target/arm/kvm.c

   1 /*
   2  * ARM implementation of KVM hooks
   3  *
   4  * Copyright Christoffer Dall 2009-2010
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   7  * See the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include "qemu/osdep.h"
  12 #include <sys/ioctl.h>
  13
  14 #include <linux/kvm.h>
  15
  16 #include "qemu-common.h"
  17 #include "qemu/timer.h"
  18 #include "qemu/error-report.h"
  19 #include "qemu/main-loop.h"
  20 #include "sysemu/sysemu.h"
  21 #include "sysemu/kvm.h"
  22 #include "sysemu/kvm_int.h"
  23 #include "kvm_arm.h"
  24 #include "cpu.h"
  25 #include "trace.h"
  26 #include "internals.h"
  27 #include "hw/pci/pci.h"
  28 #include "exec/memattrs.h"
  29 #include "exec/address-spaces.h"
  30 #include "hw/boards.h"
  31 #include "hw/irq.h"
  32 #include "qemu/log.h"
  33
  34 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
  35     KVM_CAP_LAST_INFO
  36 };
  37
  38 static bool cap_has_mp_state;
  39 static bool cap_has_inject_serror_esr;
  40
  41 static ARMHostCPUFeatures arm_host_cpu_features;
  42
  43 int kvm_arm_vcpu_init(CPUState *cs)
  44 {
  45     ARMCPU *cpu = ARM_CPU(cs);
  46     struct kvm_vcpu_init init;
  47
  48     init.target = cpu->kvm_target;
  49     memcpy(init.features, cpu->kvm_init_features, sizeof(init.features));
  50
  51     return kvm_vcpu_ioctl(cs, KVM_ARM_VCPU_INIT, &init);
  52 }
  53
  54 void kvm_arm_init_serror_injection(CPUState *cs)
  55 {
  56     cap_has_inject_serror_esr = kvm_check_extension(cs->kvm_state,
  57                                     KVM_CAP_ARM_INJECT_SERROR_ESR);
  58 }
  59
  60 bool kvm_arm_create_scratch_host_vcpu(const uint32_t *cpus_to_try,
  61                                       int *fdarray,
  62                                       struct kvm_vcpu_init *init)
  63 {
  64     int ret, kvmfd = -1, vmfd = -1, cpufd = -1;
  65
  66     kvmfd = qemu_open("/dev/kvm", O_RDWR);
  67     if (kvmfd < 0) {
  68         goto err;
  69     }
  70     vmfd = ioctl(kvmfd, KVM_CREATE_VM, 0);
  71     if (vmfd < 0) {
  72         goto err;
  73     }
  74     cpufd = ioctl(vmfd, KVM_CREATE_VCPU, 0);
  75     if (cpufd < 0) {
  76         goto err;
  77     }
  78
  79     if (!init) {
  80         /* Caller doesn't want the VCPU to be initialized, so skip it */
  81         goto finish;
  82     }
  83
  84     ret = ioctl(vmfd, KVM_ARM_PREFERRED_TARGET, init);
  85     if (ret >= 0) {
  86         ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init);
  87         if (ret < 0) {
  88             goto err;
  89         }
  90     } else if (cpus_to_try) {
  91         /* Old kernel which doesn't know about the
  92          * PREFERRED_TARGET ioctl: we know it will only support
  93          * creating one kind of guest CPU which is its preferred
  94          * CPU type.
  95          */
  96         while (*cpus_to_try != QEMU_KVM_ARM_TARGET_NONE) {
  97             init->target = *cpus_to_try++;
  98             memset(init->features, 0, sizeof(init->features));
  99             ret = ioctl(cpufd, KVM_ARM_VCPU_INIT, init);
 100             if (ret >= 0) {
 101                 break;
 102             }
 103         }
 104         if (ret < 0) {
 105             goto err;
 106         }
 107     } else {
 108         /* Treat a NULL cpus_to_try argument the same as an empty
 109          * list, which means we will fail the call since this must
 110          * be an old kernel which doesn't support PREFERRED_TARGET.
 111          */
 112         goto err;
 113     }
 114
 115 finish:
 116     fdarray[0] = kvmfd;
 117     fdarray[1] = vmfd;
 118     fdarray[2] = cpufd;
 119
 120     return true;
 121
 122 err:
 123     if (cpufd >= 0) {
 124         close(cpufd);
 125     }
 126     if (vmfd >= 0) {
 127         close(vmfd);
 128     }
 129     if (kvmfd >= 0) {
 130         close(kvmfd);
 131     }
 132
 133     return false;
 134 }
 135
 136 void kvm_arm_destroy_scratch_host_vcpu(int *fdarray)
 137 {
 138     int i;
 139
 140     for (i = 2; i >= 0; i--) {
 141         close(fdarray[i]);
 142     }
 143 }
 144
 145 void kvm_arm_set_cpu_features_from_host(ARMCPU *cpu)
 146 {
 147     CPUARMState *env = &cpu->env;
 148
 149     if (!arm_host_cpu_features.dtb_compatible) {
 150         if (!kvm_enabled() ||
 151             !kvm_arm_get_host_cpu_features(&arm_host_cpu_features)) {
 152             /* We can't report this error yet, so flag that we need to
 153              * in arm_cpu_realizefn().
 154              */
 155             cpu->kvm_target = QEMU_KVM_ARM_TARGET_NONE;
 156             cpu->host_cpu_probe_failed = true;
 157             return;
 158         }
 159     }
 160
 161     cpu->kvm_target = arm_host_cpu_features.target;
 162     cpu->dtb_compatible = arm_host_cpu_features.dtb_compatible;
 163     cpu->isar = arm_host_cpu_features.isar;
 164     env->features = arm_host_cpu_features.features;
 165 }
 166
 167 bool kvm_arm_pmu_supported(CPUState *cpu)
 168 {
 169     KVMState *s = KVM_STATE(current_machine->accelerator);
 170
 171     return kvm_check_extension(s, KVM_CAP_ARM_PMU_V3);
 172 }
 173
 174 int kvm_arm_get_max_vm_ipa_size(MachineState *ms)
 175 {
 176     KVMState *s = KVM_STATE(ms->accelerator);
 177     int ret;
 178
 179     ret = kvm_check_extension(s, KVM_CAP_ARM_VM_IPA_SIZE);
 180     return ret > 0 ? ret : 40;
 181 }
 182
 183 int kvm_arch_init(MachineState *ms, KVMState *s)
 184 {
 185     /* For ARM interrupt delivery is always asynchronous,
 186      * whether we are using an in-kernel VGIC or not.
 187      */
 188     kvm_async_interrupts_allowed = true;
 189
 190     /*
 191      * PSCI wakes up secondary cores, so we always need to
 192      * have vCPUs waiting in kernel space
 193      */
 194     kvm_halt_in_kernel_allowed = true;
 195
 196     cap_has_mp_state = kvm_check_extension(s, KVM_CAP_MP_STATE);
 197
 198     return 0;
 199 }
 200
 201 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 202 {
 203     return cpu->cpu_index;
 204 }
 205
 206 /* We track all the KVM devices which need their memory addresses
 207  * passing to the kernel in a list of these structures.
 208  * When board init is complete we run through the list and
 209  * tell the kernel the base addresses of the memory regions.
 210  * We use a MemoryListener to track mapping and unmapping of
 211  * the regions during board creation, so the board models don't
 212  * need to do anything special for the KVM case.
 213  *
 214  * Sometimes the address must be OR'ed with some other fields
 215  * (for example for KVM_VGIC_V3_ADDR_TYPE_REDIST_REGION).
 216  * @kda_addr_ormask aims at storing the value of those fields.
 217  */
 218 typedef struct KVMDevice {
 219     struct kvm_arm_device_addr kda;
 220     struct kvm_device_attr kdattr;
 221     uint64_t kda_addr_ormask;
 222     MemoryRegion *mr;
 223     QSLIST_ENTRY(KVMDevice) entries;
 224     int dev_fd;
 225 } KVMDevice;
 226
 227 static QSLIST_HEAD(, KVMDevice) kvm_devices_head;
 228
 229 static void kvm_arm_devlistener_add(MemoryListener *listener,
 230                                     MemoryRegionSection *section)
 231 {
 232     KVMDevice *kd;
 233
 234     QSLIST_FOREACH(kd, &kvm_devices_head, entries) {
 235         if (section->mr == kd->mr) {
 236             kd->kda.addr = section->offset_within_address_space;
 237         }
 238     }
 239 }
 240
 241 static void kvm_arm_devlistener_del(MemoryListener *listener,
 242                                     MemoryRegionSection *section)
 243 {
 244     KVMDevice *kd;
 245
 246     QSLIST_FOREACH(kd, &kvm_devices_head, entries) {
 247         if (section->mr == kd->mr) {
 248             kd->kda.addr = -1;
 249         }
 250     }
 251 }
 252
 253 static MemoryListener devlistener = {
 254     .region_add = kvm_arm_devlistener_add,
 255     .region_del = kvm_arm_devlistener_del,
 256 };
 257
 258 static void kvm_arm_set_device_addr(KVMDevice *kd)
 259 {
 260     struct kvm_device_attr *attr = &kd->kdattr;
 261     int ret;
 262
 263     /* If the device control API is available and we have a device fd on the
 264      * KVMDevice struct, let's use the newer API
 265      */
 266     if (kd->dev_fd >= 0) {
 267         uint64_t addr = kd->kda.addr;
 268
 269         addr |= kd->kda_addr_ormask;
 270         attr->addr = (uintptr_t)&addr;
 271         ret = kvm_device_ioctl(kd->dev_fd, KVM_SET_DEVICE_ATTR, attr);
 272     } else {
 273         ret = kvm_vm_ioctl(kvm_state, KVM_ARM_SET_DEVICE_ADDR, &kd->kda);
 274     }
 275
 276     if (ret < 0) {
 277         fprintf(stderr, "Failed to set device address: %s\n",
 278                 strerror(-ret));
 279         abort();
 280     }
 281 }
 282
 283 static void kvm_arm_machine_init_done(Notifier *notifier, void *data)
 284 {
 285     KVMDevice *kd, *tkd;
 286
 287     QSLIST_FOREACH_SAFE(kd, &kvm_devices_head, entries, tkd) {
 288         if (kd->kda.addr != -1) {
 289             kvm_arm_set_device_addr(kd);
 290         }
 291         memory_region_unref(kd->mr);
 292         QSLIST_REMOVE_HEAD(&kvm_devices_head, entries);
 293         g_free(kd);
 294     }
 295     memory_listener_unregister(&devlistener);
 296 }
 297
 298 static Notifier notify = {
 299     .notify = kvm_arm_machine_init_done,
 300 };
 301
 302 void kvm_arm_register_device(MemoryRegion *mr, uint64_t devid, uint64_t group,
 303                              uint64_t attr, int dev_fd, uint64_t addr_ormask)
 304 {
 305     KVMDevice *kd;
 306
 307     if (!kvm_irqchip_in_kernel()) {
 308         return;
 309     }
 310
 311     if (QSLIST_EMPTY(&kvm_devices_head)) {
 312         memory_listener_register(&devlistener, &address_space_memory);
 313         qemu_add_machine_init_done_notifier(&notify);
 314     }
 315     kd = g_new0(KVMDevice, 1);
 316     kd->mr = mr;
 317     kd->kda.id = devid;
 318     kd->kda.addr = -1;
 319     kd->kdattr.flags = 0;
 320     kd->kdattr.group = group;
 321     kd->kdattr.attr = attr;
 322     kd->dev_fd = dev_fd;
 323     kd->kda_addr_ormask = addr_ormask;
 324     QSLIST_INSERT_HEAD(&kvm_devices_head, kd, entries);
 325     memory_region_ref(kd->mr);
 326 }
 327
 328 static int compare_u64(const void *a, const void *b)
 329 {
 330     if (*(uint64_t *)a > *(uint64_t *)b) {
 331         return 1;
 332     }
 333     if (*(uint64_t *)a < *(uint64_t *)b) {
 334         return -1;
 335     }
 336     return 0;
 337 }
 338
 339 /* Initialize the ARMCPU cpreg list according to the kernel's
 340  * definition of what CPU registers it knows about (and throw away
 341  * the previous TCG-created cpreg list).
 342  */
 343 int kvm_arm_init_cpreg_list(ARMCPU *cpu)
 344 {
 345     struct kvm_reg_list rl;
 346     struct kvm_reg_list *rlp;
 347     int i, ret, arraylen;
 348     CPUState *cs = CPU(cpu);
 349
 350     rl.n = 0;
 351     ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, &rl);
 352     if (ret != -E2BIG) {
 353         return ret;
 354     }
 355     rlp = g_malloc(sizeof(struct kvm_reg_list) + rl.n * sizeof(uint64_t));
 356     rlp->n = rl.n;
 357     ret = kvm_vcpu_ioctl(cs, KVM_GET_REG_LIST, rlp);
 358     if (ret) {
 359         goto out;
 360     }
 361     /* Sort the list we get back from the kernel, since cpreg_tuples
 362      * must be in strictly ascending order.
 363      */
 364     qsort(&rlp->reg, rlp->n, sizeof(rlp->reg[0]), compare_u64);
 365
 366     for (i = 0, arraylen = 0; i < rlp->n; i++) {
 367         if (!kvm_arm_reg_syncs_via_cpreg_list(rlp->reg[i])) {
 368             continue;
 369         }
 370         switch (rlp->reg[i] & KVM_REG_SIZE_MASK) {
 371         case KVM_REG_SIZE_U32:
 372         case KVM_REG_SIZE_U64:
 373             break;
 374         default:
 375             fprintf(stderr, "Can't handle size of register in kernel list\n");
 376             ret = -EINVAL;
 377             goto out;
 378         }
 379
 380         arraylen++;
 381     }
 382
 383     cpu->cpreg_indexes = g_renew(uint64_t, cpu->cpreg_indexes, arraylen);
 384     cpu->cpreg_values = g_renew(uint64_t, cpu->cpreg_values, arraylen);
 385     cpu->cpreg_vmstate_indexes = g_renew(uint64_t, cpu->cpreg_vmstate_indexes,
 386                                          arraylen);
 387     cpu->cpreg_vmstate_values = g_renew(uint64_t, cpu->cpreg_vmstate_values,
 388                                         arraylen);
 389     cpu->cpreg_array_len = arraylen;
 390     cpu->cpreg_vmstate_array_len = arraylen;
 391
 392     for (i = 0, arraylen = 0; i < rlp->n; i++) {
 393         uint64_t regidx = rlp->reg[i];
 394         if (!kvm_arm_reg_syncs_via_cpreg_list(regidx)) {
 395             continue;
 396         }
 397         cpu->cpreg_indexes[arraylen] = regidx;
 398         arraylen++;
 399     }
 400     assert(cpu->cpreg_array_len == arraylen);
 401
 402     if (!write_kvmstate_to_list(cpu)) {
 403         /* Shouldn't happen unless kernel is inconsistent about
 404          * what registers exist.
 405          */
 406         fprintf(stderr, "Initial read of kernel register state failed\n");
 407         ret = -EINVAL;
 408         goto out;
 409     }
 410
 411 out:
 412     g_free(rlp);
 413     return ret;
 414 }
 415
 416 bool write_kvmstate_to_list(ARMCPU *cpu)
 417 {
 418     CPUState *cs = CPU(cpu);
 419     int i;
 420     bool ok = true;
 421
 422     for (i = 0; i < cpu->cpreg_array_len; i++) {
 423         struct kvm_one_reg r;
 424         uint64_t regidx = cpu->cpreg_indexes[i];
 425         uint32_t v32;
 426         int ret;
 427
 428         r.id = regidx;
 429
 430         switch (regidx & KVM_REG_SIZE_MASK) {
 431         case KVM_REG_SIZE_U32:
 432             r.addr = (uintptr_t)&v32;
 433             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
 434             if (!ret) {
 435                 cpu->cpreg_values[i] = v32;
 436             }
 437             break;
 438         case KVM_REG_SIZE_U64:
 439             r.addr = (uintptr_t)(cpu->cpreg_values + i);
 440             ret = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &r);
 441             break;
 442         default:
 443             abort();
 444         }
 445         if (ret) {
 446             ok = false;
 447         }
 448     }
 449     return ok;
 450 }
 451
 452 bool write_list_to_kvmstate(ARMCPU *cpu, int level)
 453 {
 454     CPUState *cs = CPU(cpu);
 455     int i;
 456     bool ok = true;
 457
 458     for (i = 0; i < cpu->cpreg_array_len; i++) {
 459         struct kvm_one_reg r;
 460         uint64_t regidx = cpu->cpreg_indexes[i];
 461         uint32_t v32;
 462         int ret;
 463
 464         if (kvm_arm_cpreg_level(regidx) > level) {
 465             continue;
 466         }
 467
 468         r.id = regidx;
 469         switch (regidx & KVM_REG_SIZE_MASK) {
 470         case KVM_REG_SIZE_U32:
 471             v32 = cpu->cpreg_values[i];
 472             r.addr = (uintptr_t)&v32;
 473             break;
 474         case KVM_REG_SIZE_U64:
 475             r.addr = (uintptr_t)(cpu->cpreg_values + i);
 476             break;
 477         default:
 478             abort();
 479         }
 480         ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &r);
 481         if (ret) {
 482             /* We might fail for "unknown register" and also for
 483              * "you tried to set a register which is constant with
 484              * a different value from what it actually contains".
 485              */
 486             ok = false;
 487         }
 488     }
 489     return ok;
 490 }
 491
 492 void kvm_arm_reset_vcpu(ARMCPU *cpu)
 493 {
 494     int ret;
 495
 496     /* Re-init VCPU so that all registers are set to
 497      * their respective reset values.
 498      */
 499     ret = kvm_arm_vcpu_init(CPU(cpu));
 500     if (ret < 0) {
 501         fprintf(stderr, "kvm_arm_vcpu_init failed: %s\n", strerror(-ret));
 502         abort();
 503     }
 504     if (!write_kvmstate_to_list(cpu)) {
 505         fprintf(stderr, "write_kvmstate_to_list failed\n");
 506         abort();
 507     }
 508     /*
 509      * Sync the reset values also into the CPUState. This is necessary
 510      * because the next thing we do will be a kvm_arch_put_registers()
 511      * which will update the list values from the CPUState before copying
 512      * the list values back to KVM. It's OK to ignore failure returns here
 513      * for the same reason we do so in kvm_arch_get_registers().
 514      */
 515     write_list_to_cpustate(cpu);
 516 }
 517
 518 /*
 519  * Update KVM's MP_STATE based on what QEMU thinks it is
 520  */
 521 int kvm_arm_sync_mpstate_to_kvm(ARMCPU *cpu)
 522 {
 523     if (cap_has_mp_state) {
 524         struct kvm_mp_state mp_state = {
 525             .mp_state = (cpu->power_state == PSCI_OFF) ?
 526             KVM_MP_STATE_STOPPED : KVM_MP_STATE_RUNNABLE
 527         };
 528         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
 529         if (ret) {
 530             fprintf(stderr, "%s: failed to set MP_STATE %d/%s\n",
 531                     __func__, ret, strerror(-ret));
 532             return -1;
 533         }
 534     }
 535
 536     return 0;
 537 }
 538
 539 /*
 540  * Sync the KVM MP_STATE into QEMU
 541  */
 542 int kvm_arm_sync_mpstate_to_qemu(ARMCPU *cpu)
 543 {
 544     if (cap_has_mp_state) {
 545         struct kvm_mp_state mp_state;
 546         int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MP_STATE, &mp_state);
 547         if (ret) {
 548             fprintf(stderr, "%s: failed to get MP_STATE %d/%s\n",
 549                     __func__, ret, strerror(-ret));
 550             abort();
 551         }
 552         cpu->power_state = (mp_state.mp_state == KVM_MP_STATE_STOPPED) ?
 553             PSCI_OFF : PSCI_ON;
 554     }
 555
 556     return 0;
 557 }
 558
 559 int kvm_put_vcpu_events(ARMCPU *cpu)
 560 {
 561     CPUARMState *env = &cpu->env;
 562     struct kvm_vcpu_events events;
 563     int ret;
 564
 565     if (!kvm_has_vcpu_events()) {
 566         return 0;
 567     }
 568
 569     memset(&events, 0, sizeof(events));
 570     events.exception.serror_pending = env->serror.pending;
 571
 572     /* Inject SError to guest with specified syndrome if host kernel
 573      * supports it, otherwise inject SError without syndrome.
 574      */
 575     if (cap_has_inject_serror_esr) {
 576         events.exception.serror_has_esr = env->serror.has_esr;
 577         events.exception.serror_esr = env->serror.esr;
 578     }
 579
 580     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_VCPU_EVENTS, &events);
 581     if (ret) {
 582         error_report("failed to put vcpu events");
 583     }
 584
 585     return ret;
 586 }
 587
 588 int kvm_get_vcpu_events(ARMCPU *cpu)
 589 {
 590     CPUARMState *env = &cpu->env;
 591     struct kvm_vcpu_events events;
 592     int ret;
 593
 594     if (!kvm_has_vcpu_events()) {
 595         return 0;
 596     }
 597
 598     memset(&events, 0, sizeof(events));
 599     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_VCPU_EVENTS, &events);
 600     if (ret) {
 601         error_report("failed to get vcpu events");
 602         return ret;
 603     }
 604
 605     env->serror.pending = events.exception.serror_pending;
 606     env->serror.has_esr = events.exception.serror_has_esr;
 607     env->serror.esr = events.exception.serror_esr;
 608
 609     return 0;
 610 }
 611
 612 void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run)
 613 {
 614 }
 615
 616 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 617 {
 618     ARMCPU *cpu;
 619     uint32_t switched_level;
 620
 621     if (kvm_irqchip_in_kernel()) {
 622         /*
 623          * We only need to sync timer states with user-space interrupt
 624          * controllers, so return early and save cycles if we don't.
 625          */
 626         return MEMTXATTRS_UNSPECIFIED;
 627     }
 628
 629     cpu = ARM_CPU(cs);
 630
 631     /* Synchronize our shadowed in-kernel device irq lines with the kvm ones */
 632     if (run->s.regs.device_irq_level != cpu->device_irq_level) {
 633         switched_level = cpu->device_irq_level ^ run->s.regs.device_irq_level;
 634
 635         qemu_mutex_lock_iothread();
 636
 637         if (switched_level & KVM_ARM_DEV_EL1_VTIMER) {
 638             qemu_set_irq(cpu->gt_timer_outputs[GTIMER_VIRT],
 639                          !!(run->s.regs.device_irq_level &
 640                             KVM_ARM_DEV_EL1_VTIMER));
 641             switched_level &= ~KVM_ARM_DEV_EL1_VTIMER;
 642         }
 643
 644         if (switched_level & KVM_ARM_DEV_EL1_PTIMER) {
 645             qemu_set_irq(cpu->gt_timer_outputs[GTIMER_PHYS],
 646                          !!(run->s.regs.device_irq_level &
 647                             KVM_ARM_DEV_EL1_PTIMER));
 648             switched_level &= ~KVM_ARM_DEV_EL1_PTIMER;
 649         }
 650
 651         if (switched_level & KVM_ARM_DEV_PMU) {
 652             qemu_set_irq(cpu->pmu_interrupt,
 653                          !!(run->s.regs.device_irq_level & KVM_ARM_DEV_PMU));
 654             switched_level &= ~KVM_ARM_DEV_PMU;
 655         }
 656
 657         if (switched_level) {
 658             qemu_log_mask(LOG_UNIMP, "%s: unhandled in-kernel device IRQ %x\n",
 659                           __func__, switched_level);
 660         }
 661
 662         /* We also mark unknown levels as processed to not waste cycles */
 663         cpu->device_irq_level = run->s.regs.device_irq_level;
 664         qemu_mutex_unlock_iothread();
 665     }
 666
 667     return MEMTXATTRS_UNSPECIFIED;
 668 }
 669
 670
 671 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
 672 {
 673     int ret = 0;
 674
 675     switch (run->exit_reason) {
 676     case KVM_EXIT_DEBUG:
 677         if (kvm_arm_handle_debug(cs, &run->debug.arch)) {
 678             ret = EXCP_DEBUG;
 679         } /* otherwise return to guest */
 680         break;
 681     default:
 682         qemu_log_mask(LOG_UNIMP, "%s: un-handled exit reason %d\n",
 683                       __func__, run->exit_reason);
 684         break;
 685     }
 686     return ret;
 687 }
 688
 689 bool kvm_arch_stop_on_emulation_error(CPUState *cs)
 690 {
 691     return true;
 692 }
 693
 694 int kvm_arch_process_async_events(CPUState *cs)
 695 {
 696     return 0;
 697 }
 698
 699 /* The #ifdef protections are until 32bit headers are imported and can
 700  * be removed once both 32 and 64 bit reach feature parity.
 701  */
 702 void kvm_arch_update_guest_debug(CPUState *cs, struct kvm_guest_debug *dbg)
 703 {
 704 #ifdef KVM_GUESTDBG_USE_SW_BP
 705     if (kvm_sw_breakpoints_active(cs)) {
 706         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
 707     }
 708 #endif
 709 #ifdef KVM_GUESTDBG_USE_HW
 710     if (kvm_arm_hw_debug_active(cs)) {
 711         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW;
 712         kvm_arm_copy_hw_debug_data(&dbg->arch);
 713     }
 714 #endif
 715 }
 716
 717 void kvm_arch_init_irq_routing(KVMState *s)
 718 {
 719 }
 720
 721 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
 722 {
 723      if (machine_kernel_irqchip_split(ms)) {
 724          perror("-machine kernel_irqchip=split is not supported on ARM.");
 725          exit(1);
 726     }
 727
 728     /* If we can create the VGIC using the newer device control API, we
 729      * let the device do this when it initializes itself, otherwise we
 730      * fall back to the old API */
 731     return kvm_check_extension(s, KVM_CAP_DEVICE_CTRL);
 732 }
 733
 734 int kvm_arm_vgic_probe(void)
 735 {
 736     if (kvm_create_device(kvm_state,
 737                           KVM_DEV_TYPE_ARM_VGIC_V3, true) == 0) {
 738         return 3;
 739     } else if (kvm_create_device(kvm_state,
 740                                  KVM_DEV_TYPE_ARM_VGIC_V2, true) == 0) {
 741         return 2;
 742     } else {
 743         return 0;
 744     }
 745 }
 746
 747 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
 748                              uint64_t address, uint32_t data, PCIDevice *dev)
 749 {
 750     AddressSpace *as = pci_device_iommu_address_space(dev);
 751     hwaddr xlat, len, doorbell_gpa;
 752     MemoryRegionSection mrs;
 753     MemoryRegion *mr;
 754     int ret = 1;
 755
 756     if (as == &address_space_memory) {
 757         return 0;
 758     }
 759
 760     /* MSI doorbell address is translated by an IOMMU */
 761
 762     rcu_read_lock();
 763     mr = address_space_translate(as, address, &xlat, &len, true,
 764                                  MEMTXATTRS_UNSPECIFIED);
 765     if (!mr) {
 766         goto unlock;
 767     }
 768     mrs = memory_region_find(mr, xlat, 1);
 769     if (!mrs.mr) {
 770         goto unlock;
 771     }
 772
 773     doorbell_gpa = mrs.offset_within_address_space;
 774     memory_region_unref(mrs.mr);
 775
 776     route->u.msi.address_lo = doorbell_gpa;
 777     route->u.msi.address_hi = doorbell_gpa >> 32;
 778
 779     trace_kvm_arm_fixup_msi_route(address, doorbell_gpa);
 780
 781     ret = 0;
 782
 783 unlock:
 784     rcu_read_unlock();
 785     return ret;
 786 }
 787
 788 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
 789                                 int vector, PCIDevice *dev)
 790 {
 791     return 0;
 792 }
 793
 794 int kvm_arch_release_virq_post(int virq)
 795 {
 796     return 0;
 797 }
 798
 799 int kvm_arch_msi_data_to_gsi(uint32_t data)
 800 {
 801     return (data - 32) & 0xffff;
 802 }