kvm-all.c

   1 /*
   2  * QEMU KVM support
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  10  * See the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include <sys/types.h>
  15 #include <sys/ioctl.h>
  16 #include <sys/mman.h>
  17 #include <stdarg.h>
  18
  19 #include <linux/kvm.h>
  20
  21 #include "qemu-common.h"
  22 #include "sysemu.h"
  23 #include "kvm.h"
  24
  25 //#define DEBUG_KVM
  26
  27 #ifdef DEBUG_KVM
  28 #define dprintf(fmt, ...) \
  29     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  30 #else
  31 #define dprintf(fmt, ...) \
  32     do { } while (0)
  33 #endif
  34
  35 typedef struct KVMSlot
  36 {
  37     target_phys_addr_t start_addr;
  38     ram_addr_t memory_size;
  39     ram_addr_t phys_offset;
  40     int slot;
  41     int flags;
  42 } KVMSlot;
  43
  44 int kvm_allowed = 0;
  45
  46 struct KVMState
  47 {
  48     KVMSlot slots[32];
  49     int fd;
  50     int vmfd;
  51 };
  52
  53 static KVMState *kvm_state;
  54
  55 static KVMSlot *kvm_alloc_slot(KVMState *s)
  56 {
  57     int i;
  58
  59     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  60         /* KVM private memory slots */
  61         if (i >= 8 && i < 12)
  62             continue;
  63         if (s->slots[i].memory_size == 0)
  64             return &s->slots[i];
  65     }
  66
  67     return NULL;
  68 }
  69
  70 static KVMSlot *kvm_lookup_slot(KVMState *s, target_phys_addr_t start_addr)
  71 {
  72     int i;
  73
  74     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  75         KVMSlot *mem = &s->slots[i];
  76
  77         if (start_addr >= mem->start_addr &&
  78             start_addr < (mem->start_addr + mem->memory_size))
  79             return mem;
  80     }
  81
  82     return NULL;
  83 }
  84
  85 int kvm_init_vcpu(CPUState *env)
  86 {
  87     KVMState *s = kvm_state;
  88     long mmap_size;
  89     int ret;
  90
  91     dprintf("kvm_init_vcpu\n");
  92
  93     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
  94     if (ret < 0) {
  95         dprintf("kvm_create_vcpu failed\n");
  96         goto err;
  97     }
  98
  99     env->kvm_fd = ret;
 100     env->kvm_state = s;
 101
 102     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 103     if (mmap_size < 0) {
 104         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 105         goto err;
 106     }
 107
 108     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 109                         env->kvm_fd, 0);
 110     if (env->kvm_run == MAP_FAILED) {
 111         ret = -errno;
 112         dprintf("mmap'ing vcpu state failed\n");
 113         goto err;
 114     }
 115
 116     ret = kvm_arch_init_vcpu(env);
 117
 118 err:
 119     return ret;
 120 }
 121
 122 int kvm_init(int smp_cpus)
 123 {
 124     KVMState *s;
 125     int ret;
 126     int i;
 127
 128     if (smp_cpus > 1)
 129         return -EINVAL;
 130
 131     s = qemu_mallocz(sizeof(KVMState));
 132     if (s == NULL)
 133         return -ENOMEM;
 134
 135     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 136         s->slots[i].slot = i;
 137
 138     s->vmfd = -1;
 139     s->fd = open("/dev/kvm", O_RDWR);
 140     if (s->fd == -1) {
 141         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 142         ret = -errno;
 143         goto err;
 144     }
 145
 146     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 147     if (ret < KVM_API_VERSION) {
 148         if (ret > 0)
 149             ret = -EINVAL;
 150         fprintf(stderr, "kvm version too old\n");
 151         goto err;
 152     }
 153
 154     if (ret > KVM_API_VERSION) {
 155         ret = -EINVAL;
 156         fprintf(stderr, "kvm version not supported\n");
 157         goto err;
 158     }
 159
 160     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 161     if (s->vmfd < 0)
 162         goto err;
 163
 164     /* initially, KVM allocated its own memory and we had to jump through
 165      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 166      * just use a user allocated buffer so we can use phys_ram_base
 167      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 168      */
 169     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_USER_MEMORY);
 170     if (ret <= 0) {
 171         if (ret == 0)
 172             ret = -EINVAL;
 173         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n");
 174         goto err;
 175     }
 176
 177     ret = kvm_arch_init(s, smp_cpus);
 178     if (ret < 0)
 179         goto err;
 180
 181     kvm_state = s;
 182
 183     return 0;
 184
 185 err:
 186     if (s) {
 187         if (s->vmfd != -1)
 188             close(s->vmfd);
 189         if (s->fd != -1)
 190             close(s->fd);
 191     }
 192     qemu_free(s);
 193
 194     return ret;
 195 }
 196
 197 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 198                          int direction, int size, uint32_t count)
 199 {
 200     int i;
 201     uint8_t *ptr = data;
 202
 203     for (i = 0; i < count; i++) {
 204         if (direction == KVM_EXIT_IO_IN) {
 205             switch (size) {
 206             case 1:
 207                 stb_p(ptr, cpu_inb(env, port));
 208                 break;
 209             case 2:
 210                 stw_p(ptr, cpu_inw(env, port));
 211                 break;
 212             case 4:
 213                 stl_p(ptr, cpu_inl(env, port));
 214                 break;
 215             }
 216         } else {
 217             switch (size) {
 218             case 1:
 219                 cpu_outb(env, port, ldub_p(ptr));
 220                 break;
 221             case 2:
 222                 cpu_outw(env, port, lduw_p(ptr));
 223                 break;
 224             case 4:
 225                 cpu_outl(env, port, ldl_p(ptr));
 226                 break;
 227             }
 228         }
 229
 230         ptr += size;
 231     }
 232
 233     return 1;
 234 }
 235
 236 int kvm_cpu_exec(CPUState *env)
 237 {
 238     struct kvm_run *run = env->kvm_run;
 239     int ret;
 240
 241     dprintf("kvm_cpu_exec()\n");
 242
 243     do {
 244         kvm_arch_pre_run(env, run);
 245
 246         if ((env->interrupt_request & CPU_INTERRUPT_EXIT)) {
 247             dprintf("interrupt exit requested\n");
 248             ret = 0;
 249             break;
 250         }
 251
 252         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 253         kvm_arch_post_run(env, run);
 254
 255         if (ret == -EINTR || ret == -EAGAIN) {
 256             dprintf("io window exit\n");
 257             ret = 0;
 258             break;
 259         }
 260
 261         if (ret < 0) {
 262             dprintf("kvm run failed %s\n", strerror(-ret));
 263             abort();
 264         }
 265
 266         ret = 0; /* exit loop */
 267         switch (run->exit_reason) {
 268         case KVM_EXIT_IO:
 269             dprintf("handle_io\n");
 270             ret = kvm_handle_io(env, run->io.port,
 271                                 (uint8_t *)run + run->io.data_offset,
 272                                 run->io.direction,
 273                                 run->io.size,
 274                                 run->io.count);
 275             break;
 276         case KVM_EXIT_MMIO:
 277             dprintf("handle_mmio\n");
 278             cpu_physical_memory_rw(run->mmio.phys_addr,
 279                                    run->mmio.data,
 280                                    run->mmio.len,
 281                                    run->mmio.is_write);
 282             ret = 1;
 283             break;
 284         case KVM_EXIT_IRQ_WINDOW_OPEN:
 285             dprintf("irq_window_open\n");
 286             break;
 287         case KVM_EXIT_SHUTDOWN:
 288             dprintf("shutdown\n");
 289             qemu_system_reset_request();
 290             ret = 1;
 291             break;
 292         case KVM_EXIT_UNKNOWN:
 293             dprintf("kvm_exit_unknown\n");
 294             break;
 295         case KVM_EXIT_FAIL_ENTRY:
 296             dprintf("kvm_exit_fail_entry\n");
 297             break;
 298         case KVM_EXIT_EXCEPTION:
 299             dprintf("kvm_exit_exception\n");
 300             break;
 301         case KVM_EXIT_DEBUG:
 302             dprintf("kvm_exit_debug\n");
 303             break;
 304         default:
 305             dprintf("kvm_arch_handle_exit\n");
 306             ret = kvm_arch_handle_exit(env, run);
 307             break;
 308         }
 309     } while (ret > 0);
 310
 311     if ((env->interrupt_request & CPU_INTERRUPT_EXIT)) {
 312         env->interrupt_request &= ~CPU_INTERRUPT_EXIT;
 313         env->exception_index = EXCP_INTERRUPT;
 314     }
 315
 316     return ret;
 317 }
 318
 319 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
 320 {
 321     struct kvm_userspace_memory_region mem;
 322
 323     mem.slot = slot->slot;
 324     mem.guest_phys_addr = slot->start_addr;
 325     mem.memory_size = slot->memory_size;
 326     mem.userspace_addr = (unsigned long)phys_ram_base + slot->phys_offset;
 327     mem.flags = slot->flags;
 328
 329     return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
 330 }
 331
 332 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 333                       ram_addr_t size,
 334                       ram_addr_t phys_offset)
 335 {
 336     KVMState *s = kvm_state;
 337     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 338     KVMSlot *mem;
 339
 340     /* KVM does not support read-only slots */
 341     phys_offset &= ~IO_MEM_ROM;
 342
 343     mem = kvm_lookup_slot(s, start_addr);
 344     if (mem) {
 345         if ((flags == IO_MEM_UNASSIGNED) || (flags >= TLB_MMIO)) {
 346             mem->memory_size = 0;
 347             mem->start_addr = start_addr;
 348             mem->phys_offset = 0;
 349             mem->flags = 0;
 350
 351             kvm_set_user_memory_region(s, mem);
 352         } else if (start_addr >= mem->start_addr &&
 353                    (start_addr + size) <= (mem->start_addr +
 354                                            mem->memory_size)) {
 355             KVMSlot slot;
 356             target_phys_addr_t mem_start;
 357             ram_addr_t mem_size, mem_offset;
 358
 359             /* Not splitting */
 360             if ((phys_offset - (start_addr - mem->start_addr)) ==
 361                 mem->phys_offset)
 362                 return;
 363
 364             /* unregister whole slot */
 365             memcpy(&slot, mem, sizeof(slot));
 366             mem->memory_size = 0;
 367             kvm_set_user_memory_region(s, mem);
 368
 369             /* register prefix slot */
 370             mem_start = slot.start_addr;
 371             mem_size = start_addr - slot.start_addr;
 372             mem_offset = slot.phys_offset;
 373             if (mem_size)
 374                 kvm_set_phys_mem(mem_start, mem_size, mem_offset);
 375
 376             /* register new slot */
 377             kvm_set_phys_mem(start_addr, size, phys_offset);
 378
 379             /* register suffix slot */
 380             mem_start = start_addr + size;
 381             mem_offset += mem_size + size;
 382             mem_size = slot.memory_size - mem_size - size;
 383             if (mem_size)
 384                 kvm_set_phys_mem(mem_start, mem_size, mem_offset);
 385
 386             return;
 387         } else {
 388             printf("Registering overlapping slot\n");
 389             abort();
 390         }
 391     }
 392     /* KVM does not need to know about this memory */
 393     if (flags >= IO_MEM_UNASSIGNED)
 394         return;
 395
 396     mem = kvm_alloc_slot(s);
 397     mem->memory_size = size;
 398     mem->start_addr = start_addr;
 399     mem->phys_offset = phys_offset;
 400     mem->flags = 0;
 401
 402     kvm_set_user_memory_region(s, mem);
 403     /* FIXME deal with errors */
 404 }
 405
 406 int kvm_ioctl(KVMState *s, int type, ...)
 407 {
 408     int ret;
 409     void *arg;
 410     va_list ap;
 411
 412     va_start(ap, type);
 413     arg = va_arg(ap, void *);
 414     va_end(ap);
 415
 416     ret = ioctl(s->fd, type, arg);
 417     if (ret == -1)
 418         ret = -errno;
 419
 420     return ret;
 421 }
 422
 423 int kvm_vm_ioctl(KVMState *s, int type, ...)
 424 {
 425     int ret;
 426     void *arg;
 427     va_list ap;
 428
 429     va_start(ap, type);
 430     arg = va_arg(ap, void *);
 431     va_end(ap);
 432
 433     ret = ioctl(s->vmfd, type, arg);
 434     if (ret == -1)
 435         ret = -errno;
 436
 437     return ret;
 438 }
 439
 440 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 441 {
 442     int ret;
 443     void *arg;
 444     va_list ap;
 445
 446     va_start(ap, type);
 447     arg = va_arg(ap, void *);
 448     va_end(ap);
 449
 450     ret = ioctl(env->kvm_fd, type, arg);
 451     if (ret == -1)
 452         ret = -errno;
 453
 454     return ret;
 455 }