MAINTAINERS: Update Xilinx Maintainership
[qemu/ar7.git] / translate-all.c
blob2a40530bbaf10e51aaa23e4b5571a98b4da9345b
1 /*
2 * Host code generation
4 * Copyright (c) 2003 Fabrice Bellard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
19 #ifdef _WIN32
20 #include <windows.h>
21 #else
22 #include <sys/types.h>
23 #include <sys/mman.h>
24 #endif
25 #include <stdarg.h>
26 #include <stdlib.h>
27 #include <stdio.h>
28 #include <string.h>
29 #include <inttypes.h>
31 #include "config.h"
33 #include "qemu-common.h"
34 #define NO_CPU_IO_DEFS
35 #include "cpu.h"
36 #include "trace.h"
37 #include "disas/disas.h"
38 #include "tcg.h"
39 #if defined(CONFIG_USER_ONLY)
40 #include "qemu.h"
41 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
42 #include <sys/param.h>
43 #if __FreeBSD_version >= 700104
44 #define HAVE_KINFO_GETVMMAP
45 #define sigqueue sigqueue_freebsd /* avoid redefinition */
46 #include <sys/time.h>
47 #include <sys/proc.h>
48 #include <machine/profile.h>
49 #define _KERNEL
50 #include <sys/user.h>
51 #undef _KERNEL
52 #undef sigqueue
53 #include <libutil.h>
54 #endif
55 #endif
56 #else
57 #include "exec/address-spaces.h"
58 #endif
60 #include "exec/cputlb.h"
61 #include "exec/tb-hash.h"
62 #include "translate-all.h"
63 #include "qemu/bitmap.h"
64 #include "qemu/timer.h"
66 //#define DEBUG_TB_INVALIDATE
67 //#define DEBUG_FLUSH
68 /* make various TB consistency checks */
69 //#define DEBUG_TB_CHECK
71 #if !defined(CONFIG_USER_ONLY)
72 /* TB consistency checks only implemented for usermode emulation. */
73 #undef DEBUG_TB_CHECK
74 #endif
76 #define SMC_BITMAP_USE_THRESHOLD 10
78 typedef struct PageDesc {
79 /* list of TBs intersecting this ram page */
80 TranslationBlock *first_tb;
81 /* in order to optimize self modifying code, we count the number
82 of lookups we do to a given page to use a bitmap */
83 unsigned int code_write_count;
84 unsigned long *code_bitmap;
85 #if defined(CONFIG_USER_ONLY)
86 unsigned long flags;
87 #endif
88 } PageDesc;
90 /* In system mode we want L1_MAP to be based on ram offsets,
91 while in user mode we want it to be based on virtual addresses. */
92 #if !defined(CONFIG_USER_ONLY)
93 #if HOST_LONG_BITS < TARGET_PHYS_ADDR_SPACE_BITS
94 # define L1_MAP_ADDR_SPACE_BITS HOST_LONG_BITS
95 #else
96 # define L1_MAP_ADDR_SPACE_BITS TARGET_PHYS_ADDR_SPACE_BITS
97 #endif
98 #else
99 # define L1_MAP_ADDR_SPACE_BITS TARGET_VIRT_ADDR_SPACE_BITS
100 #endif
102 /* Size of the L2 (and L3, etc) page tables. */
103 #define V_L2_BITS 10
104 #define V_L2_SIZE (1 << V_L2_BITS)
106 /* The bits remaining after N lower levels of page tables. */
107 #define V_L1_BITS_REM \
108 ((L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS) % V_L2_BITS)
110 #if V_L1_BITS_REM < 4
111 #define V_L1_BITS (V_L1_BITS_REM + V_L2_BITS)
112 #else
113 #define V_L1_BITS V_L1_BITS_REM
114 #endif
116 #define V_L1_SIZE ((target_ulong)1 << V_L1_BITS)
118 #define V_L1_SHIFT (L1_MAP_ADDR_SPACE_BITS - TARGET_PAGE_BITS - V_L1_BITS)
120 uintptr_t qemu_real_host_page_size;
121 uintptr_t qemu_real_host_page_mask;
122 uintptr_t qemu_host_page_size;
123 uintptr_t qemu_host_page_mask;
125 /* This is a multi-level map on the virtual address space.
126 The bottom level has pointers to PageDesc. */
127 static void *l1_map[V_L1_SIZE];
129 /* code generation context */
130 TCGContext tcg_ctx;
132 static void tb_link_page(TranslationBlock *tb, tb_page_addr_t phys_pc,
133 tb_page_addr_t phys_page2);
134 static TranslationBlock *tb_find_pc(uintptr_t tc_ptr);
136 void cpu_gen_init(void)
138 tcg_context_init(&tcg_ctx);
141 /* return non zero if the very first instruction is invalid so that
142 the virtual CPU can trigger an exception.
144 '*gen_code_size_ptr' contains the size of the generated code (host
145 code).
147 int cpu_gen_code(CPUArchState *env, TranslationBlock *tb, int *gen_code_size_ptr)
149 TCGContext *s = &tcg_ctx;
150 tcg_insn_unit *gen_code_buf;
151 int gen_code_size;
152 #ifdef CONFIG_PROFILER
153 int64_t ti;
154 #endif
156 #ifdef CONFIG_PROFILER
157 s->tb_count1++; /* includes aborted translations because of
158 exceptions */
159 ti = profile_getclock();
160 #endif
161 tcg_func_start(s);
163 gen_intermediate_code(env, tb);
165 trace_translate_block(tb, tb->pc, tb->tc_ptr);
167 /* generate machine code */
168 gen_code_buf = tb->tc_ptr;
169 tb->tb_next_offset[0] = 0xffff;
170 tb->tb_next_offset[1] = 0xffff;
171 s->tb_next_offset = tb->tb_next_offset;
172 #ifdef USE_DIRECT_JUMP
173 s->tb_jmp_offset = tb->tb_jmp_offset;
174 s->tb_next = NULL;
175 #else
176 s->tb_jmp_offset = NULL;
177 s->tb_next = tb->tb_next;
178 #endif
180 #ifdef CONFIG_PROFILER
181 s->tb_count++;
182 s->interm_time += profile_getclock() - ti;
183 s->code_time -= profile_getclock();
184 #endif
185 gen_code_size = tcg_gen_code(s, gen_code_buf);
186 *gen_code_size_ptr = gen_code_size;
187 #ifdef CONFIG_PROFILER
188 s->code_time += profile_getclock();
189 s->code_in_len += tb->size;
190 s->code_out_len += gen_code_size;
191 #endif
193 #ifdef DEBUG_DISAS
194 if (qemu_loglevel_mask(CPU_LOG_TB_OUT_ASM)) {
195 qemu_log("OUT: [size=%d]\n", gen_code_size);
196 log_disas(tb->tc_ptr, gen_code_size);
197 qemu_log("\n");
198 qemu_log_flush();
200 #endif
201 return 0;
204 /* The cpu state corresponding to 'searched_pc' is restored.
206 static int cpu_restore_state_from_tb(CPUState *cpu, TranslationBlock *tb,
207 uintptr_t searched_pc)
209 CPUArchState *env = cpu->env_ptr;
210 TCGContext *s = &tcg_ctx;
211 int j;
212 uintptr_t tc_ptr;
213 #ifdef CONFIG_PROFILER
214 int64_t ti;
215 #endif
217 #ifdef CONFIG_PROFILER
218 ti = profile_getclock();
219 #endif
220 tcg_func_start(s);
222 gen_intermediate_code_pc(env, tb);
224 if (tb->cflags & CF_USE_ICOUNT) {
225 assert(use_icount);
226 /* Reset the cycle counter to the start of the block. */
227 cpu->icount_decr.u16.low += tb->icount;
228 /* Clear the IO flag. */
229 cpu->can_do_io = 0;
232 /* find opc index corresponding to search_pc */
233 tc_ptr = (uintptr_t)tb->tc_ptr;
234 if (searched_pc < tc_ptr)
235 return -1;
237 s->tb_next_offset = tb->tb_next_offset;
238 #ifdef USE_DIRECT_JUMP
239 s->tb_jmp_offset = tb->tb_jmp_offset;
240 s->tb_next = NULL;
241 #else
242 s->tb_jmp_offset = NULL;
243 s->tb_next = tb->tb_next;
244 #endif
245 j = tcg_gen_code_search_pc(s, (tcg_insn_unit *)tc_ptr,
246 searched_pc - tc_ptr);
247 if (j < 0)
248 return -1;
249 /* now find start of instruction before */
250 while (s->gen_opc_instr_start[j] == 0) {
251 j--;
253 cpu->icount_decr.u16.low -= s->gen_opc_icount[j];
255 restore_state_to_opc(env, tb, j);
257 #ifdef CONFIG_PROFILER
258 s->restore_time += profile_getclock() - ti;
259 s->restore_count++;
260 #endif
261 return 0;
264 bool cpu_restore_state(CPUState *cpu, uintptr_t retaddr)
266 TranslationBlock *tb;
268 tb = tb_find_pc(retaddr);
269 if (tb) {
270 cpu_restore_state_from_tb(cpu, tb, retaddr);
271 if (tb->cflags & CF_NOCACHE) {
272 /* one-shot translation, invalidate it immediately */
273 cpu->current_tb = NULL;
274 tb_phys_invalidate(tb, -1);
275 tb_free(tb);
277 return true;
279 return false;
282 #ifdef _WIN32
283 static __attribute__((unused)) void map_exec(void *addr, long size)
285 DWORD old_protect;
286 VirtualProtect(addr, size,
287 PAGE_EXECUTE_READWRITE, &old_protect);
289 #else
290 static __attribute__((unused)) void map_exec(void *addr, long size)
292 unsigned long start, end, page_size;
294 page_size = getpagesize();
295 start = (unsigned long)addr;
296 start &= ~(page_size - 1);
298 end = (unsigned long)addr + size;
299 end += page_size - 1;
300 end &= ~(page_size - 1);
302 mprotect((void *)start, end - start,
303 PROT_READ | PROT_WRITE | PROT_EXEC);
305 #endif
307 void page_size_init(void)
309 /* NOTE: we can always suppose that qemu_host_page_size >=
310 TARGET_PAGE_SIZE */
311 qemu_real_host_page_size = getpagesize();
312 qemu_real_host_page_mask = ~(qemu_real_host_page_size - 1);
313 if (qemu_host_page_size == 0) {
314 qemu_host_page_size = qemu_real_host_page_size;
316 if (qemu_host_page_size < TARGET_PAGE_SIZE) {
317 qemu_host_page_size = TARGET_PAGE_SIZE;
319 qemu_host_page_mask = ~(qemu_host_page_size - 1);
322 static void page_init(void)
324 page_size_init();
325 #if defined(CONFIG_BSD) && defined(CONFIG_USER_ONLY)
327 #ifdef HAVE_KINFO_GETVMMAP
328 struct kinfo_vmentry *freep;
329 int i, cnt;
331 freep = kinfo_getvmmap(getpid(), &cnt);
332 if (freep) {
333 mmap_lock();
334 for (i = 0; i < cnt; i++) {
335 unsigned long startaddr, endaddr;
337 startaddr = freep[i].kve_start;
338 endaddr = freep[i].kve_end;
339 if (h2g_valid(startaddr)) {
340 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
342 if (h2g_valid(endaddr)) {
343 endaddr = h2g(endaddr);
344 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
345 } else {
346 #if TARGET_ABI_BITS <= L1_MAP_ADDR_SPACE_BITS
347 endaddr = ~0ul;
348 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
349 #endif
353 free(freep);
354 mmap_unlock();
356 #else
357 FILE *f;
359 last_brk = (unsigned long)sbrk(0);
361 f = fopen("/compat/linux/proc/self/maps", "r");
362 if (f) {
363 mmap_lock();
365 do {
366 unsigned long startaddr, endaddr;
367 int n;
369 n = fscanf(f, "%lx-%lx %*[^\n]\n", &startaddr, &endaddr);
371 if (n == 2 && h2g_valid(startaddr)) {
372 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
374 if (h2g_valid(endaddr)) {
375 endaddr = h2g(endaddr);
376 } else {
377 endaddr = ~0ul;
379 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
381 } while (!feof(f));
383 fclose(f);
384 mmap_unlock();
386 #endif
388 #endif
391 static PageDesc *page_find_alloc(tb_page_addr_t index, int alloc)
393 PageDesc *pd;
394 void **lp;
395 int i;
397 /* Level 1. Always allocated. */
398 lp = l1_map + ((index >> V_L1_SHIFT) & (V_L1_SIZE - 1));
400 /* Level 2..N-1. */
401 for (i = V_L1_SHIFT / V_L2_BITS - 1; i > 0; i--) {
402 void **p = *lp;
404 if (p == NULL) {
405 if (!alloc) {
406 return NULL;
408 p = g_new0(void *, V_L2_SIZE);
409 *lp = p;
412 lp = p + ((index >> (i * V_L2_BITS)) & (V_L2_SIZE - 1));
415 pd = *lp;
416 if (pd == NULL) {
417 if (!alloc) {
418 return NULL;
420 pd = g_new0(PageDesc, V_L2_SIZE);
421 *lp = pd;
424 return pd + (index & (V_L2_SIZE - 1));
427 static inline PageDesc *page_find(tb_page_addr_t index)
429 return page_find_alloc(index, 0);
432 #if !defined(CONFIG_USER_ONLY)
433 #define mmap_lock() do { } while (0)
434 #define mmap_unlock() do { } while (0)
435 #endif
437 #if defined(CONFIG_USER_ONLY)
438 /* Currently it is not recommended to allocate big chunks of data in
439 user mode. It will change when a dedicated libc will be used. */
440 /* ??? 64-bit hosts ought to have no problem mmaping data outside the
441 region in which the guest needs to run. Revisit this. */
442 #define USE_STATIC_CODE_GEN_BUFFER
443 #endif
445 /* ??? Should configure for this, not list operating systems here. */
446 #if (defined(__linux__) \
447 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) \
448 || defined(__DragonFly__) || defined(__OpenBSD__) \
449 || defined(__NetBSD__))
450 # define USE_MMAP
451 #endif
453 /* Minimum size of the code gen buffer. This number is randomly chosen,
454 but not so small that we can't have a fair number of TB's live. */
455 #define MIN_CODE_GEN_BUFFER_SIZE (1024u * 1024)
457 /* Maximum size of the code gen buffer we'd like to use. Unless otherwise
458 indicated, this is constrained by the range of direct branches on the
459 host cpu, as used by the TCG implementation of goto_tb. */
460 #if defined(__x86_64__)
461 # define MAX_CODE_GEN_BUFFER_SIZE (2ul * 1024 * 1024 * 1024)
462 #elif defined(__sparc__)
463 # define MAX_CODE_GEN_BUFFER_SIZE (2ul * 1024 * 1024 * 1024)
464 #elif defined(__aarch64__)
465 # define MAX_CODE_GEN_BUFFER_SIZE (128ul * 1024 * 1024)
466 #elif defined(__arm__)
467 # define MAX_CODE_GEN_BUFFER_SIZE (16u * 1024 * 1024)
468 #elif defined(__s390x__)
469 /* We have a +- 4GB range on the branches; leave some slop. */
470 # define MAX_CODE_GEN_BUFFER_SIZE (3ul * 1024 * 1024 * 1024)
471 #elif defined(__mips__)
472 /* We have a 256MB branch region, but leave room to make sure the
473 main executable is also within that region. */
474 # define MAX_CODE_GEN_BUFFER_SIZE (128ul * 1024 * 1024)
475 #else
476 # define MAX_CODE_GEN_BUFFER_SIZE ((size_t)-1)
477 #endif
479 #define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (32u * 1024 * 1024)
481 #define DEFAULT_CODE_GEN_BUFFER_SIZE \
482 (DEFAULT_CODE_GEN_BUFFER_SIZE_1 < MAX_CODE_GEN_BUFFER_SIZE \
483 ? DEFAULT_CODE_GEN_BUFFER_SIZE_1 : MAX_CODE_GEN_BUFFER_SIZE)
485 static inline size_t size_code_gen_buffer(size_t tb_size)
487 /* Size the buffer. */
488 if (tb_size == 0) {
489 #ifdef USE_STATIC_CODE_GEN_BUFFER
490 tb_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
491 #else
492 /* ??? Needs adjustments. */
493 /* ??? If we relax the requirement that CONFIG_USER_ONLY use the
494 static buffer, we could size this on RESERVED_VA, on the text
495 segment size of the executable, or continue to use the default. */
496 tb_size = (unsigned long)(ram_size / 4);
497 #endif
499 if (tb_size < MIN_CODE_GEN_BUFFER_SIZE) {
500 tb_size = MIN_CODE_GEN_BUFFER_SIZE;
502 if (tb_size > MAX_CODE_GEN_BUFFER_SIZE) {
503 tb_size = MAX_CODE_GEN_BUFFER_SIZE;
505 tcg_ctx.code_gen_buffer_size = tb_size;
506 return tb_size;
509 #ifdef __mips__
510 /* In order to use J and JAL within the code_gen_buffer, we require
511 that the buffer not cross a 256MB boundary. */
512 static inline bool cross_256mb(void *addr, size_t size)
514 return ((uintptr_t)addr ^ ((uintptr_t)addr + size)) & 0xf0000000;
517 /* We weren't able to allocate a buffer without crossing that boundary,
518 so make do with the larger portion of the buffer that doesn't cross.
519 Returns the new base of the buffer, and adjusts code_gen_buffer_size. */
520 static inline void *split_cross_256mb(void *buf1, size_t size1)
522 void *buf2 = (void *)(((uintptr_t)buf1 + size1) & 0xf0000000);
523 size_t size2 = buf1 + size1 - buf2;
525 size1 = buf2 - buf1;
526 if (size1 < size2) {
527 size1 = size2;
528 buf1 = buf2;
531 tcg_ctx.code_gen_buffer_size = size1;
532 return buf1;
534 #endif
536 #ifdef USE_STATIC_CODE_GEN_BUFFER
537 static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE]
538 __attribute__((aligned(CODE_GEN_ALIGN)));
540 static inline void *alloc_code_gen_buffer(void)
542 void *buf = static_code_gen_buffer;
543 #ifdef __mips__
544 if (cross_256mb(buf, tcg_ctx.code_gen_buffer_size)) {
545 buf = split_cross_256mb(buf, tcg_ctx.code_gen_buffer_size);
547 #endif
548 map_exec(buf, tcg_ctx.code_gen_buffer_size);
549 return buf;
551 #elif defined(USE_MMAP)
552 static inline void *alloc_code_gen_buffer(void)
554 int flags = MAP_PRIVATE | MAP_ANONYMOUS;
555 uintptr_t start = 0;
556 void *buf;
558 /* Constrain the position of the buffer based on the host cpu.
559 Note that these addresses are chosen in concert with the
560 addresses assigned in the relevant linker script file. */
561 # if defined(__PIE__) || defined(__PIC__)
562 /* Don't bother setting a preferred location if we're building
563 a position-independent executable. We're more likely to get
564 an address near the main executable if we let the kernel
565 choose the address. */
566 # elif defined(__x86_64__) && defined(MAP_32BIT)
567 /* Force the memory down into low memory with the executable.
568 Leave the choice of exact location with the kernel. */
569 flags |= MAP_32BIT;
570 /* Cannot expect to map more than 800MB in low memory. */
571 if (tcg_ctx.code_gen_buffer_size > 800u * 1024 * 1024) {
572 tcg_ctx.code_gen_buffer_size = 800u * 1024 * 1024;
574 # elif defined(__sparc__)
575 start = 0x40000000ul;
576 # elif defined(__s390x__)
577 start = 0x90000000ul;
578 # elif defined(__mips__)
579 /* ??? We ought to more explicitly manage layout for softmmu too. */
580 # ifdef CONFIG_USER_ONLY
581 start = 0x68000000ul;
582 # elif _MIPS_SIM == _ABI64
583 start = 0x128000000ul;
584 # else
585 start = 0x08000000ul;
586 # endif
587 # endif
589 buf = mmap((void *)start, tcg_ctx.code_gen_buffer_size,
590 PROT_WRITE | PROT_READ | PROT_EXEC, flags, -1, 0);
591 if (buf == MAP_FAILED) {
592 return NULL;
595 #ifdef __mips__
596 if (cross_256mb(buf, tcg_ctx.code_gen_buffer_size)) {
597 /* Try again, with the original still mapped, to avoid re-acquiring
598 that 256mb crossing. This time don't specify an address. */
599 size_t size2, size1 = tcg_ctx.code_gen_buffer_size;
600 void *buf2 = mmap(NULL, size1, PROT_WRITE | PROT_READ | PROT_EXEC,
601 flags, -1, 0);
602 if (buf2 != MAP_FAILED) {
603 if (!cross_256mb(buf2, size1)) {
604 /* Success! Use the new buffer. */
605 munmap(buf, size1);
606 return buf2;
608 /* Failure. Work with what we had. */
609 munmap(buf2, size1);
612 /* Split the original buffer. Free the smaller half. */
613 buf2 = split_cross_256mb(buf, size1);
614 size2 = tcg_ctx.code_gen_buffer_size;
615 munmap(buf + (buf == buf2 ? size2 : 0), size1 - size2);
616 return buf2;
618 #endif
620 return buf;
622 #else
623 static inline void *alloc_code_gen_buffer(void)
625 void *buf = g_try_malloc(tcg_ctx.code_gen_buffer_size);
627 if (buf == NULL) {
628 return NULL;
631 #ifdef __mips__
632 if (cross_256mb(buf, tcg_ctx.code_gen_buffer_size)) {
633 void *buf2 = g_malloc(tcg_ctx.code_gen_buffer_size);
634 if (buf2 != NULL && !cross_256mb(buf2, size1)) {
635 /* Success! Use the new buffer. */
636 free(buf);
637 buf = buf2;
638 } else {
639 /* Failure. Work with what we had. Since this is malloc
640 and not mmap, we can't free the other half. */
641 free(buf2);
642 buf = split_cross_256mb(buf, tcg_ctx.code_gen_buffer_size);
645 #endif
647 map_exec(buf, tcg_ctx.code_gen_buffer_size);
648 return buf;
650 #endif /* USE_STATIC_CODE_GEN_BUFFER, USE_MMAP */
652 static inline void code_gen_alloc(size_t tb_size)
654 tcg_ctx.code_gen_buffer_size = size_code_gen_buffer(tb_size);
655 tcg_ctx.code_gen_buffer = alloc_code_gen_buffer();
656 if (tcg_ctx.code_gen_buffer == NULL) {
657 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
658 exit(1);
661 qemu_madvise(tcg_ctx.code_gen_buffer, tcg_ctx.code_gen_buffer_size,
662 QEMU_MADV_HUGEPAGE);
664 /* Steal room for the prologue at the end of the buffer. This ensures
665 (via the MAX_CODE_GEN_BUFFER_SIZE limits above) that direct branches
666 from TB's to the prologue are going to be in range. It also means
667 that we don't need to mark (additional) portions of the data segment
668 as executable. */
669 tcg_ctx.code_gen_prologue = tcg_ctx.code_gen_buffer +
670 tcg_ctx.code_gen_buffer_size - 1024;
671 tcg_ctx.code_gen_buffer_size -= 1024;
673 tcg_ctx.code_gen_buffer_max_size = tcg_ctx.code_gen_buffer_size -
674 (TCG_MAX_OP_SIZE * OPC_BUF_SIZE);
675 tcg_ctx.code_gen_max_blocks = tcg_ctx.code_gen_buffer_size /
676 CODE_GEN_AVG_BLOCK_SIZE;
677 tcg_ctx.tb_ctx.tbs =
678 g_malloc(tcg_ctx.code_gen_max_blocks * sizeof(TranslationBlock));
681 /* Must be called before using the QEMU cpus. 'tb_size' is the size
682 (in bytes) allocated to the translation buffer. Zero means default
683 size. */
684 void tcg_exec_init(unsigned long tb_size)
686 cpu_gen_init();
687 code_gen_alloc(tb_size);
688 tcg_ctx.code_gen_ptr = tcg_ctx.code_gen_buffer;
689 tcg_register_jit(tcg_ctx.code_gen_buffer, tcg_ctx.code_gen_buffer_size);
690 page_init();
691 #if defined(CONFIG_SOFTMMU)
692 /* There's no guest base to take into account, so go ahead and
693 initialize the prologue now. */
694 tcg_prologue_init(&tcg_ctx);
695 #endif
698 bool tcg_enabled(void)
700 return tcg_ctx.code_gen_buffer != NULL;
703 /* Allocate a new translation block. Flush the translation buffer if
704 too many translation blocks or too much generated code. */
705 static TranslationBlock *tb_alloc(target_ulong pc)
707 TranslationBlock *tb;
709 if (tcg_ctx.tb_ctx.nb_tbs >= tcg_ctx.code_gen_max_blocks ||
710 (tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer) >=
711 tcg_ctx.code_gen_buffer_max_size) {
712 return NULL;
714 tb = &tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs++];
715 tb->pc = pc;
716 tb->cflags = 0;
717 return tb;
720 void tb_free(TranslationBlock *tb)
722 /* In practice this is mostly used for single use temporary TB
723 Ignore the hard cases and just back up if this TB happens to
724 be the last one generated. */
725 if (tcg_ctx.tb_ctx.nb_tbs > 0 &&
726 tb == &tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs - 1]) {
727 tcg_ctx.code_gen_ptr = tb->tc_ptr;
728 tcg_ctx.tb_ctx.nb_tbs--;
732 static inline void invalidate_page_bitmap(PageDesc *p)
734 if (p->code_bitmap) {
735 g_free(p->code_bitmap);
736 p->code_bitmap = NULL;
738 p->code_write_count = 0;
741 /* Set to NULL all the 'first_tb' fields in all PageDescs. */
742 static void page_flush_tb_1(int level, void **lp)
744 int i;
746 if (*lp == NULL) {
747 return;
749 if (level == 0) {
750 PageDesc *pd = *lp;
752 for (i = 0; i < V_L2_SIZE; ++i) {
753 pd[i].first_tb = NULL;
754 invalidate_page_bitmap(pd + i);
756 } else {
757 void **pp = *lp;
759 for (i = 0; i < V_L2_SIZE; ++i) {
760 page_flush_tb_1(level - 1, pp + i);
765 static void page_flush_tb(void)
767 int i;
769 for (i = 0; i < V_L1_SIZE; i++) {
770 page_flush_tb_1(V_L1_SHIFT / V_L2_BITS - 1, l1_map + i);
774 /* flush all the translation blocks */
775 /* XXX: tb_flush is currently not thread safe */
776 void tb_flush(CPUState *cpu)
778 #if defined(DEBUG_FLUSH)
779 printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n",
780 (unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer),
781 tcg_ctx.tb_ctx.nb_tbs, tcg_ctx.tb_ctx.nb_tbs > 0 ?
782 ((unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer)) /
783 tcg_ctx.tb_ctx.nb_tbs : 0);
784 #endif
785 if ((unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer)
786 > tcg_ctx.code_gen_buffer_size) {
787 cpu_abort(cpu, "Internal error: code buffer overflow\n");
789 tcg_ctx.tb_ctx.nb_tbs = 0;
791 CPU_FOREACH(cpu) {
792 memset(cpu->tb_jmp_cache, 0, sizeof(cpu->tb_jmp_cache));
795 memset(tcg_ctx.tb_ctx.tb_phys_hash, 0, sizeof(tcg_ctx.tb_ctx.tb_phys_hash));
796 page_flush_tb();
798 tcg_ctx.code_gen_ptr = tcg_ctx.code_gen_buffer;
799 /* XXX: flush processor icache at this point if cache flush is
800 expensive */
801 tcg_ctx.tb_ctx.tb_flush_count++;
804 #ifdef DEBUG_TB_CHECK
806 static void tb_invalidate_check(target_ulong address)
808 TranslationBlock *tb;
809 int i;
811 address &= TARGET_PAGE_MASK;
812 for (i = 0; i < CODE_GEN_PHYS_HASH_SIZE; i++) {
813 for (tb = tb_ctx.tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
814 if (!(address + TARGET_PAGE_SIZE <= tb->pc ||
815 address >= tb->pc + tb->size)) {
816 printf("ERROR invalidate: address=" TARGET_FMT_lx
817 " PC=%08lx size=%04x\n",
818 address, (long)tb->pc, tb->size);
824 /* verify that all the pages have correct rights for code */
825 static void tb_page_check(void)
827 TranslationBlock *tb;
828 int i, flags1, flags2;
830 for (i = 0; i < CODE_GEN_PHYS_HASH_SIZE; i++) {
831 for (tb = tcg_ctx.tb_ctx.tb_phys_hash[i]; tb != NULL;
832 tb = tb->phys_hash_next) {
833 flags1 = page_get_flags(tb->pc);
834 flags2 = page_get_flags(tb->pc + tb->size - 1);
835 if ((flags1 & PAGE_WRITE) || (flags2 & PAGE_WRITE)) {
836 printf("ERROR page flags: PC=%08lx size=%04x f1=%x f2=%x\n",
837 (long)tb->pc, tb->size, flags1, flags2);
843 #endif
845 static inline void tb_hash_remove(TranslationBlock **ptb, TranslationBlock *tb)
847 TranslationBlock *tb1;
849 for (;;) {
850 tb1 = *ptb;
851 if (tb1 == tb) {
852 *ptb = tb1->phys_hash_next;
853 break;
855 ptb = &tb1->phys_hash_next;
859 static inline void tb_page_remove(TranslationBlock **ptb, TranslationBlock *tb)
861 TranslationBlock *tb1;
862 unsigned int n1;
864 for (;;) {
865 tb1 = *ptb;
866 n1 = (uintptr_t)tb1 & 3;
867 tb1 = (TranslationBlock *)((uintptr_t)tb1 & ~3);
868 if (tb1 == tb) {
869 *ptb = tb1->page_next[n1];
870 break;
872 ptb = &tb1->page_next[n1];
876 static inline void tb_jmp_remove(TranslationBlock *tb, int n)
878 TranslationBlock *tb1, **ptb;
879 unsigned int n1;
881 ptb = &tb->jmp_next[n];
882 tb1 = *ptb;
883 if (tb1) {
884 /* find tb(n) in circular list */
885 for (;;) {
886 tb1 = *ptb;
887 n1 = (uintptr_t)tb1 & 3;
888 tb1 = (TranslationBlock *)((uintptr_t)tb1 & ~3);
889 if (n1 == n && tb1 == tb) {
890 break;
892 if (n1 == 2) {
893 ptb = &tb1->jmp_first;
894 } else {
895 ptb = &tb1->jmp_next[n1];
898 /* now we can suppress tb(n) from the list */
899 *ptb = tb->jmp_next[n];
901 tb->jmp_next[n] = NULL;
905 /* reset the jump entry 'n' of a TB so that it is not chained to
906 another TB */
907 static inline void tb_reset_jump(TranslationBlock *tb, int n)
909 tb_set_jmp_target(tb, n, (uintptr_t)(tb->tc_ptr + tb->tb_next_offset[n]));
912 /* invalidate one TB */
913 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr)
915 CPUState *cpu;
916 PageDesc *p;
917 unsigned int h, n1;
918 tb_page_addr_t phys_pc;
919 TranslationBlock *tb1, *tb2;
921 /* remove the TB from the hash list */
922 phys_pc = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
923 h = tb_phys_hash_func(phys_pc);
924 tb_hash_remove(&tcg_ctx.tb_ctx.tb_phys_hash[h], tb);
926 /* remove the TB from the page list */
927 if (tb->page_addr[0] != page_addr) {
928 p = page_find(tb->page_addr[0] >> TARGET_PAGE_BITS);
929 tb_page_remove(&p->first_tb, tb);
930 invalidate_page_bitmap(p);
932 if (tb->page_addr[1] != -1 && tb->page_addr[1] != page_addr) {
933 p = page_find(tb->page_addr[1] >> TARGET_PAGE_BITS);
934 tb_page_remove(&p->first_tb, tb);
935 invalidate_page_bitmap(p);
938 tcg_ctx.tb_ctx.tb_invalidated_flag = 1;
940 /* remove the TB from the hash list */
941 h = tb_jmp_cache_hash_func(tb->pc);
942 CPU_FOREACH(cpu) {
943 if (cpu->tb_jmp_cache[h] == tb) {
944 cpu->tb_jmp_cache[h] = NULL;
948 /* suppress this TB from the two jump lists */
949 tb_jmp_remove(tb, 0);
950 tb_jmp_remove(tb, 1);
952 /* suppress any remaining jumps to this TB */
953 tb1 = tb->jmp_first;
954 for (;;) {
955 n1 = (uintptr_t)tb1 & 3;
956 if (n1 == 2) {
957 break;
959 tb1 = (TranslationBlock *)((uintptr_t)tb1 & ~3);
960 tb2 = tb1->jmp_next[n1];
961 tb_reset_jump(tb1, n1);
962 tb1->jmp_next[n1] = NULL;
963 tb1 = tb2;
965 tb->jmp_first = (TranslationBlock *)((uintptr_t)tb | 2); /* fail safe */
967 tcg_ctx.tb_ctx.tb_phys_invalidate_count++;
970 static void build_page_bitmap(PageDesc *p)
972 int n, tb_start, tb_end;
973 TranslationBlock *tb;
975 p->code_bitmap = bitmap_new(TARGET_PAGE_SIZE);
977 tb = p->first_tb;
978 while (tb != NULL) {
979 n = (uintptr_t)tb & 3;
980 tb = (TranslationBlock *)((uintptr_t)tb & ~3);
981 /* NOTE: this is subtle as a TB may span two physical pages */
982 if (n == 0) {
983 /* NOTE: tb_end may be after the end of the page, but
984 it is not a problem */
985 tb_start = tb->pc & ~TARGET_PAGE_MASK;
986 tb_end = tb_start + tb->size;
987 if (tb_end > TARGET_PAGE_SIZE) {
988 tb_end = TARGET_PAGE_SIZE;
990 } else {
991 tb_start = 0;
992 tb_end = ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
994 bitmap_set(p->code_bitmap, tb_start, tb_end - tb_start);
995 tb = tb->page_next[n];
999 TranslationBlock *tb_gen_code(CPUState *cpu,
1000 target_ulong pc, target_ulong cs_base,
1001 int flags, int cflags)
1003 CPUArchState *env = cpu->env_ptr;
1004 TranslationBlock *tb;
1005 tb_page_addr_t phys_pc, phys_page2;
1006 target_ulong virt_page2;
1007 int code_gen_size;
1009 phys_pc = get_page_addr_code(env, pc);
1010 if (use_icount) {
1011 cflags |= CF_USE_ICOUNT;
1013 tb = tb_alloc(pc);
1014 if (!tb) {
1015 /* flush must be done */
1016 tb_flush(cpu);
1017 /* cannot fail at this point */
1018 tb = tb_alloc(pc);
1019 /* Don't forget to invalidate previous TB info. */
1020 tcg_ctx.tb_ctx.tb_invalidated_flag = 1;
1022 tb->tc_ptr = tcg_ctx.code_gen_ptr;
1023 tb->cs_base = cs_base;
1024 tb->flags = flags;
1025 tb->cflags = cflags;
1026 cpu_gen_code(env, tb, &code_gen_size);
1027 tcg_ctx.code_gen_ptr = (void *)(((uintptr_t)tcg_ctx.code_gen_ptr +
1028 code_gen_size + CODE_GEN_ALIGN - 1) & ~(CODE_GEN_ALIGN - 1));
1030 /* check next page if needed */
1031 virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
1032 phys_page2 = -1;
1033 if ((pc & TARGET_PAGE_MASK) != virt_page2) {
1034 phys_page2 = get_page_addr_code(env, virt_page2);
1036 tb_link_page(tb, phys_pc, phys_page2);
1037 return tb;
1041 * Invalidate all TBs which intersect with the target physical address range
1042 * [start;end[. NOTE: start and end may refer to *different* physical pages.
1043 * 'is_cpu_write_access' should be true if called from a real cpu write
1044 * access: the virtual CPU will exit the current TB if code is modified inside
1045 * this TB.
1047 void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end)
1049 while (start < end) {
1050 tb_invalidate_phys_page_range(start, end, 0);
1051 start &= TARGET_PAGE_MASK;
1052 start += TARGET_PAGE_SIZE;
1057 * Invalidate all TBs which intersect with the target physical address range
1058 * [start;end[. NOTE: start and end must refer to the *same* physical page.
1059 * 'is_cpu_write_access' should be true if called from a real cpu write
1060 * access: the virtual CPU will exit the current TB if code is modified inside
1061 * this TB.
1063 void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
1064 int is_cpu_write_access)
1066 TranslationBlock *tb, *tb_next, *saved_tb;
1067 CPUState *cpu = current_cpu;
1068 #if defined(TARGET_HAS_PRECISE_SMC)
1069 CPUArchState *env = NULL;
1070 #endif
1071 tb_page_addr_t tb_start, tb_end;
1072 PageDesc *p;
1073 int n;
1074 #ifdef TARGET_HAS_PRECISE_SMC
1075 int current_tb_not_found = is_cpu_write_access;
1076 TranslationBlock *current_tb = NULL;
1077 int current_tb_modified = 0;
1078 target_ulong current_pc = 0;
1079 target_ulong current_cs_base = 0;
1080 int current_flags = 0;
1081 #endif /* TARGET_HAS_PRECISE_SMC */
1083 p = page_find(start >> TARGET_PAGE_BITS);
1084 if (!p) {
1085 return;
1087 #if defined(TARGET_HAS_PRECISE_SMC)
1088 if (cpu != NULL) {
1089 env = cpu->env_ptr;
1091 #endif
1093 /* we remove all the TBs in the range [start, end[ */
1094 /* XXX: see if in some cases it could be faster to invalidate all
1095 the code */
1096 tb = p->first_tb;
1097 while (tb != NULL) {
1098 n = (uintptr_t)tb & 3;
1099 tb = (TranslationBlock *)((uintptr_t)tb & ~3);
1100 tb_next = tb->page_next[n];
1101 /* NOTE: this is subtle as a TB may span two physical pages */
1102 if (n == 0) {
1103 /* NOTE: tb_end may be after the end of the page, but
1104 it is not a problem */
1105 tb_start = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
1106 tb_end = tb_start + tb->size;
1107 } else {
1108 tb_start = tb->page_addr[1];
1109 tb_end = tb_start + ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
1111 if (!(tb_end <= start || tb_start >= end)) {
1112 #ifdef TARGET_HAS_PRECISE_SMC
1113 if (current_tb_not_found) {
1114 current_tb_not_found = 0;
1115 current_tb = NULL;
1116 if (cpu->mem_io_pc) {
1117 /* now we have a real cpu fault */
1118 current_tb = tb_find_pc(cpu->mem_io_pc);
1121 if (current_tb == tb &&
1122 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1123 /* If we are modifying the current TB, we must stop
1124 its execution. We could be more precise by checking
1125 that the modification is after the current PC, but it
1126 would require a specialized function to partially
1127 restore the CPU state */
1129 current_tb_modified = 1;
1130 cpu_restore_state_from_tb(cpu, current_tb, cpu->mem_io_pc);
1131 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1132 &current_flags);
1134 #endif /* TARGET_HAS_PRECISE_SMC */
1135 /* we need to do that to handle the case where a signal
1136 occurs while doing tb_phys_invalidate() */
1137 saved_tb = NULL;
1138 if (cpu != NULL) {
1139 saved_tb = cpu->current_tb;
1140 cpu->current_tb = NULL;
1142 tb_phys_invalidate(tb, -1);
1143 if (cpu != NULL) {
1144 cpu->current_tb = saved_tb;
1145 if (cpu->interrupt_request && cpu->current_tb) {
1146 cpu_interrupt(cpu, cpu->interrupt_request);
1150 tb = tb_next;
1152 #if !defined(CONFIG_USER_ONLY)
1153 /* if no code remaining, no need to continue to use slow writes */
1154 if (!p->first_tb) {
1155 invalidate_page_bitmap(p);
1156 tlb_unprotect_code(start);
1158 #endif
1159 #ifdef TARGET_HAS_PRECISE_SMC
1160 if (current_tb_modified) {
1161 /* we generate a block containing just the instruction
1162 modifying the memory. It will ensure that it cannot modify
1163 itself */
1164 cpu->current_tb = NULL;
1165 tb_gen_code(cpu, current_pc, current_cs_base, current_flags, 1);
1166 cpu_resume_from_signal(cpu, NULL);
1168 #endif
1171 /* len must be <= 8 and start must be a multiple of len */
1172 void tb_invalidate_phys_page_fast(tb_page_addr_t start, int len)
1174 PageDesc *p;
1176 #if 0
1177 if (1) {
1178 qemu_log("modifying code at 0x%x size=%d EIP=%x PC=%08x\n",
1179 cpu_single_env->mem_io_vaddr, len,
1180 cpu_single_env->eip,
1181 cpu_single_env->eip +
1182 (intptr_t)cpu_single_env->segs[R_CS].base);
1184 #endif
1185 p = page_find(start >> TARGET_PAGE_BITS);
1186 if (!p) {
1187 return;
1189 if (!p->code_bitmap &&
1190 ++p->code_write_count >= SMC_BITMAP_USE_THRESHOLD) {
1191 /* build code bitmap */
1192 build_page_bitmap(p);
1194 if (p->code_bitmap) {
1195 unsigned int nr;
1196 unsigned long b;
1198 nr = start & ~TARGET_PAGE_MASK;
1199 b = p->code_bitmap[BIT_WORD(nr)] >> (nr & (BITS_PER_LONG - 1));
1200 if (b & ((1 << len) - 1)) {
1201 goto do_invalidate;
1203 } else {
1204 do_invalidate:
1205 tb_invalidate_phys_page_range(start, start + len, 1);
1209 #if !defined(CONFIG_SOFTMMU)
1210 static void tb_invalidate_phys_page(tb_page_addr_t addr,
1211 uintptr_t pc, void *puc,
1212 bool locked)
1214 TranslationBlock *tb;
1215 PageDesc *p;
1216 int n;
1217 #ifdef TARGET_HAS_PRECISE_SMC
1218 TranslationBlock *current_tb = NULL;
1219 CPUState *cpu = current_cpu;
1220 CPUArchState *env = NULL;
1221 int current_tb_modified = 0;
1222 target_ulong current_pc = 0;
1223 target_ulong current_cs_base = 0;
1224 int current_flags = 0;
1225 #endif
1227 addr &= TARGET_PAGE_MASK;
1228 p = page_find(addr >> TARGET_PAGE_BITS);
1229 if (!p) {
1230 return;
1232 tb = p->first_tb;
1233 #ifdef TARGET_HAS_PRECISE_SMC
1234 if (tb && pc != 0) {
1235 current_tb = tb_find_pc(pc);
1237 if (cpu != NULL) {
1238 env = cpu->env_ptr;
1240 #endif
1241 while (tb != NULL) {
1242 n = (uintptr_t)tb & 3;
1243 tb = (TranslationBlock *)((uintptr_t)tb & ~3);
1244 #ifdef TARGET_HAS_PRECISE_SMC
1245 if (current_tb == tb &&
1246 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1247 /* If we are modifying the current TB, we must stop
1248 its execution. We could be more precise by checking
1249 that the modification is after the current PC, but it
1250 would require a specialized function to partially
1251 restore the CPU state */
1253 current_tb_modified = 1;
1254 cpu_restore_state_from_tb(cpu, current_tb, pc);
1255 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1256 &current_flags);
1258 #endif /* TARGET_HAS_PRECISE_SMC */
1259 tb_phys_invalidate(tb, addr);
1260 tb = tb->page_next[n];
1262 p->first_tb = NULL;
1263 #ifdef TARGET_HAS_PRECISE_SMC
1264 if (current_tb_modified) {
1265 /* we generate a block containing just the instruction
1266 modifying the memory. It will ensure that it cannot modify
1267 itself */
1268 cpu->current_tb = NULL;
1269 tb_gen_code(cpu, current_pc, current_cs_base, current_flags, 1);
1270 if (locked) {
1271 mmap_unlock();
1273 cpu_resume_from_signal(cpu, puc);
1275 #endif
1277 #endif
1279 /* add the tb in the target page and protect it if necessary */
1280 static inline void tb_alloc_page(TranslationBlock *tb,
1281 unsigned int n, tb_page_addr_t page_addr)
1283 PageDesc *p;
1284 #ifndef CONFIG_USER_ONLY
1285 bool page_already_protected;
1286 #endif
1288 tb->page_addr[n] = page_addr;
1289 p = page_find_alloc(page_addr >> TARGET_PAGE_BITS, 1);
1290 tb->page_next[n] = p->first_tb;
1291 #ifndef CONFIG_USER_ONLY
1292 page_already_protected = p->first_tb != NULL;
1293 #endif
1294 p->first_tb = (TranslationBlock *)((uintptr_t)tb | n);
1295 invalidate_page_bitmap(p);
1297 #if defined(CONFIG_USER_ONLY)
1298 if (p->flags & PAGE_WRITE) {
1299 target_ulong addr;
1300 PageDesc *p2;
1301 int prot;
1303 /* force the host page as non writable (writes will have a
1304 page fault + mprotect overhead) */
1305 page_addr &= qemu_host_page_mask;
1306 prot = 0;
1307 for (addr = page_addr; addr < page_addr + qemu_host_page_size;
1308 addr += TARGET_PAGE_SIZE) {
1310 p2 = page_find(addr >> TARGET_PAGE_BITS);
1311 if (!p2) {
1312 continue;
1314 prot |= p2->flags;
1315 p2->flags &= ~PAGE_WRITE;
1317 mprotect(g2h(page_addr), qemu_host_page_size,
1318 (prot & PAGE_BITS) & ~PAGE_WRITE);
1319 #ifdef DEBUG_TB_INVALIDATE
1320 printf("protecting code page: 0x" TARGET_FMT_lx "\n",
1321 page_addr);
1322 #endif
1324 #else
1325 /* if some code is already present, then the pages are already
1326 protected. So we handle the case where only the first TB is
1327 allocated in a physical page */
1328 if (!page_already_protected) {
1329 tlb_protect_code(page_addr);
1331 #endif
1334 /* add a new TB and link it to the physical page tables. phys_page2 is
1335 (-1) to indicate that only one page contains the TB. */
1336 static void tb_link_page(TranslationBlock *tb, tb_page_addr_t phys_pc,
1337 tb_page_addr_t phys_page2)
1339 unsigned int h;
1340 TranslationBlock **ptb;
1342 /* Grab the mmap lock to stop another thread invalidating this TB
1343 before we are done. */
1344 mmap_lock();
1345 /* add in the physical hash table */
1346 h = tb_phys_hash_func(phys_pc);
1347 ptb = &tcg_ctx.tb_ctx.tb_phys_hash[h];
1348 tb->phys_hash_next = *ptb;
1349 *ptb = tb;
1351 /* add in the page list */
1352 tb_alloc_page(tb, 0, phys_pc & TARGET_PAGE_MASK);
1353 if (phys_page2 != -1) {
1354 tb_alloc_page(tb, 1, phys_page2);
1355 } else {
1356 tb->page_addr[1] = -1;
1359 tb->jmp_first = (TranslationBlock *)((uintptr_t)tb | 2);
1360 tb->jmp_next[0] = NULL;
1361 tb->jmp_next[1] = NULL;
1363 /* init original jump addresses */
1364 if (tb->tb_next_offset[0] != 0xffff) {
1365 tb_reset_jump(tb, 0);
1367 if (tb->tb_next_offset[1] != 0xffff) {
1368 tb_reset_jump(tb, 1);
1371 #ifdef DEBUG_TB_CHECK
1372 tb_page_check();
1373 #endif
1374 mmap_unlock();
1377 /* find the TB 'tb' such that tb[0].tc_ptr <= tc_ptr <
1378 tb[1].tc_ptr. Return NULL if not found */
1379 static TranslationBlock *tb_find_pc(uintptr_t tc_ptr)
1381 int m_min, m_max, m;
1382 uintptr_t v;
1383 TranslationBlock *tb;
1385 if (tcg_ctx.tb_ctx.nb_tbs <= 0) {
1386 return NULL;
1388 if (tc_ptr < (uintptr_t)tcg_ctx.code_gen_buffer ||
1389 tc_ptr >= (uintptr_t)tcg_ctx.code_gen_ptr) {
1390 return NULL;
1392 /* binary search (cf Knuth) */
1393 m_min = 0;
1394 m_max = tcg_ctx.tb_ctx.nb_tbs - 1;
1395 while (m_min <= m_max) {
1396 m = (m_min + m_max) >> 1;
1397 tb = &tcg_ctx.tb_ctx.tbs[m];
1398 v = (uintptr_t)tb->tc_ptr;
1399 if (v == tc_ptr) {
1400 return tb;
1401 } else if (tc_ptr < v) {
1402 m_max = m - 1;
1403 } else {
1404 m_min = m + 1;
1407 return &tcg_ctx.tb_ctx.tbs[m_max];
1410 #if !defined(CONFIG_USER_ONLY)
1411 void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr)
1413 ram_addr_t ram_addr;
1414 MemoryRegion *mr;
1415 hwaddr l = 1;
1417 rcu_read_lock();
1418 mr = address_space_translate(as, addr, &addr, &l, false);
1419 if (!(memory_region_is_ram(mr)
1420 || memory_region_is_romd(mr))) {
1421 rcu_read_unlock();
1422 return;
1424 ram_addr = (memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK)
1425 + addr;
1426 tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
1427 rcu_read_unlock();
1429 #endif /* !defined(CONFIG_USER_ONLY) */
1431 void tb_check_watchpoint(CPUState *cpu)
1433 TranslationBlock *tb;
1435 tb = tb_find_pc(cpu->mem_io_pc);
1436 if (tb) {
1437 /* We can use retranslation to find the PC. */
1438 cpu_restore_state_from_tb(cpu, tb, cpu->mem_io_pc);
1439 tb_phys_invalidate(tb, -1);
1440 } else {
1441 /* The exception probably happened in a helper. The CPU state should
1442 have been saved before calling it. Fetch the PC from there. */
1443 CPUArchState *env = cpu->env_ptr;
1444 target_ulong pc, cs_base;
1445 tb_page_addr_t addr;
1446 int flags;
1448 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
1449 addr = get_page_addr_code(env, pc);
1450 tb_invalidate_phys_range(addr, addr + 1);
1454 #ifndef CONFIG_USER_ONLY
1455 /* mask must never be zero, except for A20 change call */
1456 static void tcg_handle_interrupt(CPUState *cpu, int mask)
1458 int old_mask;
1460 old_mask = cpu->interrupt_request;
1461 cpu->interrupt_request |= mask;
1464 * If called from iothread context, wake the target cpu in
1465 * case its halted.
1467 if (!qemu_cpu_is_self(cpu)) {
1468 qemu_cpu_kick(cpu);
1469 return;
1472 if (use_icount) {
1473 cpu->icount_decr.u16.high = 0xffff;
1474 if (!cpu->can_do_io
1475 && (mask & ~old_mask) != 0) {
1476 cpu_abort(cpu, "Raised interrupt while not in I/O function");
1478 } else {
1479 cpu->tcg_exit_req = 1;
1483 CPUInterruptHandler cpu_interrupt_handler = tcg_handle_interrupt;
1485 /* in deterministic execution mode, instructions doing device I/Os
1486 must be at the end of the TB */
1487 void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
1489 #if defined(TARGET_MIPS) || defined(TARGET_SH4)
1490 CPUArchState *env = cpu->env_ptr;
1491 #endif
1492 TranslationBlock *tb;
1493 uint32_t n, cflags;
1494 target_ulong pc, cs_base;
1495 uint64_t flags;
1497 tb = tb_find_pc(retaddr);
1498 if (!tb) {
1499 cpu_abort(cpu, "cpu_io_recompile: could not find TB for pc=%p",
1500 (void *)retaddr);
1502 n = cpu->icount_decr.u16.low + tb->icount;
1503 cpu_restore_state_from_tb(cpu, tb, retaddr);
1504 /* Calculate how many instructions had been executed before the fault
1505 occurred. */
1506 n = n - cpu->icount_decr.u16.low;
1507 /* Generate a new TB ending on the I/O insn. */
1508 n++;
1509 /* On MIPS and SH, delay slot instructions can only be restarted if
1510 they were already the first instruction in the TB. If this is not
1511 the first instruction in a TB then re-execute the preceding
1512 branch. */
1513 #if defined(TARGET_MIPS)
1514 if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
1515 env->active_tc.PC -= (env->hflags & MIPS_HFLAG_B16 ? 2 : 4);
1516 cpu->icount_decr.u16.low++;
1517 env->hflags &= ~MIPS_HFLAG_BMASK;
1519 #elif defined(TARGET_SH4)
1520 if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
1521 && n > 1) {
1522 env->pc -= 2;
1523 cpu->icount_decr.u16.low++;
1524 env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
1526 #endif
1527 /* This should never happen. */
1528 if (n > CF_COUNT_MASK) {
1529 cpu_abort(cpu, "TB too big during recompile");
1532 cflags = n | CF_LAST_IO;
1533 pc = tb->pc;
1534 cs_base = tb->cs_base;
1535 flags = tb->flags;
1536 tb_phys_invalidate(tb, -1);
1537 if (tb->cflags & CF_NOCACHE) {
1538 if (tb->orig_tb) {
1539 /* Invalidate original TB if this TB was generated in
1540 * cpu_exec_nocache() */
1541 tb_phys_invalidate(tb->orig_tb, -1);
1543 tb_free(tb);
1545 /* FIXME: In theory this could raise an exception. In practice
1546 we have already translated the block once so it's probably ok. */
1547 tb_gen_code(cpu, pc, cs_base, flags, cflags);
1548 /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
1549 the first in the TB) then we end up generating a whole new TB and
1550 repeating the fault, which is horribly inefficient.
1551 Better would be to execute just this insn uncached, or generate a
1552 second new TB. */
1553 cpu_resume_from_signal(cpu, NULL);
1556 void tb_flush_jmp_cache(CPUState *cpu, target_ulong addr)
1558 unsigned int i;
1560 /* Discard jump cache entries for any tb which might potentially
1561 overlap the flushed page. */
1562 i = tb_jmp_cache_hash_page(addr - TARGET_PAGE_SIZE);
1563 memset(&cpu->tb_jmp_cache[i], 0,
1564 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1566 i = tb_jmp_cache_hash_page(addr);
1567 memset(&cpu->tb_jmp_cache[i], 0,
1568 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1571 void dump_exec_info(FILE *f, fprintf_function cpu_fprintf)
1573 int i, target_code_size, max_target_code_size;
1574 int direct_jmp_count, direct_jmp2_count, cross_page;
1575 TranslationBlock *tb;
1577 target_code_size = 0;
1578 max_target_code_size = 0;
1579 cross_page = 0;
1580 direct_jmp_count = 0;
1581 direct_jmp2_count = 0;
1582 for (i = 0; i < tcg_ctx.tb_ctx.nb_tbs; i++) {
1583 tb = &tcg_ctx.tb_ctx.tbs[i];
1584 target_code_size += tb->size;
1585 if (tb->size > max_target_code_size) {
1586 max_target_code_size = tb->size;
1588 if (tb->page_addr[1] != -1) {
1589 cross_page++;
1591 if (tb->tb_next_offset[0] != 0xffff) {
1592 direct_jmp_count++;
1593 if (tb->tb_next_offset[1] != 0xffff) {
1594 direct_jmp2_count++;
1598 /* XXX: avoid using doubles ? */
1599 cpu_fprintf(f, "Translation buffer state:\n");
1600 cpu_fprintf(f, "gen code size %td/%zd\n",
1601 tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer,
1602 tcg_ctx.code_gen_buffer_max_size);
1603 cpu_fprintf(f, "TB count %d/%d\n",
1604 tcg_ctx.tb_ctx.nb_tbs, tcg_ctx.code_gen_max_blocks);
1605 cpu_fprintf(f, "TB avg target size %d max=%d bytes\n",
1606 tcg_ctx.tb_ctx.nb_tbs ? target_code_size /
1607 tcg_ctx.tb_ctx.nb_tbs : 0,
1608 max_target_code_size);
1609 cpu_fprintf(f, "TB avg host size %td bytes (expansion ratio: %0.1f)\n",
1610 tcg_ctx.tb_ctx.nb_tbs ? (tcg_ctx.code_gen_ptr -
1611 tcg_ctx.code_gen_buffer) /
1612 tcg_ctx.tb_ctx.nb_tbs : 0,
1613 target_code_size ? (double) (tcg_ctx.code_gen_ptr -
1614 tcg_ctx.code_gen_buffer) /
1615 target_code_size : 0);
1616 cpu_fprintf(f, "cross page TB count %d (%d%%)\n", cross_page,
1617 tcg_ctx.tb_ctx.nb_tbs ? (cross_page * 100) /
1618 tcg_ctx.tb_ctx.nb_tbs : 0);
1619 cpu_fprintf(f, "direct jump count %d (%d%%) (2 jumps=%d %d%%)\n",
1620 direct_jmp_count,
1621 tcg_ctx.tb_ctx.nb_tbs ? (direct_jmp_count * 100) /
1622 tcg_ctx.tb_ctx.nb_tbs : 0,
1623 direct_jmp2_count,
1624 tcg_ctx.tb_ctx.nb_tbs ? (direct_jmp2_count * 100) /
1625 tcg_ctx.tb_ctx.nb_tbs : 0);
1626 cpu_fprintf(f, "\nStatistics:\n");
1627 cpu_fprintf(f, "TB flush count %d\n", tcg_ctx.tb_ctx.tb_flush_count);
1628 cpu_fprintf(f, "TB invalidate count %d\n",
1629 tcg_ctx.tb_ctx.tb_phys_invalidate_count);
1630 cpu_fprintf(f, "TLB flush count %d\n", tlb_flush_count);
1631 tcg_dump_info(f, cpu_fprintf);
1634 void dump_opcount_info(FILE *f, fprintf_function cpu_fprintf)
1636 tcg_dump_op_count(f, cpu_fprintf);
1639 #else /* CONFIG_USER_ONLY */
1641 void cpu_interrupt(CPUState *cpu, int mask)
1643 cpu->interrupt_request |= mask;
1644 cpu->tcg_exit_req = 1;
1648 * Walks guest process memory "regions" one by one
1649 * and calls callback function 'fn' for each region.
1651 struct walk_memory_regions_data {
1652 walk_memory_regions_fn fn;
1653 void *priv;
1654 target_ulong start;
1655 int prot;
1658 static int walk_memory_regions_end(struct walk_memory_regions_data *data,
1659 target_ulong end, int new_prot)
1661 if (data->start != -1u) {
1662 int rc = data->fn(data->priv, data->start, end, data->prot);
1663 if (rc != 0) {
1664 return rc;
1668 data->start = (new_prot ? end : -1u);
1669 data->prot = new_prot;
1671 return 0;
1674 static int walk_memory_regions_1(struct walk_memory_regions_data *data,
1675 target_ulong base, int level, void **lp)
1677 target_ulong pa;
1678 int i, rc;
1680 if (*lp == NULL) {
1681 return walk_memory_regions_end(data, base, 0);
1684 if (level == 0) {
1685 PageDesc *pd = *lp;
1687 for (i = 0; i < V_L2_SIZE; ++i) {
1688 int prot = pd[i].flags;
1690 pa = base | (i << TARGET_PAGE_BITS);
1691 if (prot != data->prot) {
1692 rc = walk_memory_regions_end(data, pa, prot);
1693 if (rc != 0) {
1694 return rc;
1698 } else {
1699 void **pp = *lp;
1701 for (i = 0; i < V_L2_SIZE; ++i) {
1702 pa = base | ((target_ulong)i <<
1703 (TARGET_PAGE_BITS + V_L2_BITS * level));
1704 rc = walk_memory_regions_1(data, pa, level - 1, pp + i);
1705 if (rc != 0) {
1706 return rc;
1711 return 0;
1714 int walk_memory_regions(void *priv, walk_memory_regions_fn fn)
1716 struct walk_memory_regions_data data;
1717 uintptr_t i;
1719 data.fn = fn;
1720 data.priv = priv;
1721 data.start = -1u;
1722 data.prot = 0;
1724 for (i = 0; i < V_L1_SIZE; i++) {
1725 int rc = walk_memory_regions_1(&data, (target_ulong)i << (V_L1_SHIFT + TARGET_PAGE_BITS),
1726 V_L1_SHIFT / V_L2_BITS - 1, l1_map + i);
1727 if (rc != 0) {
1728 return rc;
1732 return walk_memory_regions_end(&data, 0, 0);
1735 static int dump_region(void *priv, target_ulong start,
1736 target_ulong end, unsigned long prot)
1738 FILE *f = (FILE *)priv;
1740 (void) fprintf(f, TARGET_FMT_lx"-"TARGET_FMT_lx
1741 " "TARGET_FMT_lx" %c%c%c\n",
1742 start, end, end - start,
1743 ((prot & PAGE_READ) ? 'r' : '-'),
1744 ((prot & PAGE_WRITE) ? 'w' : '-'),
1745 ((prot & PAGE_EXEC) ? 'x' : '-'));
1747 return 0;
1750 /* dump memory mappings */
1751 void page_dump(FILE *f)
1753 const int length = sizeof(target_ulong) * 2;
1754 (void) fprintf(f, "%-*s %-*s %-*s %s\n",
1755 length, "start", length, "end", length, "size", "prot");
1756 walk_memory_regions(f, dump_region);
1759 int page_get_flags(target_ulong address)
1761 PageDesc *p;
1763 p = page_find(address >> TARGET_PAGE_BITS);
1764 if (!p) {
1765 return 0;
1767 return p->flags;
1770 /* Modify the flags of a page and invalidate the code if necessary.
1771 The flag PAGE_WRITE_ORG is positioned automatically depending
1772 on PAGE_WRITE. The mmap_lock should already be held. */
1773 void page_set_flags(target_ulong start, target_ulong end, int flags)
1775 target_ulong addr, len;
1777 /* This function should never be called with addresses outside the
1778 guest address space. If this assert fires, it probably indicates
1779 a missing call to h2g_valid. */
1780 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
1781 assert(end < ((target_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
1782 #endif
1783 assert(start < end);
1785 start = start & TARGET_PAGE_MASK;
1786 end = TARGET_PAGE_ALIGN(end);
1788 if (flags & PAGE_WRITE) {
1789 flags |= PAGE_WRITE_ORG;
1792 for (addr = start, len = end - start;
1793 len != 0;
1794 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
1795 PageDesc *p = page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
1797 /* If the write protection bit is set, then we invalidate
1798 the code inside. */
1799 if (!(p->flags & PAGE_WRITE) &&
1800 (flags & PAGE_WRITE) &&
1801 p->first_tb) {
1802 tb_invalidate_phys_page(addr, 0, NULL, false);
1804 p->flags = flags;
1808 int page_check_range(target_ulong start, target_ulong len, int flags)
1810 PageDesc *p;
1811 target_ulong end;
1812 target_ulong addr;
1814 /* This function should never be called with addresses outside the
1815 guest address space. If this assert fires, it probably indicates
1816 a missing call to h2g_valid. */
1817 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
1818 assert(start < ((target_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
1819 #endif
1821 if (len == 0) {
1822 return 0;
1824 if (start + len - 1 < start) {
1825 /* We've wrapped around. */
1826 return -1;
1829 /* must do before we loose bits in the next step */
1830 end = TARGET_PAGE_ALIGN(start + len);
1831 start = start & TARGET_PAGE_MASK;
1833 for (addr = start, len = end - start;
1834 len != 0;
1835 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
1836 p = page_find(addr >> TARGET_PAGE_BITS);
1837 if (!p) {
1838 return -1;
1840 if (!(p->flags & PAGE_VALID)) {
1841 return -1;
1844 if ((flags & PAGE_READ) && !(p->flags & PAGE_READ)) {
1845 return -1;
1847 if (flags & PAGE_WRITE) {
1848 if (!(p->flags & PAGE_WRITE_ORG)) {
1849 return -1;
1851 /* unprotect the page if it was put read-only because it
1852 contains translated code */
1853 if (!(p->flags & PAGE_WRITE)) {
1854 if (!page_unprotect(addr, 0, NULL)) {
1855 return -1;
1860 return 0;
1863 /* called from signal handler: invalidate the code and unprotect the
1864 page. Return TRUE if the fault was successfully handled. */
1865 int page_unprotect(target_ulong address, uintptr_t pc, void *puc)
1867 unsigned int prot;
1868 PageDesc *p;
1869 target_ulong host_start, host_end, addr;
1871 /* Technically this isn't safe inside a signal handler. However we
1872 know this only ever happens in a synchronous SEGV handler, so in
1873 practice it seems to be ok. */
1874 mmap_lock();
1876 p = page_find(address >> TARGET_PAGE_BITS);
1877 if (!p) {
1878 mmap_unlock();
1879 return 0;
1882 /* if the page was really writable, then we change its
1883 protection back to writable */
1884 if ((p->flags & PAGE_WRITE_ORG) && !(p->flags & PAGE_WRITE)) {
1885 host_start = address & qemu_host_page_mask;
1886 host_end = host_start + qemu_host_page_size;
1888 prot = 0;
1889 for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) {
1890 p = page_find(addr >> TARGET_PAGE_BITS);
1891 p->flags |= PAGE_WRITE;
1892 prot |= p->flags;
1894 /* and since the content will be modified, we must invalidate
1895 the corresponding translated code. */
1896 tb_invalidate_phys_page(addr, pc, puc, true);
1897 #ifdef DEBUG_TB_CHECK
1898 tb_invalidate_check(addr);
1899 #endif
1901 mprotect((void *)g2h(host_start), qemu_host_page_size,
1902 prot & PAGE_BITS);
1904 mmap_unlock();
1905 return 1;
1907 mmap_unlock();
1908 return 0;
1910 #endif /* CONFIG_USER_ONLY */