kvm/user/main.c

   1 /*
   2  * Kernel-based Virtual Machine test driver
   3  *
   4  * This test driver provides a simple way of testing kvm, without a full
   5  * device model.
   6  *
   7  * Copyright (C) 2006 Qumranet
   8  *
   9  * Authors:
  10  *
  11  *  Avi Kivity <avi@qumranet.com>
  12  *  Yaniv Kamay <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the GNU LGPL license, version 2.
  15  */
  16
  17 #define _GNU_SOURCE
  18
  19 #include <libkvm.h>
  20 #include "test/x86/lib/apic.h"
  21 #include "test/x86/ioram.h"
  22
  23 #include <stdio.h>
  24 #include <unistd.h>
  25 #include <fcntl.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <semaphore.h>
  29 #include <sys/types.h>
  30 #include <errno.h>
  31 #include <pthread.h>
  32 #include <signal.h>
  33 #include <pthread.h>
  34 #include <sys/syscall.h>
  35 #include <linux/unistd.h>
  36 #include <getopt.h>
  37 #include <stdbool.h>
  38
  39 #include "iotable.h"
  40
  41 static uint8_t ioram[IORAM_LEN];
  42
  43 static int gettid(void)
  44 {
  45         return syscall(__NR_gettid);
  46 }
  47
  48 static int tkill(int pid, int sig)
  49 {
  50         return syscall(__NR_tkill, pid, sig);
  51 }
  52
  53 kvm_context_t kvm;
  54
  55 #define MAX_VCPUS 4
  56
  57 #define IPI_SIGNAL (SIGRTMIN + 4)
  58
  59 static int ncpus = 1;
  60 static sem_t init_sem;
  61 static __thread int vcpu;
  62 static int apic_ipi_vector = 0xff;
  63 static sigset_t kernel_sigmask;
  64 static sigset_t ipi_sigmask;
  65 static uint64_t memory_size = 128 * 1024 * 1024;
  66
  67 static struct io_table pio_table;
  68
  69 struct vcpu_info {
  70         pid_t tid;
  71         sem_t sipi_sem;
  72 };
  73
  74 struct vcpu_info *vcpus;
  75
  76 static uint32_t apic_sipi_addr;
  77
  78 static void apic_send_sipi(int vcpu)
  79 {
  80         sem_post(&vcpus[vcpu].sipi_sem);
  81 }
  82
  83 static void apic_send_ipi(int vcpu)
  84 {
  85         struct vcpu_info *v;
  86
  87         if (vcpu < 0 || vcpu >= ncpus)
  88                 return;
  89         v = &vcpus[vcpu];
  90         tkill(v->tid, IPI_SIGNAL);
  91 }
  92
  93 static int apic_io(void *opaque, int size, int is_write,
  94                    uint64_t addr, uint64_t *value)
  95 {
  96         if (!is_write)
  97                 *value = -1u;
  98
  99         switch (addr - APIC_BASE) {
 100         case APIC_REG_NCPU:
 101                 if (!is_write)
 102                         *value = ncpus;
 103                 break;
 104         case APIC_REG_ID:
 105                 if (!is_write)
 106                         *value = vcpu;
 107                 break;
 108         case APIC_REG_SIPI_ADDR:
 109                 if (!is_write)
 110                         *value = apic_sipi_addr;
 111                 else
 112                         apic_sipi_addr = *value;
 113                 break;
 114         case APIC_REG_SEND_SIPI:
 115                 if (is_write)
 116                         apic_send_sipi(*value);
 117                 break;
 118         case APIC_REG_IPI_VECTOR:
 119                 if (!is_write)
 120                         *value = apic_ipi_vector;
 121                 else
 122                         apic_ipi_vector = *value;
 123                 break;
 124         case APIC_REG_SEND_IPI:
 125                 if (is_write)
 126                         apic_send_ipi(*value);
 127                 break;
 128         }
 129
 130         return 0;
 131 }
 132
 133 static int apic_init(void)
 134 {
 135         return io_table_register(&pio_table, APIC_BASE,
 136                                  APIC_SIZE, apic_io, NULL);
 137 }
 138
 139 static int misc_io(void *opaque, int size, int is_write,
 140                    uint64_t addr, uint64_t *value)
 141 {
 142         static int newline = 1;
 143
 144         if (!is_write)
 145                 *value = -1;
 146
 147         switch (addr) {
 148         case 0xff: // irq injector
 149                 if (is_write) {
 150                         printf("injecting interrupt 0x%x\n", (uint8_t)*value);
 151                         kvm_inject_irq(kvm, 0, *value);
 152                 }
 153                 break;
 154         case 0xf1: // serial
 155                 if (is_write) {
 156                         if (newline)
 157                                 fputs("GUEST: ", stdout);
 158                         putchar(*value);
 159                         newline = *value == '\n';
 160                 }
 161                 break;
 162         case 0xd1:
 163                 if (!is_write)
 164                         *value = memory_size;
 165                 break;
 166         case 0xf4: // exit
 167                 if (is_write)
 168                         exit(*value);
 169                 break;
 170         }
 171
 172         return 0;
 173 }
 174
 175 static int misc_init(void)
 176 {
 177         int err;
 178
 179         err = io_table_register(&pio_table, 0xff, 1, misc_io, NULL);
 180         if (err < 0)
 181                 return err;
 182
 183         err = io_table_register(&pio_table, 0xf1, 1, misc_io, NULL);
 184         if (err < 0)
 185                 return err;
 186
 187         err = io_table_register(&pio_table, 0xf4, 1, misc_io, NULL);
 188         if (err < 0)
 189                 return err;
 190
 191         return io_table_register(&pio_table, 0xd1, 1, misc_io, NULL);
 192 }
 193
 194 static int test_inb(void *opaque, uint16_t addr, uint8_t *value)
 195 {
 196         struct io_table_entry *entry;
 197
 198         entry = io_table_lookup(&pio_table, addr);
 199         if (entry) {
 200                 uint64_t val;
 201                 entry->handler(entry->opaque, 1, 0, addr, &val);
 202                 *value = val;
 203         } else {
 204                 *value = -1;
 205                 printf("inb 0x%x\n", addr);
 206         }
 207
 208         return 0;
 209 }
 210
 211 static int test_inw(void *opaque, uint16_t addr, uint16_t *value)
 212 {
 213         struct io_table_entry *entry;
 214
 215         entry = io_table_lookup(&pio_table, addr);
 216         if (entry) {
 217                 uint64_t val;
 218                 entry->handler(entry->opaque, 2, 0, addr, &val);
 219                 *value = val;
 220         } else {
 221                 *value = -1;
 222                 printf("inw 0x%x\n", addr);
 223         }
 224
 225         return 0;
 226 }
 227
 228 static int test_inl(void *opaque, uint16_t addr, uint32_t *value)
 229 {
 230         struct io_table_entry *entry;
 231
 232         entry = io_table_lookup(&pio_table, addr);
 233         if (entry) {
 234                 uint64_t val;
 235                 entry->handler(entry->opaque, 4, 0, addr, &val);
 236                 *value = val;
 237         } else {
 238                 *value = -1;
 239                 printf("inl 0x%x\n", addr);
 240         }
 241
 242         return 0;
 243 }
 244
 245 static int test_outb(void *opaque, uint16_t addr, uint8_t value)
 246 {
 247         struct io_table_entry *entry;
 248
 249         entry = io_table_lookup(&pio_table, addr);
 250         if (entry) {
 251                 uint64_t val = value;
 252                 entry->handler(entry->opaque, 1, 1, addr, &val);
 253         } else
 254                 printf("outb $0x%x, 0x%x\n", value, addr);
 255
 256         return 0;
 257 }
 258
 259 static int test_outw(void *opaque, uint16_t addr, uint16_t value)
 260 {
 261         struct io_table_entry *entry;
 262
 263         entry = io_table_lookup(&pio_table, addr);
 264         if (entry) {
 265                 uint64_t val = value;
 266                 entry->handler(entry->opaque, 2, 1, addr, &val);
 267         } else
 268                 printf("outw $0x%x, 0x%x\n", value, addr);
 269
 270         return 0;
 271 }
 272
 273 static int test_outl(void *opaque, uint16_t addr, uint32_t value)
 274 {
 275         struct io_table_entry *entry;
 276
 277         entry = io_table_lookup(&pio_table, addr);
 278         if (entry) {
 279                 uint64_t val = value;
 280                 entry->handler(entry->opaque, 4, 1, addr, &val);
 281         } else
 282                 printf("outl $0x%x, 0x%x\n", value, addr);
 283
 284         return 0;
 285 }
 286
 287 static int test_debug(void *opaque, int vcpu)
 288 {
 289         printf("test_debug\n");
 290         return 0;
 291 }
 292
 293 static int test_halt(void *opaque, int vcpu)
 294 {
 295         int n;
 296
 297         sigwait(&ipi_sigmask, &n);
 298         kvm_inject_irq(kvm, vcpu, apic_ipi_vector);
 299         return 0;
 300 }
 301
 302 static int test_io_window(void *opaque)
 303 {
 304         return 0;
 305 }
 306
 307 static int test_try_push_interrupts(void *opaque)
 308 {
 309         return 0;
 310 }
 311
 312 static void test_post_kvm_run(void *opaque, int vcpu)
 313 {
 314 }
 315
 316 static int test_pre_kvm_run(void *opaque, int vcpu)
 317 {
 318         return 0;
 319 }
 320
 321 static int test_mem_read(void *opaque, uint64_t addr, uint8_t *data, int len)
 322 {
 323         if (addr < IORAM_BASE_PHYS || addr + len > IORAM_BASE_PHYS + IORAM_LEN)
 324                 return 1;
 325         memcpy(data, ioram + addr - IORAM_BASE_PHYS, len);
 326         return 0;
 327 }
 328
 329 static int test_mem_write(void *opaque, uint64_t addr, uint8_t *data, int len)
 330 {
 331         if (addr < IORAM_BASE_PHYS || addr + len > IORAM_BASE_PHYS + IORAM_LEN)
 332                 return 1;
 333         memcpy(ioram + addr - IORAM_BASE_PHYS, data, len);
 334         return 0;
 335 }
 336
 337 static struct kvm_callbacks test_callbacks = {
 338         .inb         = test_inb,
 339         .inw         = test_inw,
 340         .inl         = test_inl,
 341         .outb        = test_outb,
 342         .outw        = test_outw,
 343         .outl        = test_outl,
 344         .mmio_read   = test_mem_read,
 345         .mmio_write  = test_mem_write,
 346         .debug       = test_debug,
 347         .halt        = test_halt,
 348         .io_window = test_io_window,
 349         .try_push_interrupts = test_try_push_interrupts,
 350         .post_kvm_run = test_post_kvm_run,
 351         .pre_kvm_run = test_pre_kvm_run,
 352 };
 353
 354 static void load_file(void *mem, const char *fname)
 355 {
 356         int r;
 357         int fd;
 358
 359         fd = open(fname, O_RDONLY);
 360         if (fd == -1) {
 361                 perror("open");
 362                 exit(1);
 363         }
 364         while ((r = read(fd, mem, 4096)) != -1 && r != 0)
 365                 mem += r;
 366         if (r == -1) {
 367                 perror("read");
 368                 exit(1);
 369         }
 370 }
 371
 372 static void enter_32(kvm_context_t kvm)
 373 {
 374         struct kvm_regs regs = {
 375                 .rsp = 0x80000,  /* 512KB */
 376                 .rip = 0x100000, /* 1MB */
 377                 .rflags = 2,
 378         };
 379         struct kvm_sregs sregs = {
 380                 .cs = { 0, -1u,  8, 11, 1, 0, 1, 1, 0, 1, 0, 0 },
 381                 .ds = { 0, -1u, 16,  3, 1, 0, 1, 1, 0, 1, 0, 0 },
 382                 .es = { 0, -1u, 16,  3, 1, 0, 1, 1, 0, 1, 0, 0 },
 383                 .fs = { 0, -1u, 16,  3, 1, 0, 1, 1, 0, 1, 0, 0 },
 384                 .gs = { 0, -1u, 16,  3, 1, 0, 1, 1, 0, 1, 0, 0 },
 385                 .ss = { 0, -1u, 16,  3, 1, 0, 1, 1, 0, 1, 0, 0 },
 386
 387                 .tr = { 0, 10000, 24, 11, 1, 0, 0, 0, 0, 0, 0, 0 },
 388                 .ldt = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1 },
 389                 .gdt = { 0, 0 },
 390                 .idt = { 0, 0 },
 391                 .cr0 = 0x37,
 392                 .cr3 = 0,
 393                 .cr4 = 0,
 394                 .efer = 0,
 395                 .apic_base = 0,
 396                 .interrupt_bitmap = { 0 },
 397         };
 398
 399         kvm_set_regs(kvm, 0, &regs);
 400         kvm_set_sregs(kvm, 0, &sregs);
 401 }
 402
 403 static void init_vcpu(int n)
 404 {
 405         sigemptyset(&ipi_sigmask);
 406         sigaddset(&ipi_sigmask, IPI_SIGNAL);
 407         sigprocmask(SIG_UNBLOCK, &ipi_sigmask, NULL);
 408         sigprocmask(SIG_BLOCK, &ipi_sigmask, &kernel_sigmask);
 409         vcpus[n].tid = gettid();
 410         vcpu = n;
 411         kvm_set_signal_mask(kvm, n, &kernel_sigmask);
 412         sem_post(&init_sem);
 413 }
 414
 415 static void *do_create_vcpu(void *_n)
 416 {
 417         int n = (long)_n;
 418         struct kvm_regs regs;
 419
 420         kvm_create_vcpu(kvm, n);
 421         init_vcpu(n);
 422         sem_wait(&vcpus[n].sipi_sem);
 423         kvm_get_regs(kvm, n, &regs);
 424         regs.rip = apic_sipi_addr;
 425         kvm_set_regs(kvm, n, &regs);
 426         kvm_run(kvm, n);
 427         return NULL;
 428 }
 429
 430 static void start_vcpu(int n)
 431 {
 432         pthread_t thread;
 433
 434         sem_init(&vcpus[n].sipi_sem, 0, 0);
 435         pthread_create(&thread, NULL, do_create_vcpu, (void *)(long)n);
 436 }
 437
 438 static void usage(const char *progname)
 439 {
 440         fprintf(stderr,
 441 "Usage: %s [OPTIONS] [bootstrap] flatfile\n"
 442 "KVM test harness.\n"
 443 "\n"
 444 "  -s, --smp=NUM          create a VM with NUM virtual CPUs\n"
 445 "  -p, --protected-mode   start VM in protected mode\n"
 446 "  -m, --memory=NUM[GMKB] allocate NUM memory for virtual machine.  A suffix\n"
 447 "                         can be used to change the unit (default: `M')\n"
 448 "  -h, --help             display this help screen and exit\n"
 449 "\n"
 450 "Report bugs to <kvm-devel@lists.sourceforge.net>.\n"
 451                 , progname);
 452 }
 453
 454 static void sig_ignore(int sig)
 455 {
 456         write(1, "boo\n", 4);
 457 }
 458
 459 int main(int argc, char **argv)
 460 {
 461         void *vm_mem;
 462         int i;
 463         const char *sopts = "s:phm:";
 464         struct option lopts[] = {
 465                 { "smp", 1, 0, 's' },
 466                 { "protected-mode", 0, 0, 'p' },
 467                 { "memory", 1, 0, 'm' },
 468                 { "help", 0, 0, 'h' },
 469                 { 0 },
 470         };
 471         int opt_ind, ch;
 472         bool enter_protected_mode = false;
 473         int nb_args;
 474         char *endptr;
 475
 476         while ((ch = getopt_long(argc, argv, sopts, lopts, &opt_ind)) != -1) {
 477                 switch (ch) {
 478                 case 's':
 479                         ncpus = atoi(optarg);
 480                         break;
 481                 case 'p':
 482                         enter_protected_mode = true;
 483                         break;
 484                 case 'm':
 485                         memory_size = strtoull(optarg, &endptr, 0);
 486                         switch (*endptr) {
 487                         case 'G': case 'g':
 488                                 memory_size <<= 30;
 489                                 break;
 490                         case '\0':
 491                         case 'M': case 'm':
 492                                 memory_size <<= 20;
 493                                 break;
 494                         case 'K': case 'k':
 495                                 memory_size <<= 10;
 496                                 break;
 497                         default:
 498                                 fprintf(stderr,
 499                                         "Unrecongized memory suffix: %c\n",
 500                                         *endptr);
 501                                 exit(1);
 502                         }
 503                         if (memory_size == 0) {
 504                                 fprintf(stderr,
 505                                         "Invalid memory size: 0\n");
 506                                 exit(1);
 507                         }
 508                         break;
 509                 case 'h':
 510                         usage(argv[0]);
 511                         exit(0);
 512                 case '?':
 513                 default:
 514                         fprintf(stderr,
 515                                 "Try `%s --help' for more information.\n",
 516                                 argv[0]);
 517                         exit(1);
 518                 }
 519         }
 520
 521         nb_args = argc - optind;
 522         if (nb_args < 1 || nb_args > 2) {
 523                 fprintf(stderr,
 524                         "Incorrect number of arguments.\n"
 525                         "Try `%s --help' for more information.\n",
 526                         argv[0]);
 527                 exit(1);
 528         }
 529
 530         signal(IPI_SIGNAL, sig_ignore);
 531
 532         vcpus = calloc(ncpus, sizeof *vcpus);
 533         if (!vcpus) {
 534                 fprintf(stderr, "calloc failed\n");
 535                 return 1;
 536         }
 537
 538         kvm = kvm_init(&test_callbacks, 0);
 539         if (!kvm) {
 540                 fprintf(stderr, "kvm_init failed\n");
 541                 return 1;
 542         }
 543         if (kvm_create(kvm, memory_size, &vm_mem) < 0) {
 544                 kvm_finalize(kvm);
 545                 fprintf(stderr, "kvm_create failed\n");
 546                 return 1;
 547         }
 548
 549         vm_mem = kvm_create_phys_mem(kvm, 0, memory_size, 0, 1);
 550
 551         if (enter_protected_mode)
 552                 enter_32(kvm);
 553         else
 554                 load_file(vm_mem + 0xf0000, argv[optind]);
 555
 556         if (nb_args > 1)
 557                 load_file(vm_mem + 0x100000, argv[optind + 1]);
 558
 559         apic_init();
 560         misc_init();
 561
 562         sem_init(&init_sem, 0, 0);
 563         init_vcpu(0);
 564         for (i = 1; i < ncpus; ++i)
 565                 start_vcpu(i);
 566         for (i = 0; i < ncpus; ++i)
 567                 sem_wait(&init_sem);
 568
 569         kvm_run(kvm, 0);
 570
 571         return 0;
 572 }