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                 /* Disable this on POWER9 until code handles new HPTE format */
 642                 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
 643                 break;
 644 #endif
 645 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 646         case KVM_CAP_PPC_FWNMI:
 647                 r = hv_enabled;
 648                 break;
 649 #endif
 650         case KVM_CAP_PPC_HTM:
 651                 r = hv_enabled &&
 652                     (cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM_COMP);
 653                 break;
 654         default:
 655                 r = 0;
 656                 break;
 657         }
 658         return r;
 659
 660 }
 661
 662 long kvm_arch_dev_ioctl(struct file *filp,
 663                         unsigned int ioctl, unsigned long arg)
 664 {
 665         return -EINVAL;
 666 }
 667
 668 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 669                            struct kvm_memory_slot *dont)
 670 {
 671         kvmppc_core_free_memslot(kvm, free, dont);
 672 }
 673
 674 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 675                             unsigned long npages)
 676 {
 677         return kvmppc_core_create_memslot(kvm, slot, npages);
 678 }
 679
 680 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 681                                    struct kvm_memory_slot *memslot,
 682                                    const struct kvm_userspace_memory_region *mem,
 683                                    enum kvm_mr_change change)
 684 {
 685         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 686 }
 687
 688 void kvm_arch_commit_memory_region(struct kvm *kvm,
 689                                    const struct kvm_userspace_memory_region *mem,
 690                                    const struct kvm_memory_slot *old,
 691                                    const struct kvm_memory_slot *new,
 692                                    enum kvm_mr_change change)
 693 {
 694         kvmppc_core_commit_memory_region(kvm, mem, old, new);
 695 }
 696
 697 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 698                                    struct kvm_memory_slot *slot)
 699 {
 700         kvmppc_core_flush_memslot(kvm, slot);
 701 }
 702
 703 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 704 {
 705         struct kvm_vcpu *vcpu;
 706         vcpu = kvmppc_core_vcpu_create(kvm, id);
 707         if (!IS_ERR(vcpu)) {
 708                 vcpu->arch.wqp = &vcpu->wq;
 709                 kvmppc_create_vcpu_debugfs(vcpu, id);
 710         }
 711         return vcpu;
 712 }
 713
 714 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 715 {
 716 }
 717
 718 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 719 {
 720         /* Make sure we're not using the vcpu anymore */
 721         hrtimer_cancel(&vcpu->arch.dec_timer);
 722
 723         kvmppc_remove_vcpu_debugfs(vcpu);
 724
 725         switch (vcpu->arch.irq_type) {
 726         case KVMPPC_IRQ_MPIC:
 727                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 728                 break;
 729         case KVMPPC_IRQ_XICS:
 730                 if (xive_enabled())
 731                         kvmppc_xive_cleanup_vcpu(vcpu);
 732                 else
 733                         kvmppc_xics_free_icp(vcpu);
 734                 break;
 735         }
 736
 737         kvmppc_core_vcpu_free(vcpu);
 738 }
 739
 740 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 741 {
 742         kvm_arch_vcpu_free(vcpu);
 743 }
 744
 745 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 746 {
 747         return kvmppc_core_pending_dec(vcpu);
 748 }
 749
 750 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 751 {
 752         struct kvm_vcpu *vcpu;
 753
 754         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 755         kvmppc_decrementer_func(vcpu);
 756
 757         return HRTIMER_NORESTART;
 758 }
 759
 760 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 761 {
 762         int ret;
 763
 764         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 765         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 766         vcpu->arch.dec_expires = ~(u64)0;
 767
 768 #ifdef CONFIG_KVM_EXIT_TIMING
 769         mutex_init(&vcpu->arch.exit_timing_lock);
 770 #endif
 771         ret = kvmppc_subarch_vcpu_init(vcpu);
 772         return ret;
 773 }
 774
 775 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 776 {
 777         kvmppc_mmu_destroy(vcpu);
 778         kvmppc_subarch_vcpu_uninit(vcpu);
 779 }
 780
 781 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 782 {
 783 #ifdef CONFIG_BOOKE
 784         /*
 785          * vrsave (formerly usprg0) isn't used by Linux, but may
 786          * be used by the guest.
 787          *
 788          * On non-booke this is associated with Altivec and
 789          * is handled by code in book3s.c.
 790          */
 791         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 792 #endif
 793         kvmppc_core_vcpu_load(vcpu, cpu);
 794 }
 795
 796 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 797 {
 798         kvmppc_core_vcpu_put(vcpu);
 799 #ifdef CONFIG_BOOKE
 800         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 801 #endif
 802 }
 803
 804 /*
 805  * irq_bypass_add_producer and irq_bypass_del_producer are only
 806  * useful if the architecture supports PCI passthrough.
 807  * irq_bypass_stop and irq_bypass_start are not needed and so
 808  * kvm_ops are not defined for them.
 809  */
 810 bool kvm_arch_has_irq_bypass(void)
 811 {
 812         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
 813                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
 814 }
 815
 816 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
 817                                      struct irq_bypass_producer *prod)
 818 {
 819         struct kvm_kernel_irqfd *irqfd =
 820                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 821         struct kvm *kvm = irqfd->kvm;
 822
 823         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
 824                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
 825
 826         return 0;
 827 }
 828
 829 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
 830                                       struct irq_bypass_producer *prod)
 831 {
 832         struct kvm_kernel_irqfd *irqfd =
 833                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 834         struct kvm *kvm = irqfd->kvm;
 835
 836         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
 837                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
 838 }
 839
 840 #ifdef CONFIG_VSX
 841 static inline int kvmppc_get_vsr_dword_offset(int index)
 842 {
 843         int offset;
 844
 845         if ((index != 0) && (index != 1))
 846                 return -1;
 847
 848 #ifdef __BIG_ENDIAN
 849         offset =  index;
 850 #else
 851         offset = 1 - index;
 852 #endif
 853
 854         return offset;
 855 }
 856
 857 static inline int kvmppc_get_vsr_word_offset(int index)
 858 {
 859         int offset;
 860
 861         if ((index > 3) || (index < 0))
 862                 return -1;
 863
 864 #ifdef __BIG_ENDIAN
 865         offset = index;
 866 #else
 867         offset = 3 - index;
 868 #endif
 869         return offset;
 870 }
 871
 872 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
 873         u64 gpr)
 874 {
 875         union kvmppc_one_reg val;
 876         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
 877         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 878
 879         if (offset == -1)
 880                 return;
 881
 882         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 883                 val.vval = VCPU_VSX_VR(vcpu, index);
 884                 val.vsxval[offset] = gpr;
 885                 VCPU_VSX_VR(vcpu, index) = val.vval;
 886         } else {
 887                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
 888         }
 889 }
 890
 891 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
 892         u64 gpr)
 893 {
 894         union kvmppc_one_reg val;
 895         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 896
 897         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 898                 val.vval = VCPU_VSX_VR(vcpu, index);
 899                 val.vsxval[0] = gpr;
 900                 val.vsxval[1] = gpr;
 901                 VCPU_VSX_VR(vcpu, index) = val.vval;
 902         } else {
 903                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
 904                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
 905         }
 906 }
 907
 908 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
 909         u32 gpr32)
 910 {
 911         union kvmppc_one_reg val;
 912         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
 913         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 914         int dword_offset, word_offset;
 915
 916         if (offset == -1)
 917                 return;
 918
 919         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 920                 val.vval = VCPU_VSX_VR(vcpu, index);
 921                 val.vsx32val[offset] = gpr32;
 922                 VCPU_VSX_VR(vcpu, index) = val.vval;
 923         } else {
 924                 dword_offset = offset / 2;
 925                 word_offset = offset % 2;
 926                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
 927                 val.vsx32val[word_offset] = gpr32;
 928                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
 929         }
 930 }
 931 #endif /* CONFIG_VSX */
 932
 933 #ifdef CONFIG_PPC_FPU
 934 static inline u64 sp_to_dp(u32 fprs)
 935 {
 936         u64 fprd;
 937
 938         preempt_disable();
 939         enable_kernel_fp();
 940         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
 941              : "fr0");
 942         preempt_enable();
 943         return fprd;
 944 }
 945
 946 static inline u32 dp_to_sp(u64 fprd)
 947 {
 948         u32 fprs;
 949
 950         preempt_disable();
 951         enable_kernel_fp();
 952         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
 953              : "fr0");
 954         preempt_enable();
 955         return fprs;
 956 }
 957
 958 #else
 959 #define sp_to_dp(x)     (x)
 960 #define dp_to_sp(x)     (x)
 961 #endif /* CONFIG_PPC_FPU */
 962
 963 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 964                                       struct kvm_run *run)
 965 {
 966         u64 uninitialized_var(gpr);
 967
 968         if (run->mmio.len > sizeof(gpr)) {
 969                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 970                 return;
 971         }
 972
 973         if (!vcpu->arch.mmio_host_swabbed) {
 974                 switch (run->mmio.len) {
 975                 case 8: gpr = *(u64 *)run->mmio.data; break;
 976                 case 4: gpr = *(u32 *)run->mmio.data; break;
 977                 case 2: gpr = *(u16 *)run->mmio.data; break;
 978                 case 1: gpr = *(u8 *)run->mmio.data; break;
 979                 }
 980         } else {
 981                 switch (run->mmio.len) {
 982                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
 983                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
 984                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
 985                 case 1: gpr = *(u8 *)run->mmio.data; break;
 986                 }
 987         }
 988
 989         /* conversion between single and double precision */
 990         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
 991                 gpr = sp_to_dp(gpr);
 992
 993         if (vcpu->arch.mmio_sign_extend) {
 994                 switch (run->mmio.len) {
 995 #ifdef CONFIG_PPC64
 996                 case 4:
 997                         gpr = (s64)(s32)gpr;
 998                         break;
 999 #endif
1000                 case 2:
1001                         gpr = (s64)(s16)gpr;
1002                         break;
1003                 case 1:
1004                         gpr = (s64)(s8)gpr;
1005                         break;
1006                 }
1007         }
1008
1009         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1010         case KVM_MMIO_REG_GPR:
1011                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1012                 break;
1013         case KVM_MMIO_REG_FPR:
1014                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1015                 break;
1016 #ifdef CONFIG_PPC_BOOK3S
1017         case KVM_MMIO_REG_QPR:
1018                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1019                 break;
1020         case KVM_MMIO_REG_FQPR:
1021                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1022                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1023                 break;
1024 #endif
1025 #ifdef CONFIG_VSX
1026         case KVM_MMIO_REG_VSX:
1027                 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1028                         kvmppc_set_vsr_dword(vcpu, gpr);
1029                 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1030                         kvmppc_set_vsr_word(vcpu, gpr);
1031                 else if (vcpu->arch.mmio_vsx_copy_type ==
1032                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1033                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1034                 break;
1035 #endif
1036         default:
1037                 BUG();
1038         }
1039 }
1040
1041 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1042                                 unsigned int rt, unsigned int bytes,
1043                                 int is_default_endian, int sign_extend)
1044 {
1045         int idx, ret;
1046         bool host_swabbed;
1047
1048         /* Pity C doesn't have a logical XOR operator */
1049         if (kvmppc_need_byteswap(vcpu)) {
1050                 host_swabbed = is_default_endian;
1051         } else {
1052                 host_swabbed = !is_default_endian;
1053         }
1054
1055         if (bytes > sizeof(run->mmio.data)) {
1056                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1057                        run->mmio.len);
1058         }
1059
1060         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1061         run->mmio.len = bytes;
1062         run->mmio.is_write = 0;
1063
1064         vcpu->arch.io_gpr = rt;
1065         vcpu->arch.mmio_host_swabbed = host_swabbed;
1066         vcpu->mmio_needed = 1;
1067         vcpu->mmio_is_write = 0;
1068         vcpu->arch.mmio_sign_extend = sign_extend;
1069
1070         idx = srcu_read_lock(&vcpu->kvm->srcu);
1071
1072         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1073                               bytes, &run->mmio.data);
1074
1075         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1076
1077         if (!ret) {
1078                 kvmppc_complete_mmio_load(vcpu, run);
1079                 vcpu->mmio_needed = 0;
1080                 return EMULATE_DONE;
1081         }
1082
1083         return EMULATE_DO_MMIO;
1084 }
1085
1086 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1087                        unsigned int rt, unsigned int bytes,
1088                        int is_default_endian)
1089 {
1090         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1091 }
1092 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1093
1094 /* Same as above, but sign extends */
1095 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1096                         unsigned int rt, unsigned int bytes,
1097                         int is_default_endian)
1098 {
1099         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1100 }
1101
1102 #ifdef CONFIG_VSX
1103 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1104                         unsigned int rt, unsigned int bytes,
1105                         int is_default_endian, int mmio_sign_extend)
1106 {
1107         enum emulation_result emulated = EMULATE_DONE;
1108
1109         /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1110         if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1111                 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1112                 return EMULATE_FAIL;
1113         }
1114
1115         while (vcpu->arch.mmio_vsx_copy_nums) {
1116                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1117                         is_default_endian, mmio_sign_extend);
1118
1119                 if (emulated != EMULATE_DONE)
1120                         break;
1121
1122                 vcpu->arch.paddr_accessed += run->mmio.len;
1123
1124                 vcpu->arch.mmio_vsx_copy_nums--;
1125                 vcpu->arch.mmio_vsx_offset++;
1126         }
1127         return emulated;
1128 }
1129 #endif /* CONFIG_VSX */
1130
1131 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1132                         u64 val, unsigned int bytes, int is_default_endian)
1133 {
1134         void *data = run->mmio.data;
1135         int idx, ret;
1136         bool host_swabbed;
1137
1138         /* Pity C doesn't have a logical XOR operator */
1139         if (kvmppc_need_byteswap(vcpu)) {
1140                 host_swabbed = is_default_endian;
1141         } else {
1142                 host_swabbed = !is_default_endian;
1143         }
1144
1145         if (bytes > sizeof(run->mmio.data)) {
1146                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1147                        run->mmio.len);
1148         }
1149
1150         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1151         run->mmio.len = bytes;
1152         run->mmio.is_write = 1;
1153         vcpu->mmio_needed = 1;
1154         vcpu->mmio_is_write = 1;
1155
1156         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1157                 val = dp_to_sp(val);
1158
1159         /* Store the value at the lowest bytes in 'data'. */
1160         if (!host_swabbed) {
1161                 switch (bytes) {
1162                 case 8: *(u64 *)data = val; break;
1163                 case 4: *(u32 *)data = val; break;
1164                 case 2: *(u16 *)data = val; break;
1165                 case 1: *(u8  *)data = val; break;
1166                 }
1167         } else {
1168                 switch (bytes) {
1169                 case 8: *(u64 *)data = swab64(val); break;
1170                 case 4: *(u32 *)data = swab32(val); break;
1171                 case 2: *(u16 *)data = swab16(val); break;
1172                 case 1: *(u8  *)data = val; break;
1173                 }
1174         }
1175
1176         idx = srcu_read_lock(&vcpu->kvm->srcu);
1177
1178         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1179                                bytes, &run->mmio.data);
1180
1181         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1182
1183         if (!ret) {
1184                 vcpu->mmio_needed = 0;
1185                 return EMULATE_DONE;
1186         }
1187
1188         return EMULATE_DO_MMIO;
1189 }
1190 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1191
1192 #ifdef CONFIG_VSX
1193 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1194 {
1195         u32 dword_offset, word_offset;
1196         union kvmppc_one_reg reg;
1197         int vsx_offset = 0;
1198         int copy_type = vcpu->arch.mmio_vsx_copy_type;
1199         int result = 0;
1200
1201         switch (copy_type) {
1202         case KVMPPC_VSX_COPY_DWORD:
1203                 vsx_offset =
1204                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1205
1206                 if (vsx_offset == -1) {
1207                         result = -1;
1208                         break;
1209                 }
1210
1211                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1212                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1213                 } else {
1214                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1215                         *val = reg.vsxval[vsx_offset];
1216                 }
1217                 break;
1218
1219         case KVMPPC_VSX_COPY_WORD:
1220                 vsx_offset =
1221                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1222
1223                 if (vsx_offset == -1) {
1224                         result = -1;
1225                         break;
1226                 }
1227
1228                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1229                         dword_offset = vsx_offset / 2;
1230                         word_offset = vsx_offset % 2;
1231                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1232                         *val = reg.vsx32val[word_offset];
1233                 } else {
1234                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1235                         *val = reg.vsx32val[vsx_offset];
1236                 }
1237                 break;
1238
1239         default:
1240                 result = -1;
1241                 break;
1242         }
1243
1244         return result;
1245 }
1246
1247 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1248                         int rs, unsigned int bytes, int is_default_endian)
1249 {
1250         u64 val;
1251         enum emulation_result emulated = EMULATE_DONE;
1252
1253         vcpu->arch.io_gpr = rs;
1254
1255         /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1256         if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1257                 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1258                 return EMULATE_FAIL;
1259         }
1260
1261         while (vcpu->arch.mmio_vsx_copy_nums) {
1262                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1263                         return EMULATE_FAIL;
1264
1265                 emulated = kvmppc_handle_store(run, vcpu,
1266                          val, bytes, is_default_endian);
1267
1268                 if (emulated != EMULATE_DONE)
1269                         break;
1270
1271                 vcpu->arch.paddr_accessed += run->mmio.len;
1272
1273                 vcpu->arch.mmio_vsx_copy_nums--;
1274                 vcpu->arch.mmio_vsx_offset++;
1275         }
1276
1277         return emulated;
1278 }
1279
1280 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1281                         struct kvm_run *run)
1282 {
1283         enum emulation_result emulated = EMULATE_FAIL;
1284         int r;
1285
1286         vcpu->arch.paddr_accessed += run->mmio.len;
1287
1288         if (!vcpu->mmio_is_write) {
1289                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1290                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1291         } else {
1292                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1293                          vcpu->arch.io_gpr, run->mmio.len, 1);
1294         }
1295
1296         switch (emulated) {
1297         case EMULATE_DO_MMIO:
1298                 run->exit_reason = KVM_EXIT_MMIO;
1299                 r = RESUME_HOST;
1300                 break;
1301         case EMULATE_FAIL:
1302                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1303                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1304                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1305                 r = RESUME_HOST;
1306                 break;
1307         default:
1308                 r = RESUME_GUEST;
1309                 break;
1310         }
1311         return r;
1312 }
1313 #endif /* CONFIG_VSX */
1314
1315 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1316 {
1317         int r = 0;
1318         union kvmppc_one_reg val;
1319         int size;
1320
1321         size = one_reg_size(reg->id);
1322         if (size > sizeof(val))
1323                 return -EINVAL;
1324
1325         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1326         if (r == -EINVAL) {
1327                 r = 0;
1328                 switch (reg->id) {
1329 #ifdef CONFIG_ALTIVEC
1330                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1331                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1332                                 r = -ENXIO;
1333                                 break;
1334                         }
1335                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1336                         break;
1337                 case KVM_REG_PPC_VSCR:
1338                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1339                                 r = -ENXIO;
1340                                 break;
1341                         }
1342                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1343                         break;
1344                 case KVM_REG_PPC_VRSAVE:
1345                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1346                         break;
1347 #endif /* CONFIG_ALTIVEC */
1348                 default:
1349                         r = -EINVAL;
1350                         break;
1351                 }
1352         }
1353
1354         if (r)
1355                 return r;
1356
1357         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1358                 r = -EFAULT;
1359
1360         return r;
1361 }
1362
1363 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1364 {
1365         int r;
1366         union kvmppc_one_reg val;
1367         int size;
1368
1369         size = one_reg_size(reg->id);
1370         if (size > sizeof(val))
1371                 return -EINVAL;
1372
1373         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1374                 return -EFAULT;
1375
1376         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1377         if (r == -EINVAL) {
1378                 r = 0;
1379                 switch (reg->id) {
1380 #ifdef CONFIG_ALTIVEC
1381                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1382                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1383                                 r = -ENXIO;
1384                                 break;
1385                         }
1386                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1387                         break;
1388                 case KVM_REG_PPC_VSCR:
1389                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1390                                 r = -ENXIO;
1391                                 break;
1392                         }
1393                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1394                         break;
1395                 case KVM_REG_PPC_VRSAVE:
1396                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1397                                 r = -ENXIO;
1398                                 break;
1399                         }
1400                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1401                         break;
1402 #endif /* CONFIG_ALTIVEC */
1403                 default:
1404                         r = -EINVAL;
1405                         break;
1406                 }
1407         }
1408
1409         return r;
1410 }
1411
1412 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1413 {
1414         int r;
1415
1416         if (vcpu->mmio_needed) {
1417                 vcpu->mmio_needed = 0;
1418                 if (!vcpu->mmio_is_write)
1419                         kvmppc_complete_mmio_load(vcpu, run);
1420 #ifdef CONFIG_VSX
1421                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1422                         vcpu->arch.mmio_vsx_copy_nums--;
1423                         vcpu->arch.mmio_vsx_offset++;
1424                 }
1425
1426                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1427                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1428                         if (r == RESUME_HOST) {
1429                                 vcpu->mmio_needed = 1;
1430                                 return r;
1431                         }
1432                 }
1433 #endif
1434         } else if (vcpu->arch.osi_needed) {
1435                 u64 *gprs = run->osi.gprs;
1436                 int i;
1437
1438                 for (i = 0; i < 32; i++)
1439                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1440                 vcpu->arch.osi_needed = 0;
1441         } else if (vcpu->arch.hcall_needed) {
1442                 int i;
1443
1444                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1445                 for (i = 0; i < 9; ++i)
1446                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1447                 vcpu->arch.hcall_needed = 0;
1448 #ifdef CONFIG_BOOKE
1449         } else if (vcpu->arch.epr_needed) {
1450                 kvmppc_set_epr(vcpu, run->epr.epr);
1451                 vcpu->arch.epr_needed = 0;
1452 #endif
1453         }
1454
1455         kvm_sigset_activate(vcpu);
1456
1457         if (run->immediate_exit)
1458                 r = -EINTR;
1459         else
1460                 r = kvmppc_vcpu_run(run, vcpu);
1461
1462         kvm_sigset_deactivate(vcpu);
1463
1464         return r;
1465 }
1466
1467 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1468 {
1469         if (irq->irq == KVM_INTERRUPT_UNSET) {
1470                 kvmppc_core_dequeue_external(vcpu);
1471                 return 0;
1472         }
1473
1474         kvmppc_core_queue_external(vcpu, irq);
1475
1476         kvm_vcpu_kick(vcpu);
1477
1478         return 0;
1479 }
1480
1481 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1482                                      struct kvm_enable_cap *cap)
1483 {
1484         int r;
1485
1486         if (cap->flags)
1487                 return -EINVAL;
1488
1489         switch (cap->cap) {
1490         case KVM_CAP_PPC_OSI:
1491                 r = 0;
1492                 vcpu->arch.osi_enabled = true;
1493                 break;
1494         case KVM_CAP_PPC_PAPR:
1495                 r = 0;
1496                 vcpu->arch.papr_enabled = true;
1497                 break;
1498         case KVM_CAP_PPC_EPR:
1499                 r = 0;
1500                 if (cap->args[0])
1501                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1502                 else
1503                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1504                 break;
1505 #ifdef CONFIG_BOOKE
1506         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1507                 r = 0;
1508                 vcpu->arch.watchdog_enabled = true;
1509                 break;
1510 #endif
1511 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1512         case KVM_CAP_SW_TLB: {
1513                 struct kvm_config_tlb cfg;
1514                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1515
1516                 r = -EFAULT;
1517                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1518                         break;
1519
1520                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1521                 break;
1522         }
1523 #endif
1524 #ifdef CONFIG_KVM_MPIC
1525         case KVM_CAP_IRQ_MPIC: {
1526                 struct fd f;
1527                 struct kvm_device *dev;
1528
1529                 r = -EBADF;
1530                 f = fdget(cap->args[0]);
1531                 if (!f.file)
1532                         break;
1533
1534                 r = -EPERM;
1535                 dev = kvm_device_from_filp(f.file);
1536                 if (dev)
1537                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1538
1539                 fdput(f);
1540                 break;
1541         }
1542 #endif
1543 #ifdef CONFIG_KVM_XICS
1544         case KVM_CAP_IRQ_XICS: {
1545                 struct fd f;
1546                 struct kvm_device *dev;
1547
1548                 r = -EBADF;
1549                 f = fdget(cap->args[0]);
1550                 if (!f.file)
1551                         break;
1552
1553                 r = -EPERM;
1554                 dev = kvm_device_from_filp(f.file);
1555                 if (dev) {
1556                         if (xive_enabled())
1557                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1558                         else
1559                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1560                 }
1561
1562                 fdput(f);
1563                 break;
1564         }
1565 #endif /* CONFIG_KVM_XICS */
1566 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1567         case KVM_CAP_PPC_FWNMI:
1568                 r = -EINVAL;
1569                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1570                         break;
1571                 r = 0;
1572                 vcpu->kvm->arch.fwnmi_enabled = true;
1573                 break;
1574 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1575         default:
1576                 r = -EINVAL;
1577                 break;
1578         }
1579
1580         if (!r)
1581                 r = kvmppc_sanity_check(vcpu);
1582
1583         return r;
1584 }
1585
1586 bool kvm_arch_intc_initialized(struct kvm *kvm)
1587 {
1588 #ifdef CONFIG_KVM_MPIC
1589         if (kvm->arch.mpic)
1590                 return true;
1591 #endif
1592 #ifdef CONFIG_KVM_XICS
1593         if (kvm->arch.xics || kvm->arch.xive)
1594                 return true;
1595 #endif
1596         return false;
1597 }
1598
1599 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1600                                     struct kvm_mp_state *mp_state)
1601 {
1602         return -EINVAL;
1603 }
1604
1605 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1606                                     struct kvm_mp_state *mp_state)
1607 {
1608         return -EINVAL;
1609 }
1610
1611 long kvm_arch_vcpu_ioctl(struct file *filp,
1612                          unsigned int ioctl, unsigned long arg)
1613 {
1614         struct kvm_vcpu *vcpu = filp->private_data;
1615         void __user *argp = (void __user *)arg;
1616         long r;
1617
1618         switch (ioctl) {
1619         case KVM_INTERRUPT: {
1620                 struct kvm_interrupt irq;
1621                 r = -EFAULT;
1622                 if (copy_from_user(&irq, argp, sizeof(irq)))
1623                         goto out;
1624                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1625                 goto out;
1626         }
1627
1628         case KVM_ENABLE_CAP:
1629         {
1630                 struct kvm_enable_cap cap;
1631                 r = -EFAULT;
1632                 if (copy_from_user(&cap, argp, sizeof(cap)))
1633                         goto out;
1634                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1635                 break;
1636         }
1637
1638         case KVM_SET_ONE_REG:
1639         case KVM_GET_ONE_REG:
1640         {
1641                 struct kvm_one_reg reg;
1642                 r = -EFAULT;
1643                 if (copy_from_user(&reg, argp, sizeof(reg)))
1644                         goto out;
1645                 if (ioctl == KVM_SET_ONE_REG)
1646                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1647                 else
1648                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1649                 break;
1650         }
1651
1652 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1653         case KVM_DIRTY_TLB: {
1654                 struct kvm_dirty_tlb dirty;
1655                 r = -EFAULT;
1656                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1657                         goto out;
1658                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1659                 break;
1660         }
1661 #endif
1662         default:
1663                 r = -EINVAL;
1664         }
1665
1666 out:
1667         return r;
1668 }
1669
1670 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1671 {
1672         return VM_FAULT_SIGBUS;
1673 }
1674
1675 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1676 {
1677         u32 inst_nop = 0x60000000;
1678 #ifdef CONFIG_KVM_BOOKE_HV
1679         u32 inst_sc1 = 0x44000022;
1680         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1681         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1682         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1683         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1684 #else
1685         u32 inst_lis = 0x3c000000;
1686         u32 inst_ori = 0x60000000;
1687         u32 inst_sc = 0x44000002;
1688         u32 inst_imm_mask = 0xffff;
1689
1690         /*
1691          * The hypercall to get into KVM from within guest context is as
1692          * follows:
1693          *
1694          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1695          *    ori r0, KVM_SC_MAGIC_R0@l
1696          *    sc
1697          *    nop
1698          */
1699         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1700         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1701         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1702         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1703 #endif
1704
1705         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1706
1707         return 0;
1708 }
1709
1710 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1711                           bool line_status)
1712 {
1713         if (!irqchip_in_kernel(kvm))
1714                 return -ENXIO;
1715
1716         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1717                                         irq_event->irq, irq_event->level,
1718                                         line_status);
1719         return 0;
1720 }
1721
1722
1723 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1724                                    struct kvm_enable_cap *cap)
1725 {
1726         int r;
1727
1728         if (cap->flags)
1729                 return -EINVAL;
1730
1731         switch (cap->cap) {
1732 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1733         case KVM_CAP_PPC_ENABLE_HCALL: {
1734                 unsigned long hcall = cap->args[0];
1735
1736                 r = -EINVAL;
1737                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1738                     cap->args[1] > 1)
1739                         break;
1740                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1741                         break;
1742                 if (cap->args[1])
1743                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1744                 else
1745                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1746                 r = 0;
1747                 break;
1748         }
1749         case KVM_CAP_PPC_SMT: {
1750                 unsigned long mode = cap->args[0];
1751                 unsigned long flags = cap->args[1];
1752
1753                 r = -EINVAL;
1754                 if (kvm->arch.kvm_ops->set_smt_mode)
1755                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
1756                 break;
1757         }
1758 #endif
1759         default:
1760                 r = -EINVAL;
1761                 break;
1762         }
1763
1764         return r;
1765 }
1766
1767 #ifdef CONFIG_PPC_BOOK3S_64
1768 /*
1769  * These functions check whether the underlying hardware is safe
1770  * against attacks based on observing the effects of speculatively
1771  * executed instructions, and whether it supplies instructions for
1772  * use in workarounds.  The information comes from firmware, either
1773  * via the device tree on powernv platforms or from an hcall on
1774  * pseries platforms.
1775  */
1776 #ifdef CONFIG_PPC_PSERIES
1777 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1778 {
1779         struct h_cpu_char_result c;
1780         unsigned long rc;
1781
1782         if (!machine_is(pseries))
1783                 return -ENOTTY;
1784
1785         rc = plpar_get_cpu_characteristics(&c);
1786         if (rc == H_SUCCESS) {
1787                 cp->character = c.character;
1788                 cp->behaviour = c.behaviour;
1789                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
1790                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
1791                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
1792                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
1793                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
1794                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
1795                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
1796                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1797                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
1798                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
1799                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1800         }
1801         return 0;
1802 }
1803 #else
1804 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1805 {
1806         return -ENOTTY;
1807 }
1808 #endif
1809
1810 static inline bool have_fw_feat(struct device_node *fw_features,
1811                                 const char *state, const char *name)
1812 {
1813         struct device_node *np;
1814         bool r = false;
1815
1816         np = of_get_child_by_name(fw_features, name);
1817         if (np) {
1818                 r = of_property_read_bool(np, state);
1819                 of_node_put(np);
1820         }
1821         return r;
1822 }
1823
1824 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1825 {
1826         struct device_node *np, *fw_features;
1827         int r;
1828
1829         memset(cp, 0, sizeof(*cp));
1830         r = pseries_get_cpu_char(cp);
1831         if (r != -ENOTTY)
1832                 return r;
1833
1834         np = of_find_node_by_name(NULL, "ibm,opal");
1835         if (np) {
1836                 fw_features = of_get_child_by_name(np, "fw-features");
1837                 of_node_put(np);
1838                 if (!fw_features)
1839                         return 0;
1840                 if (have_fw_feat(fw_features, "enabled",
1841                                  "inst-spec-barrier-ori31,31,0"))
1842                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
1843                 if (have_fw_feat(fw_features, "enabled",
1844                                  "fw-bcctrl-serialized"))
1845                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
1846                 if (have_fw_feat(fw_features, "enabled",
1847                                  "inst-l1d-flush-ori30,30,0"))
1848                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
1849                 if (have_fw_feat(fw_features, "enabled",
1850                                  "inst-l1d-flush-trig2"))
1851                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
1852                 if (have_fw_feat(fw_features, "enabled",
1853                                  "fw-l1d-thread-split"))
1854                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
1855                 if (have_fw_feat(fw_features, "enabled",
1856                                  "fw-count-cache-disabled"))
1857                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1858                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
1859                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
1860                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
1861                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
1862                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
1863                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1864
1865                 if (have_fw_feat(fw_features, "enabled",
1866                                  "speculation-policy-favor-security"))
1867                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
1868                 if (!have_fw_feat(fw_features, "disabled",
1869                                   "needs-l1d-flush-msr-pr-0-to-1"))
1870                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
1871                 if (!have_fw_feat(fw_features, "disabled",
1872                                   "needs-spec-barrier-for-bound-checks"))
1873                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1874                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
1875                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
1876                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1877
1878                 of_node_put(fw_features);
1879         }
1880
1881         return 0;
1882 }
1883 #endif
1884
1885 long kvm_arch_vm_ioctl(struct file *filp,
1886                        unsigned int ioctl, unsigned long arg)
1887 {
1888         struct kvm *kvm __maybe_unused = filp->private_data;
1889         void __user *argp = (void __user *)arg;
1890         long r;
1891
1892         switch (ioctl) {
1893         case KVM_PPC_GET_PVINFO: {
1894                 struct kvm_ppc_pvinfo pvinfo;
1895                 memset(&pvinfo, 0, sizeof(pvinfo));
1896                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1897                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1898                         r = -EFAULT;
1899                         goto out;
1900                 }
1901
1902                 break;
1903         }
1904         case KVM_ENABLE_CAP:
1905         {
1906                 struct kvm_enable_cap cap;
1907                 r = -EFAULT;
1908                 if (copy_from_user(&cap, argp, sizeof(cap)))
1909                         goto out;
1910                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1911                 break;
1912         }
1913 #ifdef CONFIG_SPAPR_TCE_IOMMU
1914         case KVM_CREATE_SPAPR_TCE_64: {
1915                 struct kvm_create_spapr_tce_64 create_tce_64;
1916
1917                 r = -EFAULT;
1918                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1919                         goto out;
1920                 if (create_tce_64.flags) {
1921                         r = -EINVAL;
1922                         goto out;
1923                 }
1924                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1925                 goto out;
1926         }
1927         case KVM_CREATE_SPAPR_TCE: {
1928                 struct kvm_create_spapr_tce create_tce;
1929                 struct kvm_create_spapr_tce_64 create_tce_64;
1930
1931                 r = -EFAULT;
1932                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1933                         goto out;
1934
1935                 create_tce_64.liobn = create_tce.liobn;
1936                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1937                 create_tce_64.offset = 0;
1938                 create_tce_64.size = create_tce.window_size >>
1939                                 IOMMU_PAGE_SHIFT_4K;
1940                 create_tce_64.flags = 0;
1941                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1942                 goto out;
1943         }
1944 #endif
1945 #ifdef CONFIG_PPC_BOOK3S_64
1946         case KVM_PPC_GET_SMMU_INFO: {
1947                 struct kvm_ppc_smmu_info info;
1948                 struct kvm *kvm = filp->private_data;
1949
1950                 memset(&info, 0, sizeof(info));
1951                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1952                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1953                         r = -EFAULT;
1954                 break;
1955         }
1956         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1957                 struct kvm *kvm = filp->private_data;
1958
1959                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1960                 break;
1961         }
1962         case KVM_PPC_CONFIGURE_V3_MMU: {
1963                 struct kvm *kvm = filp->private_data;
1964                 struct kvm_ppc_mmuv3_cfg cfg;
1965
1966                 r = -EINVAL;
1967                 if (!kvm->arch.kvm_ops->configure_mmu)
1968                         goto out;
1969                 r = -EFAULT;
1970                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1971                         goto out;
1972                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1973                 break;
1974         }
1975         case KVM_PPC_GET_RMMU_INFO: {
1976                 struct kvm *kvm = filp->private_data;
1977                 struct kvm_ppc_rmmu_info info;
1978
1979                 r = -EINVAL;
1980                 if (!kvm->arch.kvm_ops->get_rmmu_info)
1981                         goto out;
1982                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1983                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1984                         r = -EFAULT;
1985                 break;
1986         }
1987         case KVM_PPC_GET_CPU_CHAR: {
1988                 struct kvm_ppc_cpu_char cpuchar;
1989
1990                 r = kvmppc_get_cpu_char(&cpuchar);
1991                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
1992                         r = -EFAULT;
1993                 break;
1994         }
1995         default: {
1996                 struct kvm *kvm = filp->private_data;
1997                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1998         }
1999 #else /* CONFIG_PPC_BOOK3S_64 */
2000         default:
2001                 r = -ENOTTY;
2002 #endif
2003         }
2004 out:
2005         return r;
2006 }
2007
2008 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2009 static unsigned long nr_lpids;
2010
2011 long kvmppc_alloc_lpid(void)
2012 {
2013         long lpid;
2014
2015         do {
2016                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2017                 if (lpid >= nr_lpids) {
2018                         pr_err("%s: No LPIDs free\n", __func__);
2019                         return -ENOMEM;
2020                 }
2021         } while (test_and_set_bit(lpid, lpid_inuse));
2022
2023         return lpid;
2024 }
2025 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2026
2027 void kvmppc_claim_lpid(long lpid)
2028 {
2029         set_bit(lpid, lpid_inuse);
2030 }
2031 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2032
2033 void kvmppc_free_lpid(long lpid)
2034 {
2035         clear_bit(lpid, lpid_inuse);
2036 }
2037 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2038
2039 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2040 {
2041         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2042         memset(lpid_inuse, 0, sizeof(lpid_inuse));
2043 }
2044 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2045
2046 int kvm_arch_init(void *opaque)
2047 {
2048         return 0;
2049 }
2050
2051 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);