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 #include "config.h"
  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 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
  31 #include "hw/i386/apic.h"
  32 #endif
  33
  34 /* -icount align implementation. */
  35
  36 typedef struct SyncClocks {
  37     int64_t diff_clk;
  38     int64_t last_cpu_icount;
  39     int64_t realtime_clock;
  40 } SyncClocks;
  41
  42 #if !defined(CONFIG_USER_ONLY)
  43 /* Allow the guest to have a max 3ms advance.
  44  * The difference between the 2 clocks could therefore
  45  * oscillate around 0.
  46  */
  47 #define VM_CLOCK_ADVANCE 3000000
  48 #define THRESHOLD_REDUCE 1.5
  49 #define MAX_DELAY_PRINT_RATE 2000000000LL
  50 #define MAX_NB_PRINTS 100
  51
  52 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
  53 {
  54     int64_t cpu_icount;
  55
  56     if (!icount_align_option) {
  57         return;
  58     }
  59
  60     cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
  61     sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
  62     sc->last_cpu_icount = cpu_icount;
  63
  64     if (sc->diff_clk > VM_CLOCK_ADVANCE) {
  65 #ifndef _WIN32
  66         struct timespec sleep_delay, rem_delay;
  67         sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
  68         sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
  69         if (nanosleep(&sleep_delay, &rem_delay) < 0) {
  70             sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
  71         } else {
  72             sc->diff_clk = 0;
  73         }
  74 #else
  75         Sleep(sc->diff_clk / SCALE_MS);
  76         sc->diff_clk = 0;
  77 #endif
  78     }
  79 }
  80
  81 static void print_delay(const SyncClocks *sc)
  82 {
  83     static float threshold_delay;
  84     static int64_t last_realtime_clock;
  85     static int nb_prints;
  86
  87     if (icount_align_option &&
  88         sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
  89         nb_prints < MAX_NB_PRINTS) {
  90         if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
  91             (-sc->diff_clk / (float)1000000000LL <
  92              (threshold_delay - THRESHOLD_REDUCE))) {
  93             threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
  94             printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
  95                    threshold_delay - 1,
  96                    threshold_delay);
  97             nb_prints++;
  98             last_realtime_clock = sc->realtime_clock;
  99         }
 100     }
 101 }
 102
 103 static void init_delay_params(SyncClocks *sc,
 104                               const CPUState *cpu)
 105 {
 106     if (!icount_align_option) {
 107         return;
 108     }
 109     sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
 110     sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
 111     sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
 112     if (sc->diff_clk < max_delay) {
 113         max_delay = sc->diff_clk;
 114     }
 115     if (sc->diff_clk > max_advance) {
 116         max_advance = sc->diff_clk;
 117     }
 118
 119     /* Print every 2s max if the guest is late. We limit the number
 120        of printed messages to NB_PRINT_MAX(currently 100) */
 121     print_delay(sc);
 122 }
 123 #else
 124 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
 125 {
 126 }
 127
 128 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
 129 {
 130 }
 131 #endif /* CONFIG USER ONLY */
 132
 133 /* Execute a TB, and fix up the CPU state afterwards if necessary */
 134 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, uint8_t *tb_ptr)
 135 {
 136     CPUArchState *env = cpu->env_ptr;
 137     uintptr_t next_tb;
 138
 139 #if defined(DEBUG_DISAS)
 140     if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
 141 #if defined(TARGET_I386)
 142         log_cpu_state(cpu, CPU_DUMP_CCOP);
 143 #elif defined(TARGET_M68K)
 144         /* ??? Should not modify env state for dumping.  */
 145         cpu_m68k_flush_flags(env, env->cc_op);
 146         env->cc_op = CC_OP_FLAGS;
 147         env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
 148         log_cpu_state(cpu, 0);
 149 #else
 150         log_cpu_state(cpu, 0);
 151 #endif
 152     }
 153 #endif /* DEBUG_DISAS */
 154
 155     cpu->can_do_io = !use_icount;
 156     next_tb = tcg_qemu_tb_exec(env, tb_ptr);
 157     cpu->can_do_io = 1;
 158     trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
 159                        next_tb & TB_EXIT_MASK);
 160
 161     if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
 162         /* We didn't start executing this TB (eg because the instruction
 163          * counter hit zero); we must restore the guest PC to the address
 164          * of the start of the TB.
 165          */
 166         CPUClass *cc = CPU_GET_CLASS(cpu);
 167         TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
 168         if (cc->synchronize_from_tb) {
 169             cc->synchronize_from_tb(cpu, tb);
 170         } else {
 171             assert(cc->set_pc);
 172             cc->set_pc(cpu, tb->pc);
 173         }
 174     }
 175     if ((next_tb & TB_EXIT_MASK) == TB_EXIT_REQUESTED) {
 176         /* We were asked to stop executing TBs (probably a pending
 177          * interrupt. We've now stopped, so clear the flag.
 178          */
 179         cpu->tcg_exit_req = 0;
 180     }
 181     return next_tb;
 182 }
 183
 184 /* Execute the code without caching the generated code. An interpreter
 185    could be used if available. */
 186 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
 187                              TranslationBlock *orig_tb)
 188 {
 189     TranslationBlock *tb;
 190
 191     /* Should never happen.
 192        We only end up here when an existing TB is too long.  */
 193     if (max_cycles > CF_COUNT_MASK)
 194         max_cycles = CF_COUNT_MASK;
 195
 196     tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
 197                      max_cycles | CF_NOCACHE);
 198     tb->orig_tb = tcg_ctx.tb_ctx.tb_invalidated_flag ? NULL : orig_tb;
 199     cpu->current_tb = tb;
 200     /* execute the generated code */
 201     trace_exec_tb_nocache(tb, tb->pc);
 202     cpu_tb_exec(cpu, tb->tc_ptr);
 203     cpu->current_tb = NULL;
 204     tb_phys_invalidate(tb, -1);
 205     tb_free(tb);
 206 }
 207
 208 static TranslationBlock *tb_find_physical(CPUState *cpu,
 209                                           target_ulong pc,
 210                                           target_ulong cs_base,
 211                                           uint64_t flags)
 212 {
 213     CPUArchState *env = (CPUArchState *)cpu->env_ptr;
 214     TranslationBlock *tb, **ptb1;
 215     unsigned int h;
 216     tb_page_addr_t phys_pc, phys_page1;
 217     target_ulong virt_page2;
 218
 219     tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
 220
 221     /* find translated block using physical mappings */
 222     phys_pc = get_page_addr_code(env, pc);
 223     phys_page1 = phys_pc & TARGET_PAGE_MASK;
 224     h = tb_phys_hash_func(phys_pc);
 225     ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
 226     for(;;) {
 227         tb = *ptb1;
 228         if (!tb) {
 229             return NULL;
 230         }
 231         if (tb->pc == pc &&
 232             tb->page_addr[0] == phys_page1 &&
 233             tb->cs_base == cs_base &&
 234             tb->flags == flags) {
 235             /* check next page if needed */
 236             if (tb->page_addr[1] != -1) {
 237                 tb_page_addr_t phys_page2;
 238
 239                 virt_page2 = (pc & TARGET_PAGE_MASK) +
 240                     TARGET_PAGE_SIZE;
 241                 phys_page2 = get_page_addr_code(env, virt_page2);
 242                 if (tb->page_addr[1] == phys_page2) {
 243                     break;
 244                 }
 245             } else {
 246                 break;
 247             }
 248         }
 249         ptb1 = &tb->phys_hash_next;
 250     }
 251
 252     /* Move the TB to the head of the list */
 253     *ptb1 = tb->phys_hash_next;
 254     tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
 255     tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
 256     return tb;
 257 }
 258
 259 static TranslationBlock *tb_find_slow(CPUState *cpu,
 260                                       target_ulong pc,
 261                                       target_ulong cs_base,
 262                                       uint64_t flags)
 263 {
 264     TranslationBlock *tb;
 265
 266     tb = tb_find_physical(cpu, pc, cs_base, flags);
 267     if (tb) {
 268         goto found;
 269     }
 270
 271 #ifdef CONFIG_USER_ONLY
 272     /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
 273      * taken outside tb_lock.  Since we're momentarily dropping
 274      * tb_lock, there's a chance that our desired tb has been
 275      * translated.
 276      */
 277     tb_unlock();
 278     mmap_lock();
 279     tb_lock();
 280     tb = tb_find_physical(cpu, pc, cs_base, flags);
 281     if (tb) {
 282         mmap_unlock();
 283         goto found;
 284     }
 285 #endif
 286
 287     /* if no translated code available, then translate it now */
 288     tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
 289
 290 #ifdef CONFIG_USER_ONLY
 291     mmap_unlock();
 292 #endif
 293
 294 found:
 295     /* we add the TB in the virtual pc hash table */
 296     cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
 297     return tb;
 298 }
 299
 300 static inline TranslationBlock *tb_find_fast(CPUState *cpu)
 301 {
 302     CPUArchState *env = (CPUArchState *)cpu->env_ptr;
 303     TranslationBlock *tb;
 304     target_ulong cs_base, pc;
 305     int flags;
 306
 307     /* we record a subset of the CPU state. It will
 308        always be the same before a given translated block
 309        is executed. */
 310     cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
 311     tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
 312     if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
 313                  tb->flags != flags)) {
 314         tb = tb_find_slow(cpu, pc, cs_base, flags);
 315     }
 316     return tb;
 317 }
 318
 319 static void cpu_handle_debug_exception(CPUState *cpu)
 320 {
 321     CPUClass *cc = CPU_GET_CLASS(cpu);
 322     CPUWatchpoint *wp;
 323
 324     if (!cpu->watchpoint_hit) {
 325         QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
 326             wp->flags &= ~BP_WATCHPOINT_HIT;
 327         }
 328     }
 329
 330     cc->debug_excp_handler(cpu);
 331 }
 332
 333 /* main execution loop */
 334
 335 int cpu_exec(CPUState *cpu)
 336 {
 337     CPUClass *cc = CPU_GET_CLASS(cpu);
 338 #ifdef TARGET_I386
 339     X86CPU *x86_cpu = X86_CPU(cpu);
 340     CPUArchState *env = &x86_cpu->env;
 341 #endif
 342     int ret, interrupt_request;
 343     TranslationBlock *tb;
 344     uint8_t *tc_ptr;
 345     uintptr_t next_tb;
 346     SyncClocks sc;
 347
 348     if (cpu->halted) {
 349 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
 350         if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
 351             apic_poll_irq(x86_cpu->apic_state);
 352             cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
 353         }
 354 #endif
 355         if (!cpu_has_work(cpu)) {
 356             return EXCP_HALTED;
 357         }
 358
 359         cpu->halted = 0;
 360     }
 361
 362     current_cpu = cpu;
 363     atomic_mb_set(&tcg_current_cpu, cpu);
 364     rcu_read_lock();
 365
 366     if (unlikely(atomic_mb_read(&exit_request))) {
 367         cpu->exit_request = 1;
 368     }
 369
 370     cc->cpu_exec_enter(cpu);
 371
 372     /* Calculate difference between guest clock and host clock.
 373      * This delay includes the delay of the last cycle, so
 374      * what we have to do is sleep until it is 0. As for the
 375      * advance/delay we gain here, we try to fix it next time.
 376      */
 377     init_delay_params(&sc, cpu);
 378
 379     /* prepare setjmp context for exception handling */
 380     for(;;) {
 381         if (sigsetjmp(cpu->jmp_env, 0) == 0) {
 382             /* if an exception is pending, we execute it here */
 383             if (cpu->exception_index >= 0) {
 384                 if (cpu->exception_index >= EXCP_INTERRUPT) {
 385                     /* exit request from the cpu execution loop */
 386                     ret = cpu->exception_index;
 387                     if (ret == EXCP_DEBUG) {
 388                         cpu_handle_debug_exception(cpu);
 389                     }
 390                     cpu->exception_index = -1;
 391                     break;
 392                 } else {
 393 #if defined(CONFIG_USER_ONLY)
 394                     /* if user mode only, we simulate a fake exception
 395                        which will be handled outside the cpu execution
 396                        loop */
 397 #if defined(TARGET_I386)
 398                     cc->do_interrupt(cpu);
 399 #endif
 400                     ret = cpu->exception_index;
 401                     cpu->exception_index = -1;
 402                     break;
 403 #else
 404                     cc->do_interrupt(cpu);
 405                     cpu->exception_index = -1;
 406 #endif
 407                 }
 408             }
 409
 410             next_tb = 0; /* force lookup of first TB */
 411             for(;;) {
 412                 interrupt_request = cpu->interrupt_request;
 413                 if (unlikely(interrupt_request)) {
 414                     if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
 415                         /* Mask out external interrupts for this step. */
 416                         interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
 417                     }
 418                     if (interrupt_request & CPU_INTERRUPT_DEBUG) {
 419                         cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
 420                         cpu->exception_index = EXCP_DEBUG;
 421                         cpu_loop_exit(cpu);
 422                     }
 423                     if (interrupt_request & CPU_INTERRUPT_HALT) {
 424                         cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
 425                         cpu->halted = 1;
 426                         cpu->exception_index = EXCP_HLT;
 427                         cpu_loop_exit(cpu);
 428                     }
 429 #if defined(TARGET_I386)
 430                     if (interrupt_request & CPU_INTERRUPT_INIT) {
 431                         cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
 432                         do_cpu_init(x86_cpu);
 433                         cpu->exception_index = EXCP_HALTED;
 434                         cpu_loop_exit(cpu);
 435                     }
 436 #else
 437                     if (interrupt_request & CPU_INTERRUPT_RESET) {
 438                         cpu_reset(cpu);
 439                     }
 440 #endif
 441                     /* The target hook has 3 exit conditions:
 442                        False when the interrupt isn't processed,
 443                        True when it is, and we should restart on a new TB,
 444                        and via longjmp via cpu_loop_exit.  */
 445                     if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
 446                         next_tb = 0;
 447                     }
 448                     /* Don't use the cached interrupt_request value,
 449                        do_interrupt may have updated the EXITTB flag. */
 450                     if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
 451                         cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
 452                         /* ensure that no TB jump will be modified as
 453                            the program flow was changed */
 454                         next_tb = 0;
 455                     }
 456                 }
 457                 if (unlikely(cpu->exit_request)) {
 458                     cpu->exit_request = 0;
 459                     cpu->exception_index = EXCP_INTERRUPT;
 460                     cpu_loop_exit(cpu);
 461                 }
 462                 tb_lock();
 463                 tb = tb_find_fast(cpu);
 464                 /* Note: we do it here to avoid a gcc bug on Mac OS X when
 465                    doing it in tb_find_slow */
 466                 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
 467                     /* as some TB could have been invalidated because
 468                        of memory exceptions while generating the code, we
 469                        must recompute the hash index here */
 470                     next_tb = 0;
 471                     tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
 472                 }
 473                 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
 474                     qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
 475                              tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
 476                 }
 477                 /* see if we can patch the calling TB. When the TB
 478                    spans two pages, we cannot safely do a direct
 479                    jump. */
 480                 if (next_tb != 0 && tb->page_addr[1] == -1) {
 481                     tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
 482                                 next_tb & TB_EXIT_MASK, tb);
 483                 }
 484                 tb_unlock();
 485                 if (likely(!cpu->exit_request)) {
 486                     trace_exec_tb(tb, tb->pc);
 487                     tc_ptr = tb->tc_ptr;
 488                     /* execute the generated code */
 489                     cpu->current_tb = tb;
 490                     next_tb = cpu_tb_exec(cpu, tc_ptr);
 491                     cpu->current_tb = NULL;
 492                     switch (next_tb & TB_EXIT_MASK) {
 493                     case TB_EXIT_REQUESTED:
 494                         /* Something asked us to stop executing
 495                          * chained TBs; just continue round the main
 496                          * loop. Whatever requested the exit will also
 497                          * have set something else (eg exit_request or
 498                          * interrupt_request) which we will handle
 499                          * next time around the loop.  But we need to
 500                          * ensure the tcg_exit_req read in generated code
 501                          * comes before the next read of cpu->exit_request
 502                          * or cpu->interrupt_request.
 503                          */
 504                         smp_rmb();
 505                         next_tb = 0;
 506                         break;
 507                     case TB_EXIT_ICOUNT_EXPIRED:
 508                     {
 509                         /* Instruction counter expired.  */
 510                         int insns_left = cpu->icount_decr.u32;
 511                         if (cpu->icount_extra && insns_left >= 0) {
 512                             /* Refill decrementer and continue execution.  */
 513                             cpu->icount_extra += insns_left;
 514                             insns_left = MIN(0xffff, cpu->icount_extra);
 515                             cpu->icount_extra -= insns_left;
 516                             cpu->icount_decr.u16.low = insns_left;
 517                         } else {
 518                             if (insns_left > 0) {
 519                                 /* Execute remaining instructions.  */
 520                                 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
 521                                 cpu_exec_nocache(cpu, insns_left, tb);
 522                                 align_clocks(&sc, cpu);
 523                             }
 524                             cpu->exception_index = EXCP_INTERRUPT;
 525                             next_tb = 0;
 526                             cpu_loop_exit(cpu);
 527                         }
 528                         break;
 529                     }
 530                     default:
 531                         break;
 532                     }
 533                 }
 534                 /* Try to align the host and virtual clocks
 535                    if the guest is in advance */
 536                 align_clocks(&sc, cpu);
 537                 /* reset soft MMU for next block (it can currently
 538                    only be set by a memory fault) */
 539             } /* for(;;) */
 540         } else {
 541             /* Reload env after longjmp - the compiler may have smashed all
 542              * local variables as longjmp is marked 'noreturn'. */
 543             cpu = current_cpu;
 544             cc = CPU_GET_CLASS(cpu);
 545             cpu->can_do_io = 1;
 546 #ifdef TARGET_I386
 547             x86_cpu = X86_CPU(cpu);
 548             env = &x86_cpu->env;
 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 }