vfio/pci: Rename INTx functions for easier tracing
[qemu/kevin.git] / translate-all.c
blob4a9ee33dac60481a5729d696aac3d37da887ca43
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_host_page_size;
121 uintptr_t qemu_host_page_mask;
123 /* The bottom level has pointers to PageDesc */
124 static void *l1_map[V_L1_SIZE];
126 /* code generation context */
127 TCGContext tcg_ctx;
129 /* translation block context */
130 #ifdef CONFIG_USER_ONLY
131 __thread int have_tb_lock;
132 #endif
134 void tb_lock(void)
136 #ifdef CONFIG_USER_ONLY
137 assert(!have_tb_lock);
138 qemu_mutex_lock(&tcg_ctx.tb_ctx.tb_lock);
139 have_tb_lock++;
140 #endif
143 void tb_unlock(void)
145 #ifdef CONFIG_USER_ONLY
146 assert(have_tb_lock);
147 have_tb_lock--;
148 qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock);
149 #endif
152 void tb_lock_reset(void)
154 #ifdef CONFIG_USER_ONLY
155 if (have_tb_lock) {
156 qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock);
157 have_tb_lock = 0;
159 #endif
162 static void tb_link_page(TranslationBlock *tb, tb_page_addr_t phys_pc,
163 tb_page_addr_t phys_page2);
164 static TranslationBlock *tb_find_pc(uintptr_t tc_ptr);
166 void cpu_gen_init(void)
168 tcg_context_init(&tcg_ctx);
171 /* return non zero if the very first instruction is invalid so that
172 * the virtual CPU can trigger an exception.
174 * '*gen_code_size_ptr' contains the size of the generated code (host
175 * code).
177 * Called with mmap_lock held for user-mode emulation.
179 int cpu_gen_code(CPUArchState *env, TranslationBlock *tb, int *gen_code_size_ptr)
181 TCGContext *s = &tcg_ctx;
182 tcg_insn_unit *gen_code_buf;
183 int gen_code_size;
184 #ifdef CONFIG_PROFILER
185 int64_t ti;
186 #endif
188 #ifdef CONFIG_PROFILER
189 s->tb_count1++; /* includes aborted translations because of
190 exceptions */
191 ti = profile_getclock();
192 #endif
193 tcg_func_start(s);
195 gen_intermediate_code(env, tb);
197 trace_translate_block(tb, tb->pc, tb->tc_ptr);
199 /* generate machine code */
200 gen_code_buf = tb->tc_ptr;
201 tb->tb_next_offset[0] = 0xffff;
202 tb->tb_next_offset[1] = 0xffff;
203 s->tb_next_offset = tb->tb_next_offset;
204 #ifdef USE_DIRECT_JUMP
205 s->tb_jmp_offset = tb->tb_jmp_offset;
206 s->tb_next = NULL;
207 #else
208 s->tb_jmp_offset = NULL;
209 s->tb_next = tb->tb_next;
210 #endif
212 #ifdef CONFIG_PROFILER
213 s->tb_count++;
214 s->interm_time += profile_getclock() - ti;
215 s->code_time -= profile_getclock();
216 #endif
217 gen_code_size = tcg_gen_code(s, gen_code_buf);
218 *gen_code_size_ptr = gen_code_size;
219 #ifdef CONFIG_PROFILER
220 s->code_time += profile_getclock();
221 s->code_in_len += tb->size;
222 s->code_out_len += gen_code_size;
223 #endif
225 #ifdef DEBUG_DISAS
226 if (qemu_loglevel_mask(CPU_LOG_TB_OUT_ASM)) {
227 qemu_log("OUT: [size=%d]\n", gen_code_size);
228 log_disas(tb->tc_ptr, gen_code_size);
229 qemu_log("\n");
230 qemu_log_flush();
232 #endif
233 return 0;
236 /* The cpu state corresponding to 'searched_pc' is restored.
238 static int cpu_restore_state_from_tb(CPUState *cpu, TranslationBlock *tb,
239 uintptr_t searched_pc)
241 CPUArchState *env = cpu->env_ptr;
242 TCGContext *s = &tcg_ctx;
243 int j;
244 uintptr_t tc_ptr;
245 #ifdef CONFIG_PROFILER
246 int64_t ti;
247 #endif
249 #ifdef CONFIG_PROFILER
250 ti = profile_getclock();
251 #endif
252 tcg_func_start(s);
254 gen_intermediate_code_pc(env, tb);
256 if (tb->cflags & CF_USE_ICOUNT) {
257 assert(use_icount);
258 /* Reset the cycle counter to the start of the block. */
259 cpu->icount_decr.u16.low += tb->icount;
260 /* Clear the IO flag. */
261 cpu->can_do_io = 0;
264 /* find opc index corresponding to search_pc */
265 tc_ptr = (uintptr_t)tb->tc_ptr;
266 if (searched_pc < tc_ptr)
267 return -1;
269 s->tb_next_offset = tb->tb_next_offset;
270 #ifdef USE_DIRECT_JUMP
271 s->tb_jmp_offset = tb->tb_jmp_offset;
272 s->tb_next = NULL;
273 #else
274 s->tb_jmp_offset = NULL;
275 s->tb_next = tb->tb_next;
276 #endif
277 j = tcg_gen_code_search_pc(s, (tcg_insn_unit *)tc_ptr,
278 searched_pc - tc_ptr);
279 if (j < 0)
280 return -1;
281 /* now find start of instruction before */
282 while (s->gen_opc_instr_start[j] == 0) {
283 j--;
285 cpu->icount_decr.u16.low -= s->gen_opc_icount[j];
287 restore_state_to_opc(env, tb, j);
289 #ifdef CONFIG_PROFILER
290 s->restore_time += profile_getclock() - ti;
291 s->restore_count++;
292 #endif
293 return 0;
296 bool cpu_restore_state(CPUState *cpu, uintptr_t retaddr)
298 TranslationBlock *tb;
300 tb = tb_find_pc(retaddr);
301 if (tb) {
302 cpu_restore_state_from_tb(cpu, tb, retaddr);
303 if (tb->cflags & CF_NOCACHE) {
304 /* one-shot translation, invalidate it immediately */
305 cpu->current_tb = NULL;
306 tb_phys_invalidate(tb, -1);
307 tb_free(tb);
309 return true;
311 return false;
314 #ifdef _WIN32
315 static __attribute__((unused)) void map_exec(void *addr, long size)
317 DWORD old_protect;
318 VirtualProtect(addr, size,
319 PAGE_EXECUTE_READWRITE, &old_protect);
321 #else
322 static __attribute__((unused)) void map_exec(void *addr, long size)
324 unsigned long start, end, page_size;
326 page_size = getpagesize();
327 start = (unsigned long)addr;
328 start &= ~(page_size - 1);
330 end = (unsigned long)addr + size;
331 end += page_size - 1;
332 end &= ~(page_size - 1);
334 mprotect((void *)start, end - start,
335 PROT_READ | PROT_WRITE | PROT_EXEC);
337 #endif
339 void page_size_init(void)
341 /* NOTE: we can always suppose that qemu_host_page_size >=
342 TARGET_PAGE_SIZE */
343 qemu_real_host_page_size = getpagesize();
344 qemu_real_host_page_mask = ~(qemu_real_host_page_size - 1);
345 if (qemu_host_page_size == 0) {
346 qemu_host_page_size = qemu_real_host_page_size;
348 if (qemu_host_page_size < TARGET_PAGE_SIZE) {
349 qemu_host_page_size = TARGET_PAGE_SIZE;
351 qemu_host_page_mask = ~(qemu_host_page_size - 1);
354 static void page_init(void)
356 page_size_init();
357 #if defined(CONFIG_BSD) && defined(CONFIG_USER_ONLY)
359 #ifdef HAVE_KINFO_GETVMMAP
360 struct kinfo_vmentry *freep;
361 int i, cnt;
363 freep = kinfo_getvmmap(getpid(), &cnt);
364 if (freep) {
365 mmap_lock();
366 for (i = 0; i < cnt; i++) {
367 unsigned long startaddr, endaddr;
369 startaddr = freep[i].kve_start;
370 endaddr = freep[i].kve_end;
371 if (h2g_valid(startaddr)) {
372 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
374 if (h2g_valid(endaddr)) {
375 endaddr = h2g(endaddr);
376 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
377 } else {
378 #if TARGET_ABI_BITS <= L1_MAP_ADDR_SPACE_BITS
379 endaddr = ~0ul;
380 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
381 #endif
385 free(freep);
386 mmap_unlock();
388 #else
389 FILE *f;
391 last_brk = (unsigned long)sbrk(0);
393 f = fopen("/compat/linux/proc/self/maps", "r");
394 if (f) {
395 mmap_lock();
397 do {
398 unsigned long startaddr, endaddr;
399 int n;
401 n = fscanf(f, "%lx-%lx %*[^\n]\n", &startaddr, &endaddr);
403 if (n == 2 && h2g_valid(startaddr)) {
404 startaddr = h2g(startaddr) & TARGET_PAGE_MASK;
406 if (h2g_valid(endaddr)) {
407 endaddr = h2g(endaddr);
408 } else {
409 endaddr = ~0ul;
411 page_set_flags(startaddr, endaddr, PAGE_RESERVED);
413 } while (!feof(f));
415 fclose(f);
416 mmap_unlock();
418 #endif
420 #endif
423 /* If alloc=1:
424 * Called with mmap_lock held for user-mode emulation.
426 static PageDesc *page_find_alloc(tb_page_addr_t index, int alloc)
428 PageDesc *pd;
429 void **lp;
430 int i;
432 /* Level 1. Always allocated. */
433 lp = l1_map + ((index >> V_L1_SHIFT) & (V_L1_SIZE - 1));
435 /* Level 2..N-1. */
436 for (i = V_L1_SHIFT / V_L2_BITS - 1; i > 0; i--) {
437 void **p = atomic_rcu_read(lp);
439 if (p == NULL) {
440 if (!alloc) {
441 return NULL;
443 p = g_new0(void *, V_L2_SIZE);
444 atomic_rcu_set(lp, p);
447 lp = p + ((index >> (i * V_L2_BITS)) & (V_L2_SIZE - 1));
450 pd = atomic_rcu_read(lp);
451 if (pd == NULL) {
452 if (!alloc) {
453 return NULL;
455 pd = g_new0(PageDesc, V_L2_SIZE);
456 atomic_rcu_set(lp, pd);
459 return pd + (index & (V_L2_SIZE - 1));
462 static inline PageDesc *page_find(tb_page_addr_t index)
464 return page_find_alloc(index, 0);
467 #if defined(CONFIG_USER_ONLY)
468 /* Currently it is not recommended to allocate big chunks of data in
469 user mode. It will change when a dedicated libc will be used. */
470 /* ??? 64-bit hosts ought to have no problem mmaping data outside the
471 region in which the guest needs to run. Revisit this. */
472 #define USE_STATIC_CODE_GEN_BUFFER
473 #endif
475 /* ??? Should configure for this, not list operating systems here. */
476 #if (defined(__linux__) \
477 || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) \
478 || defined(__DragonFly__) || defined(__OpenBSD__) \
479 || defined(__NetBSD__))
480 # define USE_MMAP
481 #endif
483 /* Minimum size of the code gen buffer. This number is randomly chosen,
484 but not so small that we can't have a fair number of TB's live. */
485 #define MIN_CODE_GEN_BUFFER_SIZE (1024u * 1024)
487 /* Maximum size of the code gen buffer we'd like to use. Unless otherwise
488 indicated, this is constrained by the range of direct branches on the
489 host cpu, as used by the TCG implementation of goto_tb. */
490 #if defined(__x86_64__)
491 # define MAX_CODE_GEN_BUFFER_SIZE (2ul * 1024 * 1024 * 1024)
492 #elif defined(__sparc__)
493 # define MAX_CODE_GEN_BUFFER_SIZE (2ul * 1024 * 1024 * 1024)
494 #elif defined(__aarch64__)
495 # define MAX_CODE_GEN_BUFFER_SIZE (128ul * 1024 * 1024)
496 #elif defined(__arm__)
497 # define MAX_CODE_GEN_BUFFER_SIZE (16u * 1024 * 1024)
498 #elif defined(__s390x__)
499 /* We have a +- 4GB range on the branches; leave some slop. */
500 # define MAX_CODE_GEN_BUFFER_SIZE (3ul * 1024 * 1024 * 1024)
501 #elif defined(__mips__)
502 /* We have a 256MB branch region, but leave room to make sure the
503 main executable is also within that region. */
504 # define MAX_CODE_GEN_BUFFER_SIZE (128ul * 1024 * 1024)
505 #else
506 # define MAX_CODE_GEN_BUFFER_SIZE ((size_t)-1)
507 #endif
509 #define DEFAULT_CODE_GEN_BUFFER_SIZE_1 (32u * 1024 * 1024)
511 #define DEFAULT_CODE_GEN_BUFFER_SIZE \
512 (DEFAULT_CODE_GEN_BUFFER_SIZE_1 < MAX_CODE_GEN_BUFFER_SIZE \
513 ? DEFAULT_CODE_GEN_BUFFER_SIZE_1 : MAX_CODE_GEN_BUFFER_SIZE)
515 static inline size_t size_code_gen_buffer(size_t tb_size)
517 /* Size the buffer. */
518 if (tb_size == 0) {
519 #ifdef USE_STATIC_CODE_GEN_BUFFER
520 tb_size = DEFAULT_CODE_GEN_BUFFER_SIZE;
521 #else
522 /* ??? Needs adjustments. */
523 /* ??? If we relax the requirement that CONFIG_USER_ONLY use the
524 static buffer, we could size this on RESERVED_VA, on the text
525 segment size of the executable, or continue to use the default. */
526 tb_size = (unsigned long)(ram_size / 4);
527 #endif
529 if (tb_size < MIN_CODE_GEN_BUFFER_SIZE) {
530 tb_size = MIN_CODE_GEN_BUFFER_SIZE;
532 if (tb_size > MAX_CODE_GEN_BUFFER_SIZE) {
533 tb_size = MAX_CODE_GEN_BUFFER_SIZE;
535 tcg_ctx.code_gen_buffer_size = tb_size;
536 return tb_size;
539 #ifdef __mips__
540 /* In order to use J and JAL within the code_gen_buffer, we require
541 that the buffer not cross a 256MB boundary. */
542 static inline bool cross_256mb(void *addr, size_t size)
544 return ((uintptr_t)addr ^ ((uintptr_t)addr + size)) & 0xf0000000;
547 /* We weren't able to allocate a buffer without crossing that boundary,
548 so make do with the larger portion of the buffer that doesn't cross.
549 Returns the new base of the buffer, and adjusts code_gen_buffer_size. */
550 static inline void *split_cross_256mb(void *buf1, size_t size1)
552 void *buf2 = (void *)(((uintptr_t)buf1 + size1) & 0xf0000000);
553 size_t size2 = buf1 + size1 - buf2;
555 size1 = buf2 - buf1;
556 if (size1 < size2) {
557 size1 = size2;
558 buf1 = buf2;
561 tcg_ctx.code_gen_buffer_size = size1;
562 return buf1;
564 #endif
566 #ifdef USE_STATIC_CODE_GEN_BUFFER
567 static uint8_t static_code_gen_buffer[DEFAULT_CODE_GEN_BUFFER_SIZE]
568 __attribute__((aligned(CODE_GEN_ALIGN)));
570 static inline void *alloc_code_gen_buffer(void)
572 void *buf = static_code_gen_buffer;
573 #ifdef __mips__
574 if (cross_256mb(buf, tcg_ctx.code_gen_buffer_size)) {
575 buf = split_cross_256mb(buf, tcg_ctx.code_gen_buffer_size);
577 #endif
578 map_exec(buf, tcg_ctx.code_gen_buffer_size);
579 return buf;
581 #elif defined(USE_MMAP)
582 static inline void *alloc_code_gen_buffer(void)
584 int flags = MAP_PRIVATE | MAP_ANONYMOUS;
585 uintptr_t start = 0;
586 void *buf;
588 /* Constrain the position of the buffer based on the host cpu.
589 Note that these addresses are chosen in concert with the
590 addresses assigned in the relevant linker script file. */
591 # if defined(__PIE__) || defined(__PIC__)
592 /* Don't bother setting a preferred location if we're building
593 a position-independent executable. We're more likely to get
594 an address near the main executable if we let the kernel
595 choose the address. */
596 # elif defined(__x86_64__) && defined(MAP_32BIT)
597 /* Force the memory down into low memory with the executable.
598 Leave the choice of exact location with the kernel. */
599 flags |= MAP_32BIT;
600 /* Cannot expect to map more than 800MB in low memory. */
601 if (tcg_ctx.code_gen_buffer_size > 800u * 1024 * 1024) {
602 tcg_ctx.code_gen_buffer_size = 800u * 1024 * 1024;
604 # elif defined(__sparc__)
605 start = 0x40000000ul;
606 # elif defined(__s390x__)
607 start = 0x90000000ul;
608 # elif defined(__mips__)
609 /* ??? We ought to more explicitly manage layout for softmmu too. */
610 # ifdef CONFIG_USER_ONLY
611 start = 0x68000000ul;
612 # elif _MIPS_SIM == _ABI64
613 start = 0x128000000ul;
614 # else
615 start = 0x08000000ul;
616 # endif
617 # endif
619 buf = mmap((void *)start, tcg_ctx.code_gen_buffer_size,
620 PROT_WRITE | PROT_READ | PROT_EXEC, flags, -1, 0);
621 if (buf == MAP_FAILED) {
622 return NULL;
625 #ifdef __mips__
626 if (cross_256mb(buf, tcg_ctx.code_gen_buffer_size)) {
627 /* Try again, with the original still mapped, to avoid re-acquiring
628 that 256mb crossing. This time don't specify an address. */
629 size_t size2, size1 = tcg_ctx.code_gen_buffer_size;
630 void *buf2 = mmap(NULL, size1, PROT_WRITE | PROT_READ | PROT_EXEC,
631 flags, -1, 0);
632 if (buf2 != MAP_FAILED) {
633 if (!cross_256mb(buf2, size1)) {
634 /* Success! Use the new buffer. */
635 munmap(buf, size1);
636 return buf2;
638 /* Failure. Work with what we had. */
639 munmap(buf2, size1);
642 /* Split the original buffer. Free the smaller half. */
643 buf2 = split_cross_256mb(buf, size1);
644 size2 = tcg_ctx.code_gen_buffer_size;
645 munmap(buf + (buf == buf2 ? size2 : 0), size1 - size2);
646 return buf2;
648 #endif
650 return buf;
652 #else
653 static inline void *alloc_code_gen_buffer(void)
655 void *buf = g_try_malloc(tcg_ctx.code_gen_buffer_size);
657 if (buf == NULL) {
658 return NULL;
661 #ifdef __mips__
662 if (cross_256mb(buf, tcg_ctx.code_gen_buffer_size)) {
663 void *buf2 = g_malloc(tcg_ctx.code_gen_buffer_size);
664 if (buf2 != NULL && !cross_256mb(buf2, size1)) {
665 /* Success! Use the new buffer. */
666 free(buf);
667 buf = buf2;
668 } else {
669 /* Failure. Work with what we had. Since this is malloc
670 and not mmap, we can't free the other half. */
671 free(buf2);
672 buf = split_cross_256mb(buf, tcg_ctx.code_gen_buffer_size);
675 #endif
677 map_exec(buf, tcg_ctx.code_gen_buffer_size);
678 return buf;
680 #endif /* USE_STATIC_CODE_GEN_BUFFER, USE_MMAP */
682 static inline void code_gen_alloc(size_t tb_size)
684 tcg_ctx.code_gen_buffer_size = size_code_gen_buffer(tb_size);
685 tcg_ctx.code_gen_buffer = alloc_code_gen_buffer();
686 if (tcg_ctx.code_gen_buffer == NULL) {
687 fprintf(stderr, "Could not allocate dynamic translator buffer\n");
688 exit(1);
691 qemu_madvise(tcg_ctx.code_gen_buffer, tcg_ctx.code_gen_buffer_size,
692 QEMU_MADV_HUGEPAGE);
694 /* Steal room for the prologue at the end of the buffer. This ensures
695 (via the MAX_CODE_GEN_BUFFER_SIZE limits above) that direct branches
696 from TB's to the prologue are going to be in range. It also means
697 that we don't need to mark (additional) portions of the data segment
698 as executable. */
699 tcg_ctx.code_gen_prologue = tcg_ctx.code_gen_buffer +
700 tcg_ctx.code_gen_buffer_size - 1024;
701 tcg_ctx.code_gen_buffer_size -= 1024;
703 tcg_ctx.code_gen_buffer_max_size = tcg_ctx.code_gen_buffer_size -
704 (TCG_MAX_OP_SIZE * OPC_BUF_SIZE);
705 tcg_ctx.code_gen_max_blocks = tcg_ctx.code_gen_buffer_size /
706 CODE_GEN_AVG_BLOCK_SIZE;
707 tcg_ctx.tb_ctx.tbs =
708 g_malloc(tcg_ctx.code_gen_max_blocks * sizeof(TranslationBlock));
709 qemu_mutex_init(&tcg_ctx.tb_ctx.tb_lock);
712 /* Must be called before using the QEMU cpus. 'tb_size' is the size
713 (in bytes) allocated to the translation buffer. Zero means default
714 size. */
715 void tcg_exec_init(unsigned long tb_size)
717 cpu_gen_init();
718 code_gen_alloc(tb_size);
719 tcg_ctx.code_gen_ptr = tcg_ctx.code_gen_buffer;
720 tcg_register_jit(tcg_ctx.code_gen_buffer, tcg_ctx.code_gen_buffer_size);
721 page_init();
722 #if defined(CONFIG_SOFTMMU)
723 /* There's no guest base to take into account, so go ahead and
724 initialize the prologue now. */
725 tcg_prologue_init(&tcg_ctx);
726 #endif
729 bool tcg_enabled(void)
731 return tcg_ctx.code_gen_buffer != NULL;
734 /* Allocate a new translation block. Flush the translation buffer if
735 too many translation blocks or too much generated code. */
736 static TranslationBlock *tb_alloc(target_ulong pc)
738 TranslationBlock *tb;
740 if (tcg_ctx.tb_ctx.nb_tbs >= tcg_ctx.code_gen_max_blocks ||
741 (tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer) >=
742 tcg_ctx.code_gen_buffer_max_size) {
743 return NULL;
745 tb = &tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs++];
746 tb->pc = pc;
747 tb->cflags = 0;
748 return tb;
751 void tb_free(TranslationBlock *tb)
753 /* In practice this is mostly used for single use temporary TB
754 Ignore the hard cases and just back up if this TB happens to
755 be the last one generated. */
756 if (tcg_ctx.tb_ctx.nb_tbs > 0 &&
757 tb == &tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs - 1]) {
758 tcg_ctx.code_gen_ptr = tb->tc_ptr;
759 tcg_ctx.tb_ctx.nb_tbs--;
763 static inline void invalidate_page_bitmap(PageDesc *p)
765 g_free(p->code_bitmap);
766 p->code_bitmap = NULL;
767 p->code_write_count = 0;
770 /* Set to NULL all the 'first_tb' fields in all PageDescs. */
771 static void page_flush_tb_1(int level, void **lp)
773 int i;
775 if (*lp == NULL) {
776 return;
778 if (level == 0) {
779 PageDesc *pd = *lp;
781 for (i = 0; i < V_L2_SIZE; ++i) {
782 pd[i].first_tb = NULL;
783 invalidate_page_bitmap(pd + i);
785 } else {
786 void **pp = *lp;
788 for (i = 0; i < V_L2_SIZE; ++i) {
789 page_flush_tb_1(level - 1, pp + i);
794 static void page_flush_tb(void)
796 int i;
798 for (i = 0; i < V_L1_SIZE; i++) {
799 page_flush_tb_1(V_L1_SHIFT / V_L2_BITS - 1, l1_map + i);
803 /* flush all the translation blocks */
804 /* XXX: tb_flush is currently not thread safe */
805 void tb_flush(CPUState *cpu)
807 #if defined(DEBUG_FLUSH)
808 printf("qemu: flush code_size=%ld nb_tbs=%d avg_tb_size=%ld\n",
809 (unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer),
810 tcg_ctx.tb_ctx.nb_tbs, tcg_ctx.tb_ctx.nb_tbs > 0 ?
811 ((unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer)) /
812 tcg_ctx.tb_ctx.nb_tbs : 0);
813 #endif
814 if ((unsigned long)(tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer)
815 > tcg_ctx.code_gen_buffer_size) {
816 cpu_abort(cpu, "Internal error: code buffer overflow\n");
818 tcg_ctx.tb_ctx.nb_tbs = 0;
820 CPU_FOREACH(cpu) {
821 memset(cpu->tb_jmp_cache, 0, sizeof(cpu->tb_jmp_cache));
824 memset(tcg_ctx.tb_ctx.tb_phys_hash, 0, sizeof(tcg_ctx.tb_ctx.tb_phys_hash));
825 page_flush_tb();
827 tcg_ctx.code_gen_ptr = tcg_ctx.code_gen_buffer;
828 /* XXX: flush processor icache at this point if cache flush is
829 expensive */
830 tcg_ctx.tb_ctx.tb_flush_count++;
833 #ifdef DEBUG_TB_CHECK
835 static void tb_invalidate_check(target_ulong address)
837 TranslationBlock *tb;
838 int i;
840 address &= TARGET_PAGE_MASK;
841 for (i = 0; i < CODE_GEN_PHYS_HASH_SIZE; i++) {
842 for (tb = tb_ctx.tb_phys_hash[i]; tb != NULL; tb = tb->phys_hash_next) {
843 if (!(address + TARGET_PAGE_SIZE <= tb->pc ||
844 address >= tb->pc + tb->size)) {
845 printf("ERROR invalidate: address=" TARGET_FMT_lx
846 " PC=%08lx size=%04x\n",
847 address, (long)tb->pc, tb->size);
853 /* verify that all the pages have correct rights for code */
854 static void tb_page_check(void)
856 TranslationBlock *tb;
857 int i, flags1, flags2;
859 for (i = 0; i < CODE_GEN_PHYS_HASH_SIZE; i++) {
860 for (tb = tcg_ctx.tb_ctx.tb_phys_hash[i]; tb != NULL;
861 tb = tb->phys_hash_next) {
862 flags1 = page_get_flags(tb->pc);
863 flags2 = page_get_flags(tb->pc + tb->size - 1);
864 if ((flags1 & PAGE_WRITE) || (flags2 & PAGE_WRITE)) {
865 printf("ERROR page flags: PC=%08lx size=%04x f1=%x f2=%x\n",
866 (long)tb->pc, tb->size, flags1, flags2);
872 #endif
874 static inline void tb_hash_remove(TranslationBlock **ptb, TranslationBlock *tb)
876 TranslationBlock *tb1;
878 for (;;) {
879 tb1 = *ptb;
880 if (tb1 == tb) {
881 *ptb = tb1->phys_hash_next;
882 break;
884 ptb = &tb1->phys_hash_next;
888 static inline void tb_page_remove(TranslationBlock **ptb, TranslationBlock *tb)
890 TranslationBlock *tb1;
891 unsigned int n1;
893 for (;;) {
894 tb1 = *ptb;
895 n1 = (uintptr_t)tb1 & 3;
896 tb1 = (TranslationBlock *)((uintptr_t)tb1 & ~3);
897 if (tb1 == tb) {
898 *ptb = tb1->page_next[n1];
899 break;
901 ptb = &tb1->page_next[n1];
905 static inline void tb_jmp_remove(TranslationBlock *tb, int n)
907 TranslationBlock *tb1, **ptb;
908 unsigned int n1;
910 ptb = &tb->jmp_next[n];
911 tb1 = *ptb;
912 if (tb1) {
913 /* find tb(n) in circular list */
914 for (;;) {
915 tb1 = *ptb;
916 n1 = (uintptr_t)tb1 & 3;
917 tb1 = (TranslationBlock *)((uintptr_t)tb1 & ~3);
918 if (n1 == n && tb1 == tb) {
919 break;
921 if (n1 == 2) {
922 ptb = &tb1->jmp_first;
923 } else {
924 ptb = &tb1->jmp_next[n1];
927 /* now we can suppress tb(n) from the list */
928 *ptb = tb->jmp_next[n];
930 tb->jmp_next[n] = NULL;
934 /* reset the jump entry 'n' of a TB so that it is not chained to
935 another TB */
936 static inline void tb_reset_jump(TranslationBlock *tb, int n)
938 tb_set_jmp_target(tb, n, (uintptr_t)(tb->tc_ptr + tb->tb_next_offset[n]));
941 /* invalidate one TB */
942 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr)
944 CPUState *cpu;
945 PageDesc *p;
946 unsigned int h, n1;
947 tb_page_addr_t phys_pc;
948 TranslationBlock *tb1, *tb2;
950 /* remove the TB from the hash list */
951 phys_pc = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
952 h = tb_phys_hash_func(phys_pc);
953 tb_hash_remove(&tcg_ctx.tb_ctx.tb_phys_hash[h], tb);
955 /* remove the TB from the page list */
956 if (tb->page_addr[0] != page_addr) {
957 p = page_find(tb->page_addr[0] >> TARGET_PAGE_BITS);
958 tb_page_remove(&p->first_tb, tb);
959 invalidate_page_bitmap(p);
961 if (tb->page_addr[1] != -1 && tb->page_addr[1] != page_addr) {
962 p = page_find(tb->page_addr[1] >> TARGET_PAGE_BITS);
963 tb_page_remove(&p->first_tb, tb);
964 invalidate_page_bitmap(p);
967 tcg_ctx.tb_ctx.tb_invalidated_flag = 1;
969 /* remove the TB from the hash list */
970 h = tb_jmp_cache_hash_func(tb->pc);
971 CPU_FOREACH(cpu) {
972 if (cpu->tb_jmp_cache[h] == tb) {
973 cpu->tb_jmp_cache[h] = NULL;
977 /* suppress this TB from the two jump lists */
978 tb_jmp_remove(tb, 0);
979 tb_jmp_remove(tb, 1);
981 /* suppress any remaining jumps to this TB */
982 tb1 = tb->jmp_first;
983 for (;;) {
984 n1 = (uintptr_t)tb1 & 3;
985 if (n1 == 2) {
986 break;
988 tb1 = (TranslationBlock *)((uintptr_t)tb1 & ~3);
989 tb2 = tb1->jmp_next[n1];
990 tb_reset_jump(tb1, n1);
991 tb1->jmp_next[n1] = NULL;
992 tb1 = tb2;
994 tb->jmp_first = (TranslationBlock *)((uintptr_t)tb | 2); /* fail safe */
996 tcg_ctx.tb_ctx.tb_phys_invalidate_count++;
999 static void build_page_bitmap(PageDesc *p)
1001 int n, tb_start, tb_end;
1002 TranslationBlock *tb;
1004 p->code_bitmap = bitmap_new(TARGET_PAGE_SIZE);
1006 tb = p->first_tb;
1007 while (tb != NULL) {
1008 n = (uintptr_t)tb & 3;
1009 tb = (TranslationBlock *)((uintptr_t)tb & ~3);
1010 /* NOTE: this is subtle as a TB may span two physical pages */
1011 if (n == 0) {
1012 /* NOTE: tb_end may be after the end of the page, but
1013 it is not a problem */
1014 tb_start = tb->pc & ~TARGET_PAGE_MASK;
1015 tb_end = tb_start + tb->size;
1016 if (tb_end > TARGET_PAGE_SIZE) {
1017 tb_end = TARGET_PAGE_SIZE;
1019 } else {
1020 tb_start = 0;
1021 tb_end = ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
1023 bitmap_set(p->code_bitmap, tb_start, tb_end - tb_start);
1024 tb = tb->page_next[n];
1028 /* Called with mmap_lock held for user mode emulation. */
1029 TranslationBlock *tb_gen_code(CPUState *cpu,
1030 target_ulong pc, target_ulong cs_base,
1031 int flags, int cflags)
1033 CPUArchState *env = cpu->env_ptr;
1034 TranslationBlock *tb;
1035 tb_page_addr_t phys_pc, phys_page2;
1036 target_ulong virt_page2;
1037 int code_gen_size;
1039 phys_pc = get_page_addr_code(env, pc);
1040 if (use_icount) {
1041 cflags |= CF_USE_ICOUNT;
1043 tb = tb_alloc(pc);
1044 if (!tb) {
1045 /* flush must be done */
1046 tb_flush(cpu);
1047 /* cannot fail at this point */
1048 tb = tb_alloc(pc);
1049 /* Don't forget to invalidate previous TB info. */
1050 tcg_ctx.tb_ctx.tb_invalidated_flag = 1;
1052 tb->tc_ptr = tcg_ctx.code_gen_ptr;
1053 tb->cs_base = cs_base;
1054 tb->flags = flags;
1055 tb->cflags = cflags;
1056 cpu_gen_code(env, tb, &code_gen_size);
1057 tcg_ctx.code_gen_ptr = (void *)(((uintptr_t)tcg_ctx.code_gen_ptr +
1058 code_gen_size + CODE_GEN_ALIGN - 1) & ~(CODE_GEN_ALIGN - 1));
1060 /* check next page if needed */
1061 virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
1062 phys_page2 = -1;
1063 if ((pc & TARGET_PAGE_MASK) != virt_page2) {
1064 phys_page2 = get_page_addr_code(env, virt_page2);
1066 tb_link_page(tb, phys_pc, phys_page2);
1067 return tb;
1071 * Invalidate all TBs which intersect with the target physical address range
1072 * [start;end[. NOTE: start and end may refer to *different* physical pages.
1073 * 'is_cpu_write_access' should be true if called from a real cpu write
1074 * access: the virtual CPU will exit the current TB if code is modified inside
1075 * this TB.
1077 * Called with mmap_lock held for user-mode emulation
1079 void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end)
1081 while (start < end) {
1082 tb_invalidate_phys_page_range(start, end, 0);
1083 start &= TARGET_PAGE_MASK;
1084 start += TARGET_PAGE_SIZE;
1089 * Invalidate all TBs which intersect with the target physical address range
1090 * [start;end[. NOTE: start and end must refer to the *same* physical page.
1091 * 'is_cpu_write_access' should be true if called from a real cpu write
1092 * access: the virtual CPU will exit the current TB if code is modified inside
1093 * this TB.
1095 * Called with mmap_lock held for user-mode emulation
1097 void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
1098 int is_cpu_write_access)
1100 TranslationBlock *tb, *tb_next, *saved_tb;
1101 CPUState *cpu = current_cpu;
1102 #if defined(TARGET_HAS_PRECISE_SMC)
1103 CPUArchState *env = NULL;
1104 #endif
1105 tb_page_addr_t tb_start, tb_end;
1106 PageDesc *p;
1107 int n;
1108 #ifdef TARGET_HAS_PRECISE_SMC
1109 int current_tb_not_found = is_cpu_write_access;
1110 TranslationBlock *current_tb = NULL;
1111 int current_tb_modified = 0;
1112 target_ulong current_pc = 0;
1113 target_ulong current_cs_base = 0;
1114 int current_flags = 0;
1115 #endif /* TARGET_HAS_PRECISE_SMC */
1117 p = page_find(start >> TARGET_PAGE_BITS);
1118 if (!p) {
1119 return;
1121 #if defined(TARGET_HAS_PRECISE_SMC)
1122 if (cpu != NULL) {
1123 env = cpu->env_ptr;
1125 #endif
1127 /* we remove all the TBs in the range [start, end[ */
1128 /* XXX: see if in some cases it could be faster to invalidate all
1129 the code */
1130 tb = p->first_tb;
1131 while (tb != NULL) {
1132 n = (uintptr_t)tb & 3;
1133 tb = (TranslationBlock *)((uintptr_t)tb & ~3);
1134 tb_next = tb->page_next[n];
1135 /* NOTE: this is subtle as a TB may span two physical pages */
1136 if (n == 0) {
1137 /* NOTE: tb_end may be after the end of the page, but
1138 it is not a problem */
1139 tb_start = tb->page_addr[0] + (tb->pc & ~TARGET_PAGE_MASK);
1140 tb_end = tb_start + tb->size;
1141 } else {
1142 tb_start = tb->page_addr[1];
1143 tb_end = tb_start + ((tb->pc + tb->size) & ~TARGET_PAGE_MASK);
1145 if (!(tb_end <= start || tb_start >= end)) {
1146 #ifdef TARGET_HAS_PRECISE_SMC
1147 if (current_tb_not_found) {
1148 current_tb_not_found = 0;
1149 current_tb = NULL;
1150 if (cpu->mem_io_pc) {
1151 /* now we have a real cpu fault */
1152 current_tb = tb_find_pc(cpu->mem_io_pc);
1155 if (current_tb == tb &&
1156 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1157 /* If we are modifying the current TB, we must stop
1158 its execution. We could be more precise by checking
1159 that the modification is after the current PC, but it
1160 would require a specialized function to partially
1161 restore the CPU state */
1163 current_tb_modified = 1;
1164 cpu_restore_state_from_tb(cpu, current_tb, cpu->mem_io_pc);
1165 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1166 &current_flags);
1168 #endif /* TARGET_HAS_PRECISE_SMC */
1169 /* we need to do that to handle the case where a signal
1170 occurs while doing tb_phys_invalidate() */
1171 saved_tb = NULL;
1172 if (cpu != NULL) {
1173 saved_tb = cpu->current_tb;
1174 cpu->current_tb = NULL;
1176 tb_phys_invalidate(tb, -1);
1177 if (cpu != NULL) {
1178 cpu->current_tb = saved_tb;
1179 if (cpu->interrupt_request && cpu->current_tb) {
1180 cpu_interrupt(cpu, cpu->interrupt_request);
1184 tb = tb_next;
1186 #if !defined(CONFIG_USER_ONLY)
1187 /* if no code remaining, no need to continue to use slow writes */
1188 if (!p->first_tb) {
1189 invalidate_page_bitmap(p);
1190 tlb_unprotect_code(start);
1192 #endif
1193 #ifdef TARGET_HAS_PRECISE_SMC
1194 if (current_tb_modified) {
1195 /* we generate a block containing just the instruction
1196 modifying the memory. It will ensure that it cannot modify
1197 itself */
1198 cpu->current_tb = NULL;
1199 tb_gen_code(cpu, current_pc, current_cs_base, current_flags, 1);
1200 cpu_resume_from_signal(cpu, NULL);
1202 #endif
1205 /* len must be <= 8 and start must be a multiple of len */
1206 void tb_invalidate_phys_page_fast(tb_page_addr_t start, int len)
1208 PageDesc *p;
1210 #if 0
1211 if (1) {
1212 qemu_log("modifying code at 0x%x size=%d EIP=%x PC=%08x\n",
1213 cpu_single_env->mem_io_vaddr, len,
1214 cpu_single_env->eip,
1215 cpu_single_env->eip +
1216 (intptr_t)cpu_single_env->segs[R_CS].base);
1218 #endif
1219 p = page_find(start >> TARGET_PAGE_BITS);
1220 if (!p) {
1221 return;
1223 if (!p->code_bitmap &&
1224 ++p->code_write_count >= SMC_BITMAP_USE_THRESHOLD) {
1225 /* build code bitmap */
1226 build_page_bitmap(p);
1228 if (p->code_bitmap) {
1229 unsigned int nr;
1230 unsigned long b;
1232 nr = start & ~TARGET_PAGE_MASK;
1233 b = p->code_bitmap[BIT_WORD(nr)] >> (nr & (BITS_PER_LONG - 1));
1234 if (b & ((1 << len) - 1)) {
1235 goto do_invalidate;
1237 } else {
1238 do_invalidate:
1239 tb_invalidate_phys_page_range(start, start + len, 1);
1243 #if !defined(CONFIG_SOFTMMU)
1244 /* Called with mmap_lock held. */
1245 static void tb_invalidate_phys_page(tb_page_addr_t addr,
1246 uintptr_t pc, void *puc,
1247 bool locked)
1249 TranslationBlock *tb;
1250 PageDesc *p;
1251 int n;
1252 #ifdef TARGET_HAS_PRECISE_SMC
1253 TranslationBlock *current_tb = NULL;
1254 CPUState *cpu = current_cpu;
1255 CPUArchState *env = NULL;
1256 int current_tb_modified = 0;
1257 target_ulong current_pc = 0;
1258 target_ulong current_cs_base = 0;
1259 int current_flags = 0;
1260 #endif
1262 addr &= TARGET_PAGE_MASK;
1263 p = page_find(addr >> TARGET_PAGE_BITS);
1264 if (!p) {
1265 return;
1267 tb = p->first_tb;
1268 #ifdef TARGET_HAS_PRECISE_SMC
1269 if (tb && pc != 0) {
1270 current_tb = tb_find_pc(pc);
1272 if (cpu != NULL) {
1273 env = cpu->env_ptr;
1275 #endif
1276 while (tb != NULL) {
1277 n = (uintptr_t)tb & 3;
1278 tb = (TranslationBlock *)((uintptr_t)tb & ~3);
1279 #ifdef TARGET_HAS_PRECISE_SMC
1280 if (current_tb == tb &&
1281 (current_tb->cflags & CF_COUNT_MASK) != 1) {
1282 /* If we are modifying the current TB, we must stop
1283 its execution. We could be more precise by checking
1284 that the modification is after the current PC, but it
1285 would require a specialized function to partially
1286 restore the CPU state */
1288 current_tb_modified = 1;
1289 cpu_restore_state_from_tb(cpu, current_tb, pc);
1290 cpu_get_tb_cpu_state(env, &current_pc, &current_cs_base,
1291 &current_flags);
1293 #endif /* TARGET_HAS_PRECISE_SMC */
1294 tb_phys_invalidate(tb, addr);
1295 tb = tb->page_next[n];
1297 p->first_tb = NULL;
1298 #ifdef TARGET_HAS_PRECISE_SMC
1299 if (current_tb_modified) {
1300 /* we generate a block containing just the instruction
1301 modifying the memory. It will ensure that it cannot modify
1302 itself */
1303 cpu->current_tb = NULL;
1304 tb_gen_code(cpu, current_pc, current_cs_base, current_flags, 1);
1305 if (locked) {
1306 mmap_unlock();
1308 cpu_resume_from_signal(cpu, puc);
1310 #endif
1312 #endif
1314 /* add the tb in the target page and protect it if necessary
1316 * Called with mmap_lock held for user-mode emulation.
1318 static inline void tb_alloc_page(TranslationBlock *tb,
1319 unsigned int n, tb_page_addr_t page_addr)
1321 PageDesc *p;
1322 #ifndef CONFIG_USER_ONLY
1323 bool page_already_protected;
1324 #endif
1326 tb->page_addr[n] = page_addr;
1327 p = page_find_alloc(page_addr >> TARGET_PAGE_BITS, 1);
1328 tb->page_next[n] = p->first_tb;
1329 #ifndef CONFIG_USER_ONLY
1330 page_already_protected = p->first_tb != NULL;
1331 #endif
1332 p->first_tb = (TranslationBlock *)((uintptr_t)tb | n);
1333 invalidate_page_bitmap(p);
1335 #if defined(CONFIG_USER_ONLY)
1336 if (p->flags & PAGE_WRITE) {
1337 target_ulong addr;
1338 PageDesc *p2;
1339 int prot;
1341 /* force the host page as non writable (writes will have a
1342 page fault + mprotect overhead) */
1343 page_addr &= qemu_host_page_mask;
1344 prot = 0;
1345 for (addr = page_addr; addr < page_addr + qemu_host_page_size;
1346 addr += TARGET_PAGE_SIZE) {
1348 p2 = page_find(addr >> TARGET_PAGE_BITS);
1349 if (!p2) {
1350 continue;
1352 prot |= p2->flags;
1353 p2->flags &= ~PAGE_WRITE;
1355 mprotect(g2h(page_addr), qemu_host_page_size,
1356 (prot & PAGE_BITS) & ~PAGE_WRITE);
1357 #ifdef DEBUG_TB_INVALIDATE
1358 printf("protecting code page: 0x" TARGET_FMT_lx "\n",
1359 page_addr);
1360 #endif
1362 #else
1363 /* if some code is already present, then the pages are already
1364 protected. So we handle the case where only the first TB is
1365 allocated in a physical page */
1366 if (!page_already_protected) {
1367 tlb_protect_code(page_addr);
1369 #endif
1372 /* add a new TB and link it to the physical page tables. phys_page2 is
1373 * (-1) to indicate that only one page contains the TB.
1375 * Called with mmap_lock held for user-mode emulation.
1377 static void tb_link_page(TranslationBlock *tb, tb_page_addr_t phys_pc,
1378 tb_page_addr_t phys_page2)
1380 unsigned int h;
1381 TranslationBlock **ptb;
1383 /* add in the physical hash table */
1384 h = tb_phys_hash_func(phys_pc);
1385 ptb = &tcg_ctx.tb_ctx.tb_phys_hash[h];
1386 tb->phys_hash_next = *ptb;
1387 *ptb = tb;
1389 /* add in the page list */
1390 tb_alloc_page(tb, 0, phys_pc & TARGET_PAGE_MASK);
1391 if (phys_page2 != -1) {
1392 tb_alloc_page(tb, 1, phys_page2);
1393 } else {
1394 tb->page_addr[1] = -1;
1397 tb->jmp_first = (TranslationBlock *)((uintptr_t)tb | 2);
1398 tb->jmp_next[0] = NULL;
1399 tb->jmp_next[1] = NULL;
1401 /* init original jump addresses */
1402 if (tb->tb_next_offset[0] != 0xffff) {
1403 tb_reset_jump(tb, 0);
1405 if (tb->tb_next_offset[1] != 0xffff) {
1406 tb_reset_jump(tb, 1);
1409 #ifdef DEBUG_TB_CHECK
1410 tb_page_check();
1411 #endif
1414 /* find the TB 'tb' such that tb[0].tc_ptr <= tc_ptr <
1415 tb[1].tc_ptr. Return NULL if not found */
1416 static TranslationBlock *tb_find_pc(uintptr_t tc_ptr)
1418 int m_min, m_max, m;
1419 uintptr_t v;
1420 TranslationBlock *tb;
1422 if (tcg_ctx.tb_ctx.nb_tbs <= 0) {
1423 return NULL;
1425 if (tc_ptr < (uintptr_t)tcg_ctx.code_gen_buffer ||
1426 tc_ptr >= (uintptr_t)tcg_ctx.code_gen_ptr) {
1427 return NULL;
1429 /* binary search (cf Knuth) */
1430 m_min = 0;
1431 m_max = tcg_ctx.tb_ctx.nb_tbs - 1;
1432 while (m_min <= m_max) {
1433 m = (m_min + m_max) >> 1;
1434 tb = &tcg_ctx.tb_ctx.tbs[m];
1435 v = (uintptr_t)tb->tc_ptr;
1436 if (v == tc_ptr) {
1437 return tb;
1438 } else if (tc_ptr < v) {
1439 m_max = m - 1;
1440 } else {
1441 m_min = m + 1;
1444 return &tcg_ctx.tb_ctx.tbs[m_max];
1447 #if !defined(CONFIG_USER_ONLY)
1448 void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr)
1450 ram_addr_t ram_addr;
1451 MemoryRegion *mr;
1452 hwaddr l = 1;
1454 rcu_read_lock();
1455 mr = address_space_translate(as, addr, &addr, &l, false);
1456 if (!(memory_region_is_ram(mr)
1457 || memory_region_is_romd(mr))) {
1458 rcu_read_unlock();
1459 return;
1461 ram_addr = (memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK)
1462 + addr;
1463 tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
1464 rcu_read_unlock();
1466 #endif /* !defined(CONFIG_USER_ONLY) */
1468 void tb_check_watchpoint(CPUState *cpu)
1470 TranslationBlock *tb;
1472 tb = tb_find_pc(cpu->mem_io_pc);
1473 if (tb) {
1474 /* We can use retranslation to find the PC. */
1475 cpu_restore_state_from_tb(cpu, tb, cpu->mem_io_pc);
1476 tb_phys_invalidate(tb, -1);
1477 } else {
1478 /* The exception probably happened in a helper. The CPU state should
1479 have been saved before calling it. Fetch the PC from there. */
1480 CPUArchState *env = cpu->env_ptr;
1481 target_ulong pc, cs_base;
1482 tb_page_addr_t addr;
1483 int flags;
1485 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
1486 addr = get_page_addr_code(env, pc);
1487 tb_invalidate_phys_range(addr, addr + 1);
1491 #ifndef CONFIG_USER_ONLY
1492 /* in deterministic execution mode, instructions doing device I/Os
1493 must be at the end of the TB */
1494 void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
1496 #if defined(TARGET_MIPS) || defined(TARGET_SH4)
1497 CPUArchState *env = cpu->env_ptr;
1498 #endif
1499 TranslationBlock *tb;
1500 uint32_t n, cflags;
1501 target_ulong pc, cs_base;
1502 uint64_t flags;
1504 tb = tb_find_pc(retaddr);
1505 if (!tb) {
1506 cpu_abort(cpu, "cpu_io_recompile: could not find TB for pc=%p",
1507 (void *)retaddr);
1509 n = cpu->icount_decr.u16.low + tb->icount;
1510 cpu_restore_state_from_tb(cpu, tb, retaddr);
1511 /* Calculate how many instructions had been executed before the fault
1512 occurred. */
1513 n = n - cpu->icount_decr.u16.low;
1514 /* Generate a new TB ending on the I/O insn. */
1515 n++;
1516 /* On MIPS and SH, delay slot instructions can only be restarted if
1517 they were already the first instruction in the TB. If this is not
1518 the first instruction in a TB then re-execute the preceding
1519 branch. */
1520 #if defined(TARGET_MIPS)
1521 if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
1522 env->active_tc.PC -= (env->hflags & MIPS_HFLAG_B16 ? 2 : 4);
1523 cpu->icount_decr.u16.low++;
1524 env->hflags &= ~MIPS_HFLAG_BMASK;
1526 #elif defined(TARGET_SH4)
1527 if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
1528 && n > 1) {
1529 env->pc -= 2;
1530 cpu->icount_decr.u16.low++;
1531 env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
1533 #endif
1534 /* This should never happen. */
1535 if (n > CF_COUNT_MASK) {
1536 cpu_abort(cpu, "TB too big during recompile");
1539 cflags = n | CF_LAST_IO;
1540 pc = tb->pc;
1541 cs_base = tb->cs_base;
1542 flags = tb->flags;
1543 tb_phys_invalidate(tb, -1);
1544 if (tb->cflags & CF_NOCACHE) {
1545 if (tb->orig_tb) {
1546 /* Invalidate original TB if this TB was generated in
1547 * cpu_exec_nocache() */
1548 tb_phys_invalidate(tb->orig_tb, -1);
1550 tb_free(tb);
1552 /* FIXME: In theory this could raise an exception. In practice
1553 we have already translated the block once so it's probably ok. */
1554 tb_gen_code(cpu, pc, cs_base, flags, cflags);
1555 /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
1556 the first in the TB) then we end up generating a whole new TB and
1557 repeating the fault, which is horribly inefficient.
1558 Better would be to execute just this insn uncached, or generate a
1559 second new TB. */
1560 cpu_resume_from_signal(cpu, NULL);
1563 void tb_flush_jmp_cache(CPUState *cpu, target_ulong addr)
1565 unsigned int i;
1567 /* Discard jump cache entries for any tb which might potentially
1568 overlap the flushed page. */
1569 i = tb_jmp_cache_hash_page(addr - TARGET_PAGE_SIZE);
1570 memset(&cpu->tb_jmp_cache[i], 0,
1571 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1573 i = tb_jmp_cache_hash_page(addr);
1574 memset(&cpu->tb_jmp_cache[i], 0,
1575 TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *));
1578 void dump_exec_info(FILE *f, fprintf_function cpu_fprintf)
1580 int i, target_code_size, max_target_code_size;
1581 int direct_jmp_count, direct_jmp2_count, cross_page;
1582 TranslationBlock *tb;
1584 target_code_size = 0;
1585 max_target_code_size = 0;
1586 cross_page = 0;
1587 direct_jmp_count = 0;
1588 direct_jmp2_count = 0;
1589 for (i = 0; i < tcg_ctx.tb_ctx.nb_tbs; i++) {
1590 tb = &tcg_ctx.tb_ctx.tbs[i];
1591 target_code_size += tb->size;
1592 if (tb->size > max_target_code_size) {
1593 max_target_code_size = tb->size;
1595 if (tb->page_addr[1] != -1) {
1596 cross_page++;
1598 if (tb->tb_next_offset[0] != 0xffff) {
1599 direct_jmp_count++;
1600 if (tb->tb_next_offset[1] != 0xffff) {
1601 direct_jmp2_count++;
1605 /* XXX: avoid using doubles ? */
1606 cpu_fprintf(f, "Translation buffer state:\n");
1607 cpu_fprintf(f, "gen code size %td/%zd\n",
1608 tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer,
1609 tcg_ctx.code_gen_buffer_max_size);
1610 cpu_fprintf(f, "TB count %d/%d\n",
1611 tcg_ctx.tb_ctx.nb_tbs, tcg_ctx.code_gen_max_blocks);
1612 cpu_fprintf(f, "TB avg target size %d max=%d bytes\n",
1613 tcg_ctx.tb_ctx.nb_tbs ? target_code_size /
1614 tcg_ctx.tb_ctx.nb_tbs : 0,
1615 max_target_code_size);
1616 cpu_fprintf(f, "TB avg host size %td bytes (expansion ratio: %0.1f)\n",
1617 tcg_ctx.tb_ctx.nb_tbs ? (tcg_ctx.code_gen_ptr -
1618 tcg_ctx.code_gen_buffer) /
1619 tcg_ctx.tb_ctx.nb_tbs : 0,
1620 target_code_size ? (double) (tcg_ctx.code_gen_ptr -
1621 tcg_ctx.code_gen_buffer) /
1622 target_code_size : 0);
1623 cpu_fprintf(f, "cross page TB count %d (%d%%)\n", cross_page,
1624 tcg_ctx.tb_ctx.nb_tbs ? (cross_page * 100) /
1625 tcg_ctx.tb_ctx.nb_tbs : 0);
1626 cpu_fprintf(f, "direct jump count %d (%d%%) (2 jumps=%d %d%%)\n",
1627 direct_jmp_count,
1628 tcg_ctx.tb_ctx.nb_tbs ? (direct_jmp_count * 100) /
1629 tcg_ctx.tb_ctx.nb_tbs : 0,
1630 direct_jmp2_count,
1631 tcg_ctx.tb_ctx.nb_tbs ? (direct_jmp2_count * 100) /
1632 tcg_ctx.tb_ctx.nb_tbs : 0);
1633 cpu_fprintf(f, "\nStatistics:\n");
1634 cpu_fprintf(f, "TB flush count %d\n", tcg_ctx.tb_ctx.tb_flush_count);
1635 cpu_fprintf(f, "TB invalidate count %d\n",
1636 tcg_ctx.tb_ctx.tb_phys_invalidate_count);
1637 cpu_fprintf(f, "TLB flush count %d\n", tlb_flush_count);
1638 tcg_dump_info(f, cpu_fprintf);
1641 void dump_opcount_info(FILE *f, fprintf_function cpu_fprintf)
1643 tcg_dump_op_count(f, cpu_fprintf);
1646 #else /* CONFIG_USER_ONLY */
1648 void cpu_interrupt(CPUState *cpu, int mask)
1650 cpu->interrupt_request |= mask;
1651 cpu->tcg_exit_req = 1;
1655 * Walks guest process memory "regions" one by one
1656 * and calls callback function 'fn' for each region.
1658 struct walk_memory_regions_data {
1659 walk_memory_regions_fn fn;
1660 void *priv;
1661 target_ulong start;
1662 int prot;
1665 static int walk_memory_regions_end(struct walk_memory_regions_data *data,
1666 target_ulong end, int new_prot)
1668 if (data->start != -1u) {
1669 int rc = data->fn(data->priv, data->start, end, data->prot);
1670 if (rc != 0) {
1671 return rc;
1675 data->start = (new_prot ? end : -1u);
1676 data->prot = new_prot;
1678 return 0;
1681 static int walk_memory_regions_1(struct walk_memory_regions_data *data,
1682 target_ulong base, int level, void **lp)
1684 target_ulong pa;
1685 int i, rc;
1687 if (*lp == NULL) {
1688 return walk_memory_regions_end(data, base, 0);
1691 if (level == 0) {
1692 PageDesc *pd = *lp;
1694 for (i = 0; i < V_L2_SIZE; ++i) {
1695 int prot = pd[i].flags;
1697 pa = base | (i << TARGET_PAGE_BITS);
1698 if (prot != data->prot) {
1699 rc = walk_memory_regions_end(data, pa, prot);
1700 if (rc != 0) {
1701 return rc;
1705 } else {
1706 void **pp = *lp;
1708 for (i = 0; i < V_L2_SIZE; ++i) {
1709 pa = base | ((target_ulong)i <<
1710 (TARGET_PAGE_BITS + V_L2_BITS * level));
1711 rc = walk_memory_regions_1(data, pa, level - 1, pp + i);
1712 if (rc != 0) {
1713 return rc;
1718 return 0;
1721 int walk_memory_regions(void *priv, walk_memory_regions_fn fn)
1723 struct walk_memory_regions_data data;
1724 uintptr_t i;
1726 data.fn = fn;
1727 data.priv = priv;
1728 data.start = -1u;
1729 data.prot = 0;
1731 for (i = 0; i < V_L1_SIZE; i++) {
1732 int rc = walk_memory_regions_1(&data, (target_ulong)i << (V_L1_SHIFT + TARGET_PAGE_BITS),
1733 V_L1_SHIFT / V_L2_BITS - 1, l1_map + i);
1734 if (rc != 0) {
1735 return rc;
1739 return walk_memory_regions_end(&data, 0, 0);
1742 static int dump_region(void *priv, target_ulong start,
1743 target_ulong end, unsigned long prot)
1745 FILE *f = (FILE *)priv;
1747 (void) fprintf(f, TARGET_FMT_lx"-"TARGET_FMT_lx
1748 " "TARGET_FMT_lx" %c%c%c\n",
1749 start, end, end - start,
1750 ((prot & PAGE_READ) ? 'r' : '-'),
1751 ((prot & PAGE_WRITE) ? 'w' : '-'),
1752 ((prot & PAGE_EXEC) ? 'x' : '-'));
1754 return 0;
1757 /* dump memory mappings */
1758 void page_dump(FILE *f)
1760 const int length = sizeof(target_ulong) * 2;
1761 (void) fprintf(f, "%-*s %-*s %-*s %s\n",
1762 length, "start", length, "end", length, "size", "prot");
1763 walk_memory_regions(f, dump_region);
1766 int page_get_flags(target_ulong address)
1768 PageDesc *p;
1770 p = page_find(address >> TARGET_PAGE_BITS);
1771 if (!p) {
1772 return 0;
1774 return p->flags;
1777 /* Modify the flags of a page and invalidate the code if necessary.
1778 The flag PAGE_WRITE_ORG is positioned automatically depending
1779 on PAGE_WRITE. The mmap_lock should already be held. */
1780 void page_set_flags(target_ulong start, target_ulong end, int flags)
1782 target_ulong addr, len;
1784 /* This function should never be called with addresses outside the
1785 guest address space. If this assert fires, it probably indicates
1786 a missing call to h2g_valid. */
1787 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
1788 assert(end < ((target_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
1789 #endif
1790 assert(start < end);
1792 start = start & TARGET_PAGE_MASK;
1793 end = TARGET_PAGE_ALIGN(end);
1795 if (flags & PAGE_WRITE) {
1796 flags |= PAGE_WRITE_ORG;
1799 for (addr = start, len = end - start;
1800 len != 0;
1801 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
1802 PageDesc *p = page_find_alloc(addr >> TARGET_PAGE_BITS, 1);
1804 /* If the write protection bit is set, then we invalidate
1805 the code inside. */
1806 if (!(p->flags & PAGE_WRITE) &&
1807 (flags & PAGE_WRITE) &&
1808 p->first_tb) {
1809 tb_invalidate_phys_page(addr, 0, NULL, false);
1811 p->flags = flags;
1815 int page_check_range(target_ulong start, target_ulong len, int flags)
1817 PageDesc *p;
1818 target_ulong end;
1819 target_ulong addr;
1821 /* This function should never be called with addresses outside the
1822 guest address space. If this assert fires, it probably indicates
1823 a missing call to h2g_valid. */
1824 #if TARGET_ABI_BITS > L1_MAP_ADDR_SPACE_BITS
1825 assert(start < ((target_ulong)1 << L1_MAP_ADDR_SPACE_BITS));
1826 #endif
1828 if (len == 0) {
1829 return 0;
1831 if (start + len - 1 < start) {
1832 /* We've wrapped around. */
1833 return -1;
1836 /* must do before we loose bits in the next step */
1837 end = TARGET_PAGE_ALIGN(start + len);
1838 start = start & TARGET_PAGE_MASK;
1840 for (addr = start, len = end - start;
1841 len != 0;
1842 len -= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
1843 p = page_find(addr >> TARGET_PAGE_BITS);
1844 if (!p) {
1845 return -1;
1847 if (!(p->flags & PAGE_VALID)) {
1848 return -1;
1851 if ((flags & PAGE_READ) && !(p->flags & PAGE_READ)) {
1852 return -1;
1854 if (flags & PAGE_WRITE) {
1855 if (!(p->flags & PAGE_WRITE_ORG)) {
1856 return -1;
1858 /* unprotect the page if it was put read-only because it
1859 contains translated code */
1860 if (!(p->flags & PAGE_WRITE)) {
1861 if (!page_unprotect(addr, 0, NULL)) {
1862 return -1;
1867 return 0;
1870 /* called from signal handler: invalidate the code and unprotect the
1871 page. Return TRUE if the fault was successfully handled. */
1872 int page_unprotect(target_ulong address, uintptr_t pc, void *puc)
1874 unsigned int prot;
1875 PageDesc *p;
1876 target_ulong host_start, host_end, addr;
1878 /* Technically this isn't safe inside a signal handler. However we
1879 know this only ever happens in a synchronous SEGV handler, so in
1880 practice it seems to be ok. */
1881 mmap_lock();
1883 p = page_find(address >> TARGET_PAGE_BITS);
1884 if (!p) {
1885 mmap_unlock();
1886 return 0;
1889 /* if the page was really writable, then we change its
1890 protection back to writable */
1891 if ((p->flags & PAGE_WRITE_ORG) && !(p->flags & PAGE_WRITE)) {
1892 host_start = address & qemu_host_page_mask;
1893 host_end = host_start + qemu_host_page_size;
1895 prot = 0;
1896 for (addr = host_start ; addr < host_end ; addr += TARGET_PAGE_SIZE) {
1897 p = page_find(addr >> TARGET_PAGE_BITS);
1898 p->flags |= PAGE_WRITE;
1899 prot |= p->flags;
1901 /* and since the content will be modified, we must invalidate
1902 the corresponding translated code. */
1903 tb_invalidate_phys_page(addr, pc, puc, true);
1904 #ifdef DEBUG_TB_CHECK
1905 tb_invalidate_check(addr);
1906 #endif
1908 mprotect((void *)g2h(host_start), qemu_host_page_size,
1909 prot & PAGE_BITS);
1911 mmap_unlock();
1912 return 1;
1914 mmap_unlock();
1915 return 0;
1917 #endif /* CONFIG_USER_ONLY */