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/tlbflush.h>
  37 #include <asm/cputhreads.h>
  38 #include <asm/irqflags.h>
  39 #include <asm/iommu.h>
  40 #include <asm/switch_to.h>
  41 #include <asm/xive.h>
  42 #ifdef CONFIG_PPC_PSERIES
  43 #include <asm/hvcall.h>
  44 #include <asm/plpar_wrappers.h>
  45 #endif
  46
  47 #include "timing.h"
  48 #include "irq.h"
  49 #include "../mm/mmu_decl.h"
  50
  51 #define CREATE_TRACE_POINTS
  52 #include "trace.h"
  53
  54 struct kvmppc_ops *kvmppc_hv_ops;
  55 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
  56 struct kvmppc_ops *kvmppc_pr_ops;
  57 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
  58
  59
  60 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  61 {
  62         return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
  63 }
  64
  65 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
  66 {
  67         return false;
  68 }
  69
  70 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  71 {
  72         return 1;
  73 }
  74
  75 /*
  76  * Common checks before entering the guest world.  Call with interrupts
  77  * disabled.
  78  *
  79  * returns:
  80  *
  81  * == 1 if we're ready to go into guest state
  82  * <= 0 if we need to go back to the host with return value
  83  */
  84 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  85 {
  86         int r;
  87
  88         WARN_ON(irqs_disabled());
  89         hard_irq_disable();
  90
  91         while (true) {
  92                 if (need_resched()) {
  93                         local_irq_enable();
  94                         cond_resched();
  95                         hard_irq_disable();
  96                         continue;
  97                 }
  98
  99                 if (signal_pending(current)) {
 100                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
 101                         vcpu->run->exit_reason = KVM_EXIT_INTR;
 102                         r = -EINTR;
 103                         break;
 104                 }
 105
 106                 vcpu->mode = IN_GUEST_MODE;
 107
 108                 /*
 109                  * Reading vcpu->requests must happen after setting vcpu->mode,
 110                  * so we don't miss a request because the requester sees
 111                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
 112                  * before next entering the guest (and thus doesn't IPI).
 113                  * This also orders the write to mode from any reads
 114                  * to the page tables done while the VCPU is running.
 115                  * Please see the comment in kvm_flush_remote_tlbs.
 116                  */
 117                 smp_mb();
 118
 119                 if (kvm_request_pending(vcpu)) {
 120                         /* Make sure we process requests preemptable */
 121                         local_irq_enable();
 122                         trace_kvm_check_requests(vcpu);
 123                         r = kvmppc_core_check_requests(vcpu);
 124                         hard_irq_disable();
 125                         if (r > 0)
 126                                 continue;
 127                         break;
 128                 }
 129
 130                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 131                         /* interrupts got enabled in between, so we
 132                            are back at square 1 */
 133                         continue;
 134                 }
 135
 136                 guest_enter_irqoff();
 137                 return 1;
 138         }
 139
 140         /* return to host */
 141         local_irq_enable();
 142         return r;
 143 }
 144 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 145
 146 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 147 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
 148 {
 149         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
 150         int i;
 151
 152         shared->sprg0 = swab64(shared->sprg0);
 153         shared->sprg1 = swab64(shared->sprg1);
 154         shared->sprg2 = swab64(shared->sprg2);
 155         shared->sprg3 = swab64(shared->sprg3);
 156         shared->srr0 = swab64(shared->srr0);
 157         shared->srr1 = swab64(shared->srr1);
 158         shared->dar = swab64(shared->dar);
 159         shared->msr = swab64(shared->msr);
 160         shared->dsisr = swab32(shared->dsisr);
 161         shared->int_pending = swab32(shared->int_pending);
 162         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
 163                 shared->sr[i] = swab32(shared->sr[i]);
 164 }
 165 #endif
 166
 167 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 168 {
 169         int nr = kvmppc_get_gpr(vcpu, 11);
 170         int r;
 171         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 172         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 173         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 174         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 175         unsigned long r2 = 0;
 176
 177         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
 178                 /* 32 bit mode */
 179                 param1 &= 0xffffffff;
 180                 param2 &= 0xffffffff;
 181                 param3 &= 0xffffffff;
 182                 param4 &= 0xffffffff;
 183         }
 184
 185         switch (nr) {
 186         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 187         {
 188 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 189                 /* Book3S can be little endian, find it out here */
 190                 int shared_big_endian = true;
 191                 if (vcpu->arch.intr_msr & MSR_LE)
 192                         shared_big_endian = false;
 193                 if (shared_big_endian != vcpu->arch.shared_big_endian)
 194                         kvmppc_swab_shared(vcpu);
 195                 vcpu->arch.shared_big_endian = shared_big_endian;
 196 #endif
 197
 198                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
 199                         /*
 200                          * Older versions of the Linux magic page code had
 201                          * a bug where they would map their trampoline code
 202                          * NX. If that's the case, remove !PR NX capability.
 203                          */
 204                         vcpu->arch.disable_kernel_nx = true;
 205                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 206                 }
 207
 208                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
 209                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
 210
 211 #ifdef CONFIG_PPC_64K_PAGES
 212                 /*
 213                  * Make sure our 4k magic page is in the same window of a 64k
 214                  * page within the guest and within the host's page.
 215                  */
 216                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
 217                     ((ulong)vcpu->arch.shared & 0xf000)) {
 218                         void *old_shared = vcpu->arch.shared;
 219                         ulong shared = (ulong)vcpu->arch.shared;
 220                         void *new_shared;
 221
 222                         shared &= PAGE_MASK;
 223                         shared |= vcpu->arch.magic_page_pa & 0xf000;
 224                         new_shared = (void*)shared;
 225                         memcpy(new_shared, old_shared, 0x1000);
 226                         vcpu->arch.shared = new_shared;
 227                 }
 228 #endif
 229
 230                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 231
 232                 r = EV_SUCCESS;
 233                 break;
 234         }
 235         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 236                 r = EV_SUCCESS;
 237 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 238                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 239 #endif
 240
 241                 /* Second return value is in r4 */
 242                 break;
 243         case EV_HCALL_TOKEN(EV_IDLE):
 244                 r = EV_SUCCESS;
 245                 kvm_vcpu_block(vcpu);
 246                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
 247                 break;
 248         default:
 249                 r = EV_UNIMPLEMENTED;
 250                 break;
 251         }
 252
 253         kvmppc_set_gpr(vcpu, 4, r2);
 254
 255         return r;
 256 }
 257 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 258
 259 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 260 {
 261         int r = false;
 262
 263         /* We have to know what CPU to virtualize */
 264         if (!vcpu->arch.pvr)
 265                 goto out;
 266
 267         /* PAPR only works with book3s_64 */
 268         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 269                 goto out;
 270
 271         /* HV KVM can only do PAPR mode for now */
 272         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 273                 goto out;
 274
 275 #ifdef CONFIG_KVM_BOOKE_HV
 276         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 277                 goto out;
 278 #endif
 279
 280         r = true;
 281
 282 out:
 283         vcpu->arch.sane = r;
 284         return r ? 0 : -EINVAL;
 285 }
 286 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 287
 288 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 289 {
 290         enum emulation_result er;
 291         int r;
 292
 293         er = kvmppc_emulate_loadstore(vcpu);
 294         switch (er) {
 295         case EMULATE_DONE:
 296                 /* Future optimization: only reload non-volatiles if they were
 297                  * actually modified. */
 298                 r = RESUME_GUEST_NV;
 299                 break;
 300         case EMULATE_AGAIN:
 301                 r = RESUME_GUEST;
 302                 break;
 303         case EMULATE_DO_MMIO:
 304                 run->exit_reason = KVM_EXIT_MMIO;
 305                 /* We must reload nonvolatiles because "update" load/store
 306                  * instructions modify register state. */
 307                 /* Future optimization: only reload non-volatiles if they were
 308                  * actually modified. */
 309                 r = RESUME_HOST_NV;
 310                 break;
 311         case EMULATE_FAIL:
 312         {
 313                 u32 last_inst;
 314
 315                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 316                 /* XXX Deliver Program interrupt to guest. */
 317                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
 318                 r = RESUME_HOST;
 319                 break;
 320         }
 321         default:
 322                 WARN_ON(1);
 323                 r = RESUME_GUEST;
 324         }
 325
 326         return r;
 327 }
 328 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 329
 330 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 331               bool data)
 332 {
 333         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 334         struct kvmppc_pte pte;
 335         int r;
 336
 337         vcpu->stat.st++;
 338
 339         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 340                          XLATE_WRITE, &pte);
 341         if (r < 0)
 342                 return r;
 343
 344         *eaddr = pte.raddr;
 345
 346         if (!pte.may_write)
 347                 return -EPERM;
 348
 349         /* Magic page override */
 350         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 351             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 352             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 353                 void *magic = vcpu->arch.shared;
 354                 magic += pte.eaddr & 0xfff;
 355                 memcpy(magic, ptr, size);
 356                 return EMULATE_DONE;
 357         }
 358
 359         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 360                 return EMULATE_DO_MMIO;
 361
 362         return EMULATE_DONE;
 363 }
 364 EXPORT_SYMBOL_GPL(kvmppc_st);
 365
 366 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 367                       bool data)
 368 {
 369         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 370         struct kvmppc_pte pte;
 371         int rc;
 372
 373         vcpu->stat.ld++;
 374
 375         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 376                           XLATE_READ, &pte);
 377         if (rc)
 378                 return rc;
 379
 380         *eaddr = pte.raddr;
 381
 382         if (!pte.may_read)
 383                 return -EPERM;
 384
 385         if (!data && !pte.may_execute)
 386                 return -ENOEXEC;
 387
 388         /* Magic page override */
 389         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 390             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 391             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 392                 void *magic = vcpu->arch.shared;
 393                 magic += pte.eaddr & 0xfff;
 394                 memcpy(ptr, magic, size);
 395                 return EMULATE_DONE;
 396         }
 397
 398         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
 399                 return EMULATE_DO_MMIO;
 400
 401         return EMULATE_DONE;
 402 }
 403 EXPORT_SYMBOL_GPL(kvmppc_ld);
 404
 405 int kvm_arch_hardware_enable(void)
 406 {
 407         return 0;
 408 }
 409
 410 int kvm_arch_hardware_setup(void)
 411 {
 412         return 0;
 413 }
 414
 415 void kvm_arch_check_processor_compat(void *rtn)
 416 {
 417         *(int *)rtn = kvmppc_core_check_processor_compat();
 418 }
 419
 420 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 421 {
 422         struct kvmppc_ops *kvm_ops = NULL;
 423         /*
 424          * if we have both HV and PR enabled, default is HV
 425          */
 426         if (type == 0) {
 427                 if (kvmppc_hv_ops)
 428                         kvm_ops = kvmppc_hv_ops;
 429                 else
 430                         kvm_ops = kvmppc_pr_ops;
 431                 if (!kvm_ops)
 432                         goto err_out;
 433         } else  if (type == KVM_VM_PPC_HV) {
 434                 if (!kvmppc_hv_ops)
 435                         goto err_out;
 436                 kvm_ops = kvmppc_hv_ops;
 437         } else if (type == KVM_VM_PPC_PR) {
 438                 if (!kvmppc_pr_ops)
 439                         goto err_out;
 440                 kvm_ops = kvmppc_pr_ops;
 441         } else
 442                 goto err_out;
 443
 444         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 445                 return -ENOENT;
 446
 447         kvm->arch.kvm_ops = kvm_ops;
 448         return kvmppc_core_init_vm(kvm);
 449 err_out:
 450         return -EINVAL;
 451 }
 452
 453 bool kvm_arch_has_vcpu_debugfs(void)
 454 {
 455         return false;
 456 }
 457
 458 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 459 {
 460         return 0;
 461 }
 462
 463 void kvm_arch_destroy_vm(struct kvm *kvm)
 464 {
 465         unsigned int i;
 466         struct kvm_vcpu *vcpu;
 467
 468 #ifdef CONFIG_KVM_XICS
 469         /*
 470          * We call kick_all_cpus_sync() to ensure that all
 471          * CPUs have executed any pending IPIs before we
 472          * continue and free VCPUs structures below.
 473          */
 474         if (is_kvmppc_hv_enabled(kvm))
 475                 kick_all_cpus_sync();
 476 #endif
 477
 478         kvm_for_each_vcpu(i, vcpu, kvm)
 479                 kvm_arch_vcpu_free(vcpu);
 480
 481         mutex_lock(&kvm->lock);
 482         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 483                 kvm->vcpus[i] = NULL;
 484
 485         atomic_set(&kvm->online_vcpus, 0);
 486
 487         kvmppc_core_destroy_vm(kvm);
 488
 489         mutex_unlock(&kvm->lock);
 490
 491         /* drop the module reference */
 492         module_put(kvm->arch.kvm_ops->owner);
 493 }
 494
 495 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 496 {
 497         int r;
 498         /* Assume we're using HV mode when the HV module is loaded */
 499         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 500
 501         if (kvm) {
 502                 /*
 503                  * Hooray - we know which VM type we're running on. Depend on
 504                  * that rather than the guess above.
 505                  */
 506                 hv_enabled = is_kvmppc_hv_enabled(kvm);
 507         }
 508
 509         switch (ext) {
 510 #ifdef CONFIG_BOOKE
 511         case KVM_CAP_PPC_BOOKE_SREGS:
 512         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 513         case KVM_CAP_PPC_EPR:
 514 #else
 515         case KVM_CAP_PPC_SEGSTATE:
 516         case KVM_CAP_PPC_HIOR:
 517         case KVM_CAP_PPC_PAPR:
 518 #endif
 519         case KVM_CAP_PPC_UNSET_IRQ:
 520         case KVM_CAP_PPC_IRQ_LEVEL:
 521         case KVM_CAP_ENABLE_CAP:
 522         case KVM_CAP_ENABLE_CAP_VM:
 523         case KVM_CAP_ONE_REG:
 524         case KVM_CAP_IOEVENTFD:
 525         case KVM_CAP_DEVICE_CTRL:
 526         case KVM_CAP_IMMEDIATE_EXIT:
 527                 r = 1;
 528                 break;
 529         case KVM_CAP_PPC_PAIRED_SINGLES:
 530         case KVM_CAP_PPC_OSI:
 531         case KVM_CAP_PPC_GET_PVINFO:
 532 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 533         case KVM_CAP_SW_TLB:
 534 #endif
 535                 /* We support this only for PR */
 536                 r = !hv_enabled;
 537                 break;
 538 #ifdef CONFIG_KVM_MPIC
 539         case KVM_CAP_IRQ_MPIC:
 540                 r = 1;
 541                 break;
 542 #endif
 543
 544 #ifdef CONFIG_PPC_BOOK3S_64
 545         case KVM_CAP_SPAPR_TCE:
 546         case KVM_CAP_SPAPR_TCE_64:
 547                 /* fallthrough */
 548         case KVM_CAP_SPAPR_TCE_VFIO:
 549         case KVM_CAP_PPC_RTAS:
 550         case KVM_CAP_PPC_FIXUP_HCALL:
 551         case KVM_CAP_PPC_ENABLE_HCALL:
 552 #ifdef CONFIG_KVM_XICS
 553         case KVM_CAP_IRQ_XICS:
 554 #endif
 555         case KVM_CAP_PPC_GET_CPU_CHAR:
 556                 r = 1;
 557                 break;
 558
 559         case KVM_CAP_PPC_ALLOC_HTAB:
 560                 r = hv_enabled;
 561                 break;
 562 #endif /* CONFIG_PPC_BOOK3S_64 */
 563 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 564         case KVM_CAP_PPC_SMT:
 565                 r = 0;
 566                 if (kvm) {
 567                         if (kvm->arch.emul_smt_mode > 1)
 568                                 r = kvm->arch.emul_smt_mode;
 569                         else
 570                                 r = kvm->arch.smt_mode;
 571                 } else if (hv_enabled) {
 572                         if (cpu_has_feature(CPU_FTR_ARCH_300))
 573                                 r = 1;
 574                         else
 575                                 r = threads_per_subcore;
 576                 }
 577                 break;
 578         case KVM_CAP_PPC_SMT_POSSIBLE:
 579                 r = 1;
 580                 if (hv_enabled) {
 581                         if (!cpu_has_feature(CPU_FTR_ARCH_300))
 582                                 r = ((threads_per_subcore << 1) - 1);
 583                         else
 584                                 /* P9 can emulate dbells, so allow any mode */
 585                                 r = 8 | 4 | 2 | 1;
 586                 }
 587                 break;
 588         case KVM_CAP_PPC_RMA:
 589                 r = 0;
 590                 break;
 591         case KVM_CAP_PPC_HWRNG:
 592                 r = kvmppc_hwrng_present();
 593                 break;
 594         case KVM_CAP_PPC_MMU_RADIX:
 595                 r = !!(hv_enabled && radix_enabled());
 596                 break;
 597         case KVM_CAP_PPC_MMU_HASH_V3:
 598                 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300));
 599                 break;
 600 #endif
 601         case KVM_CAP_SYNC_MMU:
 602 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 603                 r = hv_enabled;
 604 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 605                 r = 1;
 606 #else
 607                 r = 0;
 608 #endif
 609                 break;
 610 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 611         case KVM_CAP_PPC_HTAB_FD:
 612                 r = hv_enabled;
 613                 break;
 614 #endif
 615         case KVM_CAP_NR_VCPUS:
 616                 /*
 617                  * Recommending a number of CPUs is somewhat arbitrary; we
 618                  * return the number of present CPUs for -HV (since a host
 619                  * will have secondary threads "offline"), and for other KVM
 620                  * implementations just count online CPUs.
 621                  */
 622                 if (hv_enabled)
 623                         r = num_present_cpus();
 624                 else
 625                         r = num_online_cpus();
 626                 break;
 627         case KVM_CAP_NR_MEMSLOTS:
 628                 r = KVM_USER_MEM_SLOTS;
 629                 break;
 630         case KVM_CAP_MAX_VCPUS:
 631                 r = KVM_MAX_VCPUS;
 632                 break;
 633 #ifdef CONFIG_PPC_BOOK3S_64
 634         case KVM_CAP_PPC_GET_SMMU_INFO:
 635                 r = 1;
 636                 break;
 637         case KVM_CAP_SPAPR_MULTITCE:
 638                 r = 1;
 639                 break;
 640         case KVM_CAP_SPAPR_RESIZE_HPT:
 641                 r = !!hv_enabled;
 642                 break;
 643 #endif
 644 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 645         case KVM_CAP_PPC_FWNMI:
 646                 r = hv_enabled;
 647                 break;
 648 #endif
 649 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 650         case KVM_CAP_PPC_HTM:
 651                 r = hv_enabled &&
 652                     (!!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
 653                      cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
 654                 break;
 655 #endif
 656         default:
 657                 r = 0;
 658                 break;
 659         }
 660         return r;
 661
 662 }
 663
 664 long kvm_arch_dev_ioctl(struct file *filp,
 665                         unsigned int ioctl, unsigned long arg)
 666 {
 667         return -EINVAL;
 668 }
 669
 670 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 671                            struct kvm_memory_slot *dont)
 672 {
 673         kvmppc_core_free_memslot(kvm, free, dont);
 674 }
 675
 676 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 677                             unsigned long npages)
 678 {
 679         return kvmppc_core_create_memslot(kvm, slot, npages);
 680 }
 681
 682 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 683                                    struct kvm_memory_slot *memslot,
 684                                    const struct kvm_userspace_memory_region *mem,
 685                                    enum kvm_mr_change change)
 686 {
 687         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 688 }
 689
 690 void kvm_arch_commit_memory_region(struct kvm *kvm,
 691                                    const struct kvm_userspace_memory_region *mem,
 692                                    const struct kvm_memory_slot *old,
 693                                    const struct kvm_memory_slot *new,
 694                                    enum kvm_mr_change change)
 695 {
 696         kvmppc_core_commit_memory_region(kvm, mem, old, new);
 697 }
 698
 699 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 700                                    struct kvm_memory_slot *slot)
 701 {
 702         kvmppc_core_flush_memslot(kvm, slot);
 703 }
 704
 705 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 706 {
 707         struct kvm_vcpu *vcpu;
 708         vcpu = kvmppc_core_vcpu_create(kvm, id);
 709         if (!IS_ERR(vcpu)) {
 710                 vcpu->arch.wqp = &vcpu->wq;
 711                 kvmppc_create_vcpu_debugfs(vcpu, id);
 712         }
 713         return vcpu;
 714 }
 715
 716 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 717 {
 718 }
 719
 720 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 721 {
 722         /* Make sure we're not using the vcpu anymore */
 723         hrtimer_cancel(&vcpu->arch.dec_timer);
 724
 725         kvmppc_remove_vcpu_debugfs(vcpu);
 726
 727         switch (vcpu->arch.irq_type) {
 728         case KVMPPC_IRQ_MPIC:
 729                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 730                 break;
 731         case KVMPPC_IRQ_XICS:
 732                 if (xive_enabled())
 733                         kvmppc_xive_cleanup_vcpu(vcpu);
 734                 else
 735                         kvmppc_xics_free_icp(vcpu);
 736                 break;
 737         }
 738
 739         kvmppc_core_vcpu_free(vcpu);
 740 }
 741
 742 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 743 {
 744         kvm_arch_vcpu_free(vcpu);
 745 }
 746
 747 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 748 {
 749         return kvmppc_core_pending_dec(vcpu);
 750 }
 751
 752 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 753 {
 754         struct kvm_vcpu *vcpu;
 755
 756         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 757         kvmppc_decrementer_func(vcpu);
 758
 759         return HRTIMER_NORESTART;
 760 }
 761
 762 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 763 {
 764         int ret;
 765
 766         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 767         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 768         vcpu->arch.dec_expires = get_tb();
 769
 770 #ifdef CONFIG_KVM_EXIT_TIMING
 771         mutex_init(&vcpu->arch.exit_timing_lock);
 772 #endif
 773         ret = kvmppc_subarch_vcpu_init(vcpu);
 774         return ret;
 775 }
 776
 777 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 778 {
 779         kvmppc_mmu_destroy(vcpu);
 780         kvmppc_subarch_vcpu_uninit(vcpu);
 781 }
 782
 783 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 784 {
 785 #ifdef CONFIG_BOOKE
 786         /*
 787          * vrsave (formerly usprg0) isn't used by Linux, but may
 788          * be used by the guest.
 789          *
 790          * On non-booke this is associated with Altivec and
 791          * is handled by code in book3s.c.
 792          */
 793         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 794 #endif
 795         kvmppc_core_vcpu_load(vcpu, cpu);
 796 }
 797
 798 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 799 {
 800         kvmppc_core_vcpu_put(vcpu);
 801 #ifdef CONFIG_BOOKE
 802         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 803 #endif
 804 }
 805
 806 /*
 807  * irq_bypass_add_producer and irq_bypass_del_producer are only
 808  * useful if the architecture supports PCI passthrough.
 809  * irq_bypass_stop and irq_bypass_start are not needed and so
 810  * kvm_ops are not defined for them.
 811  */
 812 bool kvm_arch_has_irq_bypass(void)
 813 {
 814         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
 815                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
 816 }
 817
 818 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
 819                                      struct irq_bypass_producer *prod)
 820 {
 821         struct kvm_kernel_irqfd *irqfd =
 822                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 823         struct kvm *kvm = irqfd->kvm;
 824
 825         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
 826                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
 827
 828         return 0;
 829 }
 830
 831 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
 832                                       struct irq_bypass_producer *prod)
 833 {
 834         struct kvm_kernel_irqfd *irqfd =
 835                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 836         struct kvm *kvm = irqfd->kvm;
 837
 838         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
 839                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
 840 }
 841
 842 #ifdef CONFIG_VSX
 843 static inline int kvmppc_get_vsr_dword_offset(int index)
 844 {
 845         int offset;
 846
 847         if ((index != 0) && (index != 1))
 848                 return -1;
 849
 850 #ifdef __BIG_ENDIAN
 851         offset =  index;
 852 #else
 853         offset = 1 - index;
 854 #endif
 855
 856         return offset;
 857 }
 858
 859 static inline int kvmppc_get_vsr_word_offset(int index)
 860 {
 861         int offset;
 862
 863         if ((index > 3) || (index < 0))
 864                 return -1;
 865
 866 #ifdef __BIG_ENDIAN
 867         offset = index;
 868 #else
 869         offset = 3 - index;
 870 #endif
 871         return offset;
 872 }
 873
 874 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
 875         u64 gpr)
 876 {
 877         union kvmppc_one_reg val;
 878         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
 879         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 880
 881         if (offset == -1)
 882                 return;
 883
 884         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 885                 val.vval = VCPU_VSX_VR(vcpu, index);
 886                 val.vsxval[offset] = gpr;
 887                 VCPU_VSX_VR(vcpu, index) = val.vval;
 888         } else {
 889                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
 890         }
 891 }
 892
 893 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
 894         u64 gpr)
 895 {
 896         union kvmppc_one_reg val;
 897         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 898
 899         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 900                 val.vval = VCPU_VSX_VR(vcpu, index);
 901                 val.vsxval[0] = gpr;
 902                 val.vsxval[1] = gpr;
 903                 VCPU_VSX_VR(vcpu, index) = val.vval;
 904         } else {
 905                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
 906                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
 907         }
 908 }
 909
 910 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
 911         u32 gpr)
 912 {
 913         union kvmppc_one_reg val;
 914         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 915
 916         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 917                 val.vsx32val[0] = gpr;
 918                 val.vsx32val[1] = gpr;
 919                 val.vsx32val[2] = gpr;
 920                 val.vsx32val[3] = gpr;
 921                 VCPU_VSX_VR(vcpu, index) = val.vval;
 922         } else {
 923                 val.vsx32val[0] = gpr;
 924                 val.vsx32val[1] = gpr;
 925                 VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
 926                 VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
 927         }
 928 }
 929
 930 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
 931         u32 gpr32)
 932 {
 933         union kvmppc_one_reg val;
 934         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
 935         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 936         int dword_offset, word_offset;
 937
 938         if (offset == -1)
 939                 return;
 940
 941         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 942                 val.vval = VCPU_VSX_VR(vcpu, index);
 943                 val.vsx32val[offset] = gpr32;
 944                 VCPU_VSX_VR(vcpu, index) = val.vval;
 945         } else {
 946                 dword_offset = offset / 2;
 947                 word_offset = offset % 2;
 948                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
 949                 val.vsx32val[word_offset] = gpr32;
 950                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
 951         }
 952 }
 953 #endif /* CONFIG_VSX */
 954
 955 #ifdef CONFIG_ALTIVEC
 956 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
 957                 int index, int element_size)
 958 {
 959         int offset;
 960         int elts = sizeof(vector128)/element_size;
 961
 962         if ((index < 0) || (index >= elts))
 963                 return -1;
 964
 965         if (kvmppc_need_byteswap(vcpu))
 966                 offset = elts - index - 1;
 967         else
 968                 offset = index;
 969
 970         return offset;
 971 }
 972
 973 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
 974                 int index)
 975 {
 976         return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
 977 }
 978
 979 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
 980                 int index)
 981 {
 982         return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
 983 }
 984
 985 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
 986                 int index)
 987 {
 988         return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
 989 }
 990
 991 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
 992                 int index)
 993 {
 994         return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
 995 }
 996
 997
 998 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
 999         u64 gpr)
1000 {
1001         union kvmppc_one_reg val;
1002         int offset = kvmppc_get_vmx_dword_offset(vcpu,
1003                         vcpu->arch.mmio_vmx_offset);
1004         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1005
1006         if (offset == -1)
1007                 return;
1008
1009         val.vval = VCPU_VSX_VR(vcpu, index);
1010         val.vsxval[offset] = gpr;
1011         VCPU_VSX_VR(vcpu, index) = val.vval;
1012 }
1013
1014 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1015         u32 gpr32)
1016 {
1017         union kvmppc_one_reg val;
1018         int offset = kvmppc_get_vmx_word_offset(vcpu,
1019                         vcpu->arch.mmio_vmx_offset);
1020         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1021
1022         if (offset == -1)
1023                 return;
1024
1025         val.vval = VCPU_VSX_VR(vcpu, index);
1026         val.vsx32val[offset] = gpr32;
1027         VCPU_VSX_VR(vcpu, index) = val.vval;
1028 }
1029
1030 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1031         u16 gpr16)
1032 {
1033         union kvmppc_one_reg val;
1034         int offset = kvmppc_get_vmx_hword_offset(vcpu,
1035                         vcpu->arch.mmio_vmx_offset);
1036         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1037
1038         if (offset == -1)
1039                 return;
1040
1041         val.vval = VCPU_VSX_VR(vcpu, index);
1042         val.vsx16val[offset] = gpr16;
1043         VCPU_VSX_VR(vcpu, index) = val.vval;
1044 }
1045
1046 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1047         u8 gpr8)
1048 {
1049         union kvmppc_one_reg val;
1050         int offset = kvmppc_get_vmx_byte_offset(vcpu,
1051                         vcpu->arch.mmio_vmx_offset);
1052         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1053
1054         if (offset == -1)
1055                 return;
1056
1057         val.vval = VCPU_VSX_VR(vcpu, index);
1058         val.vsx8val[offset] = gpr8;
1059         VCPU_VSX_VR(vcpu, index) = val.vval;
1060 }
1061 #endif /* CONFIG_ALTIVEC */
1062
1063 #ifdef CONFIG_PPC_FPU
1064 static inline u64 sp_to_dp(u32 fprs)
1065 {
1066         u64 fprd;
1067
1068         preempt_disable();
1069         enable_kernel_fp();
1070         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1071              : "fr0");
1072         preempt_enable();
1073         return fprd;
1074 }
1075
1076 static inline u32 dp_to_sp(u64 fprd)
1077 {
1078         u32 fprs;
1079
1080         preempt_disable();
1081         enable_kernel_fp();
1082         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1083              : "fr0");
1084         preempt_enable();
1085         return fprs;
1086 }
1087
1088 #else
1089 #define sp_to_dp(x)     (x)
1090 #define dp_to_sp(x)     (x)
1091 #endif /* CONFIG_PPC_FPU */
1092
1093 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1094                                       struct kvm_run *run)
1095 {
1096         u64 uninitialized_var(gpr);
1097
1098         if (run->mmio.len > sizeof(gpr)) {
1099                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1100                 return;
1101         }
1102
1103         if (!vcpu->arch.mmio_host_swabbed) {
1104                 switch (run->mmio.len) {
1105                 case 8: gpr = *(u64 *)run->mmio.data; break;
1106                 case 4: gpr = *(u32 *)run->mmio.data; break;
1107                 case 2: gpr = *(u16 *)run->mmio.data; break;
1108                 case 1: gpr = *(u8 *)run->mmio.data; break;
1109                 }
1110         } else {
1111                 switch (run->mmio.len) {
1112                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1113                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1114                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1115                 case 1: gpr = *(u8 *)run->mmio.data; break;
1116                 }
1117         }
1118
1119         /* conversion between single and double precision */
1120         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1121                 gpr = sp_to_dp(gpr);
1122
1123         if (vcpu->arch.mmio_sign_extend) {
1124                 switch (run->mmio.len) {
1125 #ifdef CONFIG_PPC64
1126                 case 4:
1127                         gpr = (s64)(s32)gpr;
1128                         break;
1129 #endif
1130                 case 2:
1131                         gpr = (s64)(s16)gpr;
1132                         break;
1133                 case 1:
1134                         gpr = (s64)(s8)gpr;
1135                         break;
1136                 }
1137         }
1138
1139         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1140         case KVM_MMIO_REG_GPR:
1141                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1142                 break;
1143         case KVM_MMIO_REG_FPR:
1144                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1145                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1146
1147                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1148                 break;
1149 #ifdef CONFIG_PPC_BOOK3S
1150         case KVM_MMIO_REG_QPR:
1151                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1152                 break;
1153         case KVM_MMIO_REG_FQPR:
1154                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1155                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1156                 break;
1157 #endif
1158 #ifdef CONFIG_VSX
1159         case KVM_MMIO_REG_VSX:
1160                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1161                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1162
1163                 if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1164                         kvmppc_set_vsr_dword(vcpu, gpr);
1165                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1166                         kvmppc_set_vsr_word(vcpu, gpr);
1167                 else if (vcpu->arch.mmio_copy_type ==
1168                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1169                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1170                 else if (vcpu->arch.mmio_copy_type ==
1171                                 KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1172                         kvmppc_set_vsr_word_dump(vcpu, gpr);
1173                 break;
1174 #endif
1175 #ifdef CONFIG_ALTIVEC
1176         case KVM_MMIO_REG_VMX:
1177                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1178                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1179
1180                 if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1181                         kvmppc_set_vmx_dword(vcpu, gpr);
1182                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1183                         kvmppc_set_vmx_word(vcpu, gpr);
1184                 else if (vcpu->arch.mmio_copy_type ==
1185                                 KVMPPC_VMX_COPY_HWORD)
1186                         kvmppc_set_vmx_hword(vcpu, gpr);
1187                 else if (vcpu->arch.mmio_copy_type ==
1188                                 KVMPPC_VMX_COPY_BYTE)
1189                         kvmppc_set_vmx_byte(vcpu, gpr);
1190                 break;
1191 #endif
1192         default:
1193                 BUG();
1194         }
1195 }
1196
1197 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1198                                 unsigned int rt, unsigned int bytes,
1199                                 int is_default_endian, int sign_extend)
1200 {
1201         int idx, ret;
1202         bool host_swabbed;
1203
1204         /* Pity C doesn't have a logical XOR operator */
1205         if (kvmppc_need_byteswap(vcpu)) {
1206                 host_swabbed = is_default_endian;
1207         } else {
1208                 host_swabbed = !is_default_endian;
1209         }
1210
1211         if (bytes > sizeof(run->mmio.data)) {
1212                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1213                        run->mmio.len);
1214         }
1215
1216         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1217         run->mmio.len = bytes;
1218         run->mmio.is_write = 0;
1219
1220         vcpu->arch.io_gpr = rt;
1221         vcpu->arch.mmio_host_swabbed = host_swabbed;
1222         vcpu->mmio_needed = 1;
1223         vcpu->mmio_is_write = 0;
1224         vcpu->arch.mmio_sign_extend = sign_extend;
1225
1226         idx = srcu_read_lock(&vcpu->kvm->srcu);
1227
1228         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1229                               bytes, &run->mmio.data);
1230
1231         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1232
1233         if (!ret) {
1234                 kvmppc_complete_mmio_load(vcpu, run);
1235                 vcpu->mmio_needed = 0;
1236                 return EMULATE_DONE;
1237         }
1238
1239         return EMULATE_DO_MMIO;
1240 }
1241
1242 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1243                        unsigned int rt, unsigned int bytes,
1244                        int is_default_endian)
1245 {
1246         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1247 }
1248 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1249
1250 /* Same as above, but sign extends */
1251 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1252                         unsigned int rt, unsigned int bytes,
1253                         int is_default_endian)
1254 {
1255         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1256 }
1257
1258 #ifdef CONFIG_VSX
1259 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1260                         unsigned int rt, unsigned int bytes,
1261                         int is_default_endian, int mmio_sign_extend)
1262 {
1263         enum emulation_result emulated = EMULATE_DONE;
1264
1265         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1266         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1267                 return EMULATE_FAIL;
1268
1269         while (vcpu->arch.mmio_vsx_copy_nums) {
1270                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1271                         is_default_endian, mmio_sign_extend);
1272
1273                 if (emulated != EMULATE_DONE)
1274                         break;
1275
1276                 vcpu->arch.paddr_accessed += run->mmio.len;
1277
1278                 vcpu->arch.mmio_vsx_copy_nums--;
1279                 vcpu->arch.mmio_vsx_offset++;
1280         }
1281         return emulated;
1282 }
1283 #endif /* CONFIG_VSX */
1284
1285 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1286                         u64 val, unsigned int bytes, int is_default_endian)
1287 {
1288         void *data = run->mmio.data;
1289         int idx, ret;
1290         bool host_swabbed;
1291
1292         /* Pity C doesn't have a logical XOR operator */
1293         if (kvmppc_need_byteswap(vcpu)) {
1294                 host_swabbed = is_default_endian;
1295         } else {
1296                 host_swabbed = !is_default_endian;
1297         }
1298
1299         if (bytes > sizeof(run->mmio.data)) {
1300                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1301                        run->mmio.len);
1302         }
1303
1304         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1305         run->mmio.len = bytes;
1306         run->mmio.is_write = 1;
1307         vcpu->mmio_needed = 1;
1308         vcpu->mmio_is_write = 1;
1309
1310         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1311                 val = dp_to_sp(val);
1312
1313         /* Store the value at the lowest bytes in 'data'. */
1314         if (!host_swabbed) {
1315                 switch (bytes) {
1316                 case 8: *(u64 *)data = val; break;
1317                 case 4: *(u32 *)data = val; break;
1318                 case 2: *(u16 *)data = val; break;
1319                 case 1: *(u8  *)data = val; break;
1320                 }
1321         } else {
1322                 switch (bytes) {
1323                 case 8: *(u64 *)data = swab64(val); break;
1324                 case 4: *(u32 *)data = swab32(val); break;
1325                 case 2: *(u16 *)data = swab16(val); break;
1326                 case 1: *(u8  *)data = val; break;
1327                 }
1328         }
1329
1330         idx = srcu_read_lock(&vcpu->kvm->srcu);
1331
1332         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1333                                bytes, &run->mmio.data);
1334
1335         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1336
1337         if (!ret) {
1338                 vcpu->mmio_needed = 0;
1339                 return EMULATE_DONE;
1340         }
1341
1342         return EMULATE_DO_MMIO;
1343 }
1344 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1345
1346 #ifdef CONFIG_VSX
1347 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1348 {
1349         u32 dword_offset, word_offset;
1350         union kvmppc_one_reg reg;
1351         int vsx_offset = 0;
1352         int copy_type = vcpu->arch.mmio_copy_type;
1353         int result = 0;
1354
1355         switch (copy_type) {
1356         case KVMPPC_VSX_COPY_DWORD:
1357                 vsx_offset =
1358                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1359
1360                 if (vsx_offset == -1) {
1361                         result = -1;
1362                         break;
1363                 }
1364
1365                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1366                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1367                 } else {
1368                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1369                         *val = reg.vsxval[vsx_offset];
1370                 }
1371                 break;
1372
1373         case KVMPPC_VSX_COPY_WORD:
1374                 vsx_offset =
1375                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1376
1377                 if (vsx_offset == -1) {
1378                         result = -1;
1379                         break;
1380                 }
1381
1382                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1383                         dword_offset = vsx_offset / 2;
1384                         word_offset = vsx_offset % 2;
1385                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1386                         *val = reg.vsx32val[word_offset];
1387                 } else {
1388                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1389                         *val = reg.vsx32val[vsx_offset];
1390                 }
1391                 break;
1392
1393         default:
1394                 result = -1;
1395                 break;
1396         }
1397
1398         return result;
1399 }
1400
1401 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1402                         int rs, unsigned int bytes, int is_default_endian)
1403 {
1404         u64 val;
1405         enum emulation_result emulated = EMULATE_DONE;
1406
1407         vcpu->arch.io_gpr = rs;
1408
1409         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1410         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1411                 return EMULATE_FAIL;
1412
1413         while (vcpu->arch.mmio_vsx_copy_nums) {
1414                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1415                         return EMULATE_FAIL;
1416
1417                 emulated = kvmppc_handle_store(run, vcpu,
1418                          val, bytes, is_default_endian);
1419
1420                 if (emulated != EMULATE_DONE)
1421                         break;
1422
1423                 vcpu->arch.paddr_accessed += run->mmio.len;
1424
1425                 vcpu->arch.mmio_vsx_copy_nums--;
1426                 vcpu->arch.mmio_vsx_offset++;
1427         }
1428
1429         return emulated;
1430 }
1431
1432 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1433                         struct kvm_run *run)
1434 {
1435         enum emulation_result emulated = EMULATE_FAIL;
1436         int r;
1437
1438         vcpu->arch.paddr_accessed += run->mmio.len;
1439
1440         if (!vcpu->mmio_is_write) {
1441                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1442                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1443         } else {
1444                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1445                          vcpu->arch.io_gpr, run->mmio.len, 1);
1446         }
1447
1448         switch (emulated) {
1449         case EMULATE_DO_MMIO:
1450                 run->exit_reason = KVM_EXIT_MMIO;
1451                 r = RESUME_HOST;
1452                 break;
1453         case EMULATE_FAIL:
1454                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1455                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1456                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1457                 r = RESUME_HOST;
1458                 break;
1459         default:
1460                 r = RESUME_GUEST;
1461                 break;
1462         }
1463         return r;
1464 }
1465 #endif /* CONFIG_VSX */
1466
1467 #ifdef CONFIG_ALTIVEC
1468 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1469                 unsigned int rt, unsigned int bytes, int is_default_endian)
1470 {
1471         enum emulation_result emulated = EMULATE_DONE;
1472
1473         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1474                 return EMULATE_FAIL;
1475
1476         while (vcpu->arch.mmio_vmx_copy_nums) {
1477                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1478                                 is_default_endian, 0);
1479
1480                 if (emulated != EMULATE_DONE)
1481                         break;
1482
1483                 vcpu->arch.paddr_accessed += run->mmio.len;
1484                 vcpu->arch.mmio_vmx_copy_nums--;
1485                 vcpu->arch.mmio_vmx_offset++;
1486         }
1487
1488         return emulated;
1489 }
1490
1491 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1492 {
1493         union kvmppc_one_reg reg;
1494         int vmx_offset = 0;
1495         int result = 0;
1496
1497         vmx_offset =
1498                 kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1499
1500         if (vmx_offset == -1)
1501                 return -1;
1502
1503         reg.vval = VCPU_VSX_VR(vcpu, index);
1504         *val = reg.vsxval[vmx_offset];
1505
1506         return result;
1507 }
1508
1509 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1510 {
1511         union kvmppc_one_reg reg;
1512         int vmx_offset = 0;
1513         int result = 0;
1514
1515         vmx_offset =
1516                 kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1517
1518         if (vmx_offset == -1)
1519                 return -1;
1520
1521         reg.vval = VCPU_VSX_VR(vcpu, index);
1522         *val = reg.vsx32val[vmx_offset];
1523
1524         return result;
1525 }
1526
1527 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1528 {
1529         union kvmppc_one_reg reg;
1530         int vmx_offset = 0;
1531         int result = 0;
1532
1533         vmx_offset =
1534                 kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1535
1536         if (vmx_offset == -1)
1537                 return -1;
1538
1539         reg.vval = VCPU_VSX_VR(vcpu, index);
1540         *val = reg.vsx16val[vmx_offset];
1541
1542         return result;
1543 }
1544
1545 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1546 {
1547         union kvmppc_one_reg reg;
1548         int vmx_offset = 0;
1549         int result = 0;
1550
1551         vmx_offset =
1552                 kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1553
1554         if (vmx_offset == -1)
1555                 return -1;
1556
1557         reg.vval = VCPU_VSX_VR(vcpu, index);
1558         *val = reg.vsx8val[vmx_offset];
1559
1560         return result;
1561 }
1562
1563 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1564                 unsigned int rs, unsigned int bytes, int is_default_endian)
1565 {
1566         u64 val = 0;
1567         unsigned int index = rs & KVM_MMIO_REG_MASK;
1568         enum emulation_result emulated = EMULATE_DONE;
1569
1570         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1571                 return EMULATE_FAIL;
1572
1573         vcpu->arch.io_gpr = rs;
1574
1575         while (vcpu->arch.mmio_vmx_copy_nums) {
1576                 switch (vcpu->arch.mmio_copy_type) {
1577                 case KVMPPC_VMX_COPY_DWORD:
1578                         if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1579                                 return EMULATE_FAIL;
1580
1581                         break;
1582                 case KVMPPC_VMX_COPY_WORD:
1583                         if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1584                                 return EMULATE_FAIL;
1585                         break;
1586                 case KVMPPC_VMX_COPY_HWORD:
1587                         if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1588                                 return EMULATE_FAIL;
1589                         break;
1590                 case KVMPPC_VMX_COPY_BYTE:
1591                         if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1592                                 return EMULATE_FAIL;
1593                         break;
1594                 default:
1595                         return EMULATE_FAIL;
1596                 }
1597
1598                 emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1599                                 is_default_endian);
1600                 if (emulated != EMULATE_DONE)
1601                         break;
1602
1603                 vcpu->arch.paddr_accessed += run->mmio.len;
1604                 vcpu->arch.mmio_vmx_copy_nums--;
1605                 vcpu->arch.mmio_vmx_offset++;
1606         }
1607
1608         return emulated;
1609 }
1610
1611 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1612                 struct kvm_run *run)
1613 {
1614         enum emulation_result emulated = EMULATE_FAIL;
1615         int r;
1616
1617         vcpu->arch.paddr_accessed += run->mmio.len;
1618
1619         if (!vcpu->mmio_is_write) {
1620                 emulated = kvmppc_handle_vmx_load(run, vcpu,
1621                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1622         } else {
1623                 emulated = kvmppc_handle_vmx_store(run, vcpu,
1624                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1625         }
1626
1627         switch (emulated) {
1628         case EMULATE_DO_MMIO:
1629                 run->exit_reason = KVM_EXIT_MMIO;
1630                 r = RESUME_HOST;
1631                 break;
1632         case EMULATE_FAIL:
1633                 pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1634                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1635                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1636                 r = RESUME_HOST;
1637                 break;
1638         default:
1639                 r = RESUME_GUEST;
1640                 break;
1641         }
1642         return r;
1643 }
1644 #endif /* CONFIG_ALTIVEC */
1645
1646 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1647 {
1648         int r = 0;
1649         union kvmppc_one_reg val;
1650         int size;
1651
1652         size = one_reg_size(reg->id);
1653         if (size > sizeof(val))
1654                 return -EINVAL;
1655
1656         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1657         if (r == -EINVAL) {
1658                 r = 0;
1659                 switch (reg->id) {
1660 #ifdef CONFIG_ALTIVEC
1661                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1662                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1663                                 r = -ENXIO;
1664                                 break;
1665                         }
1666                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1667                         break;
1668                 case KVM_REG_PPC_VSCR:
1669                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1670                                 r = -ENXIO;
1671                                 break;
1672                         }
1673                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1674                         break;
1675                 case KVM_REG_PPC_VRSAVE:
1676                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1677                         break;
1678 #endif /* CONFIG_ALTIVEC */
1679                 default:
1680                         r = -EINVAL;
1681                         break;
1682                 }
1683         }
1684
1685         if (r)
1686                 return r;
1687
1688         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1689                 r = -EFAULT;
1690
1691         return r;
1692 }
1693
1694 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1695 {
1696         int r;
1697         union kvmppc_one_reg val;
1698         int size;
1699
1700         size = one_reg_size(reg->id);
1701         if (size > sizeof(val))
1702                 return -EINVAL;
1703
1704         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1705                 return -EFAULT;
1706
1707         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1708         if (r == -EINVAL) {
1709                 r = 0;
1710                 switch (reg->id) {
1711 #ifdef CONFIG_ALTIVEC
1712                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1713                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1714                                 r = -ENXIO;
1715                                 break;
1716                         }
1717                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1718                         break;
1719                 case KVM_REG_PPC_VSCR:
1720                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1721                                 r = -ENXIO;
1722                                 break;
1723                         }
1724                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1725                         break;
1726                 case KVM_REG_PPC_VRSAVE:
1727                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1728                                 r = -ENXIO;
1729                                 break;
1730                         }
1731                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1732                         break;
1733 #endif /* CONFIG_ALTIVEC */
1734                 default:
1735                         r = -EINVAL;
1736                         break;
1737                 }
1738         }
1739
1740         return r;
1741 }
1742
1743 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1744 {
1745         int r;
1746
1747         vcpu_load(vcpu);
1748
1749         if (vcpu->mmio_needed) {
1750                 vcpu->mmio_needed = 0;
1751                 if (!vcpu->mmio_is_write)
1752                         kvmppc_complete_mmio_load(vcpu, run);
1753 #ifdef CONFIG_VSX
1754                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1755                         vcpu->arch.mmio_vsx_copy_nums--;
1756                         vcpu->arch.mmio_vsx_offset++;
1757                 }
1758
1759                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1760                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1761                         if (r == RESUME_HOST) {
1762                                 vcpu->mmio_needed = 1;
1763                                 goto out;
1764                         }
1765                 }
1766 #endif
1767 #ifdef CONFIG_ALTIVEC
1768                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1769                         vcpu->arch.mmio_vmx_copy_nums--;
1770                         vcpu->arch.mmio_vmx_offset++;
1771                 }
1772
1773                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1774                         r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1775                         if (r == RESUME_HOST) {
1776                                 vcpu->mmio_needed = 1;
1777                                 goto out;
1778                         }
1779                 }
1780 #endif
1781         } else if (vcpu->arch.osi_needed) {
1782                 u64 *gprs = run->osi.gprs;
1783                 int i;
1784
1785                 for (i = 0; i < 32; i++)
1786                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1787                 vcpu->arch.osi_needed = 0;
1788         } else if (vcpu->arch.hcall_needed) {
1789                 int i;
1790
1791                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1792                 for (i = 0; i < 9; ++i)
1793                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1794                 vcpu->arch.hcall_needed = 0;
1795 #ifdef CONFIG_BOOKE
1796         } else if (vcpu->arch.epr_needed) {
1797                 kvmppc_set_epr(vcpu, run->epr.epr);
1798                 vcpu->arch.epr_needed = 0;
1799 #endif
1800         }
1801
1802         kvm_sigset_activate(vcpu);
1803
1804         if (run->immediate_exit)
1805                 r = -EINTR;
1806         else
1807                 r = kvmppc_vcpu_run(run, vcpu);
1808
1809         kvm_sigset_deactivate(vcpu);
1810
1811 #ifdef CONFIG_ALTIVEC
1812 out:
1813 #endif
1814         vcpu_put(vcpu);
1815         return r;
1816 }
1817
1818 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1819 {
1820         if (irq->irq == KVM_INTERRUPT_UNSET) {
1821                 kvmppc_core_dequeue_external(vcpu);
1822                 return 0;
1823         }
1824
1825         kvmppc_core_queue_external(vcpu, irq);
1826
1827         kvm_vcpu_kick(vcpu);
1828
1829         return 0;
1830 }
1831
1832 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1833                                      struct kvm_enable_cap *cap)
1834 {
1835         int r;
1836
1837         if (cap->flags)
1838                 return -EINVAL;
1839
1840         switch (cap->cap) {
1841         case KVM_CAP_PPC_OSI:
1842                 r = 0;
1843                 vcpu->arch.osi_enabled = true;
1844                 break;
1845         case KVM_CAP_PPC_PAPR:
1846                 r = 0;
1847                 vcpu->arch.papr_enabled = true;
1848                 break;
1849         case KVM_CAP_PPC_EPR:
1850                 r = 0;
1851                 if (cap->args[0])
1852                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1853                 else
1854                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1855                 break;
1856 #ifdef CONFIG_BOOKE
1857         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1858                 r = 0;
1859                 vcpu->arch.watchdog_enabled = true;
1860                 break;
1861 #endif
1862 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1863         case KVM_CAP_SW_TLB: {
1864                 struct kvm_config_tlb cfg;
1865                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1866
1867                 r = -EFAULT;
1868                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1869                         break;
1870
1871                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1872                 break;
1873         }
1874 #endif
1875 #ifdef CONFIG_KVM_MPIC
1876         case KVM_CAP_IRQ_MPIC: {
1877                 struct fd f;
1878                 struct kvm_device *dev;
1879
1880                 r = -EBADF;
1881                 f = fdget(cap->args[0]);
1882                 if (!f.file)
1883                         break;
1884
1885                 r = -EPERM;
1886                 dev = kvm_device_from_filp(f.file);
1887                 if (dev)
1888                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1889
1890                 fdput(f);
1891                 break;
1892         }
1893 #endif
1894 #ifdef CONFIG_KVM_XICS
1895         case KVM_CAP_IRQ_XICS: {
1896                 struct fd f;
1897                 struct kvm_device *dev;
1898
1899                 r = -EBADF;
1900                 f = fdget(cap->args[0]);
1901                 if (!f.file)
1902                         break;
1903
1904                 r = -EPERM;
1905                 dev = kvm_device_from_filp(f.file);
1906                 if (dev) {
1907                         if (xive_enabled())
1908                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1909                         else
1910                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1911                 }
1912
1913                 fdput(f);
1914                 break;
1915         }
1916 #endif /* CONFIG_KVM_XICS */
1917 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1918         case KVM_CAP_PPC_FWNMI:
1919                 r = -EINVAL;
1920                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1921                         break;
1922                 r = 0;
1923                 vcpu->kvm->arch.fwnmi_enabled = true;
1924                 break;
1925 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1926         default:
1927                 r = -EINVAL;
1928                 break;
1929         }
1930
1931         if (!r)
1932                 r = kvmppc_sanity_check(vcpu);
1933
1934         return r;
1935 }
1936
1937 bool kvm_arch_intc_initialized(struct kvm *kvm)
1938 {
1939 #ifdef CONFIG_KVM_MPIC
1940         if (kvm->arch.mpic)
1941                 return true;
1942 #endif
1943 #ifdef CONFIG_KVM_XICS
1944         if (kvm->arch.xics || kvm->arch.xive)
1945                 return true;
1946 #endif
1947         return false;
1948 }
1949
1950 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1951                                     struct kvm_mp_state *mp_state)
1952 {
1953         return -EINVAL;
1954 }
1955
1956 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1957                                     struct kvm_mp_state *mp_state)
1958 {
1959         return -EINVAL;
1960 }
1961
1962 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1963                                unsigned int ioctl, unsigned long arg)
1964 {
1965         struct kvm_vcpu *vcpu = filp->private_data;
1966         void __user *argp = (void __user *)arg;
1967
1968         if (ioctl == KVM_INTERRUPT) {
1969                 struct kvm_interrupt irq;
1970                 if (copy_from_user(&irq, argp, sizeof(irq)))
1971                         return -EFAULT;
1972                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1973         }
1974         return -ENOIOCTLCMD;
1975 }
1976
1977 long kvm_arch_vcpu_ioctl(struct file *filp,
1978                          unsigned int ioctl, unsigned long arg)
1979 {
1980         struct kvm_vcpu *vcpu = filp->private_data;
1981         void __user *argp = (void __user *)arg;
1982         long r;
1983
1984         vcpu_load(vcpu);
1985
1986         switch (ioctl) {
1987         case KVM_ENABLE_CAP:
1988         {
1989                 struct kvm_enable_cap cap;
1990                 r = -EFAULT;
1991                 if (copy_from_user(&cap, argp, sizeof(cap)))
1992                         goto out;
1993                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1994                 break;
1995         }
1996
1997         case KVM_SET_ONE_REG:
1998         case KVM_GET_ONE_REG:
1999         {
2000                 struct kvm_one_reg reg;
2001                 r = -EFAULT;
2002                 if (copy_from_user(&reg, argp, sizeof(reg)))
2003                         goto out;
2004                 if (ioctl == KVM_SET_ONE_REG)
2005                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2006                 else
2007                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2008                 break;
2009         }
2010
2011 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2012         case KVM_DIRTY_TLB: {
2013                 struct kvm_dirty_tlb dirty;
2014                 r = -EFAULT;
2015                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
2016                         goto out;
2017                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2018                 break;
2019         }
2020 #endif
2021         default:
2022                 r = -EINVAL;
2023         }
2024
2025 out:
2026         vcpu_put(vcpu);
2027         return r;
2028 }
2029
2030 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2031 {
2032         return VM_FAULT_SIGBUS;
2033 }
2034
2035 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2036 {
2037         u32 inst_nop = 0x60000000;
2038 #ifdef CONFIG_KVM_BOOKE_HV
2039         u32 inst_sc1 = 0x44000022;
2040         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2041         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2042         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2043         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2044 #else
2045         u32 inst_lis = 0x3c000000;
2046         u32 inst_ori = 0x60000000;
2047         u32 inst_sc = 0x44000002;
2048         u32 inst_imm_mask = 0xffff;
2049
2050         /*
2051          * The hypercall to get into KVM from within guest context is as
2052          * follows:
2053          *
2054          *    lis r0, r0, KVM_SC_MAGIC_R0@h
2055          *    ori r0, KVM_SC_MAGIC_R0@l
2056          *    sc
2057          *    nop
2058          */
2059         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2060         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2061         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2062         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2063 #endif
2064
2065         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2066
2067         return 0;
2068 }
2069
2070 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2071                           bool line_status)
2072 {
2073         if (!irqchip_in_kernel(kvm))
2074                 return -ENXIO;
2075
2076         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2077                                         irq_event->irq, irq_event->level,
2078                                         line_status);
2079         return 0;
2080 }
2081
2082
2083 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2084                                    struct kvm_enable_cap *cap)
2085 {
2086         int r;
2087
2088         if (cap->flags)
2089                 return -EINVAL;
2090
2091         switch (cap->cap) {
2092 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2093         case KVM_CAP_PPC_ENABLE_HCALL: {
2094                 unsigned long hcall = cap->args[0];
2095
2096                 r = -EINVAL;
2097                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2098                     cap->args[1] > 1)
2099                         break;
2100                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2101                         break;
2102                 if (cap->args[1])
2103                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2104                 else
2105                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2106                 r = 0;
2107                 break;
2108         }
2109         case KVM_CAP_PPC_SMT: {
2110                 unsigned long mode = cap->args[0];
2111                 unsigned long flags = cap->args[1];
2112
2113                 r = -EINVAL;
2114                 if (kvm->arch.kvm_ops->set_smt_mode)
2115                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2116                 break;
2117         }
2118 #endif
2119         default:
2120                 r = -EINVAL;
2121                 break;
2122         }
2123
2124         return r;
2125 }
2126
2127 #ifdef CONFIG_PPC_BOOK3S_64
2128 /*
2129  * These functions check whether the underlying hardware is safe
2130  * against attacks based on observing the effects of speculatively
2131  * executed instructions, and whether it supplies instructions for
2132  * use in workarounds.  The information comes from firmware, either
2133  * via the device tree on powernv platforms or from an hcall on
2134  * pseries platforms.
2135  */
2136 #ifdef CONFIG_PPC_PSERIES
2137 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2138 {
2139         struct h_cpu_char_result c;
2140         unsigned long rc;
2141
2142         if (!machine_is(pseries))
2143                 return -ENOTTY;
2144
2145         rc = plpar_get_cpu_characteristics(&c);
2146         if (rc == H_SUCCESS) {
2147                 cp->character = c.character;
2148                 cp->behaviour = c.behaviour;
2149                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2150                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2151                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2152                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2153                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2154                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2155                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2156                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2157                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2158                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2159                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2160         }
2161         return 0;
2162 }
2163 #else
2164 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2165 {
2166         return -ENOTTY;
2167 }
2168 #endif
2169
2170 static inline bool have_fw_feat(struct device_node *fw_features,
2171                                 const char *state, const char *name)
2172 {
2173         struct device_node *np;
2174         bool r = false;
2175
2176         np = of_get_child_by_name(fw_features, name);
2177         if (np) {
2178                 r = of_property_read_bool(np, state);
2179                 of_node_put(np);
2180         }
2181         return r;
2182 }
2183
2184 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2185 {
2186         struct device_node *np, *fw_features;
2187         int r;
2188
2189         memset(cp, 0, sizeof(*cp));
2190         r = pseries_get_cpu_char(cp);
2191         if (r != -ENOTTY)
2192                 return r;
2193
2194         np = of_find_node_by_name(NULL, "ibm,opal");
2195         if (np) {
2196                 fw_features = of_get_child_by_name(np, "fw-features");
2197                 of_node_put(np);
2198                 if (!fw_features)
2199                         return 0;
2200                 if (have_fw_feat(fw_features, "enabled",
2201                                  "inst-spec-barrier-ori31,31,0"))
2202                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2203                 if (have_fw_feat(fw_features, "enabled",
2204                                  "fw-bcctrl-serialized"))
2205                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2206                 if (have_fw_feat(fw_features, "enabled",
2207                                  "inst-l1d-flush-ori30,30,0"))
2208                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2209                 if (have_fw_feat(fw_features, "enabled",
2210                                  "inst-l1d-flush-trig2"))
2211                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2212                 if (have_fw_feat(fw_features, "enabled",
2213                                  "fw-l1d-thread-split"))
2214                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2215                 if (have_fw_feat(fw_features, "enabled",
2216                                  "fw-count-cache-disabled"))
2217                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2218                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2219                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2220                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2221                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2222                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2223                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2224
2225                 if (have_fw_feat(fw_features, "enabled",
2226                                  "speculation-policy-favor-security"))
2227                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2228                 if (!have_fw_feat(fw_features, "disabled",
2229                                   "needs-l1d-flush-msr-pr-0-to-1"))
2230                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2231                 if (!have_fw_feat(fw_features, "disabled",
2232                                   "needs-spec-barrier-for-bound-checks"))
2233                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2234                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2235                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2236                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2237
2238                 of_node_put(fw_features);
2239         }
2240
2241         return 0;
2242 }
2243 #endif
2244
2245 long kvm_arch_vm_ioctl(struct file *filp,
2246                        unsigned int ioctl, unsigned long arg)
2247 {
2248         struct kvm *kvm __maybe_unused = filp->private_data;
2249         void __user *argp = (void __user *)arg;
2250         long r;
2251
2252         switch (ioctl) {
2253         case KVM_PPC_GET_PVINFO: {
2254                 struct kvm_ppc_pvinfo pvinfo;
2255                 memset(&pvinfo, 0, sizeof(pvinfo));
2256                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2257                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2258                         r = -EFAULT;
2259                         goto out;
2260                 }
2261
2262                 break;
2263         }
2264         case KVM_ENABLE_CAP:
2265         {
2266                 struct kvm_enable_cap cap;
2267                 r = -EFAULT;
2268                 if (copy_from_user(&cap, argp, sizeof(cap)))
2269                         goto out;
2270                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
2271                 break;
2272         }
2273 #ifdef CONFIG_SPAPR_TCE_IOMMU
2274         case KVM_CREATE_SPAPR_TCE_64: {
2275                 struct kvm_create_spapr_tce_64 create_tce_64;
2276
2277                 r = -EFAULT;
2278                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2279                         goto out;
2280                 if (create_tce_64.flags) {
2281                         r = -EINVAL;
2282                         goto out;
2283                 }
2284                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2285                 goto out;
2286         }
2287         case KVM_CREATE_SPAPR_TCE: {
2288                 struct kvm_create_spapr_tce create_tce;
2289                 struct kvm_create_spapr_tce_64 create_tce_64;
2290
2291                 r = -EFAULT;
2292                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2293                         goto out;
2294
2295                 create_tce_64.liobn = create_tce.liobn;
2296                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2297                 create_tce_64.offset = 0;
2298                 create_tce_64.size = create_tce.window_size >>
2299                                 IOMMU_PAGE_SHIFT_4K;
2300                 create_tce_64.flags = 0;
2301                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2302                 goto out;
2303         }
2304 #endif
2305 #ifdef CONFIG_PPC_BOOK3S_64
2306         case KVM_PPC_GET_SMMU_INFO: {
2307                 struct kvm_ppc_smmu_info info;
2308                 struct kvm *kvm = filp->private_data;
2309
2310                 memset(&info, 0, sizeof(info));
2311                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2312                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2313                         r = -EFAULT;
2314                 break;
2315         }
2316         case KVM_PPC_RTAS_DEFINE_TOKEN: {
2317                 struct kvm *kvm = filp->private_data;
2318
2319                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2320                 break;
2321         }
2322         case KVM_PPC_CONFIGURE_V3_MMU: {
2323                 struct kvm *kvm = filp->private_data;
2324                 struct kvm_ppc_mmuv3_cfg cfg;
2325
2326                 r = -EINVAL;
2327                 if (!kvm->arch.kvm_ops->configure_mmu)
2328                         goto out;
2329                 r = -EFAULT;
2330                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
2331                         goto out;
2332                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2333                 break;
2334         }
2335         case KVM_PPC_GET_RMMU_INFO: {
2336                 struct kvm *kvm = filp->private_data;
2337                 struct kvm_ppc_rmmu_info info;
2338
2339                 r = -EINVAL;
2340                 if (!kvm->arch.kvm_ops->get_rmmu_info)
2341                         goto out;
2342                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2343                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2344                         r = -EFAULT;
2345                 break;
2346         }
2347         case KVM_PPC_GET_CPU_CHAR: {
2348                 struct kvm_ppc_cpu_char cpuchar;
2349
2350                 r = kvmppc_get_cpu_char(&cpuchar);
2351                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2352                         r = -EFAULT;
2353                 break;
2354         }
2355         default: {
2356                 struct kvm *kvm = filp->private_data;
2357                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2358         }
2359 #else /* CONFIG_PPC_BOOK3S_64 */
2360         default:
2361                 r = -ENOTTY;
2362 #endif
2363         }
2364 out:
2365         return r;
2366 }
2367
2368 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2369 static unsigned long nr_lpids;
2370
2371 long kvmppc_alloc_lpid(void)
2372 {
2373         long lpid;
2374
2375         do {
2376                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2377                 if (lpid >= nr_lpids) {
2378                         pr_err("%s: No LPIDs free\n", __func__);
2379                         return -ENOMEM;
2380                 }
2381         } while (test_and_set_bit(lpid, lpid_inuse));
2382
2383         return lpid;
2384 }
2385 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2386
2387 void kvmppc_claim_lpid(long lpid)
2388 {
2389         set_bit(lpid, lpid_inuse);
2390 }
2391 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2392
2393 void kvmppc_free_lpid(long lpid)
2394 {
2395         clear_bit(lpid, lpid_inuse);
2396 }
2397 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2398
2399 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2400 {
2401         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2402         memset(lpid_inuse, 0, sizeof(lpid_inuse));
2403 }
2404 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2405
2406 int kvm_arch_init(void *opaque)
2407 {
2408         return 0;
2409 }
2410
2411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);