >> TARGET_PAGE_BITS),
2859 0xff, size >> TARGET_PAGE_BITS);
2860
2861 if (kvm_enabled())
2862 kvm_setup_guest_memory(new_block->host, size);
2863
2864 return new_block->offset;
2865 }
2866
2867 ram_addr_t qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr)
2868 {
2869 return qemu_ram_alloc_from_ptr(size, NULL, mr);
2870 }
2871
2872 void qemu_ram_free_from_ptr(ram_addr_t addr)
2873 {
2874 RAMBlock *block;
2875
2876 QLIST_FOREACH(block, &ram_list.blocks, next) {
2877 if (addr == block->offset) {
2878 QLIST_REMOVE(block, next);
2879 g_free(block);
2880 return;
2881 }
2882 }
2883 }
2884
2885 void qemu_ram_free(ram_addr_t addr)
2886 {
2887 RAMBlock *block;
2888
2889 QLIST_FOREACH(block, &ram_list.blocks, next) {
2890 if (addr == block->offset) {
2891 QLIST_REMOVE(block, next);
2892 if (block->flags & RAM_PREALLOC_MASK) {
2893 ;
2894 } else if (mem_path) {
2895 #if defined (__linux__) && !defined(TARGET_S390X)
2896 if (block->fd) {
2897 munmap(block->host, block->length);
2898 close(block->fd);
2899 } else {
2900 qemu_vfree(block->host);
2901 }
2902 #else
2903 abort();
2904 #endif
2905 } else {
2906 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2907 munmap(block->host, block->length);
2908 #else
2909 if (xen_enabled()) {
2910 xen_invalidate_map_cache_entry(block->host);
2911 } else {
2912 qemu_vfree(block->host);
2913 }
2914 #endif
2915 }
2916 g_free(block);
2917 return;
2918 }
2919 }
2920
2921 }
2922
2923 #ifndef _WIN32
2924 void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
2925 {
2926 RAMBlock *block;
2927 ram_addr_t offset;
2928 int flags;
2929 void *area, *vaddr;
2930
2931 QLIST_FOREACH(block, &ram_list.blocks, next) {
2932 offset = addr - block->offset;
2933 if (offset < block->length) {
2934 vaddr = block->host + offset;
2935 if (block->flags & RAM_PREALLOC_MASK) {
2936 ;
2937 } else {
2938 flags = MAP_FIXED;
2939 munmap(vaddr, length);
2940 if (mem_path) {
2941 #if defined(__linux__) && !defined(TARGET_S390X)
2942 if (block->fd) {
2943 #ifdef MAP_POPULATE
2944 flags |= mem_prealloc ? MAP_POPULATE | MAP_SHARED :
2945 MAP_PRIVATE;
2946 #else
2947 flags |= MAP_PRIVATE;
2948 #endif
2949 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2950 flags, block->fd, offset);
2951 } else {
2952 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2953 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2954 flags, -1, 0);
2955 }
2956 #else
2957 abort();
2958 #endif
2959 } else {
2960 #if defined(TARGET_S390X) && defined(CONFIG_KVM)
2961 flags |= MAP_SHARED | MAP_ANONYMOUS;
2962 area = mmap(vaddr, length, PROT_EXEC|PROT_READ|PROT_WRITE,
2963 flags, -1, 0);
2964 #else
2965 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2966 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2967 flags, -1, 0);
2968 #endif
2969 }
2970 if (area != vaddr) {
2971 fprintf(stderr, "Could not remap addr: "
2972 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
2973 length, addr);
2974 exit(1);
2975 }
2976 qemu_madvise(vaddr, length, QEMU_MADV_MERGEABLE);
2977 }
2978 return;
2979 }
2980 }
2981 }
2982 #endif /* !_WIN32 */
2983
2984 /* Return a host pointer to ram allocated with qemu_ram_alloc.
2985 With the exception of the softmmu code in this file, this should
2986 only be used for local memory (e.g. video ram) that the device owns,
2987 and knows it isn't going to access beyond the end of the block.
2988
2989 It should not be used for general purpose DMA.
2990 Use cpu_physical_memory_map/cpu_physical_memory_rw instead.
2991 */
2992 void *qemu_get_ram_ptr(ram_addr_t addr)
2993 {
2994 RAMBlock *block;
2995
2996 QLIST_FOREACH(block, &ram_list.blocks, next) {
2997 if (addr - block->offset < block->length) {
2998 /* Move this entry to to start of the list. */
2999 if (block != QLIST_FIRST(&ram_list.blocks)) {
3000 QLIST_REMOVE(block, next);
3001 QLIST_INSERT_HEAD(&ram_list.blocks, block, next);
3002 }
3003 if (xen_enabled()) {
3004 /* We need to check if the requested address is in the RAM
3005 * because we don't want to map the entire memory in QEMU.
3006 * In that case just map until the end of the page.
3007 */
3008 if (block->offset == 0) {
3009 return xen_map_cache(addr, 0, 0);
3010 } else if (block->host == NULL) {
3011 block->host =
3012 xen_map_cache(block->offset, block->length, 1);
3013 }
3014 }
3015 return block->host + (addr - block->offset);
3016 }
3017 }
3018
3019 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
3020 abort();
3021
3022 return NULL;
3023 }
3024
3025 /* Return a host pointer to ram allocated with qemu_ram_alloc.
3026 * Same as qemu_get_ram_ptr but avoid reordering ramblocks.
3027 */
3028 void *qemu_safe_ram_ptr(ram_addr_t addr)
3029 {
3030 RAMBlock *block;
3031
3032 QLIST_FOREACH(block, &ram_list.blocks, next) {
3033 if (addr - block->offset < block->length) {
3034 if (xen_enabled()) {
3035 /* We need to check if the requested address is in the RAM
3036 * because we don't want to map the entire memory in QEMU.
3037 * In that case just map until the end of the page.
3038 */
3039 if (block->offset == 0) {
3040 return xen_map_cache(addr, 0, 0);
3041 } else if (block->host == NULL) {
3042 block->host =
3043 xen_map_cache(block->offset, block->length, 1);
3044 }
3045 }
3046 return block->host + (addr - block->offset);
3047 }
3048 }
3049
3050 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
3051 abort();
3052
3053 return NULL;
3054 }
3055
3056 /* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr
3057 * but takes a size argument */
3058 void *qemu_ram_ptr_length(ram_addr_t addr, ram_addr_t *size)
3059 {
3060 if (*size == 0) {
3061 return NULL;
3062 }
3063 if (xen_enabled()) {
3064 return xen_map_cache(addr, *size, 1);
3065 } else {
3066 RAMBlock *block;
3067
3068 QLIST_FOREACH(block, &ram_list.blocks, next) {
3069 if (addr - block->offset < block->length) {
3070 if (addr - block->offset + *size > block->length)
3071 *size = block->length - addr + block->offset;
3072 return block->host + (addr - block->offset);
3073 }
3074 }
3075
3076 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
3077 abort();
3078 }
3079 }
3080
3081 void qemu_put_ram_ptr(void *addr)
3082 {
3083 trace_qemu_put_ram_ptr(addr);
3084 }
3085
3086 int qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
3087 {
3088 RAMBlock *block;
3089 uint8_t *host = ptr;
3090
3091 if (xen_enabled()) {
3092 *ram_addr = xen_ram_addr_from_mapcache(ptr);
3093 return 0;
3094 }
3095
3096 QLIST_FOREACH(block, &ram_list.blocks, next) {
3097 /* This case append when the block is not mapped. */
3098 if (block->host == NULL) {
3099 continue;
3100 }
3101 if (host - block->host < block->length) {
3102 *ram_addr = block->offset + (host - block->host);
3103 return 0;
3104 }
3105 }
3106
3107 return -1;
3108 }
3109
3110 /* Some of the softmmu routines need to translate from a host pointer
3111 (typically a TLB entry) back to a ram offset. */
3112 ram_addr_t qemu_ram_addr_from_host_nofail(void *ptr)
3113 {
3114 ram_addr_t ram_addr;
3115
3116 if (qemu_ram_addr_from_host(ptr, &ram_addr)) {
3117 fprintf(stderr, "Bad ram pointer %p\n", ptr);
3118 abort();
3119 }
3120 return ram_addr;
3121 }
3122
3123 static uint64_t unassigned_mem_read(void *opaque, target_phys_addr_t addr,
3124 unsigned size)
3125 {
3126 #ifdef DEBUG_UNASSIGNED
3127 printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
3128 #endif
3129 #if defined(TARGET_ALPHA) || defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3130 cpu_unassigned_access(cpu_single_env, addr, 0, 0, 0, size);
3131 #endif
3132 return 0;
3133 }
3134
3135 static void unassigned_mem_write(void *opaque, target_phys_addr_t addr,
3136 uint64_t val, unsigned size)
3137 {
3138 #ifdef DEBUG_UNASSIGNED
3139 printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
3140 #endif
3141 #if defined(TARGET_ALPHA) || defined(TARGET_SPARC) || defined(TARGET_MICROBLAZE)
3142 cpu_unassigned_access(cpu_single_env, addr, 1, 0, 0, size);
3143 #endif
3144 }
3145
3146 static const MemoryRegionOps unassigned_mem_ops = {
3147 .read = unassigned_mem_read,
3148 .write = unassigned_mem_write,
3149 .endianness = DEVICE_NATIVE_ENDIAN,
3150 };
3151
3152 static uint64_t error_mem_read(void *opaque, target_phys_addr_t addr,
3153 unsigned size)
3154 {
3155 abort();
3156 }
3157
3158 static void error_mem_write(void *opaque, target_phys_addr_t addr,
3159 uint64_t value, unsigned size)
3160 {
3161 abort();
3162 }
3163
3164 static const MemoryRegionOps error_mem_ops = {
3165 .read = error_mem_read,
3166 .write = error_mem_write,
3167 .endianness = DEVICE_NATIVE_ENDIAN,
3168 };
3169
3170 static const MemoryRegionOps rom_mem_ops = {
3171 .read = error_mem_read,
3172 .write = unassigned_mem_write,
3173 .endianness = DEVICE_NATIVE_ENDIAN,
3174 };
3175
3176 static void notdirty_mem_write(void *opaque, target_phys_addr_t ram_addr,
3177 uint64_t val, unsigned size)
3178 {
3179 int dirty_flags;
3180 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3181 if (!(dirty_flags & CODE_DIRTY_FLAG)) {
3182 #if !defined(CONFIG_USER_ONLY)
3183 tb_invalidate_phys_page_fast(ram_addr, size);
3184 dirty_flags = cpu_physical_memory_get_dirty_flags(ram_addr);
3185 #endif
3186 }
3187 switch (size) {
3188 case 1:
3189 stb_p(qemu_get_ram_ptr(ram_addr), val);
3190 break;
3191 case 2:
3192 stw_p(qemu_get_ram_ptr(ram_addr), val);
3193 break;
3194 case 4:
3195 stl_p(qemu_get_ram_ptr(ram_addr), val);
3196 break;
3197 default:
3198 abort();
3199 }
3200 dirty_flags |= (0xff & ~CODE_DIRTY_FLAG);
3201 cpu_physical_memory_set_dirty_flags(ram_addr, dirty_flags);
3202 /* we remove the notdirty callback only if the code has been
3203 flushed */
3204 if (dirty_flags == 0xff)
3205 tlb_set_dirty(cpu_single_env, cpu_single_env->mem_io_vaddr);
3206 }
3207
3208 static const MemoryRegionOps notdirty_mem_ops = {
3209 .read = error_mem_read,
3210 .write = notdirty_mem_write,
3211 .endianness = DEVICE_NATIVE_ENDIAN,
3212 };
3213
3214 /* Generate a debug exception if a watchpoint has been hit. */
3215 static void check_watchpoint(int offset, int len_mask, int flags)
3216 {
3217 CPUState *env = cpu_single_env;
3218 target_ulong pc, cs_base;
3219 TranslationBlock *tb;
3220 target_ulong vaddr;
3221 CPUWatchpoint *wp;
3222 int cpu_flags;
3223
3224 if (env->watchpoint_hit) {
3225 /* We re-entered the check after replacing the TB. Now raise
3226 * the debug interrupt so that is will trigger after the
3227 * current instruction. */
3228 cpu_interrupt(env, CPU_INTERRUPT_DEBUG);
3229 return;
3230 }
3231 vaddr = (env->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
3232 QTAILQ_FOREACH(wp, &env->watchpoints, entry) {
3233 if ((vaddr == (wp->vaddr & len_mask) ||
3234 (vaddr & wp->len_mask) == wp->vaddr) && (wp->flags & flags)) {
3235 wp->flags |= BP_WATCHPOINT_HIT;
3236 if (!env->watchpoint_hit) {
3237 env->watchpoint_hit = wp;
3238 tb = tb_find_pc(env->mem_io_pc);
3239 if (!tb) {
3240 cpu_abort(env, "check_watchpoint: could not find TB for "
3241 "pc=%p", (void *)env->mem_io_pc);
3242 }
3243 cpu_restore_state(tb, env, env->mem_io_pc);
3244 tb_phys_invalidate(tb, -1);
3245 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
3246 env->exception_index = EXCP_DEBUG;
3247 } else {
3248 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
3249 tb_gen_code(env, pc, cs_base, cpu_flags, 1);
3250 }
3251 cpu_resume_from_signal(env, NULL);
3252 }
3253 } else {
3254 wp->flags &= ~BP_WATCHPOINT_HIT;
3255 }
3256 }
3257 }
3258
3259 /* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
3260 so these check for a hit then pass through to the normal out-of-line
3261 phys routines. */
3262 static uint32_t watch_mem_readb(void *opaque, target_phys_addr_t addr)
3263 {
3264 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x0, BP_MEM_READ);
3265 return ldub_phys(addr);
3266 }
3267
3268 static uint32_t watch_mem_readw(void *opaque, target_phys_addr_t addr)
3269 {
3270 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x1, BP_MEM_READ);
3271 return lduw_phys(addr);
3272 }
3273
3274 static uint32_t watch_mem_readl(void *opaque, target_phys_addr_t addr)
3275 {
3276 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x3, BP_MEM_READ);
3277 return ldl_phys(addr);
3278 }
3279
3280 static void watch_mem_writeb(void *opaque, target_phys_addr_t addr,
3281 uint32_t val)
3282 {
3283 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x0, BP_MEM_WRITE);
3284 stb_phys(addr, val);
3285 }
3286
3287 static void watch_mem_writew(void *opaque, target_phys_addr_t addr,
3288 uint32_t val)
3289 {
3290 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x1, BP_MEM_WRITE);
3291 stw_phys(addr, val);
3292 }
3293
3294 static void watch_mem_writel(void *opaque, target_phys_addr_t addr,
3295 uint32_t val)
3296 {
3297 check_watchpoint(addr & ~TARGET_PAGE_MASK, ~0x3, BP_MEM_WRITE);
3298 stl_phys(addr, val);
3299 }
3300
3301 static CPUReadMemoryFunc * const watch_mem_read[3] = {
3302 watch_mem_readb,
3303 watch_mem_readw,
3304 watch_mem_readl,
3305 };
3306
3307 static CPUWriteMemoryFunc * const watch_mem_write[3] = {
3308 watch_mem_writeb,
3309 watch_mem_writew,
3310 watch_mem_writel,
3311 };
3312
3313 static uint64_t subpage_read(void *opaque, target_phys_addr_t addr,
3314 unsigned len)
3315 {
3316 subpage_t *mmio = opaque;
3317 unsigned int idx = SUBPAGE_IDX(addr);
3318 #if defined(DEBUG_SUBPAGE)
3319 printf("%s: subpage %p len %d addr " TARGET_FMT_plx " idx %d\n", __func__,
3320 mmio, len, addr, idx);
3321 #endif
3322
3323 addr += mmio->region_offset[idx];
3324 idx = mmio->sub_io_index[idx];
3325 return io_mem_read(idx, addr, len);
3326 }
3327
3328 static void subpage_write(void *opaque, target_phys_addr_t addr,
3329 uint64_t value, unsigned len)
3330 {
3331 subpage_t *mmio = opaque;
3332 unsigned int idx = SUBPAGE_IDX(addr);
3333 #if defined(DEBUG_SUBPAGE)
3334 printf("%s: subpage %p len %d addr " TARGET_FMT_plx
3335 " idx %d value %"PRIx64"\n",
3336 __func__, mmio, len, addr, idx, value);
3337 #endif
3338
3339 addr += mmio->region_offset[idx];
3340 idx = mmio->sub_io_index[idx];
3341 io_mem_write(idx, addr, value, len);
3342 }
3343
3344 static const MemoryRegionOps subpage_ops = {
3345 .read = subpage_read,
3346 .write = subpage_write,
3347 .endianness = DEVICE_NATIVE_ENDIAN,
3348 };
3349
3350 static uint32_t subpage_ram_readb(void *opaque, target_phys_addr_t addr)
3351 {
3352 ram_addr_t raddr = addr;
3353 void *ptr = qemu_get_ram_ptr(raddr);
3354 return ldub_p(ptr);
3355 }
3356
3357 static void subpage_ram_writeb(void *opaque, target_phys_addr_t addr,
3358 uint32_t value)
3359 {
3360 ram_addr_t raddr = addr;
3361 void *ptr = qemu_get_ram_ptr(raddr);
3362 stb_p(ptr, value);
3363 }
3364
3365 static uint32_t subpage_ram_readw(void *opaque, target_phys_addr_t addr)
3366 {
3367 ram_addr_t raddr = addr;
3368 void *ptr = qemu_get_ram_ptr(raddr);
3369 return lduw_p(ptr);
3370 }
3371
3372 static void subpage_ram_writew(void *opaque, target_phys_addr_t addr,
3373 uint32_t value)
3374 {
3375 ram_addr_t raddr = addr;
3376 void *ptr = qemu_get_ram_ptr(raddr);
3377 stw_p(ptr, value);
3378 }
3379
3380 static uint32_t subpage_ram_readl(void *opaque, target_phys_addr_t addr)
3381 {
3382 ram_addr_t raddr = addr;
3383 void *ptr = qemu_get_ram_ptr(raddr);
3384 return ldl_p(ptr);
3385 }
3386
3387 static void subpage_ram_writel(void *opaque, target_phys_addr_t addr,
3388 uint32_t value)
3389 {
3390 ram_addr_t raddr = addr;
3391 void *ptr = qemu_get_ram_ptr(raddr);
3392 stl_p(ptr, value);
3393 }
3394
3395 static CPUReadMemoryFunc * const subpage_ram_read[] = {
3396 &subpage_ram_readb,
3397 &subpage_ram_readw,
3398 &subpage_ram_readl,
3399 };
3400
3401 static CPUWriteMemoryFunc * const subpage_ram_write[] = {
3402 &subpage_ram_writeb,
3403 &subpage_ram_writew,
3404 &subpage_ram_writel,
3405 };
3406
3407 static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
3408 ram_addr_t memory, ram_addr_t region_offset)
3409 {
3410 int idx, eidx;
3411
3412 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
3413 return -1;
3414 idx = SUBPAGE_IDX(start);
3415 eidx = SUBPAGE_IDX(end);
3416 #if defined(DEBUG_SUBPAGE)
3417 printf("%s: %p start %08x end %08x idx %08x eidx %08x mem %ld\n", __func__,
3418 mmio, start, end, idx, eidx, memory);
3419 #endif
3420 if ((memory & ~TARGET_PAGE_MASK) == io_mem_ram.ram_addr) {
3421 memory = IO_MEM_SUBPAGE_RAM;
3422 }
3423 memory = (memory >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3424 for (; idx <= eidx; idx++) {
3425 mmio->sub_io_index[idx] = memory;
3426 mmio->region_offset[idx] = region_offset;
3427 }
3428
3429 return 0;
3430 }
3431
3432 static subpage_t *subpage_init (target_phys_addr_t base, ram_addr_t *phys,
3433 ram_addr_t orig_memory,
3434 ram_addr_t region_offset)
3435 {
3436 subpage_t *mmio;
3437 int subpage_memory;
3438
3439 mmio = g_malloc0(sizeof(subpage_t));
3440
3441 mmio->base = base;
3442 memory_region_init_io(&mmio->iomem, &subpage_ops, mmio,
3443 "subpage", TARGET_PAGE_SIZE);
3444 subpage_memory = mmio->iomem.ram_addr;
3445 #if defined(DEBUG_SUBPAGE)
3446 printf("%s: %p base " TARGET_FMT_plx " len %08x %d\n", __func__,
3447 mmio, base, TARGET_PAGE_SIZE, subpage_memory);
3448 #endif
3449 *phys = subpage_memory | IO_MEM_SUBPAGE;
3450 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, orig_memory, region_offset);
3451
3452 return mmio;
3453 }
3454
3455 static int get_free_io_mem_idx(void)
3456 {
3457 int i;
3458
3459 for (i = 0; i<IO_MEM_NB_ENTRIES; i++)
3460 if (!io_mem_used[i]) {
3461 io_mem_used[i] = 1;
3462 return i;
3463 }
3464 fprintf(stderr, "RAN out out io_mem_idx, max %d !\n", IO_MEM_NB_ENTRIES);
3465 return -1;
3466 }
3467
3468 /* mem_read and mem_write are arrays of functions containing the
3469 function to access byte (index 0), word (index 1) and dword (index
3470 2). Functions can be omitted with a NULL function pointer.
3471 If io_index is non zero, the corresponding io zone is
3472 modified. If it is zero, a new io zone is allocated. The return
3473 value can be used with cpu_register_physical_memory(). (-1) is
3474 returned if error. */
3475 static int cpu_register_io_memory_fixed(int io_index,
3476 CPUReadMemoryFunc * const *mem_read,
3477 CPUWriteMemoryFunc * const *mem_write,
3478 void *opaque)
3479 {
3480 int i;
3481
3482 if (io_index <= 0) {
3483 io_index = get_free_io_mem_idx();
3484 if (io_index == -1)
3485 return io_index;
3486 } else {
3487 io_index >>= IO_MEM_SHIFT;
3488 if (io_index >= IO_MEM_NB_ENTRIES)
3489 return -1;
3490 }
3491
3492 for (i = 0; i < 3; ++i) {
3493 assert(mem_read[i]);
3494 _io_mem_read[io_index][i] = mem_read[i];
3495 }
3496 for (i = 0; i < 3; ++i) {
3497 assert(mem_write[i]);
3498 _io_mem_write[io_index][i] = mem_write[i];
3499 }
3500 io_mem_opaque[io_index] = opaque;
3501
3502 return (io_index << IO_MEM_SHIFT);
3503 }
3504
3505 int cpu_register_io_memory(CPUReadMemoryFunc * const *mem_read,
3506 CPUWriteMemoryFunc * const *mem_write,
3507 void *opaque)
3508 {
3509 return cpu_register_io_memory_fixed(0, mem_read, mem_write, opaque);
3510 }
3511
3512 void cpu_unregister_io_memory(int io_table_address)
3513 {
3514 int i;
3515 int io_index = io_table_address >> IO_MEM_SHIFT;
3516
3517 for (i=0;i < 3; i++) {
3518 _io_mem_read[io_index][i] = NULL;
3519 _io_mem_write[io_index][i] = NULL;
3520 }
3521 io_mem_opaque[io_index] = NULL;
3522 io_mem_used[io_index] = 0;
3523 }
3524
3525 static void io_mem_init(void)
3526 {
3527 int i;
3528
3529 /* Must be first: */
3530 memory_region_init_io(&io_mem_ram, &error_mem_ops, NULL, "ram", UINT64_MAX);
3531 assert(io_mem_ram.ram_addr == 0);
3532 memory_region_init_io(&io_mem_rom, &rom_mem_ops, NULL, "rom", UINT64_MAX);
3533 memory_region_init_io(&io_mem_unassigned, &unassigned_mem_ops, NULL,
3534 "unassigned", UINT64_MAX);
3535 memory_region_init_io(&io_mem_notdirty, &notdirty_mem_ops, NULL,
3536 "notdirty", UINT64_MAX);
3537 cpu_register_io_memory_fixed(IO_MEM_SUBPAGE_RAM, subpage_ram_read,
3538 subpage_ram_write, NULL);
3539 for (i=0; i<5; i++)
3540 io_mem_used[i] = 1;
3541
3542 io_mem_watch = cpu_register_io_memory(watch_mem_read,
3543 watch_mem_write, NULL);
3544 }
3545
3546 static void memory_map_init(void)
3547 {
3548 system_memory = g_malloc(sizeof(*system_memory));
3549 memory_region_init(system_memory, "system", INT64_MAX);
3550 set_system_memory_map(system_memory);
3551
3552 system_io = g_malloc(sizeof(*system_io));
3553 memory_region_init(system_io, "io", 65536);
3554 set_system_io_map(system_io);
3555 }
3556
3557 MemoryRegion *get_system_memory(void)
3558 {
3559 return system_memory;
3560 }
3561
3562 MemoryRegion *get_system_io(void)
3563 {
3564 return system_io;
3565 }
3566
3567 #endif /* !defined(CONFIG_USER_ONLY) */
3568
3569 /* physical memory access (slow version, mainly for debug) */
3570 #if defined(CONFIG_USER_ONLY)
3571 int cpu_memory_rw_debug(CPUState *env, target_ulong addr,
3572 uint8_t *buf, int len, int is_write)
3573 {
3574 int l, flags;
3575 target_ulong page;
3576 void * p;
3577
3578 while (len > 0) {
3579 page = addr & TARGET_PAGE_MASK;
3580 l = (page + TARGET_PAGE_SIZE) - addr;
3581 if (l > len)
3582 l = len;
3583 flags = page_get_flags(page);
3584 if (!(flags & PAGE_VALID))
3585 return -1;
3586 if (is_write) {
3587 if (!(flags & PAGE_WRITE))
3588 return -1;
3589 /* XXX: this code should not depend on lock_user */
3590 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
3591 return -1;
3592 memcpy(p, buf, l);
3593 unlock_user(p, addr, l);
3594 } else {
3595 if (!(flags & PAGE_READ))
3596 return -1;
3597 /* XXX: this code should not depend on lock_user */
3598 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
3599 return -1;
3600 memcpy(buf, p, l);
3601 unlock_user(p, addr, 0);
3602 }
3603 len -= l;
3604 buf += l;
3605 addr += l;
3606 }
3607 return 0;
3608 }
3609
3610 #else
3611 void cpu_physical_memory_rw(target_phys_addr_t addr, uint8_t *buf,
3612 int len, int is_write)
3613 {
3614 int l, io_index;
3615 uint8_t *ptr;
3616 uint32_t val;
3617 target_phys_addr_t page;
3618 ram_addr_t pd;
3619 PhysPageDesc p;
3620
3621 while (len > 0) {
3622 page = addr & TARGET_PAGE_MASK;
3623 l = (page + TARGET_PAGE_SIZE) - addr;
3624 if (l > len)
3625 l = len;
3626 p = phys_page_find(page >> TARGET_PAGE_BITS);
3627 pd = p.phys_offset;
3628
3629 if (is_write) {
3630 if ((pd & ~TARGET_PAGE_MASK) != io_mem_ram.ram_addr) {
3631 target_phys_addr_t addr1;
3632 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3633 addr1 = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
3634 /* XXX: could force cpu_single_env to NULL to avoid
3635 potential bugs */
3636 if (l >= 4 && ((addr1 & 3) == 0)) {
3637 /* 32 bit write access */
3638 val = ldl_p(buf);
3639 io_mem_write(io_index, addr1, val, 4);
3640 l = 4;
3641 } else if (l >= 2 && ((addr1 & 1) == 0)) {
3642 /* 16 bit write access */
3643 val = lduw_p(buf);
3644 io_mem_write(io_index, addr1, val, 2);
3645 l = 2;
3646 } else {
3647 /* 8 bit write access */
3648 val = ldub_p(buf);
3649 io_mem_write(io_index, addr1, val, 1);
3650 l = 1;
3651 }
3652 } else {
3653 ram_addr_t addr1;
3654 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3655 /* RAM case */
3656 ptr = qemu_get_ram_ptr(addr1);
3657 memcpy(ptr, buf, l);
3658 if (!cpu_physical_memory_is_dirty(addr1)) {
3659 /* invalidate code */
3660 tb_invalidate_phys_page_range(addr1, addr1 + l, 0);
3661 /* set dirty bit */
3662 cpu_physical_memory_set_dirty_flags(
3663 addr1, (0xff & ~CODE_DIRTY_FLAG));
3664 }
3665 qemu_put_ram_ptr(ptr);
3666 }
3667 } else {
3668 if (!is_ram_rom_romd(pd)) {
3669 target_phys_addr_t addr1;
3670 /* I/O case */
3671 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3672 addr1 = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
3673 if (l >= 4 && ((addr1 & 3) == 0)) {
3674 /* 32 bit read access */
3675 val = io_mem_read(io_index, addr1, 4);
3676 stl_p(buf, val);
3677 l = 4;
3678 } else if (l >= 2 && ((addr1 & 1) == 0)) {
3679 /* 16 bit read access */
3680 val = io_mem_read(io_index, addr1, 2);
3681 stw_p(buf, val);
3682 l = 2;
3683 } else {
3684 /* 8 bit read access */
3685 val = io_mem_read(io_index, addr1, 1);
3686 stb_p(buf, val);
3687 l = 1;
3688 }
3689 } else {
3690 /* RAM case */
3691 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK);
3692 memcpy(buf, ptr + (addr & ~TARGET_PAGE_MASK), l);
3693 qemu_put_ram_ptr(ptr);
3694 }
3695 }
3696 len -= l;
3697 buf += l;
3698 addr += l;
3699 }
3700 }
3701
3702 /* used for ROM loading : can write in RAM and ROM */
3703 void cpu_physical_memory_write_rom(target_phys_addr_t addr,
3704 const uint8_t *buf, int len)
3705 {
3706 int l;
3707 uint8_t *ptr;
3708 target_phys_addr_t page;
3709 unsigned long pd;
3710 PhysPageDesc p;
3711
3712 while (len > 0) {
3713 page = addr & TARGET_PAGE_MASK;
3714 l = (page + TARGET_PAGE_SIZE) - addr;
3715 if (l > len)
3716 l = len;
3717 p = phys_page_find(page >> TARGET_PAGE_BITS);
3718 pd = p.phys_offset;
3719
3720 if (!is_ram_rom_romd(pd)) {
3721 /* do nothing */
3722 } else {
3723 unsigned long addr1;
3724 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3725 /* ROM/RAM case */
3726 ptr = qemu_get_ram_ptr(addr1);
3727 memcpy(ptr, buf, l);
3728 qemu_put_ram_ptr(ptr);
3729 }
3730 len -= l;
3731 buf += l;
3732 addr += l;
3733 }
3734 }
3735
3736 typedef struct {
3737 void *buffer;
3738 target_phys_addr_t addr;
3739 target_phys_addr_t len;
3740 } BounceBuffer;
3741
3742 static BounceBuffer bounce;
3743
3744 typedef struct MapClient {
3745 void *opaque;
3746 void (*callback)(void *opaque);
3747 QLIST_ENTRY(MapClient) link;
3748 } MapClient;
3749
3750 static QLIST_HEAD(map_client_list, MapClient) map_client_list
3751 = QLIST_HEAD_INITIALIZER(map_client_list);
3752
3753 void *cpu_register_map_client(void *opaque, void (*callback)(void *opaque))
3754 {
3755 MapClient *client = g_malloc(sizeof(*client));
3756
3757 client->opaque = opaque;
3758 client->callback = callback;
3759 QLIST_INSERT_HEAD(&map_client_list, client, link);
3760 return client;
3761 }
3762
3763 void cpu_unregister_map_client(void *_client)
3764 {
3765 MapClient *client = (MapClient *)_client;
3766
3767 QLIST_REMOVE(client, link);
3768 g_free(client);
3769 }
3770
3771 static void cpu_notify_map_clients(void)
3772 {
3773 MapClient *client;
3774
3775 while (!QLIST_EMPTY(&map_client_list)) {
3776 client = QLIST_FIRST(&map_client_list);
3777 client->callback(client->opaque);
3778 cpu_unregister_map_client(client);
3779 }
3780 }
3781
3782 /* Map a physical memory region into a host virtual address.
3783 * May map a subset of the requested range, given by and returned in *plen.
3784 * May return NULL if resources needed to perform the mapping are exhausted.
3785 * Use only for reads OR writes - not for read-modify-write operations.
3786 * Use cpu_register_map_client() to know when retrying the map operation is
3787 * likely to succeed.
3788 */
3789 void *cpu_physical_memory_map(target_phys_addr_t addr,
3790 target_phys_addr_t *plen,
3791 int is_write)
3792 {
3793 target_phys_addr_t len = *plen;
3794 target_phys_addr_t todo = 0;
3795 int l;
3796 target_phys_addr_t page;
3797 unsigned long pd;
3798 PhysPageDesc p;
3799 ram_addr_t raddr = RAM_ADDR_MAX;
3800 ram_addr_t rlen;
3801 void *ret;
3802
3803 while (len > 0) {
3804 page = addr & TARGET_PAGE_MASK;
3805 l = (page + TARGET_PAGE_SIZE) - addr;
3806 if (l > len)
3807 l = len;
3808 p = phys_page_find(page >> TARGET_PAGE_BITS);
3809 pd = p.phys_offset;
3810
3811 if ((pd & ~TARGET_PAGE_MASK) != io_mem_ram.ram_addr) {
3812 if (todo || bounce.buffer) {
3813 break;
3814 }
3815 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, TARGET_PAGE_SIZE);
3816 bounce.addr = addr;
3817 bounce.len = l;
3818 if (!is_write) {
3819 cpu_physical_memory_read(addr, bounce.buffer, l);
3820 }
3821
3822 *plen = l;
3823 return bounce.buffer;
3824 }
3825 if (!todo) {
3826 raddr = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
3827 }
3828
3829 len -= l;
3830 addr += l;
3831 todo += l;
3832 }
3833 rlen = todo;
3834 ret = qemu_ram_ptr_length(raddr, &rlen);
3835 *plen = rlen;
3836 return ret;
3837 }
3838
3839 /* Unmaps a memory region previously mapped by cpu_physical_memory_map().
3840 * Will also mark the memory as dirty if is_write == 1. access_len gives
3841 * the amount of memory that was actually read or written by the caller.
3842 */
3843 void cpu_physical_memory_unmap(void *buffer, target_phys_addr_t len,
3844 int is_write, target_phys_addr_t access_len)
3845 {
3846 if (buffer != bounce.buffer) {
3847 if (is_write) {
3848 ram_addr_t addr1 = qemu_ram_addr_from_host_nofail(buffer);
3849 while (access_len) {
3850 unsigned l;
3851 l = TARGET_PAGE_SIZE;
3852 if (l > access_len)
3853 l = access_len;
3854 if (!cpu_physical_memory_is_dirty(addr1)) {
3855 /* invalidate code */
3856 tb_invalidate_phys_page_range(addr1, addr1 + l, 0);
3857 /* set dirty bit */
3858 cpu_physical_memory_set_dirty_flags(
3859 addr1, (0xff & ~CODE_DIRTY_FLAG));
3860 }
3861 addr1 += l;
3862 access_len -= l;
3863 }
3864 }
3865 if (xen_enabled()) {
3866 xen_invalidate_map_cache_entry(buffer);
3867 }
3868 return;
3869 }
3870 if (is_write) {
3871 cpu_physical_memory_write(bounce.addr, bounce.buffer, access_len);
3872 }
3873 qemu_vfree(bounce.buffer);
3874 bounce.buffer = NULL;
3875 cpu_notify_map_clients();
3876 }
3877
3878 /* warning: addr must be aligned */
3879 static inline uint32_t ldl_phys_internal(target_phys_addr_t addr,
3880 enum device_endian endian)
3881 {
3882 int io_index;
3883 uint8_t *ptr;
3884 uint32_t val;
3885 unsigned long pd;
3886 PhysPageDesc p;
3887
3888 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3889 pd = p.phys_offset;
3890
3891 if (!is_ram_rom_romd(pd)) {
3892 /* I/O case */
3893 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3894 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
3895 val = io_mem_read(io_index, addr, 4);
3896 #if defined(TARGET_WORDS_BIGENDIAN)
3897 if (endian == DEVICE_LITTLE_ENDIAN) {
3898 val = bswap32(val);
3899 }
3900 #else
3901 if (endian == DEVICE_BIG_ENDIAN) {
3902 val = bswap32(val);
3903 }
3904 #endif
3905 } else {
3906 /* RAM case */
3907 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
3908 (addr & ~TARGET_PAGE_MASK);
3909 switch (endian) {
3910 case DEVICE_LITTLE_ENDIAN:
3911 val = ldl_le_p(ptr);
3912 break;
3913 case DEVICE_BIG_ENDIAN:
3914 val = ldl_be_p(ptr);
3915 break;
3916 default:
3917 val = ldl_p(ptr);
3918 break;
3919 }
3920 }
3921 return val;
3922 }
3923
3924 uint32_t ldl_phys(target_phys_addr_t addr)
3925 {
3926 return ldl_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
3927 }
3928
3929 uint32_t ldl_le_phys(target_phys_addr_t addr)
3930 {
3931 return ldl_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
3932 }
3933
3934 uint32_t ldl_be_phys(target_phys_addr_t addr)
3935 {
3936 return ldl_phys_internal(addr, DEVICE_BIG_ENDIAN);
3937 }
3938
3939 /* warning: addr must be aligned */
3940 static inline uint64_t ldq_phys_internal(target_phys_addr_t addr,
3941 enum device_endian endian)
3942 {
3943 int io_index;
3944 uint8_t *ptr;
3945 uint64_t val;
3946 unsigned long pd;
3947 PhysPageDesc p;
3948
3949 p = phys_page_find(addr >> TARGET_PAGE_BITS);
3950 pd = p.phys_offset;
3951
3952 if (!is_ram_rom_romd(pd)) {
3953 /* I/O case */
3954 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
3955 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
3956
3957 /* XXX This is broken when device endian != cpu endian.
3958 Fix and add "endian" variable check */
3959 #ifdef TARGET_WORDS_BIGENDIAN
3960 val = io_mem_read(io_index, addr, 4) << 32;
3961 val |= io_mem_read(io_index, addr + 4, 4);
3962 #else
3963 val = io_mem_read(io_index, addr, 4);
3964 val |= io_mem_read(io_index, addr + 4, 4) << 32;
3965 #endif
3966 } else {
3967 /* RAM case */
3968 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
3969 (addr & ~TARGET_PAGE_MASK);
3970 switch (endian) {
3971 case DEVICE_LITTLE_ENDIAN:
3972 val = ldq_le_p(ptr);
3973 break;
3974 case DEVICE_BIG_ENDIAN:
3975 val = ldq_be_p(ptr);
3976 break;
3977 default:
3978 val = ldq_p(ptr);
3979 break;
3980 }
3981 }
3982 return val;
3983 }
3984
3985 uint64_t ldq_phys(target_phys_addr_t addr)
3986 {
3987 return ldq_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
3988 }
3989
3990 uint64_t ldq_le_phys(target_phys_addr_t addr)
3991 {
3992 return ldq_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
3993 }
3994
3995 uint64_t ldq_be_phys(target_phys_addr_t addr)
3996 {
3997 return ldq_phys_internal(addr, DEVICE_BIG_ENDIAN);
3998 }
3999
4000 /* XXX: optimize */
4001 uint32_t ldub_phys(target_phys_addr_t addr)
4002 {
4003 uint8_t val;
4004 cpu_physical_memory_read(addr, &val, 1);
4005 return val;
4006 }
4007
4008 /* warning: addr must be aligned */
4009 static inline uint32_t lduw_phys_internal(target_phys_addr_t addr,
4010 enum device_endian endian)
4011 {
4012 int io_index;
4013 uint8_t *ptr;
4014 uint64_t val;
4015 unsigned long pd;
4016 PhysPageDesc p;
4017
4018 p = phys_page_find(addr >> TARGET_PAGE_BITS);
4019 pd = p.phys_offset;
4020
4021 if (!is_ram_rom_romd(pd)) {
4022 /* I/O case */
4023 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
4024 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
4025 val = io_mem_read(io_index, addr, 2);
4026 #if defined(TARGET_WORDS_BIGENDIAN)
4027 if (endian == DEVICE_LITTLE_ENDIAN) {
4028 val = bswap16(val);
4029 }
4030 #else
4031 if (endian == DEVICE_BIG_ENDIAN) {
4032 val = bswap16(val);
4033 }
4034 #endif
4035 } else {
4036 /* RAM case */
4037 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
4038 (addr & ~TARGET_PAGE_MASK);
4039 switch (endian) {
4040 case DEVICE_LITTLE_ENDIAN:
4041 val = lduw_le_p(ptr);
4042 break;
4043 case DEVICE_BIG_ENDIAN:
4044 val = lduw_be_p(ptr);
4045 break;
4046 default:
4047 val = lduw_p(ptr);
4048 break;
4049 }
4050 }
4051 return val;
4052 }
4053
4054 uint32_t lduw_phys(target_phys_addr_t addr)
4055 {
4056 return lduw_phys_internal(addr, DEVICE_NATIVE_ENDIAN);
4057 }
4058
4059 uint32_t lduw_le_phys(target_phys_addr_t addr)
4060 {
4061 return lduw_phys_internal(addr, DEVICE_LITTLE_ENDIAN);
4062 }
4063
4064 uint32_t lduw_be_phys(target_phys_addr_t addr)
4065 {
4066 return lduw_phys_internal(addr, DEVICE_BIG_ENDIAN);
4067 }
4068
4069 /* warning: addr must be aligned. The ram page is not masked as dirty
4070 and the code inside is not invalidated. It is useful if the dirty
4071 bits are used to track modified PTEs */
4072 void stl_phys_notdirty(target_phys_addr_t addr, uint32_t val)
4073 {
4074 int io_index;
4075 uint8_t *ptr;
4076 unsigned long pd;
4077 PhysPageDesc p;
4078
4079 p = phys_page_find(addr >> TARGET_PAGE_BITS);
4080 pd = p.phys_offset;
4081
4082 if ((pd & ~TARGET_PAGE_MASK) != io_mem_ram.ram_addr) {
4083 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
4084 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
4085 io_mem_write(io_index, addr, val, 4);
4086 } else {
4087 unsigned long addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
4088 ptr = qemu_get_ram_ptr(addr1);
4089 stl_p(ptr, val);
4090
4091 if (unlikely(in_migration)) {
4092 if (!cpu_physical_memory_is_dirty(addr1)) {
4093 /* invalidate code */
4094 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
4095 /* set dirty bit */
4096 cpu_physical_memory_set_dirty_flags(
4097 addr1, (0xff & ~CODE_DIRTY_FLAG));
4098 }
4099 }
4100 }
4101 }
4102
4103 void stq_phys_notdirty(target_phys_addr_t addr, uint64_t val)
4104 {
4105 int io_index;
4106 uint8_t *ptr;
4107 unsigned long pd;
4108 PhysPageDesc p;
4109
4110 p = phys_page_find(addr >> TARGET_PAGE_BITS);
4111 pd = p.phys_offset;
4112
4113 if ((pd & ~TARGET_PAGE_MASK) != io_mem_ram.ram_addr) {
4114 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
4115 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
4116 #ifdef TARGET_WORDS_BIGENDIAN
4117 io_mem_write(io_index, addr, val >> 32, 4);
4118 io_mem_write(io_index, addr + 4, (uint32_t)val, 4);
4119 #else
4120 io_mem_write(io_index, addr, (uint32_t)val, 4);
4121 io_mem_write(io_index, addr + 4, val >> 32, 4);
4122 #endif
4123 } else {
4124 ptr = qemu_get_ram_ptr(pd & TARGET_PAGE_MASK) +
4125 (addr & ~TARGET_PAGE_MASK);
4126 stq_p(ptr, val);
4127 }
4128 }
4129
4130 /* warning: addr must be aligned */
4131 static inline void stl_phys_internal(target_phys_addr_t addr, uint32_t val,
4132 enum device_endian endian)
4133 {
4134 int io_index;
4135 uint8_t *ptr;
4136 unsigned long pd;
4137 PhysPageDesc p;
4138
4139 p = phys_page_find(addr >> TARGET_PAGE_BITS);
4140 pd = p.phys_offset;
4141
4142 if ((pd & ~TARGET_PAGE_MASK) != io_mem_ram.ram_addr) {
4143 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
4144 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
4145 #if defined(TARGET_WORDS_BIGENDIAN)
4146 if (endian == DEVICE_LITTLE_ENDIAN) {
4147 val = bswap32(val);
4148 }
4149 #else
4150 if (endian == DEVICE_BIG_ENDIAN) {
4151 val = bswap32(val);
4152 }
4153 #endif
4154 io_mem_write(io_index, addr, val, 4);
4155 } else {
4156 unsigned long addr1;
4157 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
4158 /* RAM case */
4159 ptr = qemu_get_ram_ptr(addr1);
4160 switch (endian) {
4161 case DEVICE_LITTLE_ENDIAN:
4162 stl_le_p(ptr, val);
4163 break;
4164 case DEVICE_BIG_ENDIAN:
4165 stl_be_p(ptr, val);
4166 break;
4167 default:
4168 stl_p(ptr, val);
4169 break;
4170 }
4171 if (!cpu_physical_memory_is_dirty(addr1)) {
4172 /* invalidate code */
4173 tb_invalidate_phys_page_range(addr1, addr1 + 4, 0);
4174 /* set dirty bit */
4175 cpu_physical_memory_set_dirty_flags(addr1,
4176 (0xff & ~CODE_DIRTY_FLAG));
4177 }
4178 }
4179 }
4180
4181 void stl_phys(target_phys_addr_t addr, uint32_t val)
4182 {
4183 stl_phys_internal(addr, val, DEVICE_NATIVE_ENDIAN);
4184 }
4185
4186 void stl_le_phys(target_phys_addr_t addr, uint32_t val)
4187 {
4188 stl_phys_internal(addr, val, DEVICE_LITTLE_ENDIAN);
4189 }
4190
4191 void stl_be_phys(target_phys_addr_t addr, uint32_t val)
4192 {
4193 stl_phys_internal(addr, val, DEVICE_BIG_ENDIAN);
4194 }
4195
4196 /* XXX: optimize */
4197 void stb_phys(target_phys_addr_t addr, uint32_t val)
4198 {
4199 uint8_t v = val;
4200 cpu_physical_memory_write(addr, &v, 1);
4201 }
4202
4203 /* warning: addr must be aligned */
4204 static inline void stw_phys_internal(target_phys_addr_t addr, uint32_t val,
4205 enum device_endian endian)
4206 {
4207 int io_index;
4208 uint8_t *ptr;
4209 unsigned long pd;
4210 PhysPageDesc p;
4211
4212 p = phys_page_find(addr >> TARGET_PAGE_BITS);
4213 pd = p.phys_offset;
4214
4215 if ((pd & ~TARGET_PAGE_MASK) != io_mem_ram.ram_addr) {
4216 io_index = (pd >> IO_MEM_SHIFT) & (IO_MEM_NB_ENTRIES - 1);
4217 addr = (addr & ~TARGET_PAGE_MASK) + p.region_offset;
4218 #if defined(TARGET_WORDS_BIGENDIAN)
4219 if (endian == DEVICE_LITTLE_ENDIAN) {
4220 val = bswap16(val);
4221 }
4222 #else
4223 if (endian == DEVICE_BIG_ENDIAN) {
4224 val = bswap16(val);
4225 }
4226 #endif
4227 io_mem_write(io_index, addr, val, 2);
4228 } else {
4229 unsigned long addr1;
4230 addr1 = (pd & TARGET_PAGE_MASK) + (addr & ~TARGET_PAGE_MASK);
4231 /* RAM case */
4232 ptr = qemu_get_ram_ptr(addr1);
4233 switch (endian) {
4234 case DEVICE_LITTLE_ENDIAN:
4235 stw_le_p(ptr, val);
4236 break;
4237 case DEVICE_BIG_ENDIAN:
4238 stw_be_p(ptr, val);
4239 break;
4240 default:
4241 stw_p(ptr, val);
4242 break;
4243 }
4244 if (!cpu_physical_memory_is_dirty(addr1)) {
4245 /* invalidate code */
4246 tb_invalidate_phys_page_range(addr1, addr1 + 2, 0);
4247 /* set dirty bit */
4248 cpu_physical_memory_set_dirty_flags(addr1,
4249 (0xff & ~CODE_DIRTY_FLAG));
4250 }
4251 }
4252 }
4253
4254 void stw_phys(target_phys_addr_t addr, uint32_t val)
4255 {
4256 stw_phys_internal(addr, val, DEVICE_NATIVE_ENDIAN);
4257 }
4258
4259 void stw_le_phys(target_phys_addr_t addr, uint32_t val)
4260 {
4261 stw_phys_internal(addr, val, DEVICE_LITTLE_ENDIAN);
4262 }
4263
4264 void stw_be_phys(target_phys_addr_t addr, uint32_t val)
4265 {
4266 stw_phys_internal(addr, val, DEVICE_BIG_ENDIAN);
4267 }
4268
4269 /* XXX: optimize */
4270 void stq_phys(target_phys_addr_t addr, uint64_t val)
4271 {
4272 val = tswap64(val);
4273 cpu_physical_memory_write(addr, &val, 8);
4274 }
4275
4276 void stq_le_phys(target_phys_addr_t addr, uint64_t val)
4277 {
4278 val = cpu_to_le64(val);
4279 cpu_physical_memory_write(addr, &val, 8);
4280 }
4281
4282 void stq_be_phys(target_phys_addr_t addr, uint64_t val)
4283 {
4284 val = cpu_to_be64(val);
4285 cpu_physical_memory_write(addr, &val, 8);
4286 }
4287
4288 /* virtual memory access for debug (includes writing to ROM) */
4289 int cpu_memory_rw_debug(CPUState *env, target_ulong addr,
4290 uint8_t *buf, int len, int is_write)
4291 {
4292 int l;
4293 target_phys_addr_t phys_addr;
4294 target_ulong page;
4295
4296 while (len > 0) {
4297 page = addr & TARGET_PAGE_MASK;
4298 phys_addr = cpu_get_phys_page_debug(env, page);
4299 /* if no physical page mapped, return an error */
4300 if (phys_addr == -1)
4301 return -1;
4302 l = (page + TARGET_PAGE_SIZE) - addr;
4303 if (l > len)
4304 l = len;
4305 phys_addr += (addr & ~TARGET_PAGE_MASK);
4306 if (is_write)
4307 cpu_physical_memory_write_rom(phys_addr, buf, l);
4308 else
4309 cpu_physical_memory_rw(phys_addr, buf, l, is_write);
4310 len -= l;
4311 buf += l;
4312 addr += l;
4313 }
4314 return 0;
4315 }
4316 #endif
4317
4318 /* in deterministic execution mode, instructions doing device I/Os
4319 must be at the end of the TB */
4320 void cpu_io_recompile(CPUState *env, void *retaddr)
4321 {
4322 TranslationBlock *tb;
4323 uint32_t n, cflags;
4324 target_ulong pc, cs_base;
4325 uint64_t flags;
4326
4327 tb = tb_find_pc((unsigned long)retaddr);
4328 if (!tb) {
4329 cpu_abort(env, "cpu_io_recompile: could not find TB for pc=%p",
4330 retaddr);
4331 }
4332 n = env->icount_decr.u16.low + tb->icount;
4333 cpu_restore_state(tb, env, (unsigned long)retaddr);
4334 /* Calculate how many instructions had been executed before the fault
4335 occurred. */
4336 n = n - env->icount_decr.u16.low;
4337 /* Generate a new TB ending on the I/O insn. */
4338 n++;
4339 /* On MIPS and SH, delay slot instructions can only be restarted if
4340 they were already the first instruction in the TB. If this is not
4341 the first instruction in a TB then re-execute the preceding
4342 branch. */
4343 #if defined(TARGET_MIPS)
4344 if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
4345 env->active_tc.PC -= 4;
4346 env->icount_decr.u16.low++;
4347 env->hflags &= ~MIPS_HFLAG_BMASK;
4348 }
4349 #elif defined(TARGET_SH4)
4350 if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
4351 && n > 1) {
4352 env->pc -= 2;
4353 env->icount_decr.u16.low++;
4354 env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
4355 }
4356 #endif
4357 /* This should never happen. */
4358 if (n > CF_COUNT_MASK)
4359 cpu_abort(env, "TB too big during recompile");
4360
4361 cflags = n | CF_LAST_IO;
4362 pc = tb->pc;
4363 cs_base = tb->cs_base;
4364 flags = tb->flags;
4365 tb_phys_invalidate(tb, -1);
4366 /* FIXME: In theory this could raise an exception. In practice
4367 we have already translated the block once so it's probably ok. */
4368 tb_gen_code(env, pc, cs_base, flags, cflags);
4369 /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
4370 the first in the TB) then we end up generating a whole new TB and
4371 repeating the fault, which is horribly inefficient.
4372 Better would be to execute just this insn uncached, or generate a
4373 second new TB. */
4374 cpu_resume_from_signal(env, NULL);
4375 }
4376
4377 #if !defined(CONFIG_USER_ONLY)
4378
4379 void dump_exec_info(FILE *f, fprintf_function cpu_fprintf)
4380 {
4381 int i, target_code_size, max_target_code_size;
4382 int direct_jmp_count, direct_jmp2_count, cross_page;
4383 TranslationBlock *tb;
4384
4385 target_code_size = 0;
4386 max_target_code_size = 0;
4387 cross_page = 0;
4388 direct_jmp_count = 0;
4389 direct_jmp2_count = 0;
4390 for(i = 0; i < nb_tbs; i++) {
4391 tb = &tbs[i];
4392 target_code_size += tb->size;
4393 if (tb->size > max_target_code_size)
4394 max_target_code_size = tb->size;
4395 if (tb->page_addr[1] != -1)
4396 cross_page++;
4397 if (tb->tb_next_offset[0] != 0xffff) {
4398 direct_jmp_count++;
4399 if (tb->tb_next_offset[1] != 0xffff) {
4400 direct_jmp2_count++;
4401 }
4402 }
4403 }
4404 /* XXX: avoid using doubles ? */
4405 cpu_fprintf(f, "Translation buffer state:\n");
4406 cpu_fprintf(f, "gen code size %td/%ld\n",
4407 code_gen_ptr - code_gen_buffer, code_gen_buffer_max_size);
4408 cpu_fprintf(f, "TB count %d/%d\n",
4409 nb_tbs, code_gen_max_blocks);
4410 cpu_fprintf(f, "TB avg target size %d max=%d bytes\n",
4411 nb_tbs ? target_code_size / nb_tbs : 0,
4412 max_target_code_size);
4413 cpu_fprintf(f, "TB avg host size %td bytes (expansion ratio: %0.1f)\n",
4414 nb_tbs ? (code_gen_ptr - code_gen_buffer) / nb_tbs : 0,
4415 target_code_size ? (double) (code_gen_ptr - code_gen_buffer) / target_code_size : 0);
4416 cpu_fprintf(f, "cross page TB count %d (%d%%)\n",
4417 cross_page,
4418 nb_tbs ? (cross_page * 100) / nb_tbs : 0);
4419 cpu_fprintf(f, "direct jump count %d (%d%%) (2 jumps=%d %d%%)\n",
4420 direct_jmp_count,
4421 nb_tbs ? (direct_jmp_count * 100) / nb_tbs : 0,
4422 direct_jmp2_count,
4423 nb_tbs ? (direct_jmp2_count * 100) / nb_tbs : 0);
4424 cpu_fprintf(f, "\nStatistics:\n");
4425 cpu_fprintf(f, "TB flush count %d\n", tb_flush_count);
4426 cpu_fprintf(f, "TB invalidate count %d\n", tb_phys_invalidate_count);
4427 cpu_fprintf(f, "TLB flush count %d\n", tlb_flush_count);
4428 tcg_dump_info(f, cpu_fprintf);
4429 }
4430
4431 /* NOTE: this function can trigger an exception */
4432 /* NOTE2: the returned address is not exactly the physical address: it
4433 is the offset relative to phys_ram_base */
4434 tb_page_addr_t get_page_addr_code(CPUState *env1, target_ulong addr)
4435 {
4436 int mmu_idx, page_index, pd;
4437 void *p;
4438
4439 page_index = (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
4440 mmu_idx = cpu_mmu_index(env1);
4441 if (unlikely(env1->tlb_table[mmu_idx][page_index].addr_code !=
4442 (addr & TARGET_PAGE_MASK))) {
4443 ldub_code(addr);
4444 }
4445 pd = env1->tlb_table[mmu_idx][page_index].addr_code & ~TARGET_PAGE_MASK;
4446 if (pd != io_mem_ram.ram_addr && pd != io_mem_rom.ram_addr
4447 && !(pd & IO_MEM_ROMD)) {
4448 #if defined(TARGET_ALPHA) || defined(TARGET_MIPS) || defined(TARGET_SPARC)
4449 cpu_unassigned_access(env1, addr, 0, 1, 0, 4);
4450 #else
4451 cpu_abort(env1, "Trying to execute code outside RAM or ROM at 0x" TARGET_FMT_lx "\n", addr);
4452 #endif
4453 }
4454 p = (void *)((uintptr_t)addr + env1->tlb_table[mmu_idx][page_index].addend);
4455 return qemu_ram_addr_from_host_nofail(p);
4456 }
4457
4458 #define MMUSUFFIX _cmmu
4459 #undef GETPC
4460 #define GETPC() NULL
4461 #define env cpu_single_env
4462 #define SOFTMMU_CODE_ACCESS
4463
4464 #define SHIFT 0
4465 #include "softmmu_template.h"
4466
4467 #define SHIFT 1
4468 #include "softmmu_template.h"
4469
4470 #define SHIFT 2
4471 #include "softmmu_template.h"
4472
4473 #define SHIFT 3
4474 #include "softmmu_template.h"
4475
4476 #undef env
4477
4478 #endif