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 <[email protected]>
  18  *          Christian Ehrhardt <[email protected]>
  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(struct kvm_vcpu *vcpu,
 911         u32 gpr32)
 912 {
 913         union kvmppc_one_reg val;
 914         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
 915         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 916         int dword_offset, word_offset;
 917
 918         if (offset == -1)
 919                 return;
 920
 921         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 922                 val.vval = VCPU_VSX_VR(vcpu, index);
 923                 val.vsx32val[offset] = gpr32;
 924                 VCPU_VSX_VR(vcpu, index) = val.vval;
 925         } else {
 926                 dword_offset = offset / 2;
 927                 word_offset = offset % 2;
 928                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
 929                 val.vsx32val[word_offset] = gpr32;
 930                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
 931         }
 932 }
 933 #endif /* CONFIG_VSX */
 934
 935 #ifdef CONFIG_ALTIVEC
 936 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
 937                 u64 gpr)
 938 {
 939         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 940         u32 hi, lo;
 941         u32 di;
 942
 943 #ifdef __BIG_ENDIAN
 944         hi = gpr >> 32;
 945         lo = gpr & 0xffffffff;
 946 #else
 947         lo = gpr >> 32;
 948         hi = gpr & 0xffffffff;
 949 #endif
 950
 951         di = 2 - vcpu->arch.mmio_vmx_copy_nums;         /* doubleword index */
 952         if (di > 1)
 953                 return;
 954
 955         if (vcpu->arch.mmio_host_swabbed)
 956                 di = 1 - di;
 957
 958         VCPU_VSX_VR(vcpu, index).u[di * 2] = hi;
 959         VCPU_VSX_VR(vcpu, index).u[di * 2 + 1] = lo;
 960 }
 961 #endif /* CONFIG_ALTIVEC */
 962
 963 #ifdef CONFIG_PPC_FPU
 964 static inline u64 sp_to_dp(u32 fprs)
 965 {
 966         u64 fprd;
 967
 968         preempt_disable();
 969         enable_kernel_fp();
 970         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
 971              : "fr0");
 972         preempt_enable();
 973         return fprd;
 974 }
 975
 976 static inline u32 dp_to_sp(u64 fprd)
 977 {
 978         u32 fprs;
 979
 980         preempt_disable();
 981         enable_kernel_fp();
 982         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
 983              : "fr0");
 984         preempt_enable();
 985         return fprs;
 986 }
 987
 988 #else
 989 #define sp_to_dp(x)     (x)
 990 #define dp_to_sp(x)     (x)
 991 #endif /* CONFIG_PPC_FPU */
 992
 993 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 994                                       struct kvm_run *run)
 995 {
 996         u64 uninitialized_var(gpr);
 997
 998         if (run->mmio.len > sizeof(gpr)) {
 999                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1000                 return;
1001         }
1002
1003         if (!vcpu->arch.mmio_host_swabbed) {
1004                 switch (run->mmio.len) {
1005                 case 8: gpr = *(u64 *)run->mmio.data; break;
1006                 case 4: gpr = *(u32 *)run->mmio.data; break;
1007                 case 2: gpr = *(u16 *)run->mmio.data; break;
1008                 case 1: gpr = *(u8 *)run->mmio.data; break;
1009                 }
1010         } else {
1011                 switch (run->mmio.len) {
1012                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1013                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1014                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1015                 case 1: gpr = *(u8 *)run->mmio.data; break;
1016                 }
1017         }
1018
1019         /* conversion between single and double precision */
1020         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1021                 gpr = sp_to_dp(gpr);
1022
1023         if (vcpu->arch.mmio_sign_extend) {
1024                 switch (run->mmio.len) {
1025 #ifdef CONFIG_PPC64
1026                 case 4:
1027                         gpr = (s64)(s32)gpr;
1028                         break;
1029 #endif
1030                 case 2:
1031                         gpr = (s64)(s16)gpr;
1032                         break;
1033                 case 1:
1034                         gpr = (s64)(s8)gpr;
1035                         break;
1036                 }
1037         }
1038
1039         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1040         case KVM_MMIO_REG_GPR:
1041                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1042                 break;
1043         case KVM_MMIO_REG_FPR:
1044                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1045                 break;
1046 #ifdef CONFIG_PPC_BOOK3S
1047         case KVM_MMIO_REG_QPR:
1048                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1049                 break;
1050         case KVM_MMIO_REG_FQPR:
1051                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1052                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1053                 break;
1054 #endif
1055 #ifdef CONFIG_VSX
1056         case KVM_MMIO_REG_VSX:
1057                 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1058                         kvmppc_set_vsr_dword(vcpu, gpr);
1059                 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1060                         kvmppc_set_vsr_word(vcpu, gpr);
1061                 else if (vcpu->arch.mmio_vsx_copy_type ==
1062                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1063                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1064                 break;
1065 #endif
1066 #ifdef CONFIG_ALTIVEC
1067         case KVM_MMIO_REG_VMX:
1068                 kvmppc_set_vmx_dword(vcpu, gpr);
1069                 break;
1070 #endif
1071         default:
1072                 BUG();
1073         }
1074 }
1075
1076 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1077                                 unsigned int rt, unsigned int bytes,
1078                                 int is_default_endian, int sign_extend)
1079 {
1080         int idx, ret;
1081         bool host_swabbed;
1082
1083         /* Pity C doesn't have a logical XOR operator */
1084         if (kvmppc_need_byteswap(vcpu)) {
1085                 host_swabbed = is_default_endian;
1086         } else {
1087                 host_swabbed = !is_default_endian;
1088         }
1089
1090         if (bytes > sizeof(run->mmio.data)) {
1091                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1092                        run->mmio.len);
1093         }
1094
1095         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1096         run->mmio.len = bytes;
1097         run->mmio.is_write = 0;
1098
1099         vcpu->arch.io_gpr = rt;
1100         vcpu->arch.mmio_host_swabbed = host_swabbed;
1101         vcpu->mmio_needed = 1;
1102         vcpu->mmio_is_write = 0;
1103         vcpu->arch.mmio_sign_extend = sign_extend;
1104
1105         idx = srcu_read_lock(&vcpu->kvm->srcu);
1106
1107         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1108                               bytes, &run->mmio.data);
1109
1110         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1111
1112         if (!ret) {
1113                 kvmppc_complete_mmio_load(vcpu, run);
1114                 vcpu->mmio_needed = 0;
1115                 return EMULATE_DONE;
1116         }
1117
1118         return EMULATE_DO_MMIO;
1119 }
1120
1121 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1122                        unsigned int rt, unsigned int bytes,
1123                        int is_default_endian)
1124 {
1125         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1126 }
1127 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1128
1129 /* Same as above, but sign extends */
1130 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1131                         unsigned int rt, unsigned int bytes,
1132                         int is_default_endian)
1133 {
1134         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1135 }
1136
1137 #ifdef CONFIG_VSX
1138 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1139                         unsigned int rt, unsigned int bytes,
1140                         int is_default_endian, int mmio_sign_extend)
1141 {
1142         enum emulation_result emulated = EMULATE_DONE;
1143
1144         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1145         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1146                 return EMULATE_FAIL;
1147
1148         while (vcpu->arch.mmio_vsx_copy_nums) {
1149                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1150                         is_default_endian, mmio_sign_extend);
1151
1152                 if (emulated != EMULATE_DONE)
1153                         break;
1154
1155                 vcpu->arch.paddr_accessed += run->mmio.len;
1156
1157                 vcpu->arch.mmio_vsx_copy_nums--;
1158                 vcpu->arch.mmio_vsx_offset++;
1159         }
1160         return emulated;
1161 }
1162 #endif /* CONFIG_VSX */
1163
1164 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1165                         u64 val, unsigned int bytes, int is_default_endian)
1166 {
1167         void *data = run->mmio.data;
1168         int idx, ret;
1169         bool host_swabbed;
1170
1171         /* Pity C doesn't have a logical XOR operator */
1172         if (kvmppc_need_byteswap(vcpu)) {
1173                 host_swabbed = is_default_endian;
1174         } else {
1175                 host_swabbed = !is_default_endian;
1176         }
1177
1178         if (bytes > sizeof(run->mmio.data)) {
1179                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1180                        run->mmio.len);
1181         }
1182
1183         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1184         run->mmio.len = bytes;
1185         run->mmio.is_write = 1;
1186         vcpu->mmio_needed = 1;
1187         vcpu->mmio_is_write = 1;
1188
1189         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1190                 val = dp_to_sp(val);
1191
1192         /* Store the value at the lowest bytes in 'data'. */
1193         if (!host_swabbed) {
1194                 switch (bytes) {
1195                 case 8: *(u64 *)data = val; break;
1196                 case 4: *(u32 *)data = val; break;
1197                 case 2: *(u16 *)data = val; break;
1198                 case 1: *(u8  *)data = val; break;
1199                 }
1200         } else {
1201                 switch (bytes) {
1202                 case 8: *(u64 *)data = swab64(val); break;
1203                 case 4: *(u32 *)data = swab32(val); break;
1204                 case 2: *(u16 *)data = swab16(val); break;
1205                 case 1: *(u8  *)data = val; break;
1206                 }
1207         }
1208
1209         idx = srcu_read_lock(&vcpu->kvm->srcu);
1210
1211         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1212                                bytes, &run->mmio.data);
1213
1214         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1215
1216         if (!ret) {
1217                 vcpu->mmio_needed = 0;
1218                 return EMULATE_DONE;
1219         }
1220
1221         return EMULATE_DO_MMIO;
1222 }
1223 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1224
1225 #ifdef CONFIG_VSX
1226 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1227 {
1228         u32 dword_offset, word_offset;
1229         union kvmppc_one_reg reg;
1230         int vsx_offset = 0;
1231         int copy_type = vcpu->arch.mmio_vsx_copy_type;
1232         int result = 0;
1233
1234         switch (copy_type) {
1235         case KVMPPC_VSX_COPY_DWORD:
1236                 vsx_offset =
1237                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1238
1239                 if (vsx_offset == -1) {
1240                         result = -1;
1241                         break;
1242                 }
1243
1244                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1245                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1246                 } else {
1247                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1248                         *val = reg.vsxval[vsx_offset];
1249                 }
1250                 break;
1251
1252         case KVMPPC_VSX_COPY_WORD:
1253                 vsx_offset =
1254                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1255
1256                 if (vsx_offset == -1) {
1257                         result = -1;
1258                         break;
1259                 }
1260
1261                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1262                         dword_offset = vsx_offset / 2;
1263                         word_offset = vsx_offset % 2;
1264                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1265                         *val = reg.vsx32val[word_offset];
1266                 } else {
1267                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1268                         *val = reg.vsx32val[vsx_offset];
1269                 }
1270                 break;
1271
1272         default:
1273                 result = -1;
1274                 break;
1275         }
1276
1277         return result;
1278 }
1279
1280 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1281                         int rs, unsigned int bytes, int is_default_endian)
1282 {
1283         u64 val;
1284         enum emulation_result emulated = EMULATE_DONE;
1285
1286         vcpu->arch.io_gpr = rs;
1287
1288         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1289         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1290                 return EMULATE_FAIL;
1291
1292         while (vcpu->arch.mmio_vsx_copy_nums) {
1293                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1294                         return EMULATE_FAIL;
1295
1296                 emulated = kvmppc_handle_store(run, vcpu,
1297                          val, bytes, is_default_endian);
1298
1299                 if (emulated != EMULATE_DONE)
1300                         break;
1301
1302                 vcpu->arch.paddr_accessed += run->mmio.len;
1303
1304                 vcpu->arch.mmio_vsx_copy_nums--;
1305                 vcpu->arch.mmio_vsx_offset++;
1306         }
1307
1308         return emulated;
1309 }
1310
1311 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1312                         struct kvm_run *run)
1313 {
1314         enum emulation_result emulated = EMULATE_FAIL;
1315         int r;
1316
1317         vcpu->arch.paddr_accessed += run->mmio.len;
1318
1319         if (!vcpu->mmio_is_write) {
1320                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1321                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1322         } else {
1323                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1324                          vcpu->arch.io_gpr, run->mmio.len, 1);
1325         }
1326
1327         switch (emulated) {
1328         case EMULATE_DO_MMIO:
1329                 run->exit_reason = KVM_EXIT_MMIO;
1330                 r = RESUME_HOST;
1331                 break;
1332         case EMULATE_FAIL:
1333                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1334                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1335                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1336                 r = RESUME_HOST;
1337                 break;
1338         default:
1339                 r = RESUME_GUEST;
1340                 break;
1341         }
1342         return r;
1343 }
1344 #endif /* CONFIG_VSX */
1345
1346 #ifdef CONFIG_ALTIVEC
1347 /* handle quadword load access in two halves */
1348 int kvmppc_handle_load128_by2x64(struct kvm_run *run, struct kvm_vcpu *vcpu,
1349                 unsigned int rt, int is_default_endian)
1350 {
1351         enum emulation_result emulated = EMULATE_DONE;
1352
1353         while (vcpu->arch.mmio_vmx_copy_nums) {
1354                 emulated = __kvmppc_handle_load(run, vcpu, rt, 8,
1355                                 is_default_endian, 0);
1356
1357                 if (emulated != EMULATE_DONE)
1358                         break;
1359
1360                 vcpu->arch.paddr_accessed += run->mmio.len;
1361                 vcpu->arch.mmio_vmx_copy_nums--;
1362         }
1363
1364         return emulated;
1365 }
1366
1367 static inline int kvmppc_get_vmx_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1368 {
1369         vector128 vrs = VCPU_VSX_VR(vcpu, rs);
1370         u32 di;
1371         u64 w0, w1;
1372
1373         di = 2 - vcpu->arch.mmio_vmx_copy_nums;         /* doubleword index */
1374         if (di > 1)
1375                 return -1;
1376
1377         if (vcpu->arch.mmio_host_swabbed)
1378                 di = 1 - di;
1379
1380         w0 = vrs.u[di * 2];
1381         w1 = vrs.u[di * 2 + 1];
1382
1383 #ifdef __BIG_ENDIAN
1384         *val = (w0 << 32) | w1;
1385 #else
1386         *val = (w1 << 32) | w0;
1387 #endif
1388         return 0;
1389 }
1390
1391 /* handle quadword store in two halves */
1392 int kvmppc_handle_store128_by2x64(struct kvm_run *run, struct kvm_vcpu *vcpu,
1393                 unsigned int rs, int is_default_endian)
1394 {
1395         u64 val = 0;
1396         enum emulation_result emulated = EMULATE_DONE;
1397
1398         vcpu->arch.io_gpr = rs;
1399
1400         while (vcpu->arch.mmio_vmx_copy_nums) {
1401                 if (kvmppc_get_vmx_data(vcpu, rs, &val) == -1)
1402                         return EMULATE_FAIL;
1403
1404                 emulated = kvmppc_handle_store(run, vcpu, val, 8,
1405                                 is_default_endian);
1406                 if (emulated != EMULATE_DONE)
1407                         break;
1408
1409                 vcpu->arch.paddr_accessed += run->mmio.len;
1410                 vcpu->arch.mmio_vmx_copy_nums--;
1411         }
1412
1413         return emulated;
1414 }
1415
1416 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1417                 struct kvm_run *run)
1418 {
1419         enum emulation_result emulated = EMULATE_FAIL;
1420         int r;
1421
1422         vcpu->arch.paddr_accessed += run->mmio.len;
1423
1424         if (!vcpu->mmio_is_write) {
1425                 emulated = kvmppc_handle_load128_by2x64(run, vcpu,
1426                                 vcpu->arch.io_gpr, 1);
1427         } else {
1428                 emulated = kvmppc_handle_store128_by2x64(run, vcpu,
1429                                 vcpu->arch.io_gpr, 1);
1430         }
1431
1432         switch (emulated) {
1433         case EMULATE_DO_MMIO:
1434                 run->exit_reason = KVM_EXIT_MMIO;
1435                 r = RESUME_HOST;
1436                 break;
1437         case EMULATE_FAIL:
1438                 pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1439                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1440                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1441                 r = RESUME_HOST;
1442                 break;
1443         default:
1444                 r = RESUME_GUEST;
1445                 break;
1446         }
1447         return r;
1448 }
1449 #endif /* CONFIG_ALTIVEC */
1450
1451 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1452 {
1453         int r = 0;
1454         union kvmppc_one_reg val;
1455         int size;
1456
1457         size = one_reg_size(reg->id);
1458         if (size > sizeof(val))
1459                 return -EINVAL;
1460
1461         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1462         if (r == -EINVAL) {
1463                 r = 0;
1464                 switch (reg->id) {
1465 #ifdef CONFIG_ALTIVEC
1466                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1467                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1468                                 r = -ENXIO;
1469                                 break;
1470                         }
1471                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1472                         break;
1473                 case KVM_REG_PPC_VSCR:
1474                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1475                                 r = -ENXIO;
1476                                 break;
1477                         }
1478                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1479                         break;
1480                 case KVM_REG_PPC_VRSAVE:
1481                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1482                         break;
1483 #endif /* CONFIG_ALTIVEC */
1484                 default:
1485                         r = -EINVAL;
1486                         break;
1487                 }
1488         }
1489
1490         if (r)
1491                 return r;
1492
1493         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1494                 r = -EFAULT;
1495
1496         return r;
1497 }
1498
1499 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1500 {
1501         int r;
1502         union kvmppc_one_reg val;
1503         int size;
1504
1505         size = one_reg_size(reg->id);
1506         if (size > sizeof(val))
1507                 return -EINVAL;
1508
1509         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1510                 return -EFAULT;
1511
1512         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1513         if (r == -EINVAL) {
1514                 r = 0;
1515                 switch (reg->id) {
1516 #ifdef CONFIG_ALTIVEC
1517                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1518                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1519                                 r = -ENXIO;
1520                                 break;
1521                         }
1522                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1523                         break;
1524                 case KVM_REG_PPC_VSCR:
1525                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1526                                 r = -ENXIO;
1527                                 break;
1528                         }
1529                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1530                         break;
1531                 case KVM_REG_PPC_VRSAVE:
1532                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1533                                 r = -ENXIO;
1534                                 break;
1535                         }
1536                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1537                         break;
1538 #endif /* CONFIG_ALTIVEC */
1539                 default:
1540                         r = -EINVAL;
1541                         break;
1542                 }
1543         }
1544
1545         return r;
1546 }
1547
1548 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1549 {
1550         int r;
1551
1552         vcpu_load(vcpu);
1553
1554         if (vcpu->mmio_needed) {
1555                 vcpu->mmio_needed = 0;
1556                 if (!vcpu->mmio_is_write)
1557                         kvmppc_complete_mmio_load(vcpu, run);
1558 #ifdef CONFIG_VSX
1559                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1560                         vcpu->arch.mmio_vsx_copy_nums--;
1561                         vcpu->arch.mmio_vsx_offset++;
1562                 }
1563
1564                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1565                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1566                         if (r == RESUME_HOST) {
1567                                 vcpu->mmio_needed = 1;
1568                                 goto out;
1569                         }
1570                 }
1571 #endif
1572 #ifdef CONFIG_ALTIVEC
1573                 if (vcpu->arch.mmio_vmx_copy_nums > 0)
1574                         vcpu->arch.mmio_vmx_copy_nums--;
1575
1576                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1577                         r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1578                         if (r == RESUME_HOST) {
1579                                 vcpu->mmio_needed = 1;
1580                                 goto out;
1581                         }
1582                 }
1583 #endif
1584         } else if (vcpu->arch.osi_needed) {
1585                 u64 *gprs = run->osi.gprs;
1586                 int i;
1587
1588                 for (i = 0; i < 32; i++)
1589                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1590                 vcpu->arch.osi_needed = 0;
1591         } else if (vcpu->arch.hcall_needed) {
1592                 int i;
1593
1594                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1595                 for (i = 0; i < 9; ++i)
1596                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1597                 vcpu->arch.hcall_needed = 0;
1598 #ifdef CONFIG_BOOKE
1599         } else if (vcpu->arch.epr_needed) {
1600                 kvmppc_set_epr(vcpu, run->epr.epr);
1601                 vcpu->arch.epr_needed = 0;
1602 #endif
1603         }
1604
1605         kvm_sigset_activate(vcpu);
1606
1607         if (run->immediate_exit)
1608                 r = -EINTR;
1609         else
1610                 r = kvmppc_vcpu_run(run, vcpu);
1611
1612         kvm_sigset_deactivate(vcpu);
1613
1614 #ifdef CONFIG_ALTIVEC
1615 out:
1616 #endif
1617         vcpu_put(vcpu);
1618         return r;
1619 }
1620
1621 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1622 {
1623         if (irq->irq == KVM_INTERRUPT_UNSET) {
1624                 kvmppc_core_dequeue_external(vcpu);
1625                 return 0;
1626         }
1627
1628         kvmppc_core_queue_external(vcpu, irq);
1629
1630         kvm_vcpu_kick(vcpu);
1631
1632         return 0;
1633 }
1634
1635 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1636                                      struct kvm_enable_cap *cap)
1637 {
1638         int r;
1639
1640         if (cap->flags)
1641                 return -EINVAL;
1642
1643         switch (cap->cap) {
1644         case KVM_CAP_PPC_OSI:
1645                 r = 0;
1646                 vcpu->arch.osi_enabled = true;
1647                 break;
1648         case KVM_CAP_PPC_PAPR:
1649                 r = 0;
1650                 vcpu->arch.papr_enabled = true;
1651                 break;
1652         case KVM_CAP_PPC_EPR:
1653                 r = 0;
1654                 if (cap->args[0])
1655                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1656                 else
1657                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1658                 break;
1659 #ifdef CONFIG_BOOKE
1660         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1661                 r = 0;
1662                 vcpu->arch.watchdog_enabled = true;
1663                 break;
1664 #endif
1665 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1666         case KVM_CAP_SW_TLB: {
1667                 struct kvm_config_tlb cfg;
1668                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1669
1670                 r = -EFAULT;
1671                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1672                         break;
1673
1674                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1675                 break;
1676         }
1677 #endif
1678 #ifdef CONFIG_KVM_MPIC
1679         case KVM_CAP_IRQ_MPIC: {
1680                 struct fd f;
1681                 struct kvm_device *dev;
1682
1683                 r = -EBADF;
1684                 f = fdget(cap->args[0]);
1685                 if (!f.file)
1686                         break;
1687
1688                 r = -EPERM;
1689                 dev = kvm_device_from_filp(f.file);
1690                 if (dev)
1691                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1692
1693                 fdput(f);
1694                 break;
1695         }
1696 #endif
1697 #ifdef CONFIG_KVM_XICS
1698         case KVM_CAP_IRQ_XICS: {
1699                 struct fd f;
1700                 struct kvm_device *dev;
1701
1702                 r = -EBADF;
1703                 f = fdget(cap->args[0]);
1704                 if (!f.file)
1705                         break;
1706
1707                 r = -EPERM;
1708                 dev = kvm_device_from_filp(f.file);
1709                 if (dev) {
1710                         if (xive_enabled())
1711                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1712                         else
1713                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1714                 }
1715
1716                 fdput(f);
1717                 break;
1718         }
1719 #endif /* CONFIG_KVM_XICS */
1720 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1721         case KVM_CAP_PPC_FWNMI:
1722                 r = -EINVAL;
1723                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1724                         break;
1725                 r = 0;
1726                 vcpu->kvm->arch.fwnmi_enabled = true;
1727                 break;
1728 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1729         default:
1730                 r = -EINVAL;
1731                 break;
1732         }
1733
1734         if (!r)
1735                 r = kvmppc_sanity_check(vcpu);
1736
1737         return r;
1738 }
1739
1740 bool kvm_arch_intc_initialized(struct kvm *kvm)
1741 {
1742 #ifdef CONFIG_KVM_MPIC
1743         if (kvm->arch.mpic)
1744                 return true;
1745 #endif
1746 #ifdef CONFIG_KVM_XICS
1747         if (kvm->arch.xics || kvm->arch.xive)
1748                 return true;
1749 #endif
1750         return false;
1751 }
1752
1753 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1754                                     struct kvm_mp_state *mp_state)
1755 {
1756         return -EINVAL;
1757 }
1758
1759 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1760                                     struct kvm_mp_state *mp_state)
1761 {
1762         return -EINVAL;
1763 }
1764
1765 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1766                                unsigned int ioctl, unsigned long arg)
1767 {
1768         struct kvm_vcpu *vcpu = filp->private_data;
1769         void __user *argp = (void __user *)arg;
1770
1771         if (ioctl == KVM_INTERRUPT) {
1772                 struct kvm_interrupt irq;
1773                 if (copy_from_user(&irq, argp, sizeof(irq)))
1774                         return -EFAULT;
1775                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1776         }
1777         return -ENOIOCTLCMD;
1778 }
1779
1780 long kvm_arch_vcpu_ioctl(struct file *filp,
1781                          unsigned int ioctl, unsigned long arg)
1782 {
1783         struct kvm_vcpu *vcpu = filp->private_data;
1784         void __user *argp = (void __user *)arg;
1785         long r;
1786
1787         vcpu_load(vcpu);
1788
1789         switch (ioctl) {
1790         case KVM_ENABLE_CAP:
1791         {
1792                 struct kvm_enable_cap cap;
1793                 r = -EFAULT;
1794                 if (copy_from_user(&cap, argp, sizeof(cap)))
1795                         goto out;
1796                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1797                 break;
1798         }
1799
1800         case KVM_SET_ONE_REG:
1801         case KVM_GET_ONE_REG:
1802         {
1803                 struct kvm_one_reg reg;
1804                 r = -EFAULT;
1805                 if (copy_from_user(&reg, argp, sizeof(reg)))
1806                         goto out;
1807                 if (ioctl == KVM_SET_ONE_REG)
1808                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1809                 else
1810                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1811                 break;
1812         }
1813
1814 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1815         case KVM_DIRTY_TLB: {
1816                 struct kvm_dirty_tlb dirty;
1817                 r = -EFAULT;
1818                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1819                         goto out;
1820                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1821                 break;
1822         }
1823 #endif
1824         default:
1825                 r = -EINVAL;
1826         }
1827
1828 out:
1829         vcpu_put(vcpu);
1830         return r;
1831 }
1832
1833 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1834 {
1835         return VM_FAULT_SIGBUS;
1836 }
1837
1838 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1839 {
1840         u32 inst_nop = 0x60000000;
1841 #ifdef CONFIG_KVM_BOOKE_HV
1842         u32 inst_sc1 = 0x44000022;
1843         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1844         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1845         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1846         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1847 #else
1848         u32 inst_lis = 0x3c000000;
1849         u32 inst_ori = 0x60000000;
1850         u32 inst_sc = 0x44000002;
1851         u32 inst_imm_mask = 0xffff;
1852
1853         /*
1854          * The hypercall to get into KVM from within guest context is as
1855          * follows:
1856          *
1857          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1858          *    ori r0, KVM_SC_MAGIC_R0@l
1859          *    sc
1860          *    nop
1861          */
1862         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1863         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1864         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1865         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1866 #endif
1867
1868         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1869
1870         return 0;
1871 }
1872
1873 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1874                           bool line_status)
1875 {
1876         if (!irqchip_in_kernel(kvm))
1877                 return -ENXIO;
1878
1879         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1880                                         irq_event->irq, irq_event->level,
1881                                         line_status);
1882         return 0;
1883 }
1884
1885
1886 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1887                                    struct kvm_enable_cap *cap)
1888 {
1889         int r;
1890
1891         if (cap->flags)
1892                 return -EINVAL;
1893
1894         switch (cap->cap) {
1895 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1896         case KVM_CAP_PPC_ENABLE_HCALL: {
1897                 unsigned long hcall = cap->args[0];
1898
1899                 r = -EINVAL;
1900                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1901                     cap->args[1] > 1)
1902                         break;
1903                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1904                         break;
1905                 if (cap->args[1])
1906                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1907                 else
1908                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1909                 r = 0;
1910                 break;
1911         }
1912         case KVM_CAP_PPC_SMT: {
1913                 unsigned long mode = cap->args[0];
1914                 unsigned long flags = cap->args[1];
1915
1916                 r = -EINVAL;
1917                 if (kvm->arch.kvm_ops->set_smt_mode)
1918                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
1919                 break;
1920         }
1921 #endif
1922         default:
1923                 r = -EINVAL;
1924                 break;
1925         }
1926
1927         return r;
1928 }
1929
1930 #ifdef CONFIG_PPC_BOOK3S_64
1931 /*
1932  * These functions check whether the underlying hardware is safe
1933  * against attacks based on observing the effects of speculatively
1934  * executed instructions, and whether it supplies instructions for
1935  * use in workarounds.  The information comes from firmware, either
1936  * via the device tree on powernv platforms or from an hcall on
1937  * pseries platforms.
1938  */
1939 #ifdef CONFIG_PPC_PSERIES
1940 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1941 {
1942         struct h_cpu_char_result c;
1943         unsigned long rc;
1944
1945         if (!machine_is(pseries))
1946                 return -ENOTTY;
1947
1948         rc = plpar_get_cpu_characteristics(&c);
1949         if (rc == H_SUCCESS) {
1950                 cp->character = c.character;
1951                 cp->behaviour = c.behaviour;
1952                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
1953                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
1954                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
1955                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
1956                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
1957                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
1958                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
1959                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1960                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
1961                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
1962                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1963         }
1964         return 0;
1965 }
1966 #else
1967 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1968 {
1969         return -ENOTTY;
1970 }
1971 #endif
1972
1973 static inline bool have_fw_feat(struct device_node *fw_features,
1974                                 const char *state, const char *name)
1975 {
1976         struct device_node *np;
1977         bool r = false;
1978
1979         np = of_get_child_by_name(fw_features, name);
1980         if (np) {
1981                 r = of_property_read_bool(np, state);
1982                 of_node_put(np);
1983         }
1984         return r;
1985 }
1986
1987 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1988 {
1989         struct device_node *np, *fw_features;
1990         int r;
1991
1992         memset(cp, 0, sizeof(*cp));
1993         r = pseries_get_cpu_char(cp);
1994         if (r != -ENOTTY)
1995                 return r;
1996
1997         np = of_find_node_by_name(NULL, "ibm,opal");
1998         if (np) {
1999                 fw_features = of_get_child_by_name(np, "fw-features");
2000                 of_node_put(np);
2001                 if (!fw_features)
2002                         return 0;
2003                 if (have_fw_feat(fw_features, "enabled",
2004                                  "inst-spec-barrier-ori31,31,0"))
2005                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2006                 if (have_fw_feat(fw_features, "enabled",
2007                                  "fw-bcctrl-serialized"))
2008                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2009                 if (have_fw_feat(fw_features, "enabled",
2010                                  "inst-l1d-flush-ori30,30,0"))
2011                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2012                 if (have_fw_feat(fw_features, "enabled",
2013                                  "inst-l1d-flush-trig2"))
2014                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2015                 if (have_fw_feat(fw_features, "enabled",
2016                                  "fw-l1d-thread-split"))
2017                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2018                 if (have_fw_feat(fw_features, "enabled",
2019                                  "fw-count-cache-disabled"))
2020                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2021                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2022                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2023                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2024                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2025                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2026                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2027
2028                 if (have_fw_feat(fw_features, "enabled",
2029                                  "speculation-policy-favor-security"))
2030                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2031                 if (!have_fw_feat(fw_features, "disabled",
2032                                   "needs-l1d-flush-msr-pr-0-to-1"))
2033                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2034                 if (!have_fw_feat(fw_features, "disabled",
2035                                   "needs-spec-barrier-for-bound-checks"))
2036                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2037                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2038                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2039                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2040
2041                 of_node_put(fw_features);
2042         }
2043
2044         return 0;
2045 }
2046 #endif
2047
2048 long kvm_arch_vm_ioctl(struct file *filp,
2049                        unsigned int ioctl, unsigned long arg)
2050 {
2051         struct kvm *kvm __maybe_unused = filp->private_data;
2052         void __user *argp = (void __user *)arg;
2053         long r;
2054
2055         switch (ioctl) {
2056         case KVM_PPC_GET_PVINFO: {
2057                 struct kvm_ppc_pvinfo pvinfo;
2058                 memset(&pvinfo, 0, sizeof(pvinfo));
2059                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2060                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2061                         r = -EFAULT;
2062                         goto out;
2063                 }
2064
2065                 break;
2066         }
2067         case KVM_ENABLE_CAP:
2068         {
2069                 struct kvm_enable_cap cap;
2070                 r = -EFAULT;
2071                 if (copy_from_user(&cap, argp, sizeof(cap)))
2072                         goto out;
2073                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
2074                 break;
2075         }
2076 #ifdef CONFIG_SPAPR_TCE_IOMMU
2077         case KVM_CREATE_SPAPR_TCE_64: {
2078                 struct kvm_create_spapr_tce_64 create_tce_64;
2079
2080                 r = -EFAULT;
2081                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2082                         goto out;
2083                 if (create_tce_64.flags) {
2084                         r = -EINVAL;
2085                         goto out;
2086                 }
2087                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2088                 goto out;
2089         }
2090         case KVM_CREATE_SPAPR_TCE: {
2091                 struct kvm_create_spapr_tce create_tce;
2092                 struct kvm_create_spapr_tce_64 create_tce_64;
2093
2094                 r = -EFAULT;
2095                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2096                         goto out;
2097
2098                 create_tce_64.liobn = create_tce.liobn;
2099                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2100                 create_tce_64.offset = 0;
2101                 create_tce_64.size = create_tce.window_size >>
2102                                 IOMMU_PAGE_SHIFT_4K;
2103                 create_tce_64.flags = 0;
2104                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2105                 goto out;
2106         }
2107 #endif
2108 #ifdef CONFIG_PPC_BOOK3S_64
2109         case KVM_PPC_GET_SMMU_INFO: {
2110                 struct kvm_ppc_smmu_info info;
2111                 struct kvm *kvm = filp->private_data;
2112
2113                 memset(&info, 0, sizeof(info));
2114                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2115                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2116                         r = -EFAULT;
2117                 break;
2118         }
2119         case KVM_PPC_RTAS_DEFINE_TOKEN: {
2120                 struct kvm *kvm = filp->private_data;
2121
2122                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2123                 break;
2124         }
2125         case KVM_PPC_CONFIGURE_V3_MMU: {
2126                 struct kvm *kvm = filp->private_data;
2127                 struct kvm_ppc_mmuv3_cfg cfg;
2128
2129                 r = -EINVAL;
2130                 if (!kvm->arch.kvm_ops->configure_mmu)
2131                         goto out;
2132                 r = -EFAULT;
2133                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
2134                         goto out;
2135                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2136                 break;
2137         }
2138         case KVM_PPC_GET_RMMU_INFO: {
2139                 struct kvm *kvm = filp->private_data;
2140                 struct kvm_ppc_rmmu_info info;
2141
2142                 r = -EINVAL;
2143                 if (!kvm->arch.kvm_ops->get_rmmu_info)
2144                         goto out;
2145                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2146                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2147                         r = -EFAULT;
2148                 break;
2149         }
2150         case KVM_PPC_GET_CPU_CHAR: {
2151                 struct kvm_ppc_cpu_char cpuchar;
2152
2153                 r = kvmppc_get_cpu_char(&cpuchar);
2154                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2155                         r = -EFAULT;
2156                 break;
2157         }
2158         default: {
2159                 struct kvm *kvm = filp->private_data;
2160                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2161         }
2162 #else /* CONFIG_PPC_BOOK3S_64 */
2163         default:
2164                 r = -ENOTTY;
2165 #endif
2166         }
2167 out:
2168         return r;
2169 }
2170
2171 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2172 static unsigned long nr_lpids;
2173
2174 long kvmppc_alloc_lpid(void)
2175 {
2176         long lpid;
2177
2178         do {
2179                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2180                 if (lpid >= nr_lpids) {
2181                         pr_err("%s: No LPIDs free\n", __func__);
2182                         return -ENOMEM;
2183                 }
2184         } while (test_and_set_bit(lpid, lpid_inuse));
2185
2186         return lpid;
2187 }
2188 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2189
2190 void kvmppc_claim_lpid(long lpid)
2191 {
2192         set_bit(lpid, lpid_inuse);
2193 }
2194 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2195
2196 void kvmppc_free_lpid(long lpid)
2197 {
2198         clear_bit(lpid, lpid_inuse);
2199 }
2200 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2201
2202 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2203 {
2204         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2205         memset(lpid_inuse, 0, sizeof(lpid_inuse));
2206 }
2207 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2208
2209 int kvm_arch_init(void *opaque)
2210 {
2211         return 0;
2212 }
2213
2214 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);