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, bool ignore_icount)
 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                          | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
 199     tb->orig_tb = tcg_ctx.tb_ctx.tb_invalidated_flag ? NULL : orig_tb;
 200     cpu->current_tb = tb;
 201     /* execute the generated code */
 202     trace_exec_tb_nocache(tb, tb->pc);
 203     cpu_tb_exec(cpu, tb->tc_ptr);
 204     cpu->current_tb = NULL;
 205     tb_phys_invalidate(tb, -1);
 206     tb_free(tb);
 207 }
 208
 209 static TranslationBlock *tb_find_physical(CPUState *cpu,
 210                                           target_ulong pc,
 211                                           target_ulong cs_base,
 212                                           uint64_t flags)
 213 {
 214     CPUArchState *env = (CPUArchState *)cpu->env_ptr;
 215     TranslationBlock *tb, **ptb1;
 216     unsigned int h;
 217     tb_page_addr_t phys_pc, phys_page1;
 218     target_ulong virt_page2;
 219
 220     tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
 221
 222     /* find translated block using physical mappings */
 223     phys_pc = get_page_addr_code(env, pc);
 224     phys_page1 = phys_pc & TARGET_PAGE_MASK;
 225     h = tb_phys_hash_func(phys_pc);
 226     ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
 227     for(;;) {
 228         tb = *ptb1;
 229         if (!tb) {
 230             return NULL;
 231         }
 232         if (tb->pc == pc &&
 233             tb->page_addr[0] == phys_page1 &&
 234             tb->cs_base == cs_base &&
 235             tb->flags == flags) {
 236             /* check next page if needed */
 237             if (tb->page_addr[1] != -1) {
 238                 tb_page_addr_t phys_page2;
 239
 240                 virt_page2 = (pc & TARGET_PAGE_MASK) +
 241                     TARGET_PAGE_SIZE;
 242                 phys_page2 = get_page_addr_code(env, virt_page2);
 243                 if (tb->page_addr[1] == phys_page2) {
 244                     break;
 245                 }
 246             } else {
 247                 break;
 248             }
 249         }
 250         ptb1 = &tb->phys_hash_next;
 251     }
 252
 253     /* Move the TB to the head of the list */
 254     *ptb1 = tb->phys_hash_next;
 255     tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
 256     tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
 257     return tb;
 258 }
 259
 260 static TranslationBlock *tb_find_slow(CPUState *cpu,
 261                                       target_ulong pc,
 262                                       target_ulong cs_base,
 263                                       uint64_t flags)
 264 {
 265     TranslationBlock *tb;
 266
 267     tb = tb_find_physical(cpu, pc, cs_base, flags);
 268     if (tb) {
 269         goto found;
 270     }
 271
 272 #ifdef CONFIG_USER_ONLY
 273     /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
 274      * taken outside tb_lock.  Since we're momentarily dropping
 275      * tb_lock, there's a chance that our desired tb has been
 276      * translated.
 277      */
 278     tb_unlock();
 279     mmap_lock();
 280     tb_lock();
 281     tb = tb_find_physical(cpu, pc, cs_base, flags);
 282     if (tb) {
 283         mmap_unlock();
 284         goto found;
 285     }
 286 #endif
 287
 288     /* if no translated code available, then translate it now */
 289     tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
 290
 291 #ifdef CONFIG_USER_ONLY
 292     mmap_unlock();
 293 #endif
 294
 295 found:
 296     /* we add the TB in the virtual pc hash table */
 297     cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
 298     return tb;
 299 }
 300
 301 static inline TranslationBlock *tb_find_fast(CPUState *cpu)
 302 {
 303     CPUArchState *env = (CPUArchState *)cpu->env_ptr;
 304     TranslationBlock *tb;
 305     target_ulong cs_base, pc;
 306     int flags;
 307
 308     /* we record a subset of the CPU state. It will
 309        always be the same before a given translated block
 310        is executed. */
 311     cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
 312     tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
 313     if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
 314                  tb->flags != flags)) {
 315         tb = tb_find_slow(cpu, pc, cs_base, flags);
 316     }
 317     return tb;
 318 }
 319
 320 static void cpu_handle_debug_exception(CPUState *cpu)
 321 {
 322     CPUClass *cc = CPU_GET_CLASS(cpu);
 323     CPUWatchpoint *wp;
 324
 325     if (!cpu->watchpoint_hit) {
 326         QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
 327             wp->flags &= ~BP_WATCHPOINT_HIT;
 328         }
 329     }
 330
 331     cc->debug_excp_handler(cpu);
 332 }
 333
 334 /* main execution loop */
 335
 336 int cpu_exec(CPUState *cpu)
 337 {
 338     CPUClass *cc = CPU_GET_CLASS(cpu);
 339 #ifdef TARGET_I386
 340     X86CPU *x86_cpu = X86_CPU(cpu);
 341     CPUArchState *env = &x86_cpu->env;
 342 #endif
 343     int ret, interrupt_request;
 344     TranslationBlock *tb;
 345     uint8_t *tc_ptr;
 346     uintptr_t next_tb;
 347     SyncClocks sc;
 348
 349     if (cpu->halted) {
 350 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
 351         if (cpu->interrupt_request & CPU_INTERRUPT_POLL) {
 352             apic_poll_irq(x86_cpu->apic_state);
 353             cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
 354         }
 355 #endif
 356         if (!cpu_has_work(cpu)) {
 357             return EXCP_HALTED;
 358         }
 359
 360         cpu->halted = 0;
 361     }
 362
 363     current_cpu = cpu;
 364     atomic_mb_set(&tcg_current_cpu, cpu);
 365     rcu_read_lock();
 366
 367     if (unlikely(atomic_mb_read(&exit_request))) {
 368         cpu->exit_request = 1;
 369     }
 370
 371     cc->cpu_exec_enter(cpu);
 372
 373     /* Calculate difference between guest clock and host clock.
 374      * This delay includes the delay of the last cycle, so
 375      * what we have to do is sleep until it is 0. As for the
 376      * advance/delay we gain here, we try to fix it next time.
 377      */
 378     init_delay_params(&sc, cpu);
 379
 380     /* prepare setjmp context for exception handling */
 381     for(;;) {
 382         if (sigsetjmp(cpu->jmp_env, 0) == 0) {
 383             /* if an exception is pending, we execute it here */
 384             if (cpu->exception_index >= 0) {
 385                 if (cpu->exception_index >= EXCP_INTERRUPT) {
 386                     /* exit request from the cpu execution loop */
 387                     ret = cpu->exception_index;
 388                     if (ret == EXCP_DEBUG) {
 389                         cpu_handle_debug_exception(cpu);
 390                     }
 391                     cpu->exception_index = -1;
 392                     break;
 393                 } else {
 394 #if defined(CONFIG_USER_ONLY)
 395                     /* if user mode only, we simulate a fake exception
 396                        which will be handled outside the cpu execution
 397                        loop */
 398 #if defined(TARGET_I386)
 399                     cc->do_interrupt(cpu);
 400 #endif
 401                     ret = cpu->exception_index;
 402                     cpu->exception_index = -1;
 403                     break;
 404 #else
 405                     cc->do_interrupt(cpu);
 406                     cpu->exception_index = -1;
 407 #endif
 408                 }
 409             }
 410
 411             next_tb = 0; /* force lookup of first TB */
 412             for(;;) {
 413                 interrupt_request = cpu->interrupt_request;
 414                 if (unlikely(interrupt_request)) {
 415                     if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
 416                         /* Mask out external interrupts for this step. */
 417                         interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
 418                     }
 419                     if (interrupt_request & CPU_INTERRUPT_DEBUG) {
 420                         cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
 421                         cpu->exception_index = EXCP_DEBUG;
 422                         cpu_loop_exit(cpu);
 423                     }
 424                     if (interrupt_request & CPU_INTERRUPT_HALT) {
 425                         cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
 426                         cpu->halted = 1;
 427                         cpu->exception_index = EXCP_HLT;
 428                         cpu_loop_exit(cpu);
 429                     }
 430 #if defined(TARGET_I386)
 431                     if (interrupt_request & CPU_INTERRUPT_INIT) {
 432                         cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
 433                         do_cpu_init(x86_cpu);
 434                         cpu->exception_index = EXCP_HALTED;
 435                         cpu_loop_exit(cpu);
 436                     }
 437 #else
 438                     if (interrupt_request & CPU_INTERRUPT_RESET) {
 439                         cpu_reset(cpu);
 440                     }
 441 #endif
 442                     /* The target hook has 3 exit conditions:
 443                        False when the interrupt isn't processed,
 444                        True when it is, and we should restart on a new TB,
 445                        and via longjmp via cpu_loop_exit.  */
 446                     if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
 447                         next_tb = 0;
 448                     }
 449                     /* Don't use the cached interrupt_request value,
 450                        do_interrupt may have updated the EXITTB flag. */
 451                     if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
 452                         cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
 453                         /* ensure that no TB jump will be modified as
 454                            the program flow was changed */
 455                         next_tb = 0;
 456                     }
 457                 }
 458                 if (unlikely(cpu->exit_request)) {
 459                     cpu->exit_request = 0;
 460                     cpu->exception_index = EXCP_INTERRUPT;
 461                     cpu_loop_exit(cpu);
 462                 }
 463                 tb_lock();
 464                 tb = tb_find_fast(cpu);
 465                 /* Note: we do it here to avoid a gcc bug on Mac OS X when
 466                    doing it in tb_find_slow */
 467                 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
 468                     /* as some TB could have been invalidated because
 469                        of memory exceptions while generating the code, we
 470                        must recompute the hash index here */
 471                     next_tb = 0;
 472                     tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
 473                 }
 474                 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
 475                     qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
 476                              tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
 477                 }
 478                 /* see if we can patch the calling TB. When the TB
 479                    spans two pages, we cannot safely do a direct
 480                    jump. */
 481                 if (next_tb != 0 && tb->page_addr[1] == -1
 482                     && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
 483                     tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
 484                                 next_tb & TB_EXIT_MASK, tb);
 485                 }
 486                 tb_unlock();
 487                 if (likely(!cpu->exit_request)) {
 488                     trace_exec_tb(tb, tb->pc);
 489                     tc_ptr = tb->tc_ptr;
 490                     /* execute the generated code */
 491                     cpu->current_tb = tb;
 492                     next_tb = cpu_tb_exec(cpu, tc_ptr);
 493                     cpu->current_tb = NULL;
 494                     switch (next_tb & TB_EXIT_MASK) {
 495                     case TB_EXIT_REQUESTED:
 496                         /* Something asked us to stop executing
 497                          * chained TBs; just continue round the main
 498                          * loop. Whatever requested the exit will also
 499                          * have set something else (eg exit_request or
 500                          * interrupt_request) which we will handle
 501                          * next time around the loop.  But we need to
 502                          * ensure the tcg_exit_req read in generated code
 503                          * comes before the next read of cpu->exit_request
 504                          * or cpu->interrupt_request.
 505                          */
 506                         smp_rmb();
 507                         next_tb = 0;
 508                         break;
 509                     case TB_EXIT_ICOUNT_EXPIRED:
 510                     {
 511                         /* Instruction counter expired.  */
 512                         int insns_left = cpu->icount_decr.u32;
 513                         if (cpu->icount_extra && insns_left >= 0) {
 514                             /* Refill decrementer and continue execution.  */
 515                             cpu->icount_extra += insns_left;
 516                             insns_left = MIN(0xffff, cpu->icount_extra);
 517                             cpu->icount_extra -= insns_left;
 518                             cpu->icount_decr.u16.low = insns_left;
 519                         } else {
 520                             if (insns_left > 0) {
 521                                 /* Execute remaining instructions.  */
 522                                 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
 523                                 cpu_exec_nocache(cpu, insns_left, tb, false);
 524                                 align_clocks(&sc, cpu);
 525                             }
 526                             cpu->exception_index = EXCP_INTERRUPT;
 527                             next_tb = 0;
 528                             cpu_loop_exit(cpu);
 529                         }
 530                         break;
 531                     }
 532                     default:
 533                         break;
 534                     }
 535                 }
 536                 /* Try to align the host and virtual clocks
 537                    if the guest is in advance */
 538                 align_clocks(&sc, cpu);
 539                 /* reset soft MMU for next block (it can currently
 540                    only be set by a memory fault) */
 541             } /* for(;;) */
 542         } else {
 543             /* Reload env after longjmp - the compiler may have smashed all
 544              * local variables as longjmp is marked 'noreturn'. */
 545             cpu = current_cpu;
 546             cc = CPU_GET_CLASS(cpu);
 547             cpu->can_do_io = 1;
 548 #ifdef TARGET_I386
 549             x86_cpu = X86_CPU(cpu);
 550             env = &x86_cpu->env;
 551 #endif
 552             tb_lock_reset();
 553         }
 554     } /* for(;;) */
 555
 556     cc->cpu_exec_exit(cpu);
 557     rcu_read_unlock();
 558
 559     /* fail safe : never use current_cpu outside cpu_exec() */
 560     current_cpu = NULL;
 561
 562     /* Does not need atomic_mb_set because a spurious wakeup is okay.  */
 563     atomic_set(&tcg_current_cpu, NULL);
 564     return ret;
 565 }