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