kqemu.c

   1 /*
   2  *  KQEMU support
   3  *
   4  *  Copyright (c) 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, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  */
  20 #include "config.h"
  21 #ifdef _WIN32
  22 #include <windows.h>
  23 #include <winioctl.h>
  24 #else
  25 #include <sys/types.h>
  26 #include <sys/mman.h>
  27 #include <sys/ioctl.h>
  28 #endif
  29 #ifdef HOST_SOLARIS
  30 #include <sys/ioccom.h>
  31 #endif
  32 #include <stdlib.h>
  33 #include <stdio.h>
  34 #include <stdarg.h>
  35 #include <string.h>
  36 #include <errno.h>
  37 #include <unistd.h>
  38 #include <inttypes.h>
  39
  40 #include "cpu.h"
  41 #include "exec-all.h"
  42
  43 #ifdef USE_KQEMU
  44
  45 #define DEBUG
  46 //#define PROFILE
  47
  48 #include <unistd.h>
  49 #include <fcntl.h>
  50 #include "kqemu.h"
  51
  52 /* compatibility stuff */
  53 #ifndef KQEMU_RET_SYSCALL
  54 #define KQEMU_RET_SYSCALL   0x0300 /* syscall insn */
  55 #endif
  56 #ifndef KQEMU_MAX_RAM_PAGES_TO_UPDATE
  57 #define KQEMU_MAX_RAM_PAGES_TO_UPDATE 512
  58 #define KQEMU_RAM_PAGES_UPDATE_ALL (KQEMU_MAX_RAM_PAGES_TO_UPDATE + 1)
  59 #endif
  60 #ifndef KQEMU_MAX_MODIFIED_RAM_PAGES
  61 #define KQEMU_MAX_MODIFIED_RAM_PAGES 512
  62 #endif
  63
  64 #ifdef _WIN32
  65 #define KQEMU_DEVICE "\\\\.\\kqemu"
  66 #else
  67 #define KQEMU_DEVICE "/dev/kqemu"
  68 #endif
  69
  70 #ifdef _WIN32
  71 #define KQEMU_INVALID_FD INVALID_HANDLE_VALUE
  72 HANDLE kqemu_fd = KQEMU_INVALID_FD;
  73 #define kqemu_closefd(x) CloseHandle(x)
  74 #else
  75 #define KQEMU_INVALID_FD -1
  76 int kqemu_fd = KQEMU_INVALID_FD;
  77 #define kqemu_closefd(x) close(x)
  78 #endif
  79
  80 /* 0 = not allowed
  81    1 = user kqemu
  82    2 = kernel kqemu
  83 */
  84 int kqemu_allowed = 1;
  85 unsigned long *pages_to_flush;
  86 unsigned int nb_pages_to_flush;
  87 unsigned long *ram_pages_to_update;
  88 unsigned int nb_ram_pages_to_update;
  89 unsigned long *modified_ram_pages;
  90 unsigned int nb_modified_ram_pages;
  91 uint8_t *modified_ram_pages_table;
  92 extern uint32_t **l1_phys_map;
  93
  94 #define cpuid(index, eax, ebx, ecx, edx) \
  95   asm volatile ("cpuid" \
  96                 : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) \
  97                 : "0" (index))
  98
  99 #ifdef __x86_64__
 100 static int is_cpuid_supported(void)
 101 {
 102     return 1;
 103 }
 104 #else
 105 static int is_cpuid_supported(void)
 106 {
 107     int v0, v1;
 108     asm volatile ("pushf\n"
 109                   "popl %0\n"
 110                   "movl %0, %1\n"
 111                   "xorl $0x00200000, %0\n"
 112                   "pushl %0\n"
 113                   "popf\n"
 114                   "pushf\n"
 115                   "popl %0\n"
 116                   : "=a" (v0), "=d" (v1)
 117                   :
 118                   : "cc");
 119     return (v0 != v1);
 120 }
 121 #endif
 122
 123 static void kqemu_update_cpuid(CPUState *env)
 124 {
 125     int critical_features_mask, features, ext_features, ext_features_mask;
 126     uint32_t eax, ebx, ecx, edx;
 127
 128     /* the following features are kept identical on the host and
 129        target cpus because they are important for user code. Strictly
 130        speaking, only SSE really matters because the OS must support
 131        it if the user code uses it. */
 132     critical_features_mask =
 133         CPUID_CMOV | CPUID_CX8 |
 134         CPUID_FXSR | CPUID_MMX | CPUID_SSE |
 135         CPUID_SSE2 | CPUID_SEP;
 136     ext_features_mask = CPUID_EXT_SSE3 | CPUID_EXT_MONITOR;
 137     if (!is_cpuid_supported()) {
 138         features = 0;
 139         ext_features = 0;
 140     } else {
 141         cpuid(1, eax, ebx, ecx, edx);
 142         features = edx;
 143         ext_features = ecx;
 144     }
 145 #ifdef __x86_64__
 146     /* NOTE: on x86_64 CPUs, SYSENTER is not supported in
 147        compatibility mode, so in order to have the best performances
 148        it is better not to use it */
 149     features &= ~CPUID_SEP;
 150 #endif
 151     env->cpuid_features = (env->cpuid_features & ~critical_features_mask) |
 152         (features & critical_features_mask);
 153     env->cpuid_ext_features = (env->cpuid_ext_features & ~ext_features_mask) |
 154         (ext_features & ext_features_mask);
 155     /* XXX: we could update more of the target CPUID state so that the
 156        non accelerated code sees exactly the same CPU features as the
 157        accelerated code */
 158 }
 159
 160 int kqemu_init(CPUState *env)
 161 {
 162     struct kqemu_init init;
 163     int ret, version;
 164 #ifdef _WIN32
 165     DWORD temp;
 166 #endif
 167
 168     if (!kqemu_allowed)
 169         return -1;
 170
 171 #ifdef _WIN32
 172     kqemu_fd = CreateFile(KQEMU_DEVICE, GENERIC_WRITE | GENERIC_READ,
 173                           FILE_SHARE_READ | FILE_SHARE_WRITE,
 174                           NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL,
 175                           NULL);
 176 #else
 177     kqemu_fd = open(KQEMU_DEVICE, O_RDWR);
 178 #endif
 179     if (kqemu_fd == KQEMU_INVALID_FD) {
 180         fprintf(stderr, "Could not open '%s' - QEMU acceleration layer not activated: %s\n",
 181                 KQEMU_DEVICE, strerror(errno));
 182         return -1;
 183     }
 184     version = 0;
 185 #ifdef _WIN32
 186     DeviceIoControl(kqemu_fd, KQEMU_GET_VERSION, NULL, 0,
 187                     &version, sizeof(version), &temp, NULL);
 188 #else
 189     ioctl(kqemu_fd, KQEMU_GET_VERSION, &version);
 190 #endif
 191     if (version != KQEMU_VERSION) {
 192         fprintf(stderr, "Version mismatch between kqemu module and qemu (%08x %08x) - disabling kqemu use\n",
 193                 version, KQEMU_VERSION);
 194         goto fail;
 195     }
 196
 197     pages_to_flush = qemu_vmalloc(KQEMU_MAX_PAGES_TO_FLUSH *
 198                                   sizeof(unsigned long));
 199     if (!pages_to_flush)
 200         goto fail;
 201
 202     ram_pages_to_update = qemu_vmalloc(KQEMU_MAX_RAM_PAGES_TO_UPDATE *
 203                                        sizeof(unsigned long));
 204     if (!ram_pages_to_update)
 205         goto fail;
 206
 207     modified_ram_pages = qemu_vmalloc(KQEMU_MAX_MODIFIED_RAM_PAGES *
 208                                       sizeof(unsigned long));
 209     if (!modified_ram_pages)
 210         goto fail;
 211     modified_ram_pages_table = qemu_mallocz(phys_ram_size >> TARGET_PAGE_BITS);
 212     if (!modified_ram_pages_table)
 213         goto fail;
 214
 215     init.ram_base = phys_ram_base;
 216     init.ram_size = phys_ram_size;
 217     init.ram_dirty = phys_ram_dirty;
 218     init.phys_to_ram_map = l1_phys_map;
 219     init.pages_to_flush = pages_to_flush;
 220 #if KQEMU_VERSION >= 0x010200
 221     init.ram_pages_to_update = ram_pages_to_update;
 222 #endif
 223 #if KQEMU_VERSION >= 0x010300
 224     init.modified_ram_pages = modified_ram_pages;
 225 #endif
 226 #ifdef _WIN32
 227     ret = DeviceIoControl(kqemu_fd, KQEMU_INIT, &init, sizeof(init),
 228                           NULL, 0, &temp, NULL) == TRUE ? 0 : -1;
 229 #else
 230     ret = ioctl(kqemu_fd, KQEMU_INIT, &init);
 231 #endif
 232     if (ret < 0) {
 233         fprintf(stderr, "Error %d while initializing QEMU acceleration layer - disabling it for now\n", ret);
 234     fail:
 235         kqemu_closefd(kqemu_fd);
 236         kqemu_fd = KQEMU_INVALID_FD;
 237         return -1;
 238     }
 239     kqemu_update_cpuid(env);
 240     env->kqemu_enabled = kqemu_allowed;
 241     nb_pages_to_flush = 0;
 242     nb_ram_pages_to_update = 0;
 243     return 0;
 244 }
 245
 246 void kqemu_flush_page(CPUState *env, target_ulong addr)
 247 {
 248 #if defined(DEBUG)
 249     if (loglevel & CPU_LOG_INT) {
 250         fprintf(logfile, "kqemu_flush_page: addr=" TARGET_FMT_lx "\n", addr);
 251     }
 252 #endif
 253     if (nb_pages_to_flush >= KQEMU_MAX_PAGES_TO_FLUSH)
 254         nb_pages_to_flush = KQEMU_FLUSH_ALL;
 255     else
 256         pages_to_flush[nb_pages_to_flush++] = addr;
 257 }
 258
 259 void kqemu_flush(CPUState *env, int global)
 260 {
 261 #ifdef DEBUG
 262     if (loglevel & CPU_LOG_INT) {
 263         fprintf(logfile, "kqemu_flush:\n");
 264     }
 265 #endif
 266     nb_pages_to_flush = KQEMU_FLUSH_ALL;
 267 }
 268
 269 void kqemu_set_notdirty(CPUState *env, ram_addr_t ram_addr)
 270 {
 271 #ifdef DEBUG
 272     if (loglevel & CPU_LOG_INT) {
 273         fprintf(logfile, "kqemu_set_notdirty: addr=%08lx\n", ram_addr);
 274     }
 275 #endif
 276     /* we only track transitions to dirty state */
 277     if (phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] != 0xff)
 278         return;
 279     if (nb_ram_pages_to_update >= KQEMU_MAX_RAM_PAGES_TO_UPDATE)
 280         nb_ram_pages_to_update = KQEMU_RAM_PAGES_UPDATE_ALL;
 281     else
 282         ram_pages_to_update[nb_ram_pages_to_update++] = ram_addr;
 283 }
 284
 285 static void kqemu_reset_modified_ram_pages(void)
 286 {
 287     int i;
 288     unsigned long page_index;
 289
 290     for(i = 0; i < nb_modified_ram_pages; i++) {
 291         page_index = modified_ram_pages[i] >> TARGET_PAGE_BITS;
 292         modified_ram_pages_table[page_index] = 0;
 293     }
 294     nb_modified_ram_pages = 0;
 295 }
 296
 297 void kqemu_modify_page(CPUState *env, ram_addr_t ram_addr)
 298 {
 299     unsigned long page_index;
 300     int ret;
 301 #ifdef _WIN32
 302     DWORD temp;
 303 #endif
 304
 305     page_index = ram_addr >> TARGET_PAGE_BITS;
 306     if (!modified_ram_pages_table[page_index]) {
 307 #if 0
 308         printf("%d: modify_page=%08lx\n", nb_modified_ram_pages, ram_addr);
 309 #endif
 310         modified_ram_pages_table[page_index] = 1;
 311         modified_ram_pages[nb_modified_ram_pages++] = ram_addr;
 312         if (nb_modified_ram_pages >= KQEMU_MAX_MODIFIED_RAM_PAGES) {
 313             /* flush */
 314 #ifdef _WIN32
 315             ret = DeviceIoControl(kqemu_fd, KQEMU_MODIFY_RAM_PAGES,
 316                                   &nb_modified_ram_pages,
 317                                   sizeof(nb_modified_ram_pages),
 318                                   NULL, 0, &temp, NULL);
 319 #else
 320             ret = ioctl(kqemu_fd, KQEMU_MODIFY_RAM_PAGES,
 321                         &nb_modified_ram_pages);
 322 #endif
 323             kqemu_reset_modified_ram_pages();
 324         }
 325     }
 326 }
 327
 328 struct fpstate {
 329     uint16_t fpuc;
 330     uint16_t dummy1;
 331     uint16_t fpus;
 332     uint16_t dummy2;
 333     uint16_t fptag;
 334     uint16_t dummy3;
 335
 336     uint32_t fpip;
 337     uint32_t fpcs;
 338     uint32_t fpoo;
 339     uint32_t fpos;
 340     uint8_t fpregs1[8 * 10];
 341 };
 342
 343 struct fpxstate {
 344     uint16_t fpuc;
 345     uint16_t fpus;
 346     uint16_t fptag;
 347     uint16_t fop;
 348     uint32_t fpuip;
 349     uint16_t cs_sel;
 350     uint16_t dummy0;
 351     uint32_t fpudp;
 352     uint16_t ds_sel;
 353     uint16_t dummy1;
 354     uint32_t mxcsr;
 355     uint32_t mxcsr_mask;
 356     uint8_t fpregs1[8 * 16];
 357     uint8_t xmm_regs[16 * 16];
 358     uint8_t dummy2[96];
 359 };
 360
 361 static struct fpxstate fpx1 __attribute__((aligned(16)));
 362
 363 static void restore_native_fp_frstor(CPUState *env)
 364 {
 365     int fptag, i, j;
 366     struct fpstate fp1, *fp = &fp1;
 367
 368     fp->fpuc = env->fpuc;
 369     fp->fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
 370     fptag = 0;
 371     for (i=7; i>=0; i--) {
 372         fptag <<= 2;
 373         if (env->fptags[i]) {
 374             fptag |= 3;
 375         } else {
 376             /* the FPU automatically computes it */
 377         }
 378     }
 379     fp->fptag = fptag;
 380     j = env->fpstt;
 381     for(i = 0;i < 8; i++) {
 382         memcpy(&fp->fpregs1[i * 10], &env->fpregs[j].d, 10);
 383         j = (j + 1) & 7;
 384     }
 385     asm volatile ("frstor %0" : "=m" (*fp));
 386 }
 387
 388 static void save_native_fp_fsave(CPUState *env)
 389 {
 390     int fptag, i, j;
 391     uint16_t fpuc;
 392     struct fpstate fp1, *fp = &fp1;
 393
 394     asm volatile ("fsave %0" : : "m" (*fp));
 395     env->fpuc = fp->fpuc;
 396     env->fpstt = (fp->fpus >> 11) & 7;
 397     env->fpus = fp->fpus & ~0x3800;
 398     fptag = fp->fptag;
 399     for(i = 0;i < 8; i++) {
 400         env->fptags[i] = ((fptag & 3) == 3);
 401         fptag >>= 2;
 402     }
 403     j = env->fpstt;
 404     for(i = 0;i < 8; i++) {
 405         memcpy(&env->fpregs[j].d, &fp->fpregs1[i * 10], 10);
 406         j = (j + 1) & 7;
 407     }
 408     /* we must restore the default rounding state */
 409     fpuc = 0x037f | (env->fpuc & (3 << 10));
 410     asm volatile("fldcw %0" : : "m" (fpuc));
 411 }
 412
 413 static void restore_native_fp_fxrstor(CPUState *env)
 414 {
 415     struct fpxstate *fp = &fpx1;
 416     int i, j, fptag;
 417
 418     fp->fpuc = env->fpuc;
 419     fp->fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
 420     fptag = 0;
 421     for(i = 0; i < 8; i++)
 422         fptag |= (env->fptags[i] << i);
 423     fp->fptag = fptag ^ 0xff;
 424
 425     j = env->fpstt;
 426     for(i = 0;i < 8; i++) {
 427         memcpy(&fp->fpregs1[i * 16], &env->fpregs[j].d, 10);
 428         j = (j + 1) & 7;
 429     }
 430     if (env->cpuid_features & CPUID_SSE) {
 431         fp->mxcsr = env->mxcsr;
 432         /* XXX: check if DAZ is not available */
 433         fp->mxcsr_mask = 0xffff;
 434         memcpy(fp->xmm_regs, env->xmm_regs, CPU_NB_REGS * 16);
 435     }
 436     asm volatile ("fxrstor %0" : "=m" (*fp));
 437 }
 438
 439 static void save_native_fp_fxsave(CPUState *env)
 440 {
 441     struct fpxstate *fp = &fpx1;
 442     int fptag, i, j;
 443     uint16_t fpuc;
 444
 445     asm volatile ("fxsave %0" : : "m" (*fp));
 446     env->fpuc = fp->fpuc;
 447     env->fpstt = (fp->fpus >> 11) & 7;
 448     env->fpus = fp->fpus & ~0x3800;
 449     fptag = fp->fptag ^ 0xff;
 450     for(i = 0;i < 8; i++) {
 451         env->fptags[i] = (fptag >> i) & 1;
 452     }
 453     j = env->fpstt;
 454     for(i = 0;i < 8; i++) {
 455         memcpy(&env->fpregs[j].d, &fp->fpregs1[i * 16], 10);
 456         j = (j + 1) & 7;
 457     }
 458     if (env->cpuid_features & CPUID_SSE) {
 459         env->mxcsr = fp->mxcsr;
 460         memcpy(env->xmm_regs, fp->xmm_regs, CPU_NB_REGS * 16);
 461     }
 462
 463     /* we must restore the default rounding state */
 464     asm volatile ("fninit");
 465     fpuc = 0x037f | (env->fpuc & (3 << 10));
 466     asm volatile("fldcw %0" : : "m" (fpuc));
 467 }
 468
 469 static int do_syscall(CPUState *env,
 470                       struct kqemu_cpu_state *kenv)
 471 {
 472     int selector;
 473
 474     selector = (env->star >> 32) & 0xffff;
 475 #ifdef __x86_64__
 476     if (env->hflags & HF_LMA_MASK) {
 477         int code64;
 478
 479         env->regs[R_ECX] = kenv->next_eip;
 480         env->regs[11] = env->eflags;
 481
 482         code64 = env->hflags & HF_CS64_MASK;
 483
 484         cpu_x86_set_cpl(env, 0);
 485         cpu_x86_load_seg_cache(env, R_CS, selector & 0xfffc,
 486                                0, 0xffffffff,
 487                                DESC_G_MASK | DESC_P_MASK |
 488                                DESC_S_MASK |
 489                                DESC_CS_MASK | DESC_R_MASK | DESC_A_MASK | DESC_L_MASK);
 490         cpu_x86_load_seg_cache(env, R_SS, (selector + 8) & 0xfffc,
 491                                0, 0xffffffff,
 492                                DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
 493                                DESC_S_MASK |
 494                                DESC_W_MASK | DESC_A_MASK);
 495         env->eflags &= ~env->fmask;
 496         if (code64)
 497             env->eip = env->lstar;
 498         else
 499             env->eip = env->cstar;
 500     } else
 501 #endif
 502     {
 503         env->regs[R_ECX] = (uint32_t)kenv->next_eip;
 504
 505         cpu_x86_set_cpl(env, 0);
 506         cpu_x86_load_seg_cache(env, R_CS, selector & 0xfffc,
 507                            0, 0xffffffff,
 508                                DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
 509                                DESC_S_MASK |
 510                                DESC_CS_MASK | DESC_R_MASK | DESC_A_MASK);
 511         cpu_x86_load_seg_cache(env, R_SS, (selector + 8) & 0xfffc,
 512                                0, 0xffffffff,
 513                                DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
 514                                DESC_S_MASK |
 515                                DESC_W_MASK | DESC_A_MASK);
 516         env->eflags &= ~(IF_MASK | RF_MASK | VM_MASK);
 517         env->eip = (uint32_t)env->star;
 518     }
 519     return 2;
 520 }
 521
 522 #ifdef CONFIG_PROFILER
 523
 524 #define PC_REC_SIZE 1
 525 #define PC_REC_HASH_BITS 16
 526 #define PC_REC_HASH_SIZE (1 << PC_REC_HASH_BITS)
 527
 528 typedef struct PCRecord {
 529     unsigned long pc;
 530     int64_t count;
 531     struct PCRecord *next;
 532 } PCRecord;
 533
 534 static PCRecord *pc_rec_hash[PC_REC_HASH_SIZE];
 535 static int nb_pc_records;
 536
 537 static void kqemu_record_pc(unsigned long pc)
 538 {
 539     unsigned long h;
 540     PCRecord **pr, *r;
 541
 542     h = pc / PC_REC_SIZE;
 543     h = h ^ (h >> PC_REC_HASH_BITS);
 544     h &= (PC_REC_HASH_SIZE - 1);
 545     pr = &pc_rec_hash[h];
 546     for(;;) {
 547         r = *pr;
 548         if (r == NULL)
 549             break;
 550         if (r->pc == pc) {
 551             r->count++;
 552             return;
 553         }
 554         pr = &r->next;
 555     }
 556     r = malloc(sizeof(PCRecord));
 557     r->count = 1;
 558     r->pc = pc;
 559     r->next = NULL;
 560     *pr = r;
 561     nb_pc_records++;
 562 }
 563
 564 static int pc_rec_cmp(const void *p1, const void *p2)
 565 {
 566     PCRecord *r1 = *(PCRecord **)p1;
 567     PCRecord *r2 = *(PCRecord **)p2;
 568     if (r1->count < r2->count)
 569         return 1;
 570     else if (r1->count == r2->count)
 571         return 0;
 572     else
 573         return -1;
 574 }
 575
 576 static void kqemu_record_flush(void)
 577 {
 578     PCRecord *r, *r_next;
 579     int h;
 580
 581     for(h = 0; h < PC_REC_HASH_SIZE; h++) {
 582         for(r = pc_rec_hash[h]; r != NULL; r = r_next) {
 583             r_next = r->next;
 584             free(r);
 585         }
 586         pc_rec_hash[h] = NULL;
 587     }
 588     nb_pc_records = 0;
 589 }
 590
 591 void kqemu_record_dump(void)
 592 {
 593     PCRecord **pr, *r;
 594     int i, h;
 595     FILE *f;
 596     int64_t total, sum;
 597
 598     pr = malloc(sizeof(PCRecord *) * nb_pc_records);
 599     i = 0;
 600     total = 0;
 601     for(h = 0; h < PC_REC_HASH_SIZE; h++) {
 602         for(r = pc_rec_hash[h]; r != NULL; r = r->next) {
 603             pr[i++] = r;
 604             total += r->count;
 605         }
 606     }
 607     qsort(pr, nb_pc_records, sizeof(PCRecord *), pc_rec_cmp);
 608
 609     f = fopen("/tmp/kqemu.stats", "w");
 610     if (!f) {
 611         perror("/tmp/kqemu.stats");
 612         exit(1);
 613     }
 614     fprintf(f, "total: %" PRId64 "\n", total);
 615     sum = 0;
 616     for(i = 0; i < nb_pc_records; i++) {
 617         r = pr[i];
 618         sum += r->count;
 619         fprintf(f, "%08lx: %" PRId64 " %0.2f%% %0.2f%%\n",
 620                 r->pc,
 621                 r->count,
 622                 (double)r->count / (double)total * 100.0,
 623                 (double)sum / (double)total * 100.0);
 624     }
 625     fclose(f);
 626     free(pr);
 627
 628     kqemu_record_flush();
 629 }
 630 #endif
 631
 632 int kqemu_cpu_exec(CPUState *env)
 633 {
 634     struct kqemu_cpu_state kcpu_state, *kenv = &kcpu_state;
 635     int ret, cpl, i;
 636 #ifdef CONFIG_PROFILER
 637     int64_t ti;
 638 #endif
 639
 640 #ifdef _WIN32
 641     DWORD temp;
 642 #endif
 643
 644 #ifdef CONFIG_PROFILER
 645     ti = profile_getclock();
 646 #endif
 647 #ifdef DEBUG
 648     if (loglevel & CPU_LOG_INT) {
 649         fprintf(logfile, "kqemu: cpu_exec: enter\n");
 650         cpu_dump_state(env, logfile, fprintf, 0);
 651     }
 652 #endif
 653     memcpy(kenv->regs, env->regs, sizeof(kenv->regs));
 654     kenv->eip = env->eip;
 655     kenv->eflags = env->eflags;
 656     memcpy(&kenv->segs, &env->segs, sizeof(env->segs));
 657     memcpy(&kenv->ldt, &env->ldt, sizeof(env->ldt));
 658     memcpy(&kenv->tr, &env->tr, sizeof(env->tr));
 659     memcpy(&kenv->gdt, &env->gdt, sizeof(env->gdt));
 660     memcpy(&kenv->idt, &env->idt, sizeof(env->idt));
 661     kenv->cr0 = env->cr[0];
 662     kenv->cr2 = env->cr[2];
 663     kenv->cr3 = env->cr[3];
 664     kenv->cr4 = env->cr[4];
 665     kenv->a20_mask = env->a20_mask;
 666 #if KQEMU_VERSION >= 0x010100
 667     kenv->efer = env->efer;
 668 #endif
 669 #if KQEMU_VERSION >= 0x010300
 670     kenv->tsc_offset = 0;
 671     kenv->star = env->star;
 672     kenv->sysenter_cs = env->sysenter_cs;
 673     kenv->sysenter_esp = env->sysenter_esp;
 674     kenv->sysenter_eip = env->sysenter_eip;
 675 #ifdef __x86_64__
 676     kenv->lstar = env->lstar;
 677     kenv->cstar = env->cstar;
 678     kenv->fmask = env->fmask;
 679     kenv->kernelgsbase = env->kernelgsbase;
 680 #endif
 681 #endif
 682     if (env->dr[7] & 0xff) {
 683         kenv->dr7 = env->dr[7];
 684         kenv->dr0 = env->dr[0];
 685         kenv->dr1 = env->dr[1];
 686         kenv->dr2 = env->dr[2];
 687         kenv->dr3 = env->dr[3];
 688     } else {
 689         kenv->dr7 = 0;
 690     }
 691     kenv->dr6 = env->dr[6];
 692     cpl = (env->hflags & HF_CPL_MASK);
 693     kenv->cpl = cpl;
 694     kenv->nb_pages_to_flush = nb_pages_to_flush;
 695 #if KQEMU_VERSION >= 0x010200
 696     kenv->user_only = (env->kqemu_enabled == 1);
 697     kenv->nb_ram_pages_to_update = nb_ram_pages_to_update;
 698 #endif
 699     nb_ram_pages_to_update = 0;
 700
 701 #if KQEMU_VERSION >= 0x010300
 702     kenv->nb_modified_ram_pages = nb_modified_ram_pages;
 703 #endif
 704     kqemu_reset_modified_ram_pages();
 705
 706     if (env->cpuid_features & CPUID_FXSR)
 707         restore_native_fp_fxrstor(env);
 708     else
 709         restore_native_fp_frstor(env);
 710
 711 #ifdef _WIN32
 712     if (DeviceIoControl(kqemu_fd, KQEMU_EXEC,
 713                         kenv, sizeof(struct kqemu_cpu_state),
 714                         kenv, sizeof(struct kqemu_cpu_state),
 715                         &temp, NULL)) {
 716         ret = kenv->retval;
 717     } else {
 718         ret = -1;
 719     }
 720 #else
 721 #if KQEMU_VERSION >= 0x010100
 722     ioctl(kqemu_fd, KQEMU_EXEC, kenv);
 723     ret = kenv->retval;
 724 #else
 725     ret = ioctl(kqemu_fd, KQEMU_EXEC, kenv);
 726 #endif
 727 #endif
 728     if (env->cpuid_features & CPUID_FXSR)
 729         save_native_fp_fxsave(env);
 730     else
 731         save_native_fp_fsave(env);
 732
 733     memcpy(env->regs, kenv->regs, sizeof(env->regs));
 734     env->eip = kenv->eip;
 735     env->eflags = kenv->eflags;
 736     memcpy(env->segs, kenv->segs, sizeof(env->segs));
 737     cpu_x86_set_cpl(env, kenv->cpl);
 738     memcpy(&env->ldt, &kenv->ldt, sizeof(env->ldt));
 739 #if 0
 740     /* no need to restore that */
 741     memcpy(env->tr, kenv->tr, sizeof(env->tr));
 742     memcpy(env->gdt, kenv->gdt, sizeof(env->gdt));
 743     memcpy(env->idt, kenv->idt, sizeof(env->idt));
 744     env->a20_mask = kenv->a20_mask;
 745 #endif
 746     env->cr[0] = kenv->cr0;
 747     env->cr[4] = kenv->cr4;
 748     env->cr[3] = kenv->cr3;
 749     env->cr[2] = kenv->cr2;
 750     env->dr[6] = kenv->dr6;
 751 #if KQEMU_VERSION >= 0x010300
 752 #ifdef __x86_64__
 753     env->kernelgsbase = kenv->kernelgsbase;
 754 #endif
 755 #endif
 756
 757     /* flush pages as indicated by kqemu */
 758     if (kenv->nb_pages_to_flush >= KQEMU_FLUSH_ALL) {
 759         tlb_flush(env, 1);
 760     } else {
 761         for(i = 0; i < kenv->nb_pages_to_flush; i++) {
 762             tlb_flush_page(env, pages_to_flush[i]);
 763         }
 764     }
 765     nb_pages_to_flush = 0;
 766
 767 #ifdef CONFIG_PROFILER
 768     kqemu_time += profile_getclock() - ti;
 769     kqemu_exec_count++;
 770 #endif
 771
 772 #if KQEMU_VERSION >= 0x010200
 773     if (kenv->nb_ram_pages_to_update > 0) {
 774         cpu_tlb_update_dirty(env);
 775     }
 776 #endif
 777
 778 #if KQEMU_VERSION >= 0x010300
 779     if (kenv->nb_modified_ram_pages > 0) {
 780         for(i = 0; i < kenv->nb_modified_ram_pages; i++) {
 781             unsigned long addr;
 782             addr = modified_ram_pages[i];
 783             tb_invalidate_phys_page_range(addr, addr + TARGET_PAGE_SIZE, 0);
 784         }
 785     }
 786 #endif
 787
 788     /* restore the hidden flags */
 789     {
 790         unsigned int new_hflags;
 791 #ifdef TARGET_X86_64
 792         if ((env->hflags & HF_LMA_MASK) &&
 793             (env->segs[R_CS].flags & DESC_L_MASK)) {
 794             /* long mode */
 795             new_hflags = HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
 796         } else
 797 #endif
 798         {
 799             /* legacy / compatibility case */
 800             new_hflags = (env->segs[R_CS].flags & DESC_B_MASK)
 801                 >> (DESC_B_SHIFT - HF_CS32_SHIFT);
 802             new_hflags |= (env->segs[R_SS].flags & DESC_B_MASK)
 803                 >> (DESC_B_SHIFT - HF_SS32_SHIFT);
 804             if (!(env->cr[0] & CR0_PE_MASK) ||
 805                    (env->eflags & VM_MASK) ||
 806                    !(env->hflags & HF_CS32_MASK)) {
 807                 /* XXX: try to avoid this test. The problem comes from the
 808                    fact that is real mode or vm86 mode we only modify the
 809                    'base' and 'selector' fields of the segment cache to go
 810                    faster. A solution may be to force addseg to one in
 811                    translate-i386.c. */
 812                 new_hflags |= HF_ADDSEG_MASK;
 813             } else {
 814                 new_hflags |= ((env->segs[R_DS].base |
 815                                 env->segs[R_ES].base |
 816                                 env->segs[R_SS].base) != 0) <<
 817                     HF_ADDSEG_SHIFT;
 818             }
 819         }
 820         env->hflags = (env->hflags &
 821            ~(HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)) |
 822             new_hflags;
 823     }
 824     /* update FPU flags */
 825     env->hflags = (env->hflags & ~(HF_MP_MASK | HF_EM_MASK | HF_TS_MASK)) |
 826         ((env->cr[0] << (HF_MP_SHIFT - 1)) & (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK));
 827     if (env->cr[4] & CR4_OSFXSR_MASK)
 828         env->hflags |= HF_OSFXSR_MASK;
 829     else
 830         env->hflags &= ~HF_OSFXSR_MASK;
 831
 832 #ifdef DEBUG
 833     if (loglevel & CPU_LOG_INT) {
 834         fprintf(logfile, "kqemu: kqemu_cpu_exec: ret=0x%x\n", ret);
 835     }
 836 #endif
 837     if (ret == KQEMU_RET_SYSCALL) {
 838         /* syscall instruction */
 839         return do_syscall(env, kenv);
 840     } else
 841     if ((ret & 0xff00) == KQEMU_RET_INT) {
 842         env->exception_index = ret & 0xff;
 843         env->error_code = 0;
 844         env->exception_is_int = 1;
 845         env->exception_next_eip = kenv->next_eip;
 846 #ifdef CONFIG_PROFILER
 847         kqemu_ret_int_count++;
 848 #endif
 849 #ifdef DEBUG
 850         if (loglevel & CPU_LOG_INT) {
 851             fprintf(logfile, "kqemu: interrupt v=%02x:\n",
 852                     env->exception_index);
 853             cpu_dump_state(env, logfile, fprintf, 0);
 854         }
 855 #endif
 856         return 1;
 857     } else if ((ret & 0xff00) == KQEMU_RET_EXCEPTION) {
 858         env->exception_index = ret & 0xff;
 859         env->error_code = kenv->error_code;
 860         env->exception_is_int = 0;
 861         env->exception_next_eip = 0;
 862 #ifdef CONFIG_PROFILER
 863         kqemu_ret_excp_count++;
 864 #endif
 865 #ifdef DEBUG
 866         if (loglevel & CPU_LOG_INT) {
 867             fprintf(logfile, "kqemu: exception v=%02x e=%04x:\n",
 868                     env->exception_index, env->error_code);
 869             cpu_dump_state(env, logfile, fprintf, 0);
 870         }
 871 #endif
 872         return 1;
 873     } else if (ret == KQEMU_RET_INTR) {
 874 #ifdef CONFIG_PROFILER
 875         kqemu_ret_intr_count++;
 876 #endif
 877 #ifdef DEBUG
 878         if (loglevel & CPU_LOG_INT) {
 879             cpu_dump_state(env, logfile, fprintf, 0);
 880         }
 881 #endif
 882         return 0;
 883     } else if (ret == KQEMU_RET_SOFTMMU) {
 884 #ifdef CONFIG_PROFILER
 885         {
 886             unsigned long pc = env->eip + env->segs[R_CS].base;
 887             kqemu_record_pc(pc);
 888         }
 889 #endif
 890 #ifdef DEBUG
 891         if (loglevel & CPU_LOG_INT) {
 892             cpu_dump_state(env, logfile, fprintf, 0);
 893         }
 894 #endif
 895         return 2;
 896     } else {
 897         cpu_dump_state(env, stderr, fprintf, 0);
 898         fprintf(stderr, "Unsupported return value: 0x%x\n", ret);
 899         exit(1);
 900     }
 901     return 0;
 902 }
 903
 904 void kqemu_cpu_interrupt(CPUState *env)
 905 {
 906 #if defined(_WIN32) && KQEMU_VERSION >= 0x010101
 907     /* cancelling the I/O request causes KQEMU to finish executing the
 908        current block and successfully returning. */
 909     CancelIo(kqemu_fd);
 910 #endif
 911 }
 912
 913 #endif