cpu-exec.c

   1 /*
   2  *  emulator main execution loop
   3  *
   4  *  Copyright (c) 2003-2005 Fabrice Bellard
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include "cpu.h"
  21 #include "trace.h"
  22 #include "disas/disas.h"
  23 #include "tcg.h"
  24 #include "qemu/atomic.h"
  25 #include "sysemu/qtest.h"
  26 #include "qemu/timer.h"
  27 #include "exec/address-spaces.h"
  28 #include "qemu/rcu.h"
  29 #include "exec/tb-hash.h"
  30
  31 #if defined(_WIN64)
  32 /* On w64, sigsetjmp is implemented by _setjmp which needs a second parameter.
  33  * If this parameter is NULL, longjump does no stack unwinding.
  34  * That is what we need for QEMU. Passing the value of register rsp (default)
  35  * lets longjmp try a stack unwinding which will crash with generated code. */
  36 #define sigsetjmp(env, savesigs) _setjmp(env, NULL)
  37 #endif
  38
  39 /* -icount align implementation. */
  40
  41 typedef struct SyncClocks {
  42     int64_t diff_clk;
  43     int64_t last_cpu_icount;
  44     int64_t realtime_clock;
  45 } SyncClocks;
  46
  47 #if !defined(CONFIG_USER_ONLY)
  48 /* Allow the guest to have a max 3ms advance.
  49  * The difference between the 2 clocks could therefore
  50  * oscillate around 0.
  51  */
  52 #define VM_CLOCK_ADVANCE 3000000
  53 #define THRESHOLD_REDUCE 1.5
  54 #define MAX_DELAY_PRINT_RATE 2000000000LL
  55 #define MAX_NB_PRINTS 100
  56
  57 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
  58 {
  59     int64_t cpu_icount;
  60
  61     if (!icount_align_option) {
  62         return;
  63     }
  64
  65     cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
  66     sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
  67     sc->last_cpu_icount = cpu_icount;
  68
  69     if (sc->diff_clk > VM_CLOCK_ADVANCE) {
  70 #ifndef _WIN32
  71         struct timespec sleep_delay, rem_delay;
  72         sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
  73         sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
  74         if (nanosleep(&sleep_delay, &rem_delay) < 0) {
  75             sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
  76         } else {
  77             sc->diff_clk = 0;
  78         }
  79 #else
  80         Sleep(sc->diff_clk / SCALE_MS);
  81         sc->diff_clk = 0;
  82 #endif
  83     }
  84 }
  85
  86 static void print_delay(const SyncClocks *sc)
  87 {
  88     static float threshold_delay;
  89     static int64_t last_realtime_clock;
  90     static int nb_prints;
  91
  92     if (icount_align_option &&
  93         sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
  94         nb_prints < MAX_NB_PRINTS) {
  95         if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
  96             (-sc->diff_clk / (float)1000000000LL <
  97              (threshold_delay - THRESHOLD_REDUCE))) {
  98             threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
  99             printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
 100                    threshold_delay - 1,
 101                    threshold_delay);
 102             nb_prints++;
 103             last_realtime_clock = sc->realtime_clock;
 104         }
 105     }
 106 }
 107
 108 static void init_delay_params(SyncClocks *sc,
 109                               const CPUState *cpu)
 110 {
 111     if (!icount_align_option) {
 112         return;
 113     }
 114     sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
 115     sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
 116     sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
 117     if (sc->diff_clk < max_delay) {
 118         max_delay = sc->diff_clk;
 119     }
 120     if (sc->diff_clk > max_advance) {
 121         max_advance = sc->diff_clk;
 122     }
 123
 124     /* Print every 2s max if the guest is late. We limit the number
 125        of printed messages to NB_PRINT_MAX(currently 100) */
 126     print_delay(sc);
 127 }
 128 #else
 129 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
 130 {
 131 }
 132
 133 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
 134 {
 135 }
 136 #endif /* CONFIG USER ONLY */
 137
 138 /* Execute a TB, and fix up the CPU state afterwards if necessary */
 139 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, uint8_t *tb_ptr)
 140 {
 141     CPUArchState *env = cpu->env_ptr;
 142     uintptr_t next_tb;
 143
 144 #if defined(DEBUG_DISAS)
 145     if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
 146 #if defined(TARGET_I386)
 147         log_cpu_state(cpu, CPU_DUMP_CCOP);
 148 #elif defined(TARGET_M68K)
 149         /* ??? Should not modify env state for dumping.  */
 150         cpu_m68k_flush_flags(env, env->cc_op);
 151         env->cc_op = CC_OP_FLAGS;
 152         env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
 153         log_cpu_state(cpu, 0);
 154 #else
 155         log_cpu_state(cpu, 0);
 156 #endif
 157     }
 158 #endif /* DEBUG_DISAS */
 159
 160     cpu->can_do_io = !use_icount;
 161     next_tb = tcg_qemu_tb_exec(env, tb_ptr);
 162     cpu->can_do_io = 1;
 163     trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
 164                        next_tb & TB_EXIT_MASK);
 165
 166     if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
 167         /* We didn't start executing this TB (eg because the instruction
 168          * counter hit zero); we must restore the guest PC to the address
 169          * of the start of the TB.
 170          */
 171         CPUClass *cc = CPU_GET_CLASS(cpu);
 172         TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
 173         if (cc->synchronize_from_tb) {
 174             cc->synchronize_from_tb(cpu, tb);
 175         } else {
 176             assert(cc->set_pc);
 177             cc->set_pc(cpu, tb->pc);
 178         }
 179     }
 180     if ((next_tb & TB_EXIT_MASK) == TB_EXIT_REQUESTED) {
 181         /* We were asked to stop executing TBs (probably a pending
 182          * interrupt. We've now stopped, so clear the flag.
 183          */
 184         cpu->tcg_exit_req = 0;
 185     }
 186     return next_tb;
 187 }
 188
 189 /* Execute the code without caching the generated code. An interpreter
 190    could be used if available. */
 191 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
 192                              TranslationBlock *orig_tb)
 193 {
 194     TranslationBlock *tb;
 195
 196     /* Should never happen.
 197        We only end up here when an existing TB is too long.  */
 198     if (max_cycles > CF_COUNT_MASK)
 199         max_cycles = CF_COUNT_MASK;
 200
 201     tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
 202                      max_cycles | CF_NOCACHE);
 203     tb->orig_tb = tcg_ctx.tb_ctx.tb_invalidated_flag ? NULL : orig_tb;
 204     cpu->current_tb = tb;
 205     /* execute the generated code */
 206     trace_exec_tb_nocache(tb, tb->pc);
 207     cpu_tb_exec(cpu, tb->tc_ptr);
 208     cpu->current_tb = NULL;
 209     tb_phys_invalidate(tb, -1);
 210     tb_free(tb);
 211 }
 212
 213 static TranslationBlock *tb_find_physical(CPUState *cpu,
 214                                           target_ulong pc,
 215                                           target_ulong cs_base,
 216                                           uint64_t flags)
 217 {
 218     CPUArchState *env = (CPUArchState *)cpu->env_ptr;
 219     TranslationBlock *tb, **ptb1;
 220     unsigned int h;
 221     tb_page_addr_t phys_pc, phys_page1;
 222     target_ulong virt_page2;
 223
 224     tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
 225
 226     /* find translated block using physical mappings */
 227     phys_pc = get_page_addr_code(env, pc);
 228     phys_page1 = phys_pc & TARGET_PAGE_MASK;
 229     h = tb_phys_hash_func(phys_pc);
 230     ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
 231     for(;;) {
 232         tb = *ptb1;
 233         if (!tb) {
 234             return NULL;
 235         }
 236         if (tb->pc == pc &&
 237             tb->page_addr[0] == phys_page1 &&
 238             tb->cs_base == cs_base &&
 239             tb->flags == flags) {
 240             /* check next page if needed */
 241             if (tb->page_addr[1] != -1) {
 242                 tb_page_addr_t phys_page2;
 243
 244                 virt_page2 = (pc & TARGET_PAGE_MASK) +
 245                     TARGET_PAGE_SIZE;
 246                 phys_page2 = get_page_addr_code(env, virt_page2);
 247                 if (tb->page_addr[1] == phys_page2) {
 248                     break;
 249                 }
 250             } else {
 251                 break;
 252             }
 253         }
 254         ptb1 = &tb->phys_hash_next;
 255     }
 256
 257     /* Move the TB to the head of the list */
 258     *ptb1 = tb->phys_hash_next;
 259     tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
 260     tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
 261     return tb;
 262 }
 263
 264 static TranslationBlock *tb_find_slow(CPUState *cpu,
 265                                       target_ulong pc,
 266                                       target_ulong cs_base,
 267                                       uint64_t flags)
 268 {
 269     TranslationBlock *tb;
 270
 271     tb = tb_find_physical(cpu, pc, cs_base, flags);
 272     if (tb) {
 273         goto found;
 274     }
 275
 276 #ifdef CONFIG_USER_ONLY
 277     /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
 278      * taken outside tb_lock.  Since we're momentarily dropping
 279      * tb_lock, there's a chance that our desired tb has been
 280      * translated.
 281      */
 282     tb_unlock();
 283     mmap_lock();
 284     tb_lock();
 285     tb = tb_find_physical(cpu, pc, cs_base, flags);
 286     if (tb) {
 287         mmap_unlock();
 288         goto found;
 289     }
 290 #endif
 291
 292     /* if no translated code available, then translate it now */
 293     tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
 294
 295 #ifdef CONFIG_USER_ONLY
 296     mmap_unlock();
 297 #endif
 298
 299 found:
 300     /* we add the TB in the virtual pc hash table */
 301     cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
 302     return tb;
 303 }
 304
 305 static inline TranslationBlock *tb_find_fast(CPUState *cpu)
 306 {
 307     CPUArchState *env = (CPUArchState *)cpu->env_ptr;
 308     TranslationBlock *tb;
 309     target_ulong cs_base, pc;
 310     int flags;
 311
 312     /* we record a subset of the CPU state. It will
 313        always be the same before a given translated block
 314        is executed. */
 315     cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
 316     tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
 317     if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
 318                  tb->flags != flags)) {
 319         tb = tb_find_slow(cpu, pc, cs_base, flags);
 320     }
 321     return tb;
 322 }
 323
 324 static void cpu_handle_debug_exception(CPUState *cpu)
 325 {
 326     CPUClass *cc = CPU_GET_CLASS(cpu);
 327     CPUWatchpoint *wp;
 328
 329     if (!cpu->watchpoint_hit) {
 330         QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
 331             wp->flags &= ~BP_WATCHPOINT_HIT;
 332         }
 333     }
 334
 335     cc->debug_excp_handler(cpu);
 336 }
 337
 338 /* main execution loop */
 339
 340 int cpu_exec(CPUState *cpu)
 341 {
 342     CPUClass *cc = CPU_GET_CLASS(cpu);
 343 #ifdef TARGET_I386
 344     X86CPU *x86_cpu = X86_CPU(cpu);
 345     CPUArchState *env = &x86_cpu->env;
 346 #endif
 347     int ret, interrupt_request;
 348     TranslationBlock *tb;
 349     uint8_t *tc_ptr;
 350     uintptr_t next_tb;
 351     SyncClocks sc;
 352
 353     if (cpu->halted) {
 354         if (!cpu_has_work(cpu)) {
 355             return EXCP_HALTED;
 356         }
 357
 358         cpu->halted = 0;
 359     }
 360
 361     current_cpu = cpu;
 362     atomic_mb_set(&tcg_current_cpu, cpu);
 363     rcu_read_lock();
 364
 365     if (unlikely(atomic_mb_read(&exit_request))) {
 366         cpu->exit_request = 1;
 367     }
 368
 369     cc->cpu_exec_enter(cpu);
 370
 371     /* Calculate difference between guest clock and host clock.
 372      * This delay includes the delay of the last cycle, so
 373      * what we have to do is sleep until it is 0. As for the
 374      * advance/delay we gain here, we try to fix it next time.
 375      */
 376     init_delay_params(&sc, cpu);
 377
 378     /* prepare setjmp context for exception handling */
 379     for(;;) {
 380         if (sigsetjmp(cpu->jmp_env, 0) == 0) {
 381             /* if an exception is pending, we execute it here */
 382             if (cpu->exception_index >= 0) {
 383                 if (cpu->exception_index >= EXCP_INTERRUPT) {
 384                     /* exit request from the cpu execution loop */
 385                     ret = cpu->exception_index;
 386                     if (ret == EXCP_DEBUG) {
 387                         cpu_handle_debug_exception(cpu);
 388                     }
 389                     cpu->exception_index = -1;
 390                     break;
 391                 } else {
 392 #if defined(CONFIG_USER_ONLY)
 393                     /* if user mode only, we simulate a fake exception
 394                        which will be handled outside the cpu execution
 395                        loop */
 396 #if defined(TARGET_I386)
 397                     cc->do_interrupt(cpu);
 398 #endif
 399                     ret = cpu->exception_index;
 400                     cpu->exception_index = -1;
 401                     break;
 402 #else
 403                     cc->do_interrupt(cpu);
 404                     cpu->exception_index = -1;
 405 #endif
 406                 }
 407             }
 408
 409             next_tb = 0; /* force lookup of first TB */
 410             for(;;) {
 411                 interrupt_request = cpu->interrupt_request;
 412                 if (unlikely(interrupt_request)) {
 413                     if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
 414                         /* Mask out external interrupts for this step. */
 415                         interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
 416                     }
 417                     if (interrupt_request & CPU_INTERRUPT_DEBUG) {
 418                         cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
 419                         cpu->exception_index = EXCP_DEBUG;
 420                         cpu_loop_exit(cpu);
 421                     }
 422                     if (interrupt_request & CPU_INTERRUPT_HALT) {
 423                         cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
 424                         cpu->halted = 1;
 425                         cpu->exception_index = EXCP_HLT;
 426                         cpu_loop_exit(cpu);
 427                     }
 428 #if defined(TARGET_I386)
 429                     if (interrupt_request & CPU_INTERRUPT_INIT) {
 430                         cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
 431                         do_cpu_init(x86_cpu);
 432                         cpu->exception_index = EXCP_HALTED;
 433                         cpu_loop_exit(cpu);
 434                     }
 435 #else
 436                     if (interrupt_request & CPU_INTERRUPT_RESET) {
 437                         cpu_reset(cpu);
 438                     }
 439 #endif
 440                     /* The target hook has 3 exit conditions:
 441                        False when the interrupt isn't processed,
 442                        True when it is, and we should restart on a new TB,
 443                        and via longjmp via cpu_loop_exit.  */
 444                     if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
 445                         next_tb = 0;
 446                     }
 447                     /* Don't use the cached interrupt_request value,
 448                        do_interrupt may have updated the EXITTB flag. */
 449                     if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
 450                         cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
 451                         /* ensure that no TB jump will be modified as
 452                            the program flow was changed */
 453                         next_tb = 0;
 454                     }
 455                 }
 456                 if (unlikely(cpu->exit_request)) {
 457                     cpu->exit_request = 0;
 458                     cpu->exception_index = EXCP_INTERRUPT;
 459                     cpu_loop_exit(cpu);
 460                 }
 461                 tb_lock();
 462                 tb = tb_find_fast(cpu);
 463                 /* Note: we do it here to avoid a gcc bug on Mac OS X when
 464                    doing it in tb_find_slow */
 465                 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
 466                     /* as some TB could have been invalidated because
 467                        of memory exceptions while generating the code, we
 468                        must recompute the hash index here */
 469                     next_tb = 0;
 470                     tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
 471                 }
 472                 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
 473                     qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
 474                              tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
 475                 }
 476                 /* see if we can patch the calling TB. When the TB
 477                    spans two pages, we cannot safely do a direct
 478                    jump. */
 479                 if (next_tb != 0 && tb->page_addr[1] == -1) {
 480                     tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
 481                                 next_tb & TB_EXIT_MASK, tb);
 482                 }
 483                 tb_unlock();
 484                 if (likely(!cpu->exit_request)) {
 485                     trace_exec_tb(tb, tb->pc);
 486                     tc_ptr = tb->tc_ptr;
 487                     /* execute the generated code */
 488                     cpu->current_tb = tb;
 489                     next_tb = cpu_tb_exec(cpu, tc_ptr);
 490                     cpu->current_tb = NULL;
 491                     switch (next_tb & TB_EXIT_MASK) {
 492                     case TB_EXIT_REQUESTED:
 493                         /* Something asked us to stop executing
 494                          * chained TBs; just continue round the main
 495                          * loop. Whatever requested the exit will also
 496                          * have set something else (eg exit_request or
 497                          * interrupt_request) which we will handle
 498                          * next time around the loop.  But we need to
 499                          * ensure the tcg_exit_req read in generated code
 500                          * comes before the next read of cpu->exit_request
 501                          * or cpu->interrupt_request.
 502                          */
 503                         smp_rmb();
 504                         next_tb = 0;
 505                         break;
 506                     case TB_EXIT_ICOUNT_EXPIRED:
 507                     {
 508                         /* Instruction counter expired.  */
 509                         int insns_left = cpu->icount_decr.u32;
 510                         if (cpu->icount_extra && insns_left >= 0) {
 511                             /* Refill decrementer and continue execution.  */
 512                             cpu->icount_extra += insns_left;
 513                             insns_left = MIN(0xffff, cpu->icount_extra);
 514                             cpu->icount_extra -= insns_left;
 515                             cpu->icount_decr.u16.low = insns_left;
 516                         } else {
 517                             if (insns_left > 0) {
 518                                 /* Execute remaining instructions.  */
 519                                 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
 520                                 cpu_exec_nocache(cpu, insns_left, tb);
 521                                 align_clocks(&sc, cpu);
 522                             }
 523                             cpu->exception_index = EXCP_INTERRUPT;
 524                             next_tb = 0;
 525                             cpu_loop_exit(cpu);
 526                         }
 527                         break;
 528                     }
 529                     default:
 530                         break;
 531                     }
 532                 }
 533                 /* Try to align the host and virtual clocks
 534                    if the guest is in advance */
 535                 align_clocks(&sc, cpu);
 536                 /* reset soft MMU for next block (it can currently
 537                    only be set by a memory fault) */
 538             } /* for(;;) */
 539         } else {
 540             /* Reload env after longjmp - the compiler may have smashed all
 541              * local variables as longjmp is marked 'noreturn'. */
 542             /* TODO: Fix comment in 2014 - then it was wrong. */
 543             g_assert(cpu == current_cpu);
 544             g_assert(cc == CPU_GET_CLASS(cpu));
 545             cpu->can_do_io = 1;
 546 #ifdef TARGET_I386
 547             g_assert(x86_cpu == X86_CPU(cpu));
 548             g_assert(env == cpu->env_ptr);
 549 #endif
 550             tb_lock_reset();
 551         }
 552     } /* for(;;) */
 553
 554     cc->cpu_exec_exit(cpu);
 555     rcu_read_unlock();
 556
 557     /* fail safe : never use current_cpu outside cpu_exec() */
 558     current_cpu = NULL;
 559
 560     /* Does not need atomic_mb_set because a spurious wakeup is okay.  */
 561     atomic_set(&tcg_current_cpu, NULL);
 562     return ret;
 563 }