arch/powerpc/kvm/powerpc.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License, version 2, as
   4  * published by the Free Software Foundation.
   5  *
   6  * This program is distributed in the hope that it will be useful,
   7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   9  * GNU General Public License for more details.
  10  *
  11  * You should have received a copy of the GNU General Public License
  12  * along with this program; if not, write to the Free Software
  13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  14  *
  15  * Copyright IBM Corp. 2007
  16  *
  17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
  18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/hrtimer.h>
  26 #include <linux/sched/signal.h>
  27 #include <linux/fs.h>
  28 #include <linux/slab.h>
  29 #include <linux/file.h>
  30 #include <linux/module.h>
  31 #include <linux/irqbypass.h>
  32 #include <linux/kvm_irqfd.h>
  33 #include <asm/cputable.h>
  34 #include <linux/uaccess.h>
  35 #include <asm/kvm_ppc.h>
  36 #include <asm/cputhreads.h>
  37 #include <asm/irqflags.h>
  38 #include <asm/iommu.h>
  39 #include <asm/switch_to.h>
  40 #include <asm/xive.h>
  41 #ifdef CONFIG_PPC_PSERIES
  42 #include <asm/hvcall.h>
  43 #include <asm/plpar_wrappers.h>
  44 #endif
  45
  46 #include "timing.h"
  47 #include "irq.h"
  48 #include "../mm/mmu_decl.h"
  49
  50 #define CREATE_TRACE_POINTS
  51 #include "trace.h"
  52
  53 struct kvmppc_ops *kvmppc_hv_ops;
  54 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
  55 struct kvmppc_ops *kvmppc_pr_ops;
  56 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
  57
  58
  59 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  60 {
  61         return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
  62 }
  63
  64 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
  65 {
  66         return false;
  67 }
  68
  69 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  70 {
  71         return 1;
  72 }
  73
  74 /*
  75  * Common checks before entering the guest world.  Call with interrupts
  76  * disabled.
  77  *
  78  * returns:
  79  *
  80  * == 1 if we're ready to go into guest state
  81  * <= 0 if we need to go back to the host with return value
  82  */
  83 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  84 {
  85         int r;
  86
  87         WARN_ON(irqs_disabled());
  88         hard_irq_disable();
  89
  90         while (true) {
  91                 if (need_resched()) {
  92                         local_irq_enable();
  93                         cond_resched();
  94                         hard_irq_disable();
  95                         continue;
  96                 }
  97
  98                 if (signal_pending(current)) {
  99                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
 100                         vcpu->run->exit_reason = KVM_EXIT_INTR;
 101                         r = -EINTR;
 102                         break;
 103                 }
 104
 105                 vcpu->mode = IN_GUEST_MODE;
 106
 107                 /*
 108                  * Reading vcpu->requests must happen after setting vcpu->mode,
 109                  * so we don't miss a request because the requester sees
 110                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
 111                  * before next entering the guest (and thus doesn't IPI).
 112                  * This also orders the write to mode from any reads
 113                  * to the page tables done while the VCPU is running.
 114                  * Please see the comment in kvm_flush_remote_tlbs.
 115                  */
 116                 smp_mb();
 117
 118                 if (kvm_request_pending(vcpu)) {
 119                         /* Make sure we process requests preemptable */
 120                         local_irq_enable();
 121                         trace_kvm_check_requests(vcpu);
 122                         r = kvmppc_core_check_requests(vcpu);
 123                         hard_irq_disable();
 124                         if (r > 0)
 125                                 continue;
 126                         break;
 127                 }
 128
 129                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 130                         /* interrupts got enabled in between, so we
 131                            are back at square 1 */
 132                         continue;
 133                 }
 134
 135                 guest_enter_irqoff();
 136                 return 1;
 137         }
 138
 139         /* return to host */
 140         local_irq_enable();
 141         return r;
 142 }
 143 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 144
 145 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 146 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
 147 {
 148         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
 149         int i;
 150
 151         shared->sprg0 = swab64(shared->sprg0);
 152         shared->sprg1 = swab64(shared->sprg1);
 153         shared->sprg2 = swab64(shared->sprg2);
 154         shared->sprg3 = swab64(shared->sprg3);
 155         shared->srr0 = swab64(shared->srr0);
 156         shared->srr1 = swab64(shared->srr1);
 157         shared->dar = swab64(shared->dar);
 158         shared->msr = swab64(shared->msr);
 159         shared->dsisr = swab32(shared->dsisr);
 160         shared->int_pending = swab32(shared->int_pending);
 161         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
 162                 shared->sr[i] = swab32(shared->sr[i]);
 163 }
 164 #endif
 165
 166 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 167 {
 168         int nr = kvmppc_get_gpr(vcpu, 11);
 169         int r;
 170         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 171         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 172         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 173         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 174         unsigned long r2 = 0;
 175
 176         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
 177                 /* 32 bit mode */
 178                 param1 &= 0xffffffff;
 179                 param2 &= 0xffffffff;
 180                 param3 &= 0xffffffff;
 181                 param4 &= 0xffffffff;
 182         }
 183
 184         switch (nr) {
 185         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 186         {
 187 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 188                 /* Book3S can be little endian, find it out here */
 189                 int shared_big_endian = true;
 190                 if (vcpu->arch.intr_msr & MSR_LE)
 191                         shared_big_endian = false;
 192                 if (shared_big_endian != vcpu->arch.shared_big_endian)
 193                         kvmppc_swab_shared(vcpu);
 194                 vcpu->arch.shared_big_endian = shared_big_endian;
 195 #endif
 196
 197                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
 198                         /*
 199                          * Older versions of the Linux magic page code had
 200                          * a bug where they would map their trampoline code
 201                          * NX. If that's the case, remove !PR NX capability.
 202                          */
 203                         vcpu->arch.disable_kernel_nx = true;
 204                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 205                 }
 206
 207                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
 208                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
 209
 210 #ifdef CONFIG_PPC_64K_PAGES
 211                 /*
 212                  * Make sure our 4k magic page is in the same window of a 64k
 213                  * page within the guest and within the host's page.
 214                  */
 215                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
 216                     ((ulong)vcpu->arch.shared & 0xf000)) {
 217                         void *old_shared = vcpu->arch.shared;
 218                         ulong shared = (ulong)vcpu->arch.shared;
 219                         void *new_shared;
 220
 221                         shared &= PAGE_MASK;
 222                         shared |= vcpu->arch.magic_page_pa & 0xf000;
 223                         new_shared = (void*)shared;
 224                         memcpy(new_shared, old_shared, 0x1000);
 225                         vcpu->arch.shared = new_shared;
 226                 }
 227 #endif
 228
 229                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 230
 231                 r = EV_SUCCESS;
 232                 break;
 233         }
 234         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 235                 r = EV_SUCCESS;
 236 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 237                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 238 #endif
 239
 240                 /* Second return value is in r4 */
 241                 break;
 242         case EV_HCALL_TOKEN(EV_IDLE):
 243                 r = EV_SUCCESS;
 244                 kvm_vcpu_block(vcpu);
 245                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
 246                 break;
 247         default:
 248                 r = EV_UNIMPLEMENTED;
 249                 break;
 250         }
 251
 252         kvmppc_set_gpr(vcpu, 4, r2);
 253
 254         return r;
 255 }
 256 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 257
 258 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 259 {
 260         int r = false;
 261
 262         /* We have to know what CPU to virtualize */
 263         if (!vcpu->arch.pvr)
 264                 goto out;
 265
 266         /* PAPR only works with book3s_64 */
 267         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 268                 goto out;
 269
 270         /* HV KVM can only do PAPR mode for now */
 271         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 272                 goto out;
 273
 274 #ifdef CONFIG_KVM_BOOKE_HV
 275         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 276                 goto out;
 277 #endif
 278
 279         r = true;
 280
 281 out:
 282         vcpu->arch.sane = r;
 283         return r ? 0 : -EINVAL;
 284 }
 285 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 286
 287 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 288 {
 289         enum emulation_result er;
 290         int r;
 291
 292         er = kvmppc_emulate_loadstore(vcpu);
 293         switch (er) {
 294         case EMULATE_DONE:
 295                 /* Future optimization: only reload non-volatiles if they were
 296                  * actually modified. */
 297                 r = RESUME_GUEST_NV;
 298                 break;
 299         case EMULATE_AGAIN:
 300                 r = RESUME_GUEST;
 301                 break;
 302         case EMULATE_DO_MMIO:
 303                 run->exit_reason = KVM_EXIT_MMIO;
 304                 /* We must reload nonvolatiles because "update" load/store
 305                  * instructions modify register state. */
 306                 /* Future optimization: only reload non-volatiles if they were
 307                  * actually modified. */
 308                 r = RESUME_HOST_NV;
 309                 break;
 310         case EMULATE_FAIL:
 311         {
 312                 u32 last_inst;
 313
 314                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 315                 /* XXX Deliver Program interrupt to guest. */
 316                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
 317                 r = RESUME_HOST;
 318                 break;
 319         }
 320         default:
 321                 WARN_ON(1);
 322                 r = RESUME_GUEST;
 323         }
 324
 325         return r;
 326 }
 327 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 328
 329 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 330               bool data)
 331 {
 332         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 333         struct kvmppc_pte pte;
 334         int r;
 335
 336         vcpu->stat.st++;
 337
 338         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 339                          XLATE_WRITE, &pte);
 340         if (r < 0)
 341                 return r;
 342
 343         *eaddr = pte.raddr;
 344
 345         if (!pte.may_write)
 346                 return -EPERM;
 347
 348         /* Magic page override */
 349         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 350             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 351             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 352                 void *magic = vcpu->arch.shared;
 353                 magic += pte.eaddr & 0xfff;
 354                 memcpy(magic, ptr, size);
 355                 return EMULATE_DONE;
 356         }
 357
 358         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 359                 return EMULATE_DO_MMIO;
 360
 361         return EMULATE_DONE;
 362 }
 363 EXPORT_SYMBOL_GPL(kvmppc_st);
 364
 365 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 366                       bool data)
 367 {
 368         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 369         struct kvmppc_pte pte;
 370         int rc;
 371
 372         vcpu->stat.ld++;
 373
 374         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 375                           XLATE_READ, &pte);
 376         if (rc)
 377                 return rc;
 378
 379         *eaddr = pte.raddr;
 380
 381         if (!pte.may_read)
 382                 return -EPERM;
 383
 384         if (!data && !pte.may_execute)
 385                 return -ENOEXEC;
 386
 387         /* Magic page override */
 388         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 389             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 390             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 391                 void *magic = vcpu->arch.shared;
 392                 magic += pte.eaddr & 0xfff;
 393                 memcpy(ptr, magic, size);
 394                 return EMULATE_DONE;
 395         }
 396
 397         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
 398                 return EMULATE_DO_MMIO;
 399
 400         return EMULATE_DONE;
 401 }
 402 EXPORT_SYMBOL_GPL(kvmppc_ld);
 403
 404 int kvm_arch_hardware_enable(void)
 405 {
 406         return 0;
 407 }
 408
 409 int kvm_arch_hardware_setup(void)
 410 {
 411         return 0;
 412 }
 413
 414 void kvm_arch_check_processor_compat(void *rtn)
 415 {
 416         *(int *)rtn = kvmppc_core_check_processor_compat();
 417 }
 418
 419 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 420 {
 421         struct kvmppc_ops *kvm_ops = NULL;
 422         /*
 423          * if we have both HV and PR enabled, default is HV
 424          */
 425         if (type == 0) {
 426                 if (kvmppc_hv_ops)
 427                         kvm_ops = kvmppc_hv_ops;
 428                 else
 429                         kvm_ops = kvmppc_pr_ops;
 430                 if (!kvm_ops)
 431                         goto err_out;
 432         } else  if (type == KVM_VM_PPC_HV) {
 433                 if (!kvmppc_hv_ops)
 434                         goto err_out;
 435                 kvm_ops = kvmppc_hv_ops;
 436         } else if (type == KVM_VM_PPC_PR) {
 437                 if (!kvmppc_pr_ops)
 438                         goto err_out;
 439                 kvm_ops = kvmppc_pr_ops;
 440         } else
 441                 goto err_out;
 442
 443         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 444                 return -ENOENT;
 445
 446         kvm->arch.kvm_ops = kvm_ops;
 447         return kvmppc_core_init_vm(kvm);
 448 err_out:
 449         return -EINVAL;
 450 }
 451
 452 bool kvm_arch_has_vcpu_debugfs(void)
 453 {
 454         return false;
 455 }
 456
 457 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 458 {
 459         return 0;
 460 }
 461
 462 void kvm_arch_destroy_vm(struct kvm *kvm)
 463 {
 464         unsigned int i;
 465         struct kvm_vcpu *vcpu;
 466
 467 #ifdef CONFIG_KVM_XICS
 468         /*
 469          * We call kick_all_cpus_sync() to ensure that all
 470          * CPUs have executed any pending IPIs before we
 471          * continue and free VCPUs structures below.
 472          */
 473         if (is_kvmppc_hv_enabled(kvm))
 474                 kick_all_cpus_sync();
 475 #endif
 476
 477         kvm_for_each_vcpu(i, vcpu, kvm)
 478                 kvm_arch_vcpu_free(vcpu);
 479
 480         mutex_lock(&kvm->lock);
 481         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 482                 kvm->vcpus[i] = NULL;
 483
 484         atomic_set(&kvm->online_vcpus, 0);
 485
 486         kvmppc_core_destroy_vm(kvm);
 487
 488         mutex_unlock(&kvm->lock);
 489
 490         /* drop the module reference */
 491         module_put(kvm->arch.kvm_ops->owner);
 492 }
 493
 494 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 495 {
 496         int r;
 497         /* Assume we're using HV mode when the HV module is loaded */
 498         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 499
 500         if (kvm) {
 501                 /*
 502                  * Hooray - we know which VM type we're running on. Depend on
 503                  * that rather than the guess above.
 504                  */
 505                 hv_enabled = is_kvmppc_hv_enabled(kvm);
 506         }
 507
 508         switch (ext) {
 509 #ifdef CONFIG_BOOKE
 510         case KVM_CAP_PPC_BOOKE_SREGS:
 511         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 512         case KVM_CAP_PPC_EPR:
 513 #else
 514         case KVM_CAP_PPC_SEGSTATE:
 515         case KVM_CAP_PPC_HIOR:
 516         case KVM_CAP_PPC_PAPR:
 517 #endif
 518         case KVM_CAP_PPC_UNSET_IRQ:
 519         case KVM_CAP_PPC_IRQ_LEVEL:
 520         case KVM_CAP_ENABLE_CAP:
 521         case KVM_CAP_ENABLE_CAP_VM:
 522         case KVM_CAP_ONE_REG:
 523         case KVM_CAP_IOEVENTFD:
 524         case KVM_CAP_DEVICE_CTRL:
 525         case KVM_CAP_IMMEDIATE_EXIT:
 526                 r = 1;
 527                 break;
 528         case KVM_CAP_PPC_PAIRED_SINGLES:
 529         case KVM_CAP_PPC_OSI:
 530         case KVM_CAP_PPC_GET_PVINFO:
 531 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 532         case KVM_CAP_SW_TLB:
 533 #endif
 534                 /* We support this only for PR */
 535                 r = !hv_enabled;
 536                 break;
 537 #ifdef CONFIG_KVM_MPIC
 538         case KVM_CAP_IRQ_MPIC:
 539                 r = 1;
 540                 break;
 541 #endif
 542
 543 #ifdef CONFIG_PPC_BOOK3S_64
 544         case KVM_CAP_SPAPR_TCE:
 545         case KVM_CAP_SPAPR_TCE_64:
 546                 /* fallthrough */
 547         case KVM_CAP_SPAPR_TCE_VFIO:
 548         case KVM_CAP_PPC_RTAS:
 549         case KVM_CAP_PPC_FIXUP_HCALL:
 550         case KVM_CAP_PPC_ENABLE_HCALL:
 551 #ifdef CONFIG_KVM_XICS
 552         case KVM_CAP_IRQ_XICS:
 553 #endif
 554         case KVM_CAP_PPC_GET_CPU_CHAR:
 555                 r = 1;
 556                 break;
 557
 558         case KVM_CAP_PPC_ALLOC_HTAB:
 559                 r = hv_enabled;
 560                 break;
 561 #endif /* CONFIG_PPC_BOOK3S_64 */
 562 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 563         case KVM_CAP_PPC_SMT:
 564                 r = 0;
 565                 if (kvm) {
 566                         if (kvm->arch.emul_smt_mode > 1)
 567                                 r = kvm->arch.emul_smt_mode;
 568                         else
 569                                 r = kvm->arch.smt_mode;
 570                 } else if (hv_enabled) {
 571                         if (cpu_has_feature(CPU_FTR_ARCH_300))
 572                                 r = 1;
 573                         else
 574                                 r = threads_per_subcore;
 575                 }
 576                 break;
 577         case KVM_CAP_PPC_SMT_POSSIBLE:
 578                 r = 1;
 579                 if (hv_enabled) {
 580                         if (!cpu_has_feature(CPU_FTR_ARCH_300))
 581                                 r = ((threads_per_subcore << 1) - 1);
 582                         else
 583                                 /* P9 can emulate dbells, so allow any mode */
 584                                 r = 8 | 4 | 2 | 1;
 585                 }
 586                 break;
 587         case KVM_CAP_PPC_RMA:
 588                 r = 0;
 589                 break;
 590         case KVM_CAP_PPC_HWRNG:
 591                 r = kvmppc_hwrng_present();
 592                 break;
 593         case KVM_CAP_PPC_MMU_RADIX:
 594                 r = !!(hv_enabled && radix_enabled());
 595                 break;
 596         case KVM_CAP_PPC_MMU_HASH_V3:
 597                 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300));
 598                 break;
 599 #endif
 600         case KVM_CAP_SYNC_MMU:
 601 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 602                 r = hv_enabled;
 603 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 604                 r = 1;
 605 #else
 606                 r = 0;
 607 #endif
 608                 break;
 609 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 610         case KVM_CAP_PPC_HTAB_FD:
 611                 r = hv_enabled;
 612                 break;
 613 #endif
 614         case KVM_CAP_NR_VCPUS:
 615                 /*
 616                  * Recommending a number of CPUs is somewhat arbitrary; we
 617                  * return the number of present CPUs for -HV (since a host
 618                  * will have secondary threads "offline"), and for other KVM
 619                  * implementations just count online CPUs.
 620                  */
 621                 if (hv_enabled)
 622                         r = num_present_cpus();
 623                 else
 624                         r = num_online_cpus();
 625                 break;
 626         case KVM_CAP_NR_MEMSLOTS:
 627                 r = KVM_USER_MEM_SLOTS;
 628                 break;
 629         case KVM_CAP_MAX_VCPUS:
 630                 r = KVM_MAX_VCPUS;
 631                 break;
 632 #ifdef CONFIG_PPC_BOOK3S_64
 633         case KVM_CAP_PPC_GET_SMMU_INFO:
 634                 r = 1;
 635                 break;
 636         case KVM_CAP_SPAPR_MULTITCE:
 637                 r = 1;
 638                 break;
 639         case KVM_CAP_SPAPR_RESIZE_HPT:
 640                 r = !!hv_enabled;
 641                 break;
 642 #endif
 643 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 644         case KVM_CAP_PPC_FWNMI:
 645                 r = hv_enabled;
 646                 break;
 647 #endif
 648 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 649         case KVM_CAP_PPC_HTM:
 650                 r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
 651                      (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
 652                 break;
 653 #endif
 654         default:
 655                 r = 0;
 656                 break;
 657         }
 658         return r;
 659
 660 }
 661
 662 long kvm_arch_dev_ioctl(struct file *filp,
 663                         unsigned int ioctl, unsigned long arg)
 664 {
 665         return -EINVAL;
 666 }
 667
 668 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 669                            struct kvm_memory_slot *dont)
 670 {
 671         kvmppc_core_free_memslot(kvm, free, dont);
 672 }
 673
 674 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 675                             unsigned long npages)
 676 {
 677         return kvmppc_core_create_memslot(kvm, slot, npages);
 678 }
 679
 680 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 681                                    struct kvm_memory_slot *memslot,
 682                                    const struct kvm_userspace_memory_region *mem,
 683                                    enum kvm_mr_change change)
 684 {
 685         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 686 }
 687
 688 void kvm_arch_commit_memory_region(struct kvm *kvm,
 689                                    const struct kvm_userspace_memory_region *mem,
 690                                    const struct kvm_memory_slot *old,
 691                                    const struct kvm_memory_slot *new,
 692                                    enum kvm_mr_change change)
 693 {
 694         kvmppc_core_commit_memory_region(kvm, mem, old, new);
 695 }
 696
 697 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 698                                    struct kvm_memory_slot *slot)
 699 {
 700         kvmppc_core_flush_memslot(kvm, slot);
 701 }
 702
 703 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 704 {
 705         struct kvm_vcpu *vcpu;
 706         vcpu = kvmppc_core_vcpu_create(kvm, id);
 707         if (!IS_ERR(vcpu)) {
 708                 vcpu->arch.wqp = &vcpu->wq;
 709                 kvmppc_create_vcpu_debugfs(vcpu, id);
 710         }
 711         return vcpu;
 712 }
 713
 714 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 715 {
 716 }
 717
 718 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 719 {
 720         /* Make sure we're not using the vcpu anymore */
 721         hrtimer_cancel(&vcpu->arch.dec_timer);
 722
 723         kvmppc_remove_vcpu_debugfs(vcpu);
 724
 725         switch (vcpu->arch.irq_type) {
 726         case KVMPPC_IRQ_MPIC:
 727                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 728                 break;
 729         case KVMPPC_IRQ_XICS:
 730                 if (xive_enabled())
 731                         kvmppc_xive_cleanup_vcpu(vcpu);
 732                 else
 733                         kvmppc_xics_free_icp(vcpu);
 734                 break;
 735         }
 736
 737         kvmppc_core_vcpu_free(vcpu);
 738 }
 739
 740 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 741 {
 742         kvm_arch_vcpu_free(vcpu);
 743 }
 744
 745 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 746 {
 747         return kvmppc_core_pending_dec(vcpu);
 748 }
 749
 750 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 751 {
 752         struct kvm_vcpu *vcpu;
 753
 754         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 755         kvmppc_decrementer_func(vcpu);
 756
 757         return HRTIMER_NORESTART;
 758 }
 759
 760 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 761 {
 762         int ret;
 763
 764         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 765         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 766         vcpu->arch.dec_expires = get_tb();
 767
 768 #ifdef CONFIG_KVM_EXIT_TIMING
 769         mutex_init(&vcpu->arch.exit_timing_lock);
 770 #endif
 771         ret = kvmppc_subarch_vcpu_init(vcpu);
 772         return ret;
 773 }
 774
 775 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 776 {
 777         kvmppc_mmu_destroy(vcpu);
 778         kvmppc_subarch_vcpu_uninit(vcpu);
 779 }
 780
 781 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 782 {
 783 #ifdef CONFIG_BOOKE
 784         /*
 785          * vrsave (formerly usprg0) isn't used by Linux, but may
 786          * be used by the guest.
 787          *
 788          * On non-booke this is associated with Altivec and
 789          * is handled by code in book3s.c.
 790          */
 791         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 792 #endif
 793         kvmppc_core_vcpu_load(vcpu, cpu);
 794 }
 795
 796 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 797 {
 798         kvmppc_core_vcpu_put(vcpu);
 799 #ifdef CONFIG_BOOKE
 800         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 801 #endif
 802 }
 803
 804 /*
 805  * irq_bypass_add_producer and irq_bypass_del_producer are only
 806  * useful if the architecture supports PCI passthrough.
 807  * irq_bypass_stop and irq_bypass_start are not needed and so
 808  * kvm_ops are not defined for them.
 809  */
 810 bool kvm_arch_has_irq_bypass(void)
 811 {
 812         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
 813                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
 814 }
 815
 816 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
 817                                      struct irq_bypass_producer *prod)
 818 {
 819         struct kvm_kernel_irqfd *irqfd =
 820                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 821         struct kvm *kvm = irqfd->kvm;
 822
 823         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
 824                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
 825
 826         return 0;
 827 }
 828
 829 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
 830                                       struct irq_bypass_producer *prod)
 831 {
 832         struct kvm_kernel_irqfd *irqfd =
 833                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 834         struct kvm *kvm = irqfd->kvm;
 835
 836         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
 837                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
 838 }
 839
 840 #ifdef CONFIG_VSX
 841 static inline int kvmppc_get_vsr_dword_offset(int index)
 842 {
 843         int offset;
 844
 845         if ((index != 0) && (index != 1))
 846                 return -1;
 847
 848 #ifdef __BIG_ENDIAN
 849         offset =  index;
 850 #else
 851         offset = 1 - index;
 852 #endif
 853
 854         return offset;
 855 }
 856
 857 static inline int kvmppc_get_vsr_word_offset(int index)
 858 {
 859         int offset;
 860
 861         if ((index > 3) || (index < 0))
 862                 return -1;
 863
 864 #ifdef __BIG_ENDIAN
 865         offset = index;
 866 #else
 867         offset = 3 - index;
 868 #endif
 869         return offset;
 870 }
 871
 872 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
 873         u64 gpr)
 874 {
 875         union kvmppc_one_reg val;
 876         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
 877         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 878
 879         if (offset == -1)
 880                 return;
 881
 882         if (index >= 32) {
 883                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
 884                 val.vsxval[offset] = gpr;
 885                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
 886         } else {
 887                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
 888         }
 889 }
 890
 891 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
 892         u64 gpr)
 893 {
 894         union kvmppc_one_reg val;
 895         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 896
 897         if (index >= 32) {
 898                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
 899                 val.vsxval[0] = gpr;
 900                 val.vsxval[1] = gpr;
 901                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
 902         } else {
 903                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
 904                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
 905         }
 906 }
 907
 908 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
 909         u32 gpr)
 910 {
 911         union kvmppc_one_reg val;
 912         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 913
 914         if (index >= 32) {
 915                 val.vsx32val[0] = gpr;
 916                 val.vsx32val[1] = gpr;
 917                 val.vsx32val[2] = gpr;
 918                 val.vsx32val[3] = gpr;
 919                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
 920         } else {
 921                 val.vsx32val[0] = gpr;
 922                 val.vsx32val[1] = gpr;
 923                 VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
 924                 VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
 925         }
 926 }
 927
 928 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
 929         u32 gpr32)
 930 {
 931         union kvmppc_one_reg val;
 932         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
 933         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 934         int dword_offset, word_offset;
 935
 936         if (offset == -1)
 937                 return;
 938
 939         if (index >= 32) {
 940                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
 941                 val.vsx32val[offset] = gpr32;
 942                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
 943         } else {
 944                 dword_offset = offset / 2;
 945                 word_offset = offset % 2;
 946                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
 947                 val.vsx32val[word_offset] = gpr32;
 948                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
 949         }
 950 }
 951 #endif /* CONFIG_VSX */
 952
 953 #ifdef CONFIG_ALTIVEC
 954 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
 955                 int index, int element_size)
 956 {
 957         int offset;
 958         int elts = sizeof(vector128)/element_size;
 959
 960         if ((index < 0) || (index >= elts))
 961                 return -1;
 962
 963         if (kvmppc_need_byteswap(vcpu))
 964                 offset = elts - index - 1;
 965         else
 966                 offset = index;
 967
 968         return offset;
 969 }
 970
 971 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
 972                 int index)
 973 {
 974         return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
 975 }
 976
 977 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
 978                 int index)
 979 {
 980         return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
 981 }
 982
 983 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
 984                 int index)
 985 {
 986         return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
 987 }
 988
 989 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
 990                 int index)
 991 {
 992         return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
 993 }
 994
 995
 996 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
 997         u64 gpr)
 998 {
 999         union kvmppc_one_reg val;
1000         int offset = kvmppc_get_vmx_dword_offset(vcpu,
1001                         vcpu->arch.mmio_vmx_offset);
1002         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1003
1004         if (offset == -1)
1005                 return;
1006
1007         val.vval = VCPU_VSX_VR(vcpu, index);
1008         val.vsxval[offset] = gpr;
1009         VCPU_VSX_VR(vcpu, index) = val.vval;
1010 }
1011
1012 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1013         u32 gpr32)
1014 {
1015         union kvmppc_one_reg val;
1016         int offset = kvmppc_get_vmx_word_offset(vcpu,
1017                         vcpu->arch.mmio_vmx_offset);
1018         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1019
1020         if (offset == -1)
1021                 return;
1022
1023         val.vval = VCPU_VSX_VR(vcpu, index);
1024         val.vsx32val[offset] = gpr32;
1025         VCPU_VSX_VR(vcpu, index) = val.vval;
1026 }
1027
1028 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1029         u16 gpr16)
1030 {
1031         union kvmppc_one_reg val;
1032         int offset = kvmppc_get_vmx_hword_offset(vcpu,
1033                         vcpu->arch.mmio_vmx_offset);
1034         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1035
1036         if (offset == -1)
1037                 return;
1038
1039         val.vval = VCPU_VSX_VR(vcpu, index);
1040         val.vsx16val[offset] = gpr16;
1041         VCPU_VSX_VR(vcpu, index) = val.vval;
1042 }
1043
1044 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1045         u8 gpr8)
1046 {
1047         union kvmppc_one_reg val;
1048         int offset = kvmppc_get_vmx_byte_offset(vcpu,
1049                         vcpu->arch.mmio_vmx_offset);
1050         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1051
1052         if (offset == -1)
1053                 return;
1054
1055         val.vval = VCPU_VSX_VR(vcpu, index);
1056         val.vsx8val[offset] = gpr8;
1057         VCPU_VSX_VR(vcpu, index) = val.vval;
1058 }
1059 #endif /* CONFIG_ALTIVEC */
1060
1061 #ifdef CONFIG_PPC_FPU
1062 static inline u64 sp_to_dp(u32 fprs)
1063 {
1064         u64 fprd;
1065
1066         preempt_disable();
1067         enable_kernel_fp();
1068         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1069              : "fr0");
1070         preempt_enable();
1071         return fprd;
1072 }
1073
1074 static inline u32 dp_to_sp(u64 fprd)
1075 {
1076         u32 fprs;
1077
1078         preempt_disable();
1079         enable_kernel_fp();
1080         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1081              : "fr0");
1082         preempt_enable();
1083         return fprs;
1084 }
1085
1086 #else
1087 #define sp_to_dp(x)     (x)
1088 #define dp_to_sp(x)     (x)
1089 #endif /* CONFIG_PPC_FPU */
1090
1091 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1092                                       struct kvm_run *run)
1093 {
1094         u64 uninitialized_var(gpr);
1095
1096         if (run->mmio.len > sizeof(gpr)) {
1097                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1098                 return;
1099         }
1100
1101         if (!vcpu->arch.mmio_host_swabbed) {
1102                 switch (run->mmio.len) {
1103                 case 8: gpr = *(u64 *)run->mmio.data; break;
1104                 case 4: gpr = *(u32 *)run->mmio.data; break;
1105                 case 2: gpr = *(u16 *)run->mmio.data; break;
1106                 case 1: gpr = *(u8 *)run->mmio.data; break;
1107                 }
1108         } else {
1109                 switch (run->mmio.len) {
1110                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1111                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1112                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1113                 case 1: gpr = *(u8 *)run->mmio.data; break;
1114                 }
1115         }
1116
1117         /* conversion between single and double precision */
1118         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1119                 gpr = sp_to_dp(gpr);
1120
1121         if (vcpu->arch.mmio_sign_extend) {
1122                 switch (run->mmio.len) {
1123 #ifdef CONFIG_PPC64
1124                 case 4:
1125                         gpr = (s64)(s32)gpr;
1126                         break;
1127 #endif
1128                 case 2:
1129                         gpr = (s64)(s16)gpr;
1130                         break;
1131                 case 1:
1132                         gpr = (s64)(s8)gpr;
1133                         break;
1134                 }
1135         }
1136
1137         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1138         case KVM_MMIO_REG_GPR:
1139                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1140                 break;
1141         case KVM_MMIO_REG_FPR:
1142                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1143                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1144
1145                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1146                 break;
1147 #ifdef CONFIG_PPC_BOOK3S
1148         case KVM_MMIO_REG_QPR:
1149                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1150                 break;
1151         case KVM_MMIO_REG_FQPR:
1152                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1153                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1154                 break;
1155 #endif
1156 #ifdef CONFIG_VSX
1157         case KVM_MMIO_REG_VSX:
1158                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1159                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1160
1161                 if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1162                         kvmppc_set_vsr_dword(vcpu, gpr);
1163                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1164                         kvmppc_set_vsr_word(vcpu, gpr);
1165                 else if (vcpu->arch.mmio_copy_type ==
1166                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1167                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1168                 else if (vcpu->arch.mmio_copy_type ==
1169                                 KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1170                         kvmppc_set_vsr_word_dump(vcpu, gpr);
1171                 break;
1172 #endif
1173 #ifdef CONFIG_ALTIVEC
1174         case KVM_MMIO_REG_VMX:
1175                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1176                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1177
1178                 if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1179                         kvmppc_set_vmx_dword(vcpu, gpr);
1180                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1181                         kvmppc_set_vmx_word(vcpu, gpr);
1182                 else if (vcpu->arch.mmio_copy_type ==
1183                                 KVMPPC_VMX_COPY_HWORD)
1184                         kvmppc_set_vmx_hword(vcpu, gpr);
1185                 else if (vcpu->arch.mmio_copy_type ==
1186                                 KVMPPC_VMX_COPY_BYTE)
1187                         kvmppc_set_vmx_byte(vcpu, gpr);
1188                 break;
1189 #endif
1190         default:
1191                 BUG();
1192         }
1193 }
1194
1195 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1196                                 unsigned int rt, unsigned int bytes,
1197                                 int is_default_endian, int sign_extend)
1198 {
1199         int idx, ret;
1200         bool host_swabbed;
1201
1202         /* Pity C doesn't have a logical XOR operator */
1203         if (kvmppc_need_byteswap(vcpu)) {
1204                 host_swabbed = is_default_endian;
1205         } else {
1206                 host_swabbed = !is_default_endian;
1207         }
1208
1209         if (bytes > sizeof(run->mmio.data)) {
1210                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1211                        run->mmio.len);
1212         }
1213
1214         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1215         run->mmio.len = bytes;
1216         run->mmio.is_write = 0;
1217
1218         vcpu->arch.io_gpr = rt;
1219         vcpu->arch.mmio_host_swabbed = host_swabbed;
1220         vcpu->mmio_needed = 1;
1221         vcpu->mmio_is_write = 0;
1222         vcpu->arch.mmio_sign_extend = sign_extend;
1223
1224         idx = srcu_read_lock(&vcpu->kvm->srcu);
1225
1226         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1227                               bytes, &run->mmio.data);
1228
1229         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1230
1231         if (!ret) {
1232                 kvmppc_complete_mmio_load(vcpu, run);
1233                 vcpu->mmio_needed = 0;
1234                 return EMULATE_DONE;
1235         }
1236
1237         return EMULATE_DO_MMIO;
1238 }
1239
1240 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1241                        unsigned int rt, unsigned int bytes,
1242                        int is_default_endian)
1243 {
1244         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1245 }
1246 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1247
1248 /* Same as above, but sign extends */
1249 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1250                         unsigned int rt, unsigned int bytes,
1251                         int is_default_endian)
1252 {
1253         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1254 }
1255
1256 #ifdef CONFIG_VSX
1257 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1258                         unsigned int rt, unsigned int bytes,
1259                         int is_default_endian, int mmio_sign_extend)
1260 {
1261         enum emulation_result emulated = EMULATE_DONE;
1262
1263         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1264         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1265                 return EMULATE_FAIL;
1266
1267         while (vcpu->arch.mmio_vsx_copy_nums) {
1268                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1269                         is_default_endian, mmio_sign_extend);
1270
1271                 if (emulated != EMULATE_DONE)
1272                         break;
1273
1274                 vcpu->arch.paddr_accessed += run->mmio.len;
1275
1276                 vcpu->arch.mmio_vsx_copy_nums--;
1277                 vcpu->arch.mmio_vsx_offset++;
1278         }
1279         return emulated;
1280 }
1281 #endif /* CONFIG_VSX */
1282
1283 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1284                         u64 val, unsigned int bytes, int is_default_endian)
1285 {
1286         void *data = run->mmio.data;
1287         int idx, ret;
1288         bool host_swabbed;
1289
1290         /* Pity C doesn't have a logical XOR operator */
1291         if (kvmppc_need_byteswap(vcpu)) {
1292                 host_swabbed = is_default_endian;
1293         } else {
1294                 host_swabbed = !is_default_endian;
1295         }
1296
1297         if (bytes > sizeof(run->mmio.data)) {
1298                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1299                        run->mmio.len);
1300         }
1301
1302         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1303         run->mmio.len = bytes;
1304         run->mmio.is_write = 1;
1305         vcpu->mmio_needed = 1;
1306         vcpu->mmio_is_write = 1;
1307
1308         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1309                 val = dp_to_sp(val);
1310
1311         /* Store the value at the lowest bytes in 'data'. */
1312         if (!host_swabbed) {
1313                 switch (bytes) {
1314                 case 8: *(u64 *)data = val; break;
1315                 case 4: *(u32 *)data = val; break;
1316                 case 2: *(u16 *)data = val; break;
1317                 case 1: *(u8  *)data = val; break;
1318                 }
1319         } else {
1320                 switch (bytes) {
1321                 case 8: *(u64 *)data = swab64(val); break;
1322                 case 4: *(u32 *)data = swab32(val); break;
1323                 case 2: *(u16 *)data = swab16(val); break;
1324                 case 1: *(u8  *)data = val; break;
1325                 }
1326         }
1327
1328         idx = srcu_read_lock(&vcpu->kvm->srcu);
1329
1330         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1331                                bytes, &run->mmio.data);
1332
1333         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1334
1335         if (!ret) {
1336                 vcpu->mmio_needed = 0;
1337                 return EMULATE_DONE;
1338         }
1339
1340         return EMULATE_DO_MMIO;
1341 }
1342 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1343
1344 #ifdef CONFIG_VSX
1345 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1346 {
1347         u32 dword_offset, word_offset;
1348         union kvmppc_one_reg reg;
1349         int vsx_offset = 0;
1350         int copy_type = vcpu->arch.mmio_copy_type;
1351         int result = 0;
1352
1353         switch (copy_type) {
1354         case KVMPPC_VSX_COPY_DWORD:
1355                 vsx_offset =
1356                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1357
1358                 if (vsx_offset == -1) {
1359                         result = -1;
1360                         break;
1361                 }
1362
1363                 if (rs < 32) {
1364                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1365                 } else {
1366                         reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1367                         *val = reg.vsxval[vsx_offset];
1368                 }
1369                 break;
1370
1371         case KVMPPC_VSX_COPY_WORD:
1372                 vsx_offset =
1373                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1374
1375                 if (vsx_offset == -1) {
1376                         result = -1;
1377                         break;
1378                 }
1379
1380                 if (rs < 32) {
1381                         dword_offset = vsx_offset / 2;
1382                         word_offset = vsx_offset % 2;
1383                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1384                         *val = reg.vsx32val[word_offset];
1385                 } else {
1386                         reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1387                         *val = reg.vsx32val[vsx_offset];
1388                 }
1389                 break;
1390
1391         default:
1392                 result = -1;
1393                 break;
1394         }
1395
1396         return result;
1397 }
1398
1399 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1400                         int rs, unsigned int bytes, int is_default_endian)
1401 {
1402         u64 val;
1403         enum emulation_result emulated = EMULATE_DONE;
1404
1405         vcpu->arch.io_gpr = rs;
1406
1407         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1408         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1409                 return EMULATE_FAIL;
1410
1411         while (vcpu->arch.mmio_vsx_copy_nums) {
1412                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1413                         return EMULATE_FAIL;
1414
1415                 emulated = kvmppc_handle_store(run, vcpu,
1416                          val, bytes, is_default_endian);
1417
1418                 if (emulated != EMULATE_DONE)
1419                         break;
1420
1421                 vcpu->arch.paddr_accessed += run->mmio.len;
1422
1423                 vcpu->arch.mmio_vsx_copy_nums--;
1424                 vcpu->arch.mmio_vsx_offset++;
1425         }
1426
1427         return emulated;
1428 }
1429
1430 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1431                         struct kvm_run *run)
1432 {
1433         enum emulation_result emulated = EMULATE_FAIL;
1434         int r;
1435
1436         vcpu->arch.paddr_accessed += run->mmio.len;
1437
1438         if (!vcpu->mmio_is_write) {
1439                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1440                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1441         } else {
1442                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1443                          vcpu->arch.io_gpr, run->mmio.len, 1);
1444         }
1445
1446         switch (emulated) {
1447         case EMULATE_DO_MMIO:
1448                 run->exit_reason = KVM_EXIT_MMIO;
1449                 r = RESUME_HOST;
1450                 break;
1451         case EMULATE_FAIL:
1452                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1453                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1454                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1455                 r = RESUME_HOST;
1456                 break;
1457         default:
1458                 r = RESUME_GUEST;
1459                 break;
1460         }
1461         return r;
1462 }
1463 #endif /* CONFIG_VSX */
1464
1465 #ifdef CONFIG_ALTIVEC
1466 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1467                 unsigned int rt, unsigned int bytes, int is_default_endian)
1468 {
1469         enum emulation_result emulated = EMULATE_DONE;
1470
1471         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1472                 return EMULATE_FAIL;
1473
1474         while (vcpu->arch.mmio_vmx_copy_nums) {
1475                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1476                                 is_default_endian, 0);
1477
1478                 if (emulated != EMULATE_DONE)
1479                         break;
1480
1481                 vcpu->arch.paddr_accessed += run->mmio.len;
1482                 vcpu->arch.mmio_vmx_copy_nums--;
1483                 vcpu->arch.mmio_vmx_offset++;
1484         }
1485
1486         return emulated;
1487 }
1488
1489 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1490 {
1491         union kvmppc_one_reg reg;
1492         int vmx_offset = 0;
1493         int result = 0;
1494
1495         vmx_offset =
1496                 kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1497
1498         if (vmx_offset == -1)
1499                 return -1;
1500
1501         reg.vval = VCPU_VSX_VR(vcpu, index);
1502         *val = reg.vsxval[vmx_offset];
1503
1504         return result;
1505 }
1506
1507 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1508 {
1509         union kvmppc_one_reg reg;
1510         int vmx_offset = 0;
1511         int result = 0;
1512
1513         vmx_offset =
1514                 kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1515
1516         if (vmx_offset == -1)
1517                 return -1;
1518
1519         reg.vval = VCPU_VSX_VR(vcpu, index);
1520         *val = reg.vsx32val[vmx_offset];
1521
1522         return result;
1523 }
1524
1525 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1526 {
1527         union kvmppc_one_reg reg;
1528         int vmx_offset = 0;
1529         int result = 0;
1530
1531         vmx_offset =
1532                 kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1533
1534         if (vmx_offset == -1)
1535                 return -1;
1536
1537         reg.vval = VCPU_VSX_VR(vcpu, index);
1538         *val = reg.vsx16val[vmx_offset];
1539
1540         return result;
1541 }
1542
1543 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1544 {
1545         union kvmppc_one_reg reg;
1546         int vmx_offset = 0;
1547         int result = 0;
1548
1549         vmx_offset =
1550                 kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1551
1552         if (vmx_offset == -1)
1553                 return -1;
1554
1555         reg.vval = VCPU_VSX_VR(vcpu, index);
1556         *val = reg.vsx8val[vmx_offset];
1557
1558         return result;
1559 }
1560
1561 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1562                 unsigned int rs, unsigned int bytes, int is_default_endian)
1563 {
1564         u64 val = 0;
1565         unsigned int index = rs & KVM_MMIO_REG_MASK;
1566         enum emulation_result emulated = EMULATE_DONE;
1567
1568         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1569                 return EMULATE_FAIL;
1570
1571         vcpu->arch.io_gpr = rs;
1572
1573         while (vcpu->arch.mmio_vmx_copy_nums) {
1574                 switch (vcpu->arch.mmio_copy_type) {
1575                 case KVMPPC_VMX_COPY_DWORD:
1576                         if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1577                                 return EMULATE_FAIL;
1578
1579                         break;
1580                 case KVMPPC_VMX_COPY_WORD:
1581                         if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1582                                 return EMULATE_FAIL;
1583                         break;
1584                 case KVMPPC_VMX_COPY_HWORD:
1585                         if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1586                                 return EMULATE_FAIL;
1587                         break;
1588                 case KVMPPC_VMX_COPY_BYTE:
1589                         if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1590                                 return EMULATE_FAIL;
1591                         break;
1592                 default:
1593                         return EMULATE_FAIL;
1594                 }
1595
1596                 emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1597                                 is_default_endian);
1598                 if (emulated != EMULATE_DONE)
1599                         break;
1600
1601                 vcpu->arch.paddr_accessed += run->mmio.len;
1602                 vcpu->arch.mmio_vmx_copy_nums--;
1603                 vcpu->arch.mmio_vmx_offset++;
1604         }
1605
1606         return emulated;
1607 }
1608
1609 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1610                 struct kvm_run *run)
1611 {
1612         enum emulation_result emulated = EMULATE_FAIL;
1613         int r;
1614
1615         vcpu->arch.paddr_accessed += run->mmio.len;
1616
1617         if (!vcpu->mmio_is_write) {
1618                 emulated = kvmppc_handle_vmx_load(run, vcpu,
1619                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1620         } else {
1621                 emulated = kvmppc_handle_vmx_store(run, vcpu,
1622                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1623         }
1624
1625         switch (emulated) {
1626         case EMULATE_DO_MMIO:
1627                 run->exit_reason = KVM_EXIT_MMIO;
1628                 r = RESUME_HOST;
1629                 break;
1630         case EMULATE_FAIL:
1631                 pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1632                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1633                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1634                 r = RESUME_HOST;
1635                 break;
1636         default:
1637                 r = RESUME_GUEST;
1638                 break;
1639         }
1640         return r;
1641 }
1642 #endif /* CONFIG_ALTIVEC */
1643
1644 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1645 {
1646         int r = 0;
1647         union kvmppc_one_reg val;
1648         int size;
1649
1650         size = one_reg_size(reg->id);
1651         if (size > sizeof(val))
1652                 return -EINVAL;
1653
1654         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1655         if (r == -EINVAL) {
1656                 r = 0;
1657                 switch (reg->id) {
1658 #ifdef CONFIG_ALTIVEC
1659                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1660                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1661                                 r = -ENXIO;
1662                                 break;
1663                         }
1664                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1665                         break;
1666                 case KVM_REG_PPC_VSCR:
1667                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1668                                 r = -ENXIO;
1669                                 break;
1670                         }
1671                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1672                         break;
1673                 case KVM_REG_PPC_VRSAVE:
1674                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1675                         break;
1676 #endif /* CONFIG_ALTIVEC */
1677                 default:
1678                         r = -EINVAL;
1679                         break;
1680                 }
1681         }
1682
1683         if (r)
1684                 return r;
1685
1686         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1687                 r = -EFAULT;
1688
1689         return r;
1690 }
1691
1692 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1693 {
1694         int r;
1695         union kvmppc_one_reg val;
1696         int size;
1697
1698         size = one_reg_size(reg->id);
1699         if (size > sizeof(val))
1700                 return -EINVAL;
1701
1702         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1703                 return -EFAULT;
1704
1705         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1706         if (r == -EINVAL) {
1707                 r = 0;
1708                 switch (reg->id) {
1709 #ifdef CONFIG_ALTIVEC
1710                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1711                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1712                                 r = -ENXIO;
1713                                 break;
1714                         }
1715                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1716                         break;
1717                 case KVM_REG_PPC_VSCR:
1718                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1719                                 r = -ENXIO;
1720                                 break;
1721                         }
1722                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1723                         break;
1724                 case KVM_REG_PPC_VRSAVE:
1725                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1726                                 r = -ENXIO;
1727                                 break;
1728                         }
1729                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1730                         break;
1731 #endif /* CONFIG_ALTIVEC */
1732                 default:
1733                         r = -EINVAL;
1734                         break;
1735                 }
1736         }
1737
1738         return r;
1739 }
1740
1741 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1742 {
1743         int r;
1744
1745         vcpu_load(vcpu);
1746
1747         if (vcpu->mmio_needed) {
1748                 vcpu->mmio_needed = 0;
1749                 if (!vcpu->mmio_is_write)
1750                         kvmppc_complete_mmio_load(vcpu, run);
1751 #ifdef CONFIG_VSX
1752                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1753                         vcpu->arch.mmio_vsx_copy_nums--;
1754                         vcpu->arch.mmio_vsx_offset++;
1755                 }
1756
1757                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1758                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1759                         if (r == RESUME_HOST) {
1760                                 vcpu->mmio_needed = 1;
1761                                 goto out;
1762                         }
1763                 }
1764 #endif
1765 #ifdef CONFIG_ALTIVEC
1766                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1767                         vcpu->arch.mmio_vmx_copy_nums--;
1768                         vcpu->arch.mmio_vmx_offset++;
1769                 }
1770
1771                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1772                         r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1773                         if (r == RESUME_HOST) {
1774                                 vcpu->mmio_needed = 1;
1775                                 goto out;
1776                         }
1777                 }
1778 #endif
1779         } else if (vcpu->arch.osi_needed) {
1780                 u64 *gprs = run->osi.gprs;
1781                 int i;
1782
1783                 for (i = 0; i < 32; i++)
1784                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1785                 vcpu->arch.osi_needed = 0;
1786         } else if (vcpu->arch.hcall_needed) {
1787                 int i;
1788
1789                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1790                 for (i = 0; i < 9; ++i)
1791                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1792                 vcpu->arch.hcall_needed = 0;
1793 #ifdef CONFIG_BOOKE
1794         } else if (vcpu->arch.epr_needed) {
1795                 kvmppc_set_epr(vcpu, run->epr.epr);
1796                 vcpu->arch.epr_needed = 0;
1797 #endif
1798         }
1799
1800         kvm_sigset_activate(vcpu);
1801
1802         if (run->immediate_exit)
1803                 r = -EINTR;
1804         else
1805                 r = kvmppc_vcpu_run(run, vcpu);
1806
1807         kvm_sigset_deactivate(vcpu);
1808
1809 #ifdef CONFIG_ALTIVEC
1810 out:
1811 #endif
1812         vcpu_put(vcpu);
1813         return r;
1814 }
1815
1816 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1817 {
1818         if (irq->irq == KVM_INTERRUPT_UNSET) {
1819                 kvmppc_core_dequeue_external(vcpu);
1820                 return 0;
1821         }
1822
1823         kvmppc_core_queue_external(vcpu, irq);
1824
1825         kvm_vcpu_kick(vcpu);
1826
1827         return 0;
1828 }
1829
1830 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1831                                      struct kvm_enable_cap *cap)
1832 {
1833         int r;
1834
1835         if (cap->flags)
1836                 return -EINVAL;
1837
1838         switch (cap->cap) {
1839         case KVM_CAP_PPC_OSI:
1840                 r = 0;
1841                 vcpu->arch.osi_enabled = true;
1842                 break;
1843         case KVM_CAP_PPC_PAPR:
1844                 r = 0;
1845                 vcpu->arch.papr_enabled = true;
1846                 break;
1847         case KVM_CAP_PPC_EPR:
1848                 r = 0;
1849                 if (cap->args[0])
1850                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1851                 else
1852                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1853                 break;
1854 #ifdef CONFIG_BOOKE
1855         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1856                 r = 0;
1857                 vcpu->arch.watchdog_enabled = true;
1858                 break;
1859 #endif
1860 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1861         case KVM_CAP_SW_TLB: {
1862                 struct kvm_config_tlb cfg;
1863                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1864
1865                 r = -EFAULT;
1866                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1867                         break;
1868
1869                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1870                 break;
1871         }
1872 #endif
1873 #ifdef CONFIG_KVM_MPIC
1874         case KVM_CAP_IRQ_MPIC: {
1875                 struct fd f;
1876                 struct kvm_device *dev;
1877
1878                 r = -EBADF;
1879                 f = fdget(cap->args[0]);
1880                 if (!f.file)
1881                         break;
1882
1883                 r = -EPERM;
1884                 dev = kvm_device_from_filp(f.file);
1885                 if (dev)
1886                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1887
1888                 fdput(f);
1889                 break;
1890         }
1891 #endif
1892 #ifdef CONFIG_KVM_XICS
1893         case KVM_CAP_IRQ_XICS: {
1894                 struct fd f;
1895                 struct kvm_device *dev;
1896
1897                 r = -EBADF;
1898                 f = fdget(cap->args[0]);
1899                 if (!f.file)
1900                         break;
1901
1902                 r = -EPERM;
1903                 dev = kvm_device_from_filp(f.file);
1904                 if (dev) {
1905                         if (xive_enabled())
1906                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1907                         else
1908                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1909                 }
1910
1911                 fdput(f);
1912                 break;
1913         }
1914 #endif /* CONFIG_KVM_XICS */
1915 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1916         case KVM_CAP_PPC_FWNMI:
1917                 r = -EINVAL;
1918                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1919                         break;
1920                 r = 0;
1921                 vcpu->kvm->arch.fwnmi_enabled = true;
1922                 break;
1923 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1924         default:
1925                 r = -EINVAL;
1926                 break;
1927         }
1928
1929         if (!r)
1930                 r = kvmppc_sanity_check(vcpu);
1931
1932         return r;
1933 }
1934
1935 bool kvm_arch_intc_initialized(struct kvm *kvm)
1936 {
1937 #ifdef CONFIG_KVM_MPIC
1938         if (kvm->arch.mpic)
1939                 return true;
1940 #endif
1941 #ifdef CONFIG_KVM_XICS
1942         if (kvm->arch.xics || kvm->arch.xive)
1943                 return true;
1944 #endif
1945         return false;
1946 }
1947
1948 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1949                                     struct kvm_mp_state *mp_state)
1950 {
1951         return -EINVAL;
1952 }
1953
1954 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1955                                     struct kvm_mp_state *mp_state)
1956 {
1957         return -EINVAL;
1958 }
1959
1960 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1961                                unsigned int ioctl, unsigned long arg)
1962 {
1963         struct kvm_vcpu *vcpu = filp->private_data;
1964         void __user *argp = (void __user *)arg;
1965
1966         if (ioctl == KVM_INTERRUPT) {
1967                 struct kvm_interrupt irq;
1968                 if (copy_from_user(&irq, argp, sizeof(irq)))
1969                         return -EFAULT;
1970                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1971         }
1972         return -ENOIOCTLCMD;
1973 }
1974
1975 long kvm_arch_vcpu_ioctl(struct file *filp,
1976                          unsigned int ioctl, unsigned long arg)
1977 {
1978         struct kvm_vcpu *vcpu = filp->private_data;
1979         void __user *argp = (void __user *)arg;
1980         long r;
1981
1982         switch (ioctl) {
1983         case KVM_ENABLE_CAP:
1984         {
1985                 struct kvm_enable_cap cap;
1986                 r = -EFAULT;
1987                 vcpu_load(vcpu);
1988                 if (copy_from_user(&cap, argp, sizeof(cap)))
1989                         goto out;
1990                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1991                 vcpu_put(vcpu);
1992                 break;
1993         }
1994
1995         case KVM_SET_ONE_REG:
1996         case KVM_GET_ONE_REG:
1997         {
1998                 struct kvm_one_reg reg;
1999                 r = -EFAULT;
2000                 if (copy_from_user(&reg, argp, sizeof(reg)))
2001                         goto out;
2002                 if (ioctl == KVM_SET_ONE_REG)
2003                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2004                 else
2005                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2006                 break;
2007         }
2008
2009 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2010         case KVM_DIRTY_TLB: {
2011                 struct kvm_dirty_tlb dirty;
2012                 r = -EFAULT;
2013                 vcpu_load(vcpu);
2014                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
2015                         goto out;
2016                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2017                 vcpu_put(vcpu);
2018                 break;
2019         }
2020 #endif
2021         default:
2022                 r = -EINVAL;
2023         }
2024
2025 out:
2026         return r;
2027 }
2028
2029 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2030 {
2031         return VM_FAULT_SIGBUS;
2032 }
2033
2034 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2035 {
2036         u32 inst_nop = 0x60000000;
2037 #ifdef CONFIG_KVM_BOOKE_HV
2038         u32 inst_sc1 = 0x44000022;
2039         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2040         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2041         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2042         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2043 #else
2044         u32 inst_lis = 0x3c000000;
2045         u32 inst_ori = 0x60000000;
2046         u32 inst_sc = 0x44000002;
2047         u32 inst_imm_mask = 0xffff;
2048
2049         /*
2050          * The hypercall to get into KVM from within guest context is as
2051          * follows:
2052          *
2053          *    lis r0, r0, KVM_SC_MAGIC_R0@h
2054          *    ori r0, KVM_SC_MAGIC_R0@l
2055          *    sc
2056          *    nop
2057          */
2058         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2059         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2060         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2061         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2062 #endif
2063
2064         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2065
2066         return 0;
2067 }
2068
2069 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2070                           bool line_status)
2071 {
2072         if (!irqchip_in_kernel(kvm))
2073                 return -ENXIO;
2074
2075         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2076                                         irq_event->irq, irq_event->level,
2077                                         line_status);
2078         return 0;
2079 }
2080
2081
2082 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2083                                    struct kvm_enable_cap *cap)
2084 {
2085         int r;
2086
2087         if (cap->flags)
2088                 return -EINVAL;
2089
2090         switch (cap->cap) {
2091 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2092         case KVM_CAP_PPC_ENABLE_HCALL: {
2093                 unsigned long hcall = cap->args[0];
2094
2095                 r = -EINVAL;
2096                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2097                     cap->args[1] > 1)
2098                         break;
2099                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2100                         break;
2101                 if (cap->args[1])
2102                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2103                 else
2104                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2105                 r = 0;
2106                 break;
2107         }
2108         case KVM_CAP_PPC_SMT: {
2109                 unsigned long mode = cap->args[0];
2110                 unsigned long flags = cap->args[1];
2111
2112                 r = -EINVAL;
2113                 if (kvm->arch.kvm_ops->set_smt_mode)
2114                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2115                 break;
2116         }
2117 #endif
2118         default:
2119                 r = -EINVAL;
2120                 break;
2121         }
2122
2123         return r;
2124 }
2125
2126 #ifdef CONFIG_PPC_BOOK3S_64
2127 /*
2128  * These functions check whether the underlying hardware is safe
2129  * against attacks based on observing the effects of speculatively
2130  * executed instructions, and whether it supplies instructions for
2131  * use in workarounds.  The information comes from firmware, either
2132  * via the device tree on powernv platforms or from an hcall on
2133  * pseries platforms.
2134  */
2135 #ifdef CONFIG_PPC_PSERIES
2136 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2137 {
2138         struct h_cpu_char_result c;
2139         unsigned long rc;
2140
2141         if (!machine_is(pseries))
2142                 return -ENOTTY;
2143
2144         rc = plpar_get_cpu_characteristics(&c);
2145         if (rc == H_SUCCESS) {
2146                 cp->character = c.character;
2147                 cp->behaviour = c.behaviour;
2148                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2149                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2150                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2151                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2152                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2153                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2154                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2155                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2156                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2157                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2158                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2159         }
2160         return 0;
2161 }
2162 #else
2163 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2164 {
2165         return -ENOTTY;
2166 }
2167 #endif
2168
2169 static inline bool have_fw_feat(struct device_node *fw_features,
2170                                 const char *state, const char *name)
2171 {
2172         struct device_node *np;
2173         bool r = false;
2174
2175         np = of_get_child_by_name(fw_features, name);
2176         if (np) {
2177                 r = of_property_read_bool(np, state);
2178                 of_node_put(np);
2179         }
2180         return r;
2181 }
2182
2183 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2184 {
2185         struct device_node *np, *fw_features;
2186         int r;
2187
2188         memset(cp, 0, sizeof(*cp));
2189         r = pseries_get_cpu_char(cp);
2190         if (r != -ENOTTY)
2191                 return r;
2192
2193         np = of_find_node_by_name(NULL, "ibm,opal");
2194         if (np) {
2195                 fw_features = of_get_child_by_name(np, "fw-features");
2196                 of_node_put(np);
2197                 if (!fw_features)
2198                         return 0;
2199                 if (have_fw_feat(fw_features, "enabled",
2200                                  "inst-spec-barrier-ori31,31,0"))
2201                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2202                 if (have_fw_feat(fw_features, "enabled",
2203                                  "fw-bcctrl-serialized"))
2204                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2205                 if (have_fw_feat(fw_features, "enabled",
2206                                  "inst-l1d-flush-ori30,30,0"))
2207                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2208                 if (have_fw_feat(fw_features, "enabled",
2209                                  "inst-l1d-flush-trig2"))
2210                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2211                 if (have_fw_feat(fw_features, "enabled",
2212                                  "fw-l1d-thread-split"))
2213                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2214                 if (have_fw_feat(fw_features, "enabled",
2215                                  "fw-count-cache-disabled"))
2216                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2217                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2218                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2219                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2220                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2221                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2222                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2223
2224                 if (have_fw_feat(fw_features, "enabled",
2225                                  "speculation-policy-favor-security"))
2226                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2227                 if (!have_fw_feat(fw_features, "disabled",
2228                                   "needs-l1d-flush-msr-pr-0-to-1"))
2229                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2230                 if (!have_fw_feat(fw_features, "disabled",
2231                                   "needs-spec-barrier-for-bound-checks"))
2232                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2233                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2234                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2235                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2236
2237                 of_node_put(fw_features);
2238         }
2239
2240         return 0;
2241 }
2242 #endif
2243
2244 long kvm_arch_vm_ioctl(struct file *filp,
2245                        unsigned int ioctl, unsigned long arg)
2246 {
2247         struct kvm *kvm __maybe_unused = filp->private_data;
2248         void __user *argp = (void __user *)arg;
2249         long r;
2250
2251         switch (ioctl) {
2252         case KVM_PPC_GET_PVINFO: {
2253                 struct kvm_ppc_pvinfo pvinfo;
2254                 memset(&pvinfo, 0, sizeof(pvinfo));
2255                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2256                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2257                         r = -EFAULT;
2258                         goto out;
2259                 }
2260
2261                 break;
2262         }
2263         case KVM_ENABLE_CAP:
2264         {
2265                 struct kvm_enable_cap cap;
2266                 r = -EFAULT;
2267                 if (copy_from_user(&cap, argp, sizeof(cap)))
2268                         goto out;
2269                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
2270                 break;
2271         }
2272 #ifdef CONFIG_SPAPR_TCE_IOMMU
2273         case KVM_CREATE_SPAPR_TCE_64: {
2274                 struct kvm_create_spapr_tce_64 create_tce_64;
2275
2276                 r = -EFAULT;
2277                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2278                         goto out;
2279                 if (create_tce_64.flags) {
2280                         r = -EINVAL;
2281                         goto out;
2282                 }
2283                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2284                 goto out;
2285         }
2286         case KVM_CREATE_SPAPR_TCE: {
2287                 struct kvm_create_spapr_tce create_tce;
2288                 struct kvm_create_spapr_tce_64 create_tce_64;
2289
2290                 r = -EFAULT;
2291                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2292                         goto out;
2293
2294                 create_tce_64.liobn = create_tce.liobn;
2295                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2296                 create_tce_64.offset = 0;
2297                 create_tce_64.size = create_tce.window_size >>
2298                                 IOMMU_PAGE_SHIFT_4K;
2299                 create_tce_64.flags = 0;
2300                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2301                 goto out;
2302         }
2303 #endif
2304 #ifdef CONFIG_PPC_BOOK3S_64
2305         case KVM_PPC_GET_SMMU_INFO: {
2306                 struct kvm_ppc_smmu_info info;
2307                 struct kvm *kvm = filp->private_data;
2308
2309                 memset(&info, 0, sizeof(info));
2310                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2311                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2312                         r = -EFAULT;
2313                 break;
2314         }
2315         case KVM_PPC_RTAS_DEFINE_TOKEN: {
2316                 struct kvm *kvm = filp->private_data;
2317
2318                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2319                 break;
2320         }
2321         case KVM_PPC_CONFIGURE_V3_MMU: {
2322                 struct kvm *kvm = filp->private_data;
2323                 struct kvm_ppc_mmuv3_cfg cfg;
2324
2325                 r = -EINVAL;
2326                 if (!kvm->arch.kvm_ops->configure_mmu)
2327                         goto out;
2328                 r = -EFAULT;
2329                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
2330                         goto out;
2331                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2332                 break;
2333         }
2334         case KVM_PPC_GET_RMMU_INFO: {
2335                 struct kvm *kvm = filp->private_data;
2336                 struct kvm_ppc_rmmu_info info;
2337
2338                 r = -EINVAL;
2339                 if (!kvm->arch.kvm_ops->get_rmmu_info)
2340                         goto out;
2341                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2342                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2343                         r = -EFAULT;
2344                 break;
2345         }
2346         case KVM_PPC_GET_CPU_CHAR: {
2347                 struct kvm_ppc_cpu_char cpuchar;
2348
2349                 r = kvmppc_get_cpu_char(&cpuchar);
2350                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2351                         r = -EFAULT;
2352                 break;
2353         }
2354         default: {
2355                 struct kvm *kvm = filp->private_data;
2356                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2357         }
2358 #else /* CONFIG_PPC_BOOK3S_64 */
2359         default:
2360                 r = -ENOTTY;
2361 #endif
2362         }
2363 out:
2364         return r;
2365 }
2366
2367 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2368 static unsigned long nr_lpids;
2369
2370 long kvmppc_alloc_lpid(void)
2371 {
2372         long lpid;
2373
2374         do {
2375                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2376                 if (lpid >= nr_lpids) {
2377                         pr_err("%s: No LPIDs free\n", __func__);
2378                         return -ENOMEM;
2379                 }
2380         } while (test_and_set_bit(lpid, lpid_inuse));
2381
2382         return lpid;
2383 }
2384 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2385
2386 void kvmppc_claim_lpid(long lpid)
2387 {
2388         set_bit(lpid, lpid_inuse);
2389 }
2390 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2391
2392 void kvmppc_free_lpid(long lpid)
2393 {
2394         clear_bit(lpid, lpid_inuse);
2395 }
2396 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2397
2398 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2399 {
2400         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2401         memset(lpid_inuse, 0, sizeof(lpid_inuse));
2402 }
2403 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2404
2405 int kvm_arch_init(void *opaque)
2406 {
2407         return 0;
2408 }
2409
2410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);